bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
// SPDX-License-Identifier: GPL-2.0
|
|
|
|
/*
|
|
|
|
* Just-In-Time compiler for eBPF filters on IA32 (32bit x86)
|
|
|
|
*
|
|
|
|
* Author: Wang YanQing (udknight@gmail.com)
|
|
|
|
* The code based on code and ideas from:
|
|
|
|
* Eric Dumazet (eric.dumazet@gmail.com)
|
|
|
|
* and from:
|
|
|
|
* Shubham Bansal <illusionist.neo@gmail.com>
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/netdevice.h>
|
|
|
|
#include <linux/filter.h>
|
|
|
|
#include <linux/if_vlan.h>
|
|
|
|
#include <asm/cacheflush.h>
|
|
|
|
#include <asm/set_memory.h>
|
|
|
|
#include <asm/nospec-branch.h>
|
|
|
|
#include <linux/bpf.h>
|
|
|
|
|
|
|
|
/*
|
|
|
|
* eBPF prog stack layout:
|
|
|
|
*
|
|
|
|
* high
|
|
|
|
* original ESP => +-----+
|
|
|
|
* | | callee saved registers
|
|
|
|
* +-----+
|
|
|
|
* | ... | eBPF JIT scratch space
|
|
|
|
* BPF_FP,IA32_EBP => +-----+
|
|
|
|
* | ... | eBPF prog stack
|
|
|
|
* +-----+
|
|
|
|
* |RSVD | JIT scratchpad
|
|
|
|
* current ESP => +-----+
|
|
|
|
* | |
|
|
|
|
* | ... | Function call stack
|
|
|
|
* | |
|
|
|
|
* +-----+
|
|
|
|
* low
|
|
|
|
*
|
|
|
|
* The callee saved registers:
|
|
|
|
*
|
|
|
|
* high
|
|
|
|
* original ESP => +------------------+ \
|
|
|
|
* | ebp | |
|
|
|
|
* current EBP => +------------------+ } callee saved registers
|
|
|
|
* | ebx,esi,edi | |
|
|
|
|
* +------------------+ /
|
|
|
|
* low
|
|
|
|
*/
|
|
|
|
|
|
|
|
static u8 *emit_code(u8 *ptr, u32 bytes, unsigned int len)
|
|
|
|
{
|
|
|
|
if (len == 1)
|
|
|
|
*ptr = bytes;
|
|
|
|
else if (len == 2)
|
|
|
|
*(u16 *)ptr = bytes;
|
|
|
|
else {
|
|
|
|
*(u32 *)ptr = bytes;
|
|
|
|
barrier();
|
|
|
|
}
|
|
|
|
return ptr + len;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define EMIT(bytes, len) \
|
|
|
|
do { prog = emit_code(prog, bytes, len); cnt += len; } while (0)
|
|
|
|
|
|
|
|
#define EMIT1(b1) EMIT(b1, 1)
|
|
|
|
#define EMIT2(b1, b2) EMIT((b1) + ((b2) << 8), 2)
|
|
|
|
#define EMIT3(b1, b2, b3) EMIT((b1) + ((b2) << 8) + ((b3) << 16), 3)
|
|
|
|
#define EMIT4(b1, b2, b3, b4) \
|
|
|
|
EMIT((b1) + ((b2) << 8) + ((b3) << 16) + ((b4) << 24), 4)
|
|
|
|
|
|
|
|
#define EMIT1_off32(b1, off) \
|
|
|
|
do { EMIT1(b1); EMIT(off, 4); } while (0)
|
|
|
|
#define EMIT2_off32(b1, b2, off) \
|
|
|
|
do { EMIT2(b1, b2); EMIT(off, 4); } while (0)
|
|
|
|
#define EMIT3_off32(b1, b2, b3, off) \
|
|
|
|
do { EMIT3(b1, b2, b3); EMIT(off, 4); } while (0)
|
|
|
|
#define EMIT4_off32(b1, b2, b3, b4, off) \
|
|
|
|
do { EMIT4(b1, b2, b3, b4); EMIT(off, 4); } while (0)
|
|
|
|
|
|
|
|
#define jmp_label(label, jmp_insn_len) (label - cnt - jmp_insn_len)
|
|
|
|
|
|
|
|
static bool is_imm8(int value)
|
|
|
|
{
|
|
|
|
return value <= 127 && value >= -128;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool is_simm32(s64 value)
|
|
|
|
{
|
|
|
|
return value == (s64) (s32) value;
|
|
|
|
}
|
|
|
|
|
|
|
|
#define STACK_OFFSET(k) (k)
|
|
|
|
#define TCALL_CNT (MAX_BPF_JIT_REG + 0) /* Tail Call Count */
|
|
|
|
|
|
|
|
#define IA32_EAX (0x0)
|
|
|
|
#define IA32_EBX (0x3)
|
|
|
|
#define IA32_ECX (0x1)
|
|
|
|
#define IA32_EDX (0x2)
|
|
|
|
#define IA32_ESI (0x6)
|
|
|
|
#define IA32_EDI (0x7)
|
|
|
|
#define IA32_EBP (0x5)
|
|
|
|
#define IA32_ESP (0x4)
|
|
|
|
|
|
|
|
/*
|
|
|
|
* List of x86 cond jumps opcodes (. + s8)
|
|
|
|
* Add 0x10 (and an extra 0x0f) to generate far jumps (. + s32)
|
|
|
|
*/
|
|
|
|
#define IA32_JB 0x72
|
|
|
|
#define IA32_JAE 0x73
|
|
|
|
#define IA32_JE 0x74
|
|
|
|
#define IA32_JNE 0x75
|
|
|
|
#define IA32_JBE 0x76
|
|
|
|
#define IA32_JA 0x77
|
|
|
|
#define IA32_JL 0x7C
|
|
|
|
#define IA32_JGE 0x7D
|
|
|
|
#define IA32_JLE 0x7E
|
|
|
|
#define IA32_JG 0x7F
|
|
|
|
|
2019-04-27 15:28:26 +07:00
|
|
|
#define COND_JMP_OPCODE_INVALID (0xFF)
|
|
|
|
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/*
|
|
|
|
* Map eBPF registers to IA32 32bit registers or stack scratch space.
|
|
|
|
*
|
|
|
|
* 1. All the registers, R0-R10, are mapped to scratch space on stack.
|
|
|
|
* 2. We need two 64 bit temp registers to do complex operations on eBPF
|
|
|
|
* registers.
|
|
|
|
* 3. For performance reason, the BPF_REG_AX for blinding constant, is
|
|
|
|
* mapped to real hardware register pair, IA32_ESI and IA32_EDI.
|
|
|
|
*
|
|
|
|
* As the eBPF registers are all 64 bit registers and IA32 has only 32 bit
|
|
|
|
* registers, we have to map each eBPF registers with two IA32 32 bit regs
|
|
|
|
* or scratch memory space and we have to build eBPF 64 bit register from those.
|
|
|
|
*
|
|
|
|
* We use IA32_EAX, IA32_EDX, IA32_ECX, IA32_EBX as temporary registers.
|
|
|
|
*/
|
|
|
|
static const u8 bpf2ia32[][2] = {
|
|
|
|
/* Return value from in-kernel function, and exit value from eBPF */
|
|
|
|
[BPF_REG_0] = {STACK_OFFSET(0), STACK_OFFSET(4)},
|
|
|
|
|
|
|
|
/* The arguments from eBPF program to in-kernel function */
|
|
|
|
/* Stored on stack scratch space */
|
|
|
|
[BPF_REG_1] = {STACK_OFFSET(8), STACK_OFFSET(12)},
|
|
|
|
[BPF_REG_2] = {STACK_OFFSET(16), STACK_OFFSET(20)},
|
|
|
|
[BPF_REG_3] = {STACK_OFFSET(24), STACK_OFFSET(28)},
|
|
|
|
[BPF_REG_4] = {STACK_OFFSET(32), STACK_OFFSET(36)},
|
|
|
|
[BPF_REG_5] = {STACK_OFFSET(40), STACK_OFFSET(44)},
|
|
|
|
|
|
|
|
/* Callee saved registers that in-kernel function will preserve */
|
|
|
|
/* Stored on stack scratch space */
|
|
|
|
[BPF_REG_6] = {STACK_OFFSET(48), STACK_OFFSET(52)},
|
|
|
|
[BPF_REG_7] = {STACK_OFFSET(56), STACK_OFFSET(60)},
|
|
|
|
[BPF_REG_8] = {STACK_OFFSET(64), STACK_OFFSET(68)},
|
|
|
|
[BPF_REG_9] = {STACK_OFFSET(72), STACK_OFFSET(76)},
|
|
|
|
|
|
|
|
/* Read only Frame Pointer to access Stack */
|
|
|
|
[BPF_REG_FP] = {STACK_OFFSET(80), STACK_OFFSET(84)},
|
|
|
|
|
|
|
|
/* Temporary register for blinding constants. */
|
|
|
|
[BPF_REG_AX] = {IA32_ESI, IA32_EDI},
|
|
|
|
|
|
|
|
/* Tail call count. Stored on stack scratch space. */
|
|
|
|
[TCALL_CNT] = {STACK_OFFSET(88), STACK_OFFSET(92)},
|
|
|
|
};
|
|
|
|
|
|
|
|
#define dst_lo dst[0]
|
|
|
|
#define dst_hi dst[1]
|
|
|
|
#define src_lo src[0]
|
|
|
|
#define src_hi src[1]
|
|
|
|
|
|
|
|
#define STACK_ALIGNMENT 8
|
|
|
|
/*
|
|
|
|
* Stack space for BPF_REG_1, BPF_REG_2, BPF_REG_3, BPF_REG_4,
|
|
|
|
* BPF_REG_5, BPF_REG_6, BPF_REG_7, BPF_REG_8, BPF_REG_9,
|
|
|
|
* BPF_REG_FP, BPF_REG_AX and Tail call counts.
