The JIT compiler emits ARM 32 bit instructions. Currently, It supports
eBPF only. Classic BPF is supported because of the conversion by BPF core.
This patch is essentially changing the current implementation of JIT compiler
of Berkeley Packet Filter from classic to internal with 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
Implementation is using scratch space to emulate 64 bit eBPF ISA on 32 bit
ARM because of deficiency of general purpose registers on ARM. Currently,
only LITTLE ENDIAN machines are supported in this eBPF JIT Compiler.
Tested on ARMv7 with QEMU by me (Shubham Bansal).
Testing results on ARMv7:
1) test_bpf: Summary: 341 PASSED, 0 FAILED, [312/333 JIT'ed]
2) test_tag: OK (40945 tests)
3) test_progs: Summary: 30 PASSED, 0 FAILED
4) test_lpm: OK
5) test_lru_map: OK
Above tests are all done with following flags enabled discreatly.
1) bpf_jit_enable=1
a) CONFIG_FRAME_POINTER enabled
b) CONFIG_FRAME_POINTER disabled
2) bpf_jit_enable=1 and bpf_jit_harden=2
a) CONFIG_FRAME_POINTER enabled
b) CONFIG_FRAME_POINTER disabled
See Documentation/networking/filter.txt for more information.
Signed-off-by: Shubham Bansal <illusionist.neo@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
For ARMv7 with UDIV instruction support, generate an UDIV instruction
followed by an MLS instruction.
For other ARM variants, generate code calling a C wrapper similar to
the jit_udiv() function used for BPF_ALU | BPF_DIV instructions.
Some performance numbers reported by the test_bpf module (the duration
per filter run is reported in nanoseconds, between "jitted:<x>" and
"PASS":
ARMv7 QEMU nojit: test_bpf: #3 DIV_MOD_KX jited:0 2196 PASS
ARMv7 QEMU jit: test_bpf: #3 DIV_MOD_KX jited:1 104 PASS
ARMv5 QEMU nojit: test_bpf: #3 DIV_MOD_KX jited:0 2176 PASS
ARMv5 QEMU jit: test_bpf: #3 DIV_MOD_KX jited:1 1104 PASS
ARMv5 kirkwood nojit: test_bpf: #3 DIV_MOD_KX jited:0 1103 PASS
ARMv5 kirkwood jit: test_bpf: #3 DIV_MOD_KX jited:1 311 PASS
Signed-off-by: Nicolas Schichan <nschichan@freebox.fr>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Will Deacon pointed out, that the currently used opcode for filling holes,
that is 0xe7ffffff, seems not robust enough ...
$ echo 0xffffffe7 | xxd -r > test.bin
$ arm-linux-gnueabihf-objdump -m arm -D -b binary test.bin
...
0: e7ffffff udf #65535 ; 0xffff
... while for Thumb, it ends up as ...
0: ffff e7ff vqshl.u64 q15, <illegal reg q15.5>, #63
... which is a bit fragile. The ARM specification defines some *permanently*
guaranteed undefined instruction (UDF) space, for example for ARM in ARMv7-AR,
section A5.4 and for Thumb in ARMv7-M, section A5.2.6.
Similarly, ptrace, kprobes, kgdb, bug and uprobes make use of such instruction
as well to trap. Given mentioned section from the specification, we can find
such a universe as (where 'x' denotes 'don't care'):
ARM: xxxx 0111 1111 xxxx xxxx xxxx 1111 xxxx
Thumb: 1101 1110 xxxx xxxx
We therefore should use a more robust opcode that fits both. Russell King
suggested that we can even reuse a single 32-bit word, that is, 0xe7fddef1
which will fault if executed in ARM *or* Thumb mode as done in f928d4f2a8
("ARM: poison the vectors page"). That will still hold our requirements:
$ echo 0xf1defde7 | xxd -r > test.bin
$ arm-unknown-linux-gnueabi-objdump -m arm -D -b binary test.bin
...
0: e7fddef1 udf #56801 ; 0xdde1
$ echo 0xf1defde7f1defde7f1defde7 | xxd -r > test.bin
$ arm-unknown-linux-gnueabi-objdump -marm -Mforce-thumb -D -b binary test.bin
...
0: def1 udf #241 ; 0xf1
2: e7fd b.n 0x0
4: def1 udf #241 ; 0xf1
6: e7fd b.n 0x4
8: def1 udf #241 ; 0xf1
a: e7fd b.n 0x8
So on ARM 0xe7fddef1 conforms to the above UDF pattern, and the low 16 bit
likewise correspond to UDF in Thumb case. The 0xe7fd part is an unconditional
branch back to the UDF instruction.
Signed-off-by: Daniel Borkmann <dborkman@redhat.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Mircea Gherzan <mgherzan@gmail.com>
Cc: Alexei Starovoitov <ast@plumgrid.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch is a follow-up for patch "filter: add XOR instruction for use
with X/K" that implements BPF ARM JIT parts for the BPF XOR operation.
Signed-off-by: Daniel Borkmann <daniel.borkmann@tik.ee.ethz.ch>
Cc: Mircea Gherzan <mgherzan@gmail.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Mircea Gherzan <mgherzan@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
JIT support for the XOR operation introduced by the commit
ffe06c17af.
Signed-off-by: Mircea Gherzan <mgherzan@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
Based of Matt Evans's PPC64 implementation.
The compiler generates ARM instructions but interworking is
supported for Thumb2 kernels.
Supports both little and big endian. Unaligned loads are emitted
for ARMv6+. Not all the BPF opcodes that deal with ancillary data
are supported. The scratch memory of the filter lives on the stack.
Hardware integer division is used if it is available.
Enabled in the same way as for x86-64 and PPC64:
echo 1 > /proc/sys/net/core/bpf_jit_enable
A value greater than 1 enables opcode output.
Signed-off-by: Mircea Gherzan <mgherzan@gmail.com>
Acked-by: David S. Miller <davem@davemloft.net>
Acked-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Russell King <rmk+kernel@arm.linux.org.uk>