|
|
|
|
*/
|
|
|
|
#define SCRATCH_SIZE 96
|
|
|
|
|
|
|
|
/* Total stack size used in JITed code */
|
2018-05-04 06:08:23 +07:00
|
|
|
#define _STACK_SIZE (stack_depth + SCRATCH_SIZE)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
#define STACK_SIZE ALIGN(_STACK_SIZE, STACK_ALIGNMENT)
|
|
|
|
|
|
|
|
/* Get the offset of eBPF REGISTERs stored on scratch space. */
|
|
|
|
#define STACK_VAR(off) (off)
|
|
|
|
|
|
|
|
/* Encode 'dst_reg' register into IA32 opcode 'byte' */
|
|
|
|
static u8 add_1reg(u8 byte, u32 dst_reg)
|
|
|
|
{
|
|
|
|
return byte + dst_reg;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Encode 'dst_reg' and 'src_reg' registers into IA32 opcode 'byte' */
|
|
|
|
static u8 add_2reg(u8 byte, u32 dst_reg, u32 src_reg)
|
|
|
|
{
|
|
|
|
return byte + dst_reg + (src_reg << 3);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void jit_fill_hole(void *area, unsigned int size)
|
|
|
|
{
|
|
|
|
/* Fill whole space with int3 instructions */
|
|
|
|
memset(area, 0xcc, size);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void emit_ia32_mov_i(const u8 dst, const u32 val, bool dstk,
|
|
|
|
u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
if (val == 0) {
|
|
|
|
/* xor eax,eax */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, IA32_EAX, IA32_EAX));
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
} else {
|
|
|
|
EMIT3_off32(0xC7, add_1reg(0x40, IA32_EBP),
|
|
|
|
STACK_VAR(dst), val);
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
if (val == 0)
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dst, dst));
|
|
|
|
else
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, dst),
|
|
|
|
val);
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = imm (4 bytes)*/
|
|
|
|
static inline void emit_ia32_mov_r(const u8 dst, const u8 src, bool dstk,
|
|
|
|
bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 sreg = sstk ? IA32_EAX : src;
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(src));
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, sreg), STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
/* mov dst,sreg */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst, sreg));
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = src */
|
|
|
|
static inline void emit_ia32_mov_r64(const bool is64, const u8 dst[],
|
|
|
|
const u8 src[], bool dstk,
|
2019-05-25 05:25:26 +07:00
|
|
|
bool sstk, u8 **pprog,
|
|
|
|
const struct bpf_prog_aux *aux)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
{
|
|
|
|
emit_ia32_mov_r(dst_lo, src_lo, dstk, sstk, pprog);
|
|
|
|
if (is64)
|
|
|
|
/* complete 8 byte move */
|
|
|
|
emit_ia32_mov_r(dst_hi, src_hi, dstk, sstk, pprog);
|
2019-05-25 05:25:26 +07:00
|
|
|
else if (!aux->verifier_zext)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* zero out high 4 bytes */
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, pprog);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Sign extended move */
|
|
|
|
static inline void emit_ia32_mov_i64(const bool is64, const u8 dst[],
|
|
|
|
const u32 val, bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u32 hi = 0;
|
|
|
|
|
|
|
|
if (is64 && (val & (1<<31)))
|
|
|
|
hi = (u32)~0;
|
|
|
|
emit_ia32_mov_i(dst_lo, val, dstk, pprog);
|
|
|
|
emit_ia32_mov_i(dst_hi, hi, dstk, pprog);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ALU operation (32 bit)
|
|
|
|
* dst = dst * src
|
|
|
|
*/
|
|
|
|
static inline void emit_ia32_mul_r(const u8 dst, const u8 src, bool dstk,
|
|
|
|
bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 sreg = sstk ? IA32_ECX : src;
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(src));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
/* mov eax,dst */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst, IA32_EAX));
|
|
|
|
|
|
|
|
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, sreg));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
/* mov dst,eax */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst, IA32_EAX));
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void emit_ia32_to_le_r64(const u8 dst[], s32 val,
|
2019-05-25 05:25:26 +07:00
|
|
|
bool dstk, u8 **pprog,
|
|
|
|
const struct bpf_prog_aux *aux)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk && val != 64) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
switch (val) {
|
|
|
|
case 16:
|
|
|
|
/*
|
|
|
|
* Emit 'movzwl eax,ax' to zero extend 16-bit
|
|
|
|
* into 64 bit
|
|
|
|
*/
|
|
|
|
EMIT2(0x0F, 0xB7);
|
|
|
|
EMIT1(add_2reg(0xC0, dreg_lo, dreg_lo));
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!aux->verifier_zext)
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case 32:
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!aux->verifier_zext)
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case 64:
|
|
|
|
/* nop */
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk && val != 64) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void emit_ia32_to_be_r64(const u8 dst[], s32 val,
|
2019-05-25 05:25:26 +07:00
|
|
|
bool dstk, u8 **pprog,
|
|
|
|
const struct bpf_prog_aux *aux)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
switch (val) {
|
|
|
|
case 16:
|
|
|
|
/* Emit 'ror %ax, 8' to swap lower 2 bytes */
|
|
|
|
EMIT1(0x66);
|
|
|
|
EMIT3(0xC1, add_1reg(0xC8, dreg_lo), 8);
|
|
|
|
|
|
|
|
EMIT2(0x0F, 0xB7);
|
|
|
|
EMIT1(add_2reg(0xC0, dreg_lo, dreg_lo));
|
|
|
|
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!aux->verifier_zext)
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case 32:
|
|
|
|
/* Emit 'bswap eax' to swap lower 4 bytes */
|
|
|
|
EMIT1(0x0F);
|
|
|
|
EMIT1(add_1reg(0xC8, dreg_lo));
|
|
|
|
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!aux->verifier_zext)
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case 64:
|
|
|
|
/* Emit 'bswap eax' to swap lower 4 bytes */
|
|
|
|
EMIT1(0x0F);
|
|
|
|
EMIT1(add_1reg(0xC8, dreg_lo));
|
|
|
|
|
|
|
|
/* Emit 'bswap edx' to swap lower 4 bytes */
|
|
|
|
EMIT1(0x0F);
|
|
|
|
EMIT1(add_1reg(0xC8, dreg_hi));
|
|
|
|
|
|
|
|
/* mov ecx,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, IA32_ECX, dreg_hi));
|
|
|
|
/* mov dreg_hi,dreg_lo */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_hi, dreg_lo));
|
|
|
|
/* mov dreg_lo,ecx */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, IA32_ECX));
|
|
|
|
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ALU operation (32 bit)
|
|
|
|
* dst = dst (div|mod) src
|
|
|
|
*/
|
|
|
|
static inline void emit_ia32_div_mod_r(const u8 op, const u8 dst, const u8 src,
|
|
|
|
bool dstk, bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src));
|
|
|
|
else if (src != IA32_ECX)
|
|
|
|
/* mov ecx,src */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src, IA32_ECX));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
/* mov eax,dst */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst, IA32_EAX));
|
|
|
|
|
|
|
|
/* xor edx,edx */
|
|
|
|
EMIT2(0x31, add_2reg(0xC0, IA32_EDX, IA32_EDX));
|
|
|
|
/* div ecx */
|
|
|
|
EMIT2(0xF7, add_1reg(0xF0, IA32_ECX));
|
|
|
|
|
|
|
|
if (op == BPF_MOD) {
|
|
|
|
if (dstk)
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst, IA32_EDX));
|
|
|
|
} else {
|
|
|
|
if (dstk)
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
else
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst, IA32_EAX));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ALU operation (32 bit)
|
|
|
|
* dst = dst (shift) src
|
|
|
|
*/
|
|
|
|
static inline void emit_ia32_shift_r(const u8 op, const u8 dst, const u8 src,
|
|
|
|
bool dstk, bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg = dstk ? IA32_EAX : dst;
|
|
|
|
u8 b2;
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(dst));
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(src));
|
|
|
|
else if (src != IA32_ECX)
|
|
|
|
/* mov ecx,src */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src, IA32_ECX));
|
|
|
|
|
|
|
|
switch (op) {
|
|
|
|
case BPF_LSH:
|
|
|
|
b2 = 0xE0; break;
|
|
|
|
case BPF_RSH:
|
|
|
|
b2 = 0xE8; break;
|
|
|
|
case BPF_ARSH:
|
|
|
|
b2 = 0xF8; break;
|
|
|
|
default:
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
EMIT2(0xD3, add_1reg(b2, dreg));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],dreg */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg), STACK_VAR(dst));
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ALU operation (32 bit)
|
|
|
|
* dst = dst (op) src
|
|
|
|
*/
|
|
|
|
static inline void emit_ia32_alu_r(const bool is64, const bool hi, const u8 op,
|
|
|
|
const u8 dst, const u8 src, bool dstk,
|
|
|
|
bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 sreg = sstk ? IA32_EAX : src;
|
|
|
|
u8 dreg = dstk ? IA32_EDX : dst;
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(src));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX), STACK_VAR(dst));
|
|
|
|
|
|
|
|
switch (BPF_OP(op)) {
|
|
|
|
/* dst = dst + src */
|
|
|
|
case BPF_ADD:
|
|
|
|
if (hi && is64)
|
|
|
|
EMIT2(0x11, add_2reg(0xC0, dreg, sreg));
|
|
|
|
else
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst - src */
|
|
|
|
case BPF_SUB:
|
|
|
|
if (hi && is64)
|
|
|
|
EMIT2(0x19, add_2reg(0xC0, dreg, sreg));
|
|
|
|
else
|
|
|
|
EMIT2(0x29, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst | src */
|
|
|
|
case BPF_OR:
|
|
|
|
EMIT2(0x09, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst & src */
|
|
|
|
case BPF_AND:
|
|
|
|
EMIT2(0x21, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst ^ src */
|
|
|
|
case BPF_XOR:
|
|
|
|
EMIT2(0x31, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],dreg */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ALU operation (64 bit) */
|
|
|
|
static inline void emit_ia32_alu_r64(const bool is64, const u8 op,
|
|
|
|
const u8 dst[], const u8 src[],
|
|
|
|
bool dstk, bool sstk,
|
2019-05-25 05:25:26 +07:00
|
|
|
u8 **pprog, const struct bpf_prog_aux *aux)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
|
|
|
|
emit_ia32_alu_r(is64, false, op, dst_lo, src_lo, dstk, sstk, &prog);
|
|
|
|
if (is64)
|
|
|
|
emit_ia32_alu_r(is64, true, op, dst_hi, src_hi, dstk, sstk,
|
|
|
|
&prog);
|
2019-05-25 05:25:26 +07:00
|
|
|
else if (!aux->verifier_zext)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* ALU operation (32 bit)
|
|
|
|
* dst = dst (op) val
|
|
|
|
*/
|
|
|
|
static inline void emit_ia32_alu_i(const bool is64, const bool hi, const u8 op,
|
|
|
|
const u8 dst, const s32 val, bool dstk,
|
|
|
|
u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg = dstk ? IA32_EAX : dst;
|
|
|
|
u8 sreg = IA32_EDX;
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(dst));
|
|
|
|
|
|
|
|
if (!is_imm8(val))
|
|
|
|
/* mov edx,imm32*/
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EDX), val);
|
|
|
|
|
|
|
|
switch (op) {
|
|
|
|
/* dst = dst + val */
|
|
|
|
case BPF_ADD:
|
|
|
|
if (hi && is64) {
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xD0, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x11, add_2reg(0xC0, dreg, sreg));
|
|
|
|
} else {
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xC0, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, dreg, sreg));
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* dst = dst - val */
|
|
|
|
case BPF_SUB:
|
|
|
|
if (hi && is64) {
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xD8, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x19, add_2reg(0xC0, dreg, sreg));
|
|
|
|
} else {
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xE8, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x29, add_2reg(0xC0, dreg, sreg));
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* dst = dst | val */
|
|
|
|
case BPF_OR:
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xC8, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x09, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst & val */
|
|
|
|
case BPF_AND:
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xE0, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x21, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
/* dst = dst ^ val */
|
|
|
|
case BPF_XOR:
|
|
|
|
if (is_imm8(val))
|
|
|
|
EMIT3(0x83, add_1reg(0xF0, dreg), val);
|
|
|
|
else
|
|
|
|
EMIT2(0x31, add_2reg(0xC0, dreg, sreg));
|
|
|
|
break;
|
|
|
|
case BPF_NEG:
|
|
|
|
EMIT2(0xF7, add_1reg(0xD8, dreg));
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],dreg */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg),
|
|
|
|
STACK_VAR(dst));
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* ALU operation (64 bit) */
|
|
|
|
static inline void emit_ia32_alu_i64(const bool is64, const u8 op,
|
|
|
|
const u8 dst[], const u32 val,
|
2019-05-25 05:25:26 +07:00
|
|
|
bool dstk, u8 **pprog,
|
|
|
|
const struct bpf_prog_aux *aux)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
u32 hi = 0;
|
|
|
|
|
|
|
|
if (is64 && (val & (1<<31)))
|
|
|
|
hi = (u32)~0;
|
|
|
|
|
|
|
|
emit_ia32_alu_i(is64, false, op, dst_lo, val, dstk, &prog);
|
|
|
|
if (is64)
|
|
|
|
emit_ia32_alu_i(is64, true, op, dst_hi, hi, dstk, &prog);
|
2019-05-25 05:25:26 +07:00
|
|
|
else if (!aux->verifier_zext)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = ~dst (64 bit) */
|
|
|
|
static inline void emit_ia32_neg64(const u8 dst[], bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
2019-04-28 09:33:02 +07:00
|
|
|
/* neg dreg_lo */
|
|
|
|
EMIT2(0xF7, add_1reg(0xD8, dreg_lo));
|
|
|
|
/* adc dreg_hi,0x0 */
|
|
|
|
EMIT3(0x83, add_1reg(0xD0, dreg_hi), 0x00);
|
|
|
|
/* neg dreg_hi */
|
|
|
|
EMIT2(0xF7, add_1reg(0xD8, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst << src */
|
|
|
|
static inline void emit_ia32_lsh_r64(const u8 dst[], const u8 src[],
|
|
|
|
bool dstk, bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mov ecx,src_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_lo, IA32_ECX));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* shld dreg_hi,dreg_lo,cl */
|
|
|
|
EMIT3(0x0F, 0xA5, add_2reg(0xC0, dreg_hi, dreg_lo));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* shl dreg_lo,cl */
|
|
|
|
EMIT2(0xD3, add_1reg(0xE0, dreg_lo));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* if ecx >= 32, mov dreg_lo into dreg_hi and clear dreg_lo */
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* cmp ecx,32 */
|
|
|
|
EMIT3(0x83, add_1reg(0xF8, IA32_ECX), 32);
|
|
|
|
/* skip the next two instructions (4 bytes) when < 32 */
|
|
|
|
EMIT2(IA32_JB, 4);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
/* mov dreg_hi,dreg_lo */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_hi, dreg_lo));
|
|
|
|
/* xor dreg_lo,dreg_lo */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_lo, dreg_lo));
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
/* out: */
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst >> src (signed)*/
|
|
|
|
static inline void emit_ia32_arsh_r64(const u8 dst[], const u8 src[],
|
|
|
|
bool dstk, bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mov ecx,src_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_lo, IA32_ECX));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* shrd dreg_lo,dreg_hi,cl */
|
|
|
|
EMIT3(0x0F, 0xAD, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
/* sar dreg_hi,cl */
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
EMIT2(0xD3, add_1reg(0xF8, dreg_hi));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* if ecx >= 32, mov dreg_hi to dreg_lo and set/clear dreg_hi depending on sign */
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* cmp ecx,32 */
|
|
|
|
EMIT3(0x83, add_1reg(0xF8, IA32_ECX), 32);
|
|
|
|
/* skip the next two instructions (5 bytes) when < 32 */
|
|
|
|
EMIT2(IA32_JB, 5);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
/* mov dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, dreg_hi));
|
2019-06-29 12:57:49 +07:00
|
|
|
/* sar dreg_hi,31 */
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
EMIT3(0xC1, add_1reg(0xF8, dreg_hi), 31);
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
/* out: */
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst >> src */
|
|
|
|
static inline void emit_ia32_rsh_r64(const u8 dst[], const u8 src[], bool dstk,
|
|
|
|
bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mov ecx,src_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_lo, IA32_ECX));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* shrd dreg_lo,dreg_hi,cl */
|
|
|
|
EMIT3(0x0F, 0xAD, add_2reg(0xC0, dreg_lo, dreg_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* shr dreg_hi,cl */
|
|
|
|
EMIT2(0xD3, add_1reg(0xE8, dreg_hi));
|
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* if ecx >= 32, mov dreg_hi to dreg_lo and clear dreg_hi */
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
2019-06-29 12:57:49 +07:00
|
|
|
/* cmp ecx,32 */
|
|
|
|
EMIT3(0x83, add_1reg(0xF8, IA32_ECX), 32);
|
|
|
|
/* skip the next two instructions (4 bytes) when < 32 */
|
|
|
|
EMIT2(IA32_JB, 4);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
/* mov dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
/* out: */
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst << val */
|
|
|
|
static inline void emit_ia32_lsh_i64(const u8 dst[], const u32 val,
|
|
|
|
bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
/* Do LSH operation */
|
|
|
|
if (val < 32) {
|
2019-06-29 12:57:50 +07:00
|
|
|
/* shld dreg_hi,dreg_lo,imm8 */
|
|
|
|
EMIT4(0x0F, 0xA4, add_2reg(0xC0, dreg_hi, dreg_lo), val);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* shl dreg_lo,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xE0, dreg_lo), val);
|
|
|
|
} else if (val >= 32 && val < 64) {
|
|
|
|
u32 value = val - 32;
|
|
|
|
|
|
|
|
/* shl dreg_lo,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xE0, dreg_lo), value);
|
|
|
|
/* mov dreg_hi,dreg_lo */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_hi, dreg_lo));
|
|
|
|
/* xor dreg_lo,dreg_lo */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_lo, dreg_lo));
|
|
|
|
} else {
|
|
|
|
/* xor dreg_lo,dreg_lo */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_lo, dreg_lo));
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst >> val */
|
|
|
|
static inline void emit_ia32_rsh_i64(const u8 dst[], const u32 val,
|
|
|
|
bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Do RSH operation */
|
|
|
|
if (val < 32) {
|
2019-06-29 12:57:50 +07:00
|
|
|
/* shrd dreg_lo,dreg_hi,imm8 */
|
|
|
|
EMIT4(0x0F, 0xAC, add_2reg(0xC0, dreg_lo, dreg_hi), val);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* shr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xE8, dreg_hi), val);
|
|
|
|
} else if (val >= 32 && val < 64) {
|
|
|
|
u32 value = val - 32;
|
|
|
|
|
|
|
|
/* shr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xE8, dreg_hi), value);
|
|
|
|
/* mov dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
|
|
|
} else {
|
|
|
|
/* xor dreg_lo,dreg_lo */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_lo, dreg_lo));
|
|
|
|
/* xor dreg_hi,dreg_hi */
|
|
|
|
EMIT2(0x33, add_2reg(0xC0, dreg_hi, dreg_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* dst = dst >> val (signed) */
|
|
|
|
static inline void emit_ia32_arsh_i64(const u8 dst[], const u32 val,
|
|
|
|
bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
/* Do RSH operation */
|
|
|
|
if (val < 32) {
|
2019-06-29 12:57:50 +07:00
|
|
|
/* shrd dreg_lo,dreg_hi,imm8 */
|
|
|
|
EMIT4(0x0F, 0xAC, add_2reg(0xC0, dreg_lo, dreg_hi), val);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* ashr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xF8, dreg_hi), val);
|
|
|
|
} else if (val >= 32 && val < 64) {
|
|
|
|
u32 value = val - 32;
|
|
|
|
|
|
|
|
/* ashr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xF8, dreg_hi), value);
|
|
|
|
/* mov dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
|
|
|
|
/* ashr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xF8, dreg_hi), 31);
|
|
|
|
} else {
|
|
|
|
/* ashr dreg_hi,imm8 */
|
|
|
|
EMIT3(0xC1, add_1reg(0xF8, dreg_hi), 31);
|
|
|
|
/* mov dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],dreg_lo */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_lo),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],dreg_hi */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, dreg_hi),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void emit_ia32_mul_r64(const u8 dst[], const u8 src[], bool dstk,
|
|
|
|
bool sstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
else
|
|
|
|
/* mov eax,dst_hi */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst_hi, IA32_EAX));
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mul src_lo */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, src_lo));
|
|
|
|
|
|
|
|
/* mov ecx,eax */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, IA32_ECX, IA32_EAX));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mov eax,dst_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(src_hi));
|
|
|
|
else
|
|
|
|
/* mul src_hi */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, src_hi));
|
|
|
|
|
|
|
|
/* add eax,eax */
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, IA32_ECX, IA32_EAX));
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mov eax,dst_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mul src_lo */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, src_lo));
|
|
|
|
|
|
|
|
/* add ecx,edx */
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, IA32_ECX, IA32_EDX));
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],ecx */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
} else {
|
|
|
|
/* mov dst_lo,eax */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
/* mov dst_hi,ecx */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst_hi, IA32_ECX));
|
|
|
|
}
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void emit_ia32_mul_i64(const u8 dst[], const u32 val,
|
|
|
|
bool dstk, u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
u32 hi;
|
|
|
|
|
|
|
|
hi = val & (1<<31) ? (u32)~0 : 0;
|
|
|
|
/* movl eax,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EAX), val);
|
|
|
|
if (dstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(dst_hi));
|
|
|
|
else
|
|
|
|
/* mul dst_hi */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, dst_hi));
|
|
|
|
|
|
|
|
/* mov ecx,eax */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, IA32_ECX, IA32_EAX));
|
|
|
|
|
|
|
|
/* movl eax,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EAX), hi);
|
|
|
|
if (dstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mul dst_lo */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, dst_lo));
|
|
|
|
/* add ecx,eax */
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, IA32_ECX, IA32_EAX));
|
|
|
|
|
|
|
|
/* movl eax,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EAX), val);
|
|
|
|
if (dstk)
|
|
|
|
/* mul dword ptr [ebp+off] */
|
|
|
|
EMIT3(0xF7, add_1reg(0x60, IA32_EBP), STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mul dst_lo */
|
|
|
|
EMIT2(0xF7, add_1reg(0xE0, dst_lo));
|
|
|
|
|
|
|
|
/* add ecx,edx */
|
|
|
|
EMIT2(0x01, add_2reg(0xC0, IA32_ECX, IA32_EDX));
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
/* mov dword ptr [ebp+off],ecx */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
} else {
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
/* mov dword ptr [ebp+off],ecx */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst_hi, IA32_ECX));
|
|
|
|
}
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int bpf_size_to_x86_bytes(int bpf_size)
|
|
|
|
{
|
|
|
|
if (bpf_size == BPF_W)
|
|
|
|
return 4;
|
|
|
|
else if (bpf_size == BPF_H)
|
|
|
|
return 2;
|
|
|
|
else if (bpf_size == BPF_B)
|
|
|
|
return 1;
|
|
|
|
else if (bpf_size == BPF_DW)
|
|
|
|
return 4; /* imm32 */
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
struct jit_context {
|
|
|
|
int cleanup_addr; /* Epilogue code offset */
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Maximum number of bytes emitted while JITing one eBPF insn */
|
|
|
|
#define BPF_MAX_INSN_SIZE 128
|
|
|
|
#define BPF_INSN_SAFETY 64
|
|
|
|
|
|
|
|
#define PROLOGUE_SIZE 35
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Emit prologue code for BPF program and check it's size.
|
|
|
|
* bpf_tail_call helper will skip it while jumping into another program.
|
|
|
|
*/
|
|
|
|
static void emit_prologue(u8 **pprog, u32 stack_depth)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
const u8 *r1 = bpf2ia32[BPF_REG_1];
|
|
|
|
const u8 fplo = bpf2ia32[BPF_REG_FP][0];
|
|
|
|
const u8 fphi = bpf2ia32[BPF_REG_FP][1];
|
|
|
|
const u8 *tcc = bpf2ia32[TCALL_CNT];
|
|
|
|
|
|
|
|
/* push ebp */
|
|
|
|
EMIT1(0x55);
|
|
|
|
/* mov ebp,esp */
|
|
|
|
EMIT2(0x89, 0xE5);
|
|
|
|
/* push edi */
|
|
|
|
EMIT1(0x57);
|
|
|
|
/* push esi */
|
|
|
|
EMIT1(0x56);
|
|
|
|
/* push ebx */
|
|
|
|
EMIT1(0x53);
|
|
|
|
|
|
|
|
/* sub esp,STACK_SIZE */
|
|
|
|
EMIT2_off32(0x81, 0xEC, STACK_SIZE);
|
2018-07-25 23:46:11 +07:00
|
|
|
/* sub ebp,SCRATCH_SIZE+12*/
|
|
|
|
EMIT3(0x83, add_1reg(0xE8, IA32_EBP), SCRATCH_SIZE + 12);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* xor ebx,ebx */
|
|
|
|
EMIT2(0x31, add_2reg(0xC0, IA32_EBX, IA32_EBX));
|
|
|
|
|
|
|
|
/* Set up BPF prog stack base register */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBP), STACK_VAR(fplo));
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(fphi));
|
|
|
|
|
|
|
|
/* Move BPF_CTX (EAX) to BPF_REG_R1 */
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(r1[0]));
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(r1[1]));
|
|
|
|
|
|
|
|
/* Initialize Tail Count */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(tcc[0]));
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(tcc[1]));
|
|
|
|
|
|
|
|
BUILD_BUG_ON(cnt != PROLOGUE_SIZE);
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Emit epilogue code for BPF program */
|
|
|
|
static void emit_epilogue(u8 **pprog, u32 stack_depth)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
const u8 *r0 = bpf2ia32[BPF_REG_0];
|
|
|
|
int cnt = 0;
|
|
|
|
|
|
|
|
/* mov eax,dword ptr [ebp+off]*/
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(r0[0]));
|
|
|
|
/* mov edx,dword ptr [ebp+off]*/
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX), STACK_VAR(r0[1]));
|
|
|
|
|
2018-07-25 23:46:11 +07:00
|
|
|
/* add ebp,SCRATCH_SIZE+12*/
|
|
|
|
EMIT3(0x83, add_1reg(0xC0, IA32_EBP), SCRATCH_SIZE + 12);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
/* mov ebx,dword ptr [ebp-12]*/
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EBX), -12);
|
|
|
|
/* mov esi,dword ptr [ebp-8]*/
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ESI), -8);
|
|
|
|
/* mov edi,dword ptr [ebp-4]*/
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDI), -4);
|
|
|
|
|
|
|
|
EMIT1(0xC9); /* leave */
|
|
|
|
EMIT1(0xC3); /* ret */
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Generate the following code:
|
|
|
|
* ... bpf_tail_call(void *ctx, struct bpf_array *array, u64 index) ...
|
|
|
|
* if (index >= array->map.max_entries)
|
|
|
|
* goto out;
|
|
|
|
* if (++tail_call_cnt > MAX_TAIL_CALL_CNT)
|
|
|
|
* goto out;
|
|
|
|
* prog = array->ptrs[index];
|
|
|
|
* if (prog == NULL)
|
|
|
|
* goto out;
|
|
|
|
* goto *(prog->bpf_func + prologue_size);
|
|
|
|
* out:
|
|
|
|
*/
|
|
|
|
static void emit_bpf_tail_call(u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
const u8 *r1 = bpf2ia32[BPF_REG_1];
|
|
|
|
const u8 *r2 = bpf2ia32[BPF_REG_2];
|
|
|
|
const u8 *r3 = bpf2ia32[BPF_REG_3];
|
|
|
|
const u8 *tcc = bpf2ia32[TCALL_CNT];
|
|
|
|
u32 lo, hi;
|
|
|
|
static int jmp_label1 = -1;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* if (index >= array->map.max_entries)
|
|
|
|
* goto out;
|
|
|
|
*/
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(r2[0]));
|
|
|
|
/* mov edx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX), STACK_VAR(r3[0]));
|
|
|
|
|
|
|
|
/* cmp dword ptr [eax+off],edx */
|
|
|
|
EMIT3(0x39, add_2reg(0x40, IA32_EAX, IA32_EDX),
|
|
|
|
offsetof(struct bpf_array, map.max_entries));
|
|
|
|
/* jbe out */
|
|
|
|
EMIT2(IA32_JBE, jmp_label(jmp_label1, 2));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* if (tail_call_cnt > MAX_TAIL_CALL_CNT)
|
|
|
|
* goto out;
|
|
|
|
*/
|
|
|
|
lo = (u32)MAX_TAIL_CALL_CNT;
|
|
|
|
hi = (u32)((u64)MAX_TAIL_CALL_CNT >> 32);
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(tcc[0]));
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(tcc[1]));
|
|
|
|
|
|
|
|
/* cmp edx,hi */
|
|
|
|
EMIT3(0x83, add_1reg(0xF8, IA32_EBX), hi);
|
|
|
|
EMIT2(IA32_JNE, 3);
|
|
|
|
/* cmp ecx,lo */
|
|
|
|
EMIT3(0x83, add_1reg(0xF8, IA32_ECX), lo);
|
|
|
|
|
|
|
|
/* ja out */
|
|
|
|
EMIT2(IA32_JAE, jmp_label(jmp_label1, 2));
|
|
|
|
|
|
|
|
/* add eax,0x1 */
|
|
|
|
EMIT3(0x83, add_1reg(0xC0, IA32_ECX), 0x01);
|
|
|
|
/* adc ebx,0x0 */
|
|
|
|
EMIT3(0x83, add_1reg(0xD0, IA32_EBX), 0x00);
|
|
|
|
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(tcc[0]));
|
|
|
|
/* mov dword ptr [ebp+off],edx */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EBX), STACK_VAR(tcc[1]));
|
|
|
|
|
|
|
|
/* prog = array->ptrs[index]; */
|
|
|
|
/* mov edx, [eax + edx * 4 + offsetof(...)] */
|
|
|
|
EMIT3_off32(0x8B, 0x94, 0x90, offsetof(struct bpf_array, ptrs));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* if (prog == NULL)
|
|
|
|
* goto out;
|
|
|
|
*/
|
|
|
|
/* test edx,edx */
|
|
|
|
EMIT2(0x85, add_2reg(0xC0, IA32_EDX, IA32_EDX));
|
|
|
|
/* je out */
|
|
|
|
EMIT2(IA32_JE, jmp_label(jmp_label1, 2));
|
|
|
|
|
|
|
|
/* goto *(prog->bpf_func + prologue_size); */
|
|
|
|
/* mov edx, dword ptr [edx + 32] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EDX, IA32_EDX),
|
|
|
|
offsetof(struct bpf_prog, bpf_func));
|
|
|
|
/* add edx,prologue_size */
|
|
|
|
EMIT3(0x83, add_1reg(0xC0, IA32_EDX), PROLOGUE_SIZE);
|
|
|
|
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX), STACK_VAR(r1[0]));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Now we're ready to jump into next BPF program:
|
|
|
|
* eax == ctx (1st arg)
|
|
|
|
* edx == prog->bpf_func + prologue_size
|
|
|
|
*/
|
|
|
|
RETPOLINE_EDX_BPF_JIT();
|
|
|
|
|
|
|
|
if (jmp_label1 == -1)
|
|
|
|
jmp_label1 = cnt;
|
|
|
|
|
|
|
|
/* out: */
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Push the scratch stack register on top of the stack. */
|
|
|
|
static inline void emit_push_r64(const u8 src[], u8 **pprog)
|
|
|
|
{
|
|
|
|
u8 *prog = *pprog;
|
|
|
|
int cnt = 0;
|
|
|
|
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(src_hi));
|
|
|
|
/* push ecx */
|
|
|
|
EMIT1(0x51);
|
|
|
|
|
|
|
|
/* mov ecx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX), STACK_VAR(src_lo));
|
|
|
|
/* push ecx */
|
|
|
|
EMIT1(0x51);
|
|
|
|
|
|
|
|
*pprog = prog;
|
|
|
|
}
|
|
|
|
|
2019-04-27 15:28:26 +07:00
|
|
|
static u8 get_cond_jmp_opcode(const u8 op, bool is_cmp_lo)
|
|
|
|
{
|
|
|
|
u8 jmp_cond;
|
|
|
|
|
|
|
|
/* Convert BPF opcode to x86 */
|
|
|
|
switch (op) {
|
|
|
|
case BPF_JEQ:
|
|
|
|
jmp_cond = IA32_JE;
|
|
|
|
break;
|
|
|
|
case BPF_JSET:
|
|
|
|
case BPF_JNE:
|
|
|
|
jmp_cond = IA32_JNE;
|
|
|
|
break;
|
|
|
|
case BPF_JGT:
|
|
|
|
/* GT is unsigned '>', JA in x86 */
|
|
|
|
jmp_cond = IA32_JA;
|
|
|
|
break;
|
|
|
|
case BPF_JLT:
|
|
|
|
/* LT is unsigned '<', JB in x86 */
|
|
|
|
jmp_cond = IA32_JB;
|
|
|
|
break;
|
|
|
|
case BPF_JGE:
|
|
|
|
/* GE is unsigned '>=', JAE in x86 */
|
|
|
|
jmp_cond = IA32_JAE;
|
|
|
|
break;
|
|
|
|
case BPF_JLE:
|
|
|
|
/* LE is unsigned '<=', JBE in x86 */
|
|
|
|
jmp_cond = IA32_JBE;
|
|
|
|
break;
|
|
|
|
case BPF_JSGT:
|
|
|
|
if (!is_cmp_lo)
|
|
|
|
/* Signed '>', GT in x86 */
|
|
|
|
jmp_cond = IA32_JG;
|
|
|
|
else
|
|
|
|
/* GT is unsigned '>', JA in x86 */
|
|
|
|
jmp_cond = IA32_JA;
|
|
|
|
break;
|
|
|
|
case BPF_JSLT:
|
|
|
|
if (!is_cmp_lo)
|
|
|
|
/* Signed '<', LT in x86 */
|
|
|
|
jmp_cond = IA32_JL;
|
|
|
|
else
|
|
|
|
/* LT is unsigned '<', JB in x86 */
|
|
|
|
jmp_cond = IA32_JB;
|
|
|
|
break;
|
|
|
|
case BPF_JSGE:
|
|
|
|
if (!is_cmp_lo)
|
|
|
|
/* Signed '>=', GE in x86 */
|
|
|
|
jmp_cond = IA32_JGE;
|
|
|
|
else
|
|
|
|
/* GE is unsigned '>=', JAE in x86 */
|
|
|
|
jmp_cond = IA32_JAE;
|
|
|
|
break;
|
|
|
|
case BPF_JSLE:
|
|
|
|
if (!is_cmp_lo)
|
|
|
|
/* Signed '<=', LE in x86 */
|
|
|
|
jmp_cond = IA32_JLE;
|
|
|
|
else
|
|
|
|
/* LE is unsigned '<=', JBE in x86 */
|
|
|
|
jmp_cond = IA32_JBE;
|
|
|
|
break;
|
|
|
|
default: /* to silence GCC warning */
|
|
|
|
jmp_cond = COND_JMP_OPCODE_INVALID;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
return jmp_cond;
|
|
|
|
}
|
|
|
|
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
|
|
|
|
int oldproglen, struct jit_context *ctx)
|
|
|
|
{
|
|
|
|
struct bpf_insn *insn = bpf_prog->insnsi;
|
|
|
|
int insn_cnt = bpf_prog->len;
|
|
|
|
bool seen_exit = false;
|
|
|
|
u8 temp[BPF_MAX_INSN_SIZE + BPF_INSN_SAFETY];
|
|
|
|
int i, cnt = 0;
|
|
|
|
int proglen = 0;
|
|
|
|
u8 *prog = temp;
|
|
|
|
|
|
|
|
emit_prologue(&prog, bpf_prog->aux->stack_depth);
|
|
|
|
|
|
|
|
for (i = 0; i < insn_cnt; i++, insn++) {
|
|
|
|
const s32 imm32 = insn->imm;
|
|
|
|
const bool is64 = BPF_CLASS(insn->code) == BPF_ALU64;
|
|
|
|
const bool dstk = insn->dst_reg == BPF_REG_AX ? false : true;
|
|
|
|
const bool sstk = insn->src_reg == BPF_REG_AX ? false : true;
|
|
|
|
const u8 code = insn->code;
|
|
|
|
const u8 *dst = bpf2ia32[insn->dst_reg];
|
|
|
|
const u8 *src = bpf2ia32[insn->src_reg];
|
|
|
|
const u8 *r0 = bpf2ia32[BPF_REG_0];
|
|
|
|
s64 jmp_offset;
|
|
|
|
u8 jmp_cond;
|
|
|
|
int ilen;
|
|
|
|
u8 *func;
|
|
|
|
|
|
|
|
switch (code) {
|
|
|
|
/* ALU operations */
|
|
|
|
/* dst = src */
|
|
|
|
case BPF_ALU | BPF_MOV | BPF_K:
|
|
|
|
case BPF_ALU | BPF_MOV | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_MOV | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_MOV | BPF_X:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
2019-05-25 05:25:26 +07:00
|
|
|
if (imm32 == 1) {
|
|
|
|
/* Special mov32 for zext. */
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
emit_ia32_mov_r64(is64, dst, src, dstk, sstk,
|
|
|
|
&prog, bpf_prog->aux);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
/* Sign-extend immediate value to dst reg */
|
|
|
|
emit_ia32_mov_i64(is64, dst, imm32,
|
|
|
|
dstk, &prog);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* dst = dst + src/imm */
|
|
|
|
/* dst = dst - src/imm */
|
|
|
|
/* dst = dst | src/imm */
|
|
|
|
/* dst = dst & src/imm */
|
|
|
|
/* dst = dst ^ src/imm */
|
|
|
|
/* dst = dst * src/imm */
|
|
|
|
/* dst = dst << src */
|
|
|
|
/* dst = dst >> src */
|
|
|
|
case BPF_ALU | BPF_ADD | BPF_K:
|
|
|
|
case BPF_ALU | BPF_ADD | BPF_X:
|
|
|
|
case BPF_ALU | BPF_SUB | BPF_K:
|
|
|
|
case BPF_ALU | BPF_SUB | BPF_X:
|
|
|
|
case BPF_ALU | BPF_OR | BPF_K:
|
|
|
|
case BPF_ALU | BPF_OR | BPF_X:
|
|
|
|
case BPF_ALU | BPF_AND | BPF_K:
|
|
|
|
case BPF_ALU | BPF_AND | BPF_X:
|
|
|
|
case BPF_ALU | BPF_XOR | BPF_K:
|
|
|
|
case BPF_ALU | BPF_XOR | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_ADD | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_ADD | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_SUB | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_SUB | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_OR | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_OR | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_AND | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_AND | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_XOR | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_XOR | BPF_X:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
|
|
|
emit_ia32_alu_r64(is64, BPF_OP(code), dst,
|
2019-05-25 05:25:26 +07:00
|
|
|
src, dstk, sstk, &prog,
|
|
|
|
bpf_prog->aux);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
emit_ia32_alu_i64(is64, BPF_OP(code), dst,
|
2019-05-25 05:25:26 +07:00
|
|
|
imm32, dstk, &prog,
|
|
|
|
bpf_prog->aux);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case BPF_ALU | BPF_MUL | BPF_K:
|
|
|
|
case BPF_ALU | BPF_MUL | BPF_X:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
|
|
|
emit_ia32_mul_r(dst_lo, src_lo, dstk,
|
|
|
|
sstk, &prog);
|
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
/* mov ecx,imm32*/
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX),
|
|
|
|
imm32);
|
|
|
|
emit_ia32_mul_r(dst_lo, IA32_ECX, dstk,
|
|
|
|
false, &prog);
|
|
|
|
break;
|
|
|
|
}
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case BPF_ALU | BPF_LSH | BPF_X:
|
|
|
|
case BPF_ALU | BPF_RSH | BPF_X:
|
|
|
|
case BPF_ALU | BPF_ARSH | BPF_K:
|
|
|
|
case BPF_ALU | BPF_ARSH | BPF_X:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
|
|
|
emit_ia32_shift_r(BPF_OP(code), dst_lo, src_lo,
|
|
|
|
dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
/* mov ecx,imm32*/
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX),
|
|
|
|
imm32);
|
|
|
|
emit_ia32_shift_r(BPF_OP(code), dst_lo,
|
|
|
|
IA32_ECX, dstk, false,
|
|
|
|
&prog);
|
|
|
|
break;
|
|
|
|
}
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
/* dst = dst / src(imm) */
|
|
|
|
/* dst = dst % src(imm) */
|
|
|
|
case BPF_ALU | BPF_DIV | BPF_K:
|
|
|
|
case BPF_ALU | BPF_DIV | BPF_X:
|
|
|
|
case BPF_ALU | BPF_MOD | BPF_K:
|
|
|
|
case BPF_ALU | BPF_MOD | BPF_X:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
|
|
|
emit_ia32_div_mod_r(BPF_OP(code), dst_lo,
|
|
|
|
src_lo, dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
/* mov ecx,imm32*/
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX),
|
|
|
|
imm32);
|
|
|
|
emit_ia32_div_mod_r(BPF_OP(code), dst_lo,
|
|
|
|
IA32_ECX, dstk, false,
|
|
|
|
&prog);
|
|
|
|
break;
|
|
|
|
}
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
case BPF_ALU64 | BPF_DIV | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_DIV | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_MOD | BPF_K:
|
|
|
|
case BPF_ALU64 | BPF_MOD | BPF_X:
|
|
|
|
goto notyet;
|
|
|
|
/* dst = dst >> imm */
|
|
|
|
/* dst = dst << imm */
|
|
|
|
case BPF_ALU | BPF_RSH | BPF_K:
|
|
|
|
case BPF_ALU | BPF_LSH | BPF_K:
|
|
|
|
if (unlikely(imm32 > 31))
|
|
|
|
return -EINVAL;
|
|
|
|
/* mov ecx,imm32*/
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX), imm32);
|
|
|
|
emit_ia32_shift_r(BPF_OP(code), dst_lo, IA32_ECX, dstk,
|
|
|
|
false, &prog);
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
/* dst = dst << imm */
|
|
|
|
case BPF_ALU64 | BPF_LSH | BPF_K:
|
|
|
|
if (unlikely(imm32 > 63))
|
|
|
|
return -EINVAL;
|
|
|
|
emit_ia32_lsh_i64(dst, imm32, dstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst >> imm */
|
|
|
|
case BPF_ALU64 | BPF_RSH | BPF_K:
|
|
|
|
if (unlikely(imm32 > 63))
|
|
|
|
return -EINVAL;
|
|
|
|
emit_ia32_rsh_i64(dst, imm32, dstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst << src */
|
|
|
|
case BPF_ALU64 | BPF_LSH | BPF_X:
|
|
|
|
emit_ia32_lsh_r64(dst, src, dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst >> src */
|
|
|
|
case BPF_ALU64 | BPF_RSH | BPF_X:
|
|
|
|
emit_ia32_rsh_r64(dst, src, dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst >> src (signed) */
|
|
|
|
case BPF_ALU64 | BPF_ARSH | BPF_X:
|
|
|
|
emit_ia32_arsh_r64(dst, src, dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst >> imm (signed) */
|
|
|
|
case BPF_ALU64 | BPF_ARSH | BPF_K:
|
|
|
|
if (unlikely(imm32 > 63))
|
|
|
|
return -EINVAL;
|
|
|
|
emit_ia32_arsh_i64(dst, imm32, dstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = ~dst */
|
|
|
|
case BPF_ALU | BPF_NEG:
|
|
|
|
emit_ia32_alu_i(is64, false, BPF_OP(code),
|
|
|
|
dst_lo, 0, dstk, &prog);
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
emit_ia32_mov_i(dst_hi, 0, dstk, &prog);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
/* dst = ~dst (64 bit) */
|
|
|
|
case BPF_ALU64 | BPF_NEG:
|
|
|
|
emit_ia32_neg64(dst, dstk, &prog);
|
|
|
|
break;
|
|
|
|
/* dst = dst * src/imm */
|
|
|
|
case BPF_ALU64 | BPF_MUL | BPF_X:
|
|
|
|
case BPF_ALU64 | BPF_MUL | BPF_K:
|
|
|
|
switch (BPF_SRC(code)) {
|
|
|
|
case BPF_X:
|
|
|
|
emit_ia32_mul_r64(dst, src, dstk, sstk, &prog);
|
|
|
|
break;
|
|
|
|
case BPF_K:
|
|
|
|
emit_ia32_mul_i64(dst, imm32, dstk, &prog);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* dst = htole(dst) */
|
|
|
|
case BPF_ALU | BPF_END | BPF_FROM_LE:
|
2019-05-25 05:25:26 +07:00
|
|
|
emit_ia32_to_le_r64(dst, imm32, dstk, &prog,
|
|
|
|
bpf_prog->aux);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
/* dst = htobe(dst) */
|
|
|
|
case BPF_ALU | BPF_END | BPF_FROM_BE:
|
2019-05-25 05:25:26 +07:00
|
|
|
emit_ia32_to_be_r64(dst, imm32, dstk, &prog,
|
|
|
|
bpf_prog->aux);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
/* dst = imm64 */
|
|
|
|
case BPF_LD | BPF_IMM | BPF_DW: {
|
|
|
|
s32 hi, lo = imm32;
|
|
|
|
|
|
|
|
hi = insn[1].imm;
|
|
|
|
emit_ia32_mov_i(dst_lo, lo, dstk, &prog);
|
|
|
|
emit_ia32_mov_i(dst_hi, hi, dstk, &prog);
|
|
|
|
insn++;
|
|
|
|
i++;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* ST: *(u8*)(dst_reg + off) = imm */
|
|
|
|
case BPF_ST | BPF_MEM | BPF_H:
|
|
|
|
case BPF_ST | BPF_MEM | BPF_B:
|
|
|
|
case BPF_ST | BPF_MEM | BPF_W:
|
|
|
|
case BPF_ST | BPF_MEM | BPF_DW:
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mov eax,dst_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
|
|
|
|
switch (BPF_SIZE(code)) {
|
|
|
|
case BPF_B:
|
|
|
|
EMIT(0xC6, 1); break;
|
|
|
|
case BPF_H:
|
|
|
|
EMIT2(0x66, 0xC7); break;
|
|
|
|
case BPF_W:
|
|
|
|
case BPF_DW:
|
|
|
|
EMIT(0xC7, 1); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_imm8(insn->off))
|
|
|
|
EMIT2(add_1reg(0x40, IA32_EAX), insn->off);
|
|
|
|
else
|
|
|
|
EMIT1_off32(add_1reg(0x80, IA32_EAX),
|
|
|
|
insn->off);
|
|
|
|
EMIT(imm32, bpf_size_to_x86_bytes(BPF_SIZE(code)));
|
|
|
|
|
|
|
|
if (BPF_SIZE(code) == BPF_DW) {
|
|
|
|
u32 hi;
|
|
|
|
|
|
|
|
hi = imm32 & (1<<31) ? (u32)~0 : 0;
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0x80, IA32_EAX),
|
|
|
|
insn->off + 4);
|
|
|
|
EMIT(hi, 4);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* STX: *(u8*)(dst_reg + off) = src_reg */
|
|
|
|
case BPF_STX | BPF_MEM | BPF_B:
|
|
|
|
case BPF_STX | BPF_MEM | BPF_H:
|
|
|
|
case BPF_STX | BPF_MEM | BPF_W:
|
|
|
|
case BPF_STX | BPF_MEM | BPF_DW:
|
|
|
|
if (dstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mov eax,dst_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, dst_lo, IA32_EAX));
|
|
|
|
|
|
|
|
if (sstk)
|
|
|
|
/* mov edx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mov edx,src_lo */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_lo, IA32_EDX));
|
|
|
|
|
|
|
|
switch (BPF_SIZE(code)) {
|
|
|
|
case BPF_B:
|
|
|
|
EMIT(0x88, 1); break;
|
|
|
|
case BPF_H:
|
|
|
|
EMIT2(0x66, 0x89); break;
|
|
|
|
case BPF_W:
|
|
|
|
case BPF_DW:
|
|
|
|
EMIT(0x89, 1); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_imm8(insn->off))
|
|
|
|
EMIT2(add_2reg(0x40, IA32_EAX, IA32_EDX),
|
|
|
|
insn->off);
|
|
|
|
else
|
|
|
|
EMIT1_off32(add_2reg(0x80, IA32_EAX, IA32_EDX),
|
|
|
|
insn->off);
|
|
|
|
|
|
|
|
if (BPF_SIZE(code) == BPF_DW) {
|
|
|
|
if (sstk)
|
|
|
|
/* mov edi,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(src_hi));
|
|
|
|
else
|
|
|
|
/* mov edi,src_hi */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_hi,
|
|
|
|
IA32_EDX));
|
|
|
|
EMIT1(0x89);
|
|
|
|
if (is_imm8(insn->off + 4)) {
|
|
|
|
EMIT2(add_2reg(0x40, IA32_EAX,
|
|
|
|
IA32_EDX),
|
|
|
|
insn->off + 4);
|
|
|
|
} else {
|
|
|
|
EMIT1(add_2reg(0x80, IA32_EAX,
|
|
|
|
IA32_EDX));
|
|
|
|
EMIT(insn->off + 4, 4);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* LDX: dst_reg = *(u8*)(src_reg + off) */
|
|
|
|
case BPF_LDX | BPF_MEM | BPF_B:
|
|
|
|
case BPF_LDX | BPF_MEM | BPF_H:
|
|
|
|
case BPF_LDX | BPF_MEM | BPF_W:
|
|
|
|
case BPF_LDX | BPF_MEM | BPF_DW:
|
|
|
|
if (sstk)
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
else
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT2(0x8B, add_2reg(0xC0, src_lo, IA32_EAX));
|
|
|
|
|
|
|
|
switch (BPF_SIZE(code)) {
|
|
|
|
case BPF_B:
|
|
|
|
EMIT2(0x0F, 0xB6); break;
|
|
|
|
case BPF_H:
|
|
|
|
EMIT2(0x0F, 0xB7); break;
|
|
|
|
case BPF_W:
|
|
|
|
case BPF_DW:
|
|
|
|
EMIT(0x8B, 1); break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (is_imm8(insn->off))
|
|
|
|
EMIT2(add_2reg(0x40, IA32_EAX, IA32_EDX),
|
|
|
|
insn->off);
|
|
|
|
else
|
|
|
|
EMIT1_off32(add_2reg(0x80, IA32_EAX, IA32_EDX),
|
|
|
|
insn->off);
|
|
|
|
|
|
|
|
if (dstk)
|
|
|
|
/* mov dword ptr [ebp+off],edx */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
else
|
|
|
|
/* mov dst_lo,edx */
|
|
|
|
EMIT2(0x89, add_2reg(0xC0, dst_lo, IA32_EDX));
|
|
|
|
switch (BPF_SIZE(code)) {
|
|
|
|
case BPF_B:
|
|
|
|
case BPF_H:
|
|
|
|
case BPF_W:
|
2019-05-25 05:25:26 +07:00
|
|
|
if (!bpf_prog->aux->verifier_zext)
|
|
|
|
break;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0xC7, add_1reg(0x40, IA32_EBP),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
EMIT(0x0, 4);
|
|
|
|
} else {
|
|
|
|
EMIT3(0xC7, add_1reg(0xC0, dst_hi), 0);
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case BPF_DW:
|
|
|
|
EMIT2_off32(0x8B,
|
|
|
|
add_2reg(0x80, IA32_EAX, IA32_EDX),
|
|
|
|
insn->off + 4);
|
|
|
|
if (dstk)
|
|
|
|
EMIT3(0x89,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
else
|
|
|
|
EMIT2(0x89,
|
|
|
|
add_2reg(0xC0, dst_hi, IA32_EDX));
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* call */
|
|
|
|
case BPF_JMP | BPF_CALL:
|
|
|
|
{
|
|
|
|
const u8 *r1 = bpf2ia32[BPF_REG_1];
|
|
|
|
const u8 *r2 = bpf2ia32[BPF_REG_2];
|
|
|
|
const u8 *r3 = bpf2ia32[BPF_REG_3];
|
|
|
|
const u8 *r4 = bpf2ia32[BPF_REG_4];
|
|
|
|
const u8 *r5 = bpf2ia32[BPF_REG_5];
|
|
|
|
|
|
|
|
if (insn->src_reg == BPF_PSEUDO_CALL)
|
|
|
|
goto notyet;
|
|
|
|
|
|
|
|
func = (u8 *) __bpf_call_base + imm32;
|
|
|
|
jmp_offset = func - (image + addrs[i]);
|
|
|
|
|
|
|
|
if (!imm32 || !is_simm32(jmp_offset)) {
|
|
|
|
pr_err("unsupported BPF func %d addr %p image %p\n",
|
|
|
|
imm32, func, image);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* mov eax,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(r1[0]));
|
|
|
|
/* mov edx,dword ptr [ebp+off] */
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(r1[1]));
|
|
|
|
|
|
|
|
emit_push_r64(r5, &prog);
|
|
|
|
emit_push_r64(r4, &prog);
|
|
|
|
emit_push_r64(r3, &prog);
|
|
|
|
emit_push_r64(r2, &prog);
|
|
|
|
|
|
|
|
EMIT1_off32(0xE8, jmp_offset + 9);
|
|
|
|
|
|
|
|
/* mov dword ptr [ebp+off],eax */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(r0[0]));
|
|
|
|
/* mov dword ptr [ebp+off],edx */
|
|
|
|
EMIT3(0x89, add_2reg(0x40, IA32_EBP, IA32_EDX),
|
|
|
|
STACK_VAR(r0[1]));
|
|
|
|
|
|
|
|
/* add esp,32 */
|
|
|
|
EMIT3(0x83, add_1reg(0xC0, IA32_ESP), 32);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BPF_JMP | BPF_TAIL_CALL:
|
|
|
|
emit_bpf_tail_call(&prog);
|
|
|
|
break;
|
|
|
|
|
|
|
|
/* cond jump */
|
|
|
|
case BPF_JMP | BPF_JEQ | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JNE | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JGT | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JLT | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JGE | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JLE | BPF_X:
|
2019-01-27 00:26:07 +07:00
|
|
|
case BPF_JMP32 | BPF_JEQ | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JNE | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JGT | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JLT | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JGE | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JLE | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JSGT | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JSLE | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JSLT | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JSGE | BPF_X: {
|
|
|
|
bool is_jmp64 = BPF_CLASS(insn->code) == BPF_JMP;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = sstk ? IA32_ECX : src_lo;
|
|
|
|
u8 sreg_hi = sstk ? IA32_EBX : src_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EBX),
|
|
|
|
STACK_VAR(src_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64) {
|
|
|
|
/* cmp dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_hi, sreg_hi));
|
|
|
|
EMIT2(IA32_JNE, 2);
|
|
|
|
}
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* cmp dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_lo, sreg_lo));
|
|
|
|
goto emit_cond_jmp;
|
|
|
|
}
|
2019-04-27 15:28:26 +07:00
|
|
|
case BPF_JMP | BPF_JSGT | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JSLE | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JSLT | BPF_X:
|
|
|
|
case BPF_JMP | BPF_JSGE | BPF_X: {
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = sstk ? IA32_ECX : src_lo;
|
|
|
|
u8 sreg_hi = sstk ? IA32_EBX : src_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EBX),
|
|
|
|
STACK_VAR(src_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* cmp dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_hi, sreg_hi));
|
|
|
|
EMIT2(IA32_JNE, 10);
|
|
|
|
/* cmp dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_lo, sreg_lo));
|
|
|
|
goto emit_cond_jmp_signed;
|
|
|
|
}
|
2019-01-27 00:26:07 +07:00
|
|
|
case BPF_JMP | BPF_JSET | BPF_X:
|
|
|
|
case BPF_JMP32 | BPF_JSET | BPF_X: {
|
|
|
|
bool is_jmp64 = BPF_CLASS(insn->code) == BPF_JMP;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = sstk ? IA32_ECX : src_lo;
|
|
|
|
u8 sreg_hi = sstk ? IA32_EBX : src_hi;
|
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
if (sstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_ECX),
|
|
|
|
STACK_VAR(src_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EBX),
|
|
|
|
STACK_VAR(src_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
/* and dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x23, add_2reg(0xC0, sreg_lo, dreg_lo));
|
|
|
|
/* and dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x23, add_2reg(0xC0, sreg_hi, dreg_hi));
|
|
|
|
/* or dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x09, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
goto emit_cond_jmp;
|
|
|
|
}
|
2019-01-27 00:26:07 +07:00
|
|
|
case BPF_JMP | BPF_JSET | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JSET | BPF_K: {
|
|
|
|
bool is_jmp64 = BPF_CLASS(insn->code) == BPF_JMP;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = IA32_ECX;
|
|
|
|
u8 sreg_hi = IA32_EBX;
|
2019-01-27 00:26:07 +07:00
|
|
|
u32 hi;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* mov ecx,imm32 */
|
2019-01-27 00:26:07 +07:00
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, sreg_lo), imm32);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
/* and dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x23, add_2reg(0xC0, sreg_lo, dreg_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64) {
|
|
|
|
hi = imm32 & (1 << 31) ? (u32)~0 : 0;
|
|
|
|
/* mov ebx,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, sreg_hi), hi);
|
|
|
|
/* and dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x23, add_2reg(0xC0, sreg_hi, dreg_hi));
|
|
|
|
/* or dreg_lo,dreg_hi */
|
|
|
|
EMIT2(0x09, add_2reg(0xC0, dreg_lo, dreg_hi));
|
|
|
|
}
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
goto emit_cond_jmp;
|
|
|
|
}
|
|
|
|
case BPF_JMP | BPF_JEQ | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JNE | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JGT | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JLT | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JGE | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JLE | BPF_K:
|
2019-01-27 00:26:07 +07:00
|
|
|
case BPF_JMP32 | BPF_JEQ | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JNE | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JGT | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JLT | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JGE | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JLE | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JSGT | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JSLE | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JSLT | BPF_K:
|
|
|
|
case BPF_JMP32 | BPF_JSGE | BPF_K: {
|
|
|
|
bool is_jmp64 = BPF_CLASS(insn->code) == BPF_JMP;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = IA32_ECX;
|
|
|
|
u8 sreg_hi = IA32_EBX;
|
2019-01-27 00:26:07 +07:00
|
|
|
u32 hi;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64)
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
}
|
|
|
|
|
|
|
|
/* mov ecx,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX), imm32);
|
2019-01-27 00:26:07 +07:00
|
|
|
if (is_jmp64) {
|
|
|
|
hi = imm32 & (1 << 31) ? (u32)~0 : 0;
|
|
|
|
/* mov ebx,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EBX), hi);
|
|
|
|
/* cmp dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_hi, sreg_hi));
|
|
|
|
EMIT2(IA32_JNE, 2);
|
|
|
|
}
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
/* cmp dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_lo, sreg_lo));
|
|
|
|
|
2019-04-27 15:28:26 +07:00
|
|
|
emit_cond_jmp: jmp_cond = get_cond_jmp_opcode(BPF_OP(code), false);
|
|
|
|
if (jmp_cond == COND_JMP_OPCODE_INVALID)
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
return -EFAULT;
|
|
|
|
jmp_offset = addrs[i + insn->off] - addrs[i];
|
|
|
|
if (is_imm8(jmp_offset)) {
|
|
|
|
EMIT2(jmp_cond, jmp_offset);
|
|
|
|
} else if (is_simm32(jmp_offset)) {
|
|
|
|
EMIT2_off32(0x0F, jmp_cond + 0x10, jmp_offset);
|
|
|
|
} else {
|
|
|
|
pr_err("cond_jmp gen bug %llx\n", jmp_offset);
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
2019-04-27 15:28:26 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BPF_JMP | BPF_JSGT | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JSLE | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JSLT | BPF_K:
|
|
|
|
case BPF_JMP | BPF_JSGE | BPF_K: {
|
|
|
|
u8 dreg_lo = dstk ? IA32_EAX : dst_lo;
|
|
|
|
u8 dreg_hi = dstk ? IA32_EDX : dst_hi;
|
|
|
|
u8 sreg_lo = IA32_ECX;
|
|
|
|
u8 sreg_hi = IA32_EBX;
|
|
|
|
u32 hi;
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
|
2019-04-27 15:28:26 +07:00
|
|
|
if (dstk) {
|
|
|
|
EMIT3(0x8B, add_2reg(0x40, IA32_EBP, IA32_EAX),
|
|
|
|
STACK_VAR(dst_lo));
|
|
|
|
EMIT3(0x8B,
|
|
|
|
add_2reg(0x40, IA32_EBP,
|
|
|
|
IA32_EDX),
|
|
|
|
STACK_VAR(dst_hi));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* mov ecx,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_ECX), imm32);
|
|
|
|
hi = imm32 & (1 << 31) ? (u32)~0 : 0;
|
|
|
|
/* mov ebx,imm32 */
|
|
|
|
EMIT2_off32(0xC7, add_1reg(0xC0, IA32_EBX), hi);
|
|
|
|
/* cmp dreg_hi,sreg_hi */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_hi, sreg_hi));
|
|
|
|
EMIT2(IA32_JNE, 10);
|
|
|
|
/* cmp dreg_lo,sreg_lo */
|
|
|
|
EMIT2(0x39, add_2reg(0xC0, dreg_lo, sreg_lo));
|
|
|
|
|
|
|
|
/*
|
|
|
|
* For simplicity of branch offset computation,
|
|
|
|
* let's use fixed jump coding here.
|
|
|
|
*/
|
|
|
|
emit_cond_jmp_signed: /* Check the condition for low 32-bit comparison */
|
|
|
|
jmp_cond = get_cond_jmp_opcode(BPF_OP(code), true);
|
|
|
|
if (jmp_cond == COND_JMP_OPCODE_INVALID)
|
|
|
|
return -EFAULT;
|
|
|
|
jmp_offset = addrs[i + insn->off] - addrs[i] + 8;
|
|
|
|
if (is_simm32(jmp_offset)) {
|
|
|
|
EMIT2_off32(0x0F, jmp_cond + 0x10, jmp_offset);
|
|
|
|
} else {
|
|
|
|
pr_err("cond_jmp gen bug %llx\n", jmp_offset);
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
EMIT2(0xEB, 6);
|
|
|
|
|
|
|
|
/* Check the condition for high 32-bit comparison */
|
|
|
|
jmp_cond = get_cond_jmp_opcode(BPF_OP(code), false);
|
|
|
|
if (jmp_cond == COND_JMP_OPCODE_INVALID)
|
|
|
|
return -EFAULT;
|
|
|
|
jmp_offset = addrs[i + insn->off] - addrs[i];
|
|
|
|
if (is_simm32(jmp_offset)) {
|
|
|
|
EMIT2_off32(0x0F, jmp_cond + 0x10, jmp_offset);
|
|
|
|
} else {
|
|
|
|
pr_err("cond_jmp gen bug %llx\n", jmp_offset);
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
case BPF_JMP | BPF_JA:
|
|
|
|
if (insn->off == -1)
|
|
|
|
/* -1 jmp instructions will always jump
|
|
|
|
* backwards two bytes. Explicitly handling
|
|
|
|
* this case avoids wasting too many passes
|
|
|
|
* when there are long sequences of replaced
|
|
|
|
* dead code.
|
|
|
|
*/
|
|
|
|
jmp_offset = -2;
|
|
|
|
else
|
|
|
|
jmp_offset = addrs[i + insn->off] - addrs[i];
|
|
|
|
|
|
|
|
if (!jmp_offset)
|
|
|
|
/* Optimize out nop jumps */
|
|
|
|
break;
|
|
|
|
emit_jmp:
|
|
|
|
if (is_imm8(jmp_offset)) {
|
|
|
|
EMIT2(0xEB, jmp_offset);
|
|
|
|
} else if (is_simm32(jmp_offset)) {
|
|
|
|
EMIT1_off32(0xE9, jmp_offset);
|
|
|
|
} else {
|
|
|
|
pr_err("jmp gen bug %llx\n", jmp_offset);
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
/* STX XADD: lock *(u32 *)(dst + off) += src */
|
|
|
|
case BPF_STX | BPF_XADD | BPF_W:
|
|
|
|
/* STX XADD: lock *(u64 *)(dst + off) += src */
|
|
|
|
case BPF_STX | BPF_XADD | BPF_DW:
|
|
|
|
goto notyet;
|
|
|
|
case BPF_JMP | BPF_EXIT:
|
|
|
|
if (seen_exit) {
|
|
|
|
jmp_offset = ctx->cleanup_addr - addrs[i];
|
|
|
|
goto emit_jmp;
|
|
|
|
}
|
|
|
|
seen_exit = true;
|
|
|
|
/* Update cleanup_addr */
|
|
|
|
ctx->cleanup_addr = proglen;
|
|
|
|
emit_epilogue(&prog, bpf_prog->aux->stack_depth);
|
|
|
|
break;
|
|
|
|
notyet:
|
|
|
|
pr_info_once("*** NOT YET: opcode %02x ***\n", code);
|
|
|
|
return -EFAULT;
|
|
|
|
default:
|
|
|
|
/*
|
|
|
|
* This error will be seen if new instruction was added
|
|
|
|
* to interpreter, but not to JIT or if there is junk in
|
|
|
|
* bpf_prog
|
|
|
|
*/
|
|
|
|
pr_err("bpf_jit: unknown opcode %02x\n", code);
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
ilen = prog - temp;
|
|
|
|
if (ilen > BPF_MAX_INSN_SIZE) {
|
|
|
|
pr_err("bpf_jit: fatal insn size error\n");
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (image) {
|
|
|
|
if (unlikely(proglen + ilen > oldproglen)) {
|
|
|
|
pr_err("bpf_jit: fatal error\n");
|
|
|
|
return -EFAULT;
|
|
|
|
}
|
|
|
|
memcpy(image + proglen, temp, ilen);
|
|
|
|
}
|
|
|
|
proglen += ilen;
|
|
|
|
addrs[i] = proglen;
|
|
|
|
prog = temp;
|
|
|
|
}
|
|
|
|
return proglen;
|
|
|
|
}
|
|
|
|
|
2019-05-25 05:25:26 +07:00
|
|
|
bool bpf_jit_needs_zext(void)
|
|
|
|
{
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
struct bpf_prog *bpf_int_jit_compile(struct bpf_prog *prog)
|
|
|
|
{
|
|
|
|
struct bpf_binary_header *header = NULL;
|
|
|
|
struct bpf_prog *tmp, *orig_prog = prog;
|
|
|
|
int proglen, oldproglen = 0;
|
|
|
|
struct jit_context ctx = {};
|
|
|
|
bool tmp_blinded = false;
|
|
|
|
u8 *image = NULL;
|
|
|
|
int *addrs;
|
|
|
|
int pass;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
if (!prog->jit_requested)
|
|
|
|
return orig_prog;
|
|
|
|
|
|
|
|
tmp = bpf_jit_blind_constants(prog);
|
|
|
|
/*
|
|
|
|
* If blinding was requested and we failed during blinding,
|
|
|
|
* we must fall back to the interpreter.
|
|
|
|
*/
|
|
|
|
if (IS_ERR(tmp))
|
|
|
|
return orig_prog;
|
|
|
|
if (tmp != prog) {
|
|
|
|
tmp_blinded = true;
|
|
|
|
prog = tmp;
|
|
|
|
}
|
|
|
|
|
treewide: kmalloc() -> kmalloc_array()
The kmalloc() function has a 2-factor argument form, kmalloc_array(). This
patch replaces cases of:
kmalloc(a * b, gfp)
with:
kmalloc_array(a * b, gfp)
as well as handling cases of:
kmalloc(a * b * c, gfp)
with:
kmalloc(array3_size(a, b, c), gfp)
as it's slightly less ugly than:
kmalloc_array(array_size(a, b), c, gfp)
This does, however, attempt to ignore constant size factors like:
kmalloc(4 * 1024, gfp)
though any constants defined via macros get caught up in the conversion.
Any factors with a sizeof() of "unsigned char", "char", and "u8" were
dropped, since they're redundant.
The tools/ directory was manually excluded, since it has its own
implementation of kmalloc().
The Coccinelle script used for this was:
// Fix redundant parens around sizeof().
@@
type TYPE;
expression THING, E;
@@
(
kmalloc(
- (sizeof(TYPE)) * E
+ sizeof(TYPE) * E
, ...)
|
kmalloc(
- (sizeof(THING)) * E
+ sizeof(THING) * E
, ...)
)
// Drop single-byte sizes and redundant parens.
@@
expression COUNT;
typedef u8;
typedef __u8;
@@
(
kmalloc(
- sizeof(u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * (COUNT)
+ COUNT
, ...)
|
kmalloc(
- sizeof(u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(__u8) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(char) * COUNT
+ COUNT
, ...)
|
kmalloc(
- sizeof(unsigned char) * COUNT
+ COUNT
, ...)
)
// 2-factor product with sizeof(type/expression) and identifier or constant.
@@
type TYPE;
expression THING;
identifier COUNT_ID;
constant COUNT_CONST;
@@
(
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_ID)
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_ID
+ COUNT_ID, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (COUNT_CONST)
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * COUNT_CONST
+ COUNT_CONST, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_ID)
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_ID
+ COUNT_ID, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (COUNT_CONST)
+ COUNT_CONST, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * COUNT_CONST
+ COUNT_CONST, sizeof(THING)
, ...)
)
// 2-factor product, only identifiers.
@@
identifier SIZE, COUNT;
@@
- kmalloc
+ kmalloc_array
(
- SIZE * COUNT
+ COUNT, SIZE
, ...)
// 3-factor product with 1 sizeof(type) or sizeof(expression), with
// redundant parens removed.
@@
expression THING;
identifier STRIDE, COUNT;
type TYPE;
@@
(
kmalloc(
- sizeof(TYPE) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(TYPE) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(TYPE))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * (COUNT) * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * (STRIDE)
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
|
kmalloc(
- sizeof(THING) * COUNT * STRIDE
+ array3_size(COUNT, STRIDE, sizeof(THING))
, ...)
)
// 3-factor product with 2 sizeof(variable), with redundant parens removed.
@@
expression THING1, THING2;
identifier COUNT;
type TYPE1, TYPE2;
@@
(
kmalloc(
- sizeof(TYPE1) * sizeof(TYPE2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(TYPE2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(THING1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(THING1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * COUNT
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
|
kmalloc(
- sizeof(TYPE1) * sizeof(THING2) * (COUNT)
+ array3_size(COUNT, sizeof(TYPE1), sizeof(THING2))
, ...)
)
// 3-factor product, only identifiers, with redundant parens removed.
@@
identifier STRIDE, SIZE, COUNT;
@@
(
kmalloc(
- (COUNT) * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * STRIDE * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- (COUNT) * (STRIDE) * (SIZE)
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
|
kmalloc(
- COUNT * STRIDE * SIZE
+ array3_size(COUNT, STRIDE, SIZE)
, ...)
)
// Any remaining multi-factor products, first at least 3-factor products,
// when they're not all constants...
@@
expression E1, E2, E3;
constant C1, C2, C3;
@@
(
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(
- (E1) * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * E3
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- (E1) * (E2) * (E3)
+ array3_size(E1, E2, E3)
, ...)
|
kmalloc(
- E1 * E2 * E3
+ array3_size(E1, E2, E3)
, ...)
)
// And then all remaining 2 factors products when they're not all constants,
// keeping sizeof() as the second factor argument.
@@
expression THING, E1, E2;
type TYPE;
constant C1, C2, C3;
@@
(
kmalloc(sizeof(THING) * C2, ...)
|
kmalloc(sizeof(TYPE) * C2, ...)
|
kmalloc(C1 * C2 * C3, ...)
|
kmalloc(C1 * C2, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * (E2)
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(TYPE) * E2
+ E2, sizeof(TYPE)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * (E2)
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- sizeof(THING) * E2
+ E2, sizeof(THING)
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * E2
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- (E1) * (E2)
+ E1, E2
, ...)
|
- kmalloc
+ kmalloc_array
(
- E1 * E2
+ E1, E2
, ...)
)
Signed-off-by: Kees Cook <keescook@chromium.org>
2018-06-13 03:55:00 +07:00
|
|
|
addrs = kmalloc_array(prog->len, sizeof(*addrs), GFP_KERNEL);
|
bpf, x86_32: add eBPF JIT compiler for ia32
The JIT compiler emits ia32 bit instructions. Currently, It supports eBPF
only. Classic BPF is supported because of the conversion by BPF core.
Almost all instructions from eBPF ISA supported except the following:
BPF_ALU64 | BPF_DIV | BPF_K
BPF_ALU64 | BPF_DIV | BPF_X
BPF_ALU64 | BPF_MOD | BPF_K
BPF_ALU64 | BPF_MOD | BPF_X
BPF_STX | BPF_XADD | BPF_W
BPF_STX | BPF_XADD | BPF_DW
It doesn't support BPF_JMP|BPF_CALL with BPF_PSEUDO_CALL at the moment.
IA32 has few general purpose registers, EAX|EDX|ECX|EBX|ESI|EDI. I use
EAX|EDX|ECX|EBX as temporary registers to simulate instructions in eBPF
ISA, and allocate ESI|EDI to BPF_REG_AX for constant blinding, all others
eBPF registers, R0-R10, are simulated through scratch space on stack.
The reasons behind the hardware registers allocation policy are:
1:MUL need EAX:EDX, shift operation need ECX, so they aren't fit
for general eBPF 64bit register simulation.
2:We need at least 4 registers to simulate most eBPF ISA operations
on registers operands instead of on register&memory operands.
3:We need to put BPF_REG_AX on hardware registers, or constant blinding
will degrade jit performance heavily.
Tested on PC (Intel(R) Core(TM) i5-5200U CPU).
Testing results on i5-5200U:
1) test_bpf: Summary: 349 PASSED, 0 FAILED, [319/341 JIT'ed]
2) test_progs: Summary: 83 PASSED, 0 FAILED.
3) test_lpm: OK
4) test_lru_map: OK
5) test_verifier: Summary: 828 PASSED, 0 FAILED.
Above tests are all done in following two conditions separately:
1:bpf_jit_enable=1 and bpf_jit_harden=0
2:bpf_jit_enable=1 and bpf_jit_harden=2
Below are some numbers for this jit implementation:
Note:
I run test_progs in kselftest 100 times continuously for every condition,
the numbers are in format: total/times=avg.
The numbers that test_bpf reports show almost the same relation.
a:jit_enable=0 and jit_harden=0 b:jit_enable=1 and jit_harden=0
test_pkt_access:PASS:ipv4:15622/100=156 test_pkt_access:PASS:ipv4:10674/100=106
test_pkt_access:PASS:ipv6:9130/100=91 test_pkt_access:PASS:ipv6:4855/100=48
test_xdp:PASS:ipv4:240198/100=2401 test_xdp:PASS:ipv4:138912/100=1389
test_xdp:PASS:ipv6:137326/100=1373 test_xdp:PASS:ipv6:68542/100=685
test_l4lb:PASS:ipv4:61100/100=611 test_l4lb:PASS:ipv4:37302/100=373
test_l4lb:PASS:ipv6:101000/100=1010 test_l4lb:PASS:ipv6:55030/100=550
c:jit_enable=1 and jit_harden=2
test_pkt_access:PASS:ipv4:10558/100=105
test_pkt_access:PASS:ipv6:5092/100=50
test_xdp:PASS:ipv4:131902/100=1319
test_xdp:PASS:ipv6:77932/100=779
test_l4lb:PASS:ipv4:38924/100=389
test_l4lb:PASS:ipv6:57520/100=575
The numbers show we get 30%~50% improvement.
See Documentation/networking/filter.txt for more information.
Changelog:
Changes v5-v6:
1:Add do {} while (0) to RETPOLINE_RAX_BPF_JIT for
consistence reason.
2:Clean up non-standard comments, reported by Daniel Borkmann.
3:Fix a memory leak issue, repoted by Daniel Borkmann.
Changes v4-v5:
1:Delete is_on_stack, BPF_REG_AX is the only one
on real hardware registers, so just check with
it.
2:Apply commit 1612a981b766 ("bpf, x64: fix JIT emission
for dead code"), suggested by Daniel Borkmann.
Changes v3-v4:
1:Fix changelog in commit.
I install llvm-6.0, then test_progs willn't report errors.
I submit another patch:
"bpf: fix misaligned access for BPF_PROG_TYPE_PERF_EVENT program type on x86_32 platform"
to fix another problem, after that patch, test_verifier willn't report errors too.
2:Fix clear r0[1] twice unnecessarily in *BPF_IND|BPF_ABS* simulation.
Changes v2-v3:
1:Move BPF_REG_AX to real hardware registers for performance reason.
3:Using bpf_load_pointer instead of bpf_jit32.S, suggested by Daniel Borkmann.
4:Delete partial codes in 1c2a088a6626, suggested by Daniel Borkmann.
5:Some bug fixes and comments improvement.
Changes v1-v2:
1:Fix bug in emit_ia32_neg64.
2:Fix bug in emit_ia32_arsh_r64.
3:Delete filename in top level comment, suggested by Thomas Gleixner.
4:Delete unnecessary boiler plate text, suggested by Thomas Gleixner.
5:Rewrite some words in changelog.
6:CodingSytle improvement and a little more comments.
Signed-off-by: Wang YanQing <udknight@gmail.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
2018-05-03 13:10:43 +07:00
|
|
|
if (!addrs) {
|
|
|
|
prog = orig_prog;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Before first pass, make a rough estimation of addrs[]
|
|
|
|
* each BPF instruction is translated to less than 64 bytes
|
|
|
|
*/
|
|
|
|
for (proglen = 0, i = 0; i < prog->len; i++) {
|
|
|
|
proglen += 64;
|
|
|
|
addrs[i] = proglen;
|
|
|
|
}
|
|
|
|
ctx.cleanup_addr = proglen;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* JITed image shrinks with every pass and the loop iterates
|
|
|
|
* until the image stops shrinking. Very large BPF programs
|
|
|
|
* may converge on the last pass. In such case do one more
|
|
|
|
* pass to emit the final image.
|
|
|
|
*/
|
|
|
|
for (pass = 0; pass < 20 || image; pass++) {
|
|
|
|
proglen = do_jit(prog, addrs, image, oldproglen, &ctx);
|
|
|
|
if (proglen <= 0) {
|
|
|
|
out_image:
|
|
|
|
image = NULL;
|
|
|
|
if (header)
|
|
|
|
bpf_jit_binary_free(header);
|
|
|
|
prog = orig_prog;
|
|
|
|
goto out_addrs;
|
|
|
|
}
|
|
|
|
if (image) {
|
|
|
|
if (proglen != oldproglen) {
|
|
|
|
pr_err("bpf_jit: proglen=%d != oldproglen=%d\n",
|
|
|
|
proglen, oldproglen);
|
|
|
|
goto out_image;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (proglen == oldproglen) {
|
|
|
|
header = bpf_jit_binary_alloc(proglen, &image,
|
|
|
|
1, jit_fill_hole);
|
|
|
|
if (!header) {
|
|
|
|
prog = orig_prog;
|
|
|
|
goto out_addrs;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
oldproglen = proglen;
|
|
|
|
cond_resched();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bpf_jit_enable > 1)
|
|
|
|
bpf_jit_dump(prog->len, proglen, pass + 1, image);
|
|
|
|
|
|
|
|
if (image) {
|
|
|
|
bpf_jit_binary_lock_ro(header);
|
|
|
|
prog->bpf_func = (void *)image;
|
|
|
|
prog->jited = 1;
|
|
|
|
prog->jited_len = proglen;
|
|
|
|
} else {
|
|
|
|
prog = orig_prog;
|
|
|
|
}
|
|
|
|
|
|
|
|
out_addrs:
|
|
|
|
kfree(addrs);
|
|
|
|
out:
|
|
|
|
if (tmp_blinded)
|
|
|
|
bpf_jit_prog_release_other(prog, prog == orig_prog ?
|
|
|
|
tmp : orig_prog);
|
|
|
|
return prog;
|
|
|
|
}
|