linux_dsm_epyc7002/drivers/gpu/drm/vc4/vc4_validate_shaders.c
Jonas Pfeil c778cc5df9 drm/vc4: Add fragment shader threading support
FS threading brings performance improvements of 0-20% in glmark2.

The validation code checks for thread switch signals and ensures that
the registers of the other thread are not touched, and that our clamps
are not live across thread switches.  It also checks that the
threading and branching instructions do not interfere.

(Original patch by Jonas, changes by anholt for style cleanup,
removing validation the kernel doesn't need to do, and adding the flag
for userspace).

v2: Minor style fixes from checkpatch.

Signed-off-by: Jonas Pfeil <pfeiljonas@gmx.de>
Signed-off-by: Eric Anholt <eric@anholt.net>
2016-11-16 13:25:26 -08:00

945 lines
27 KiB
C

/*
* Copyright © 2014 Broadcom
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice (including the next
* paragraph) shall be included in all copies or substantial portions of the
* Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
/**
* DOC: Shader validator for VC4.
*
* The VC4 has no IOMMU between it and system memory, so a user with
* access to execute shaders could escalate privilege by overwriting
* system memory (using the VPM write address register in the
* general-purpose DMA mode) or reading system memory it shouldn't
* (reading it as a texture, or uniform data, or vertex data).
*
* This walks over a shader BO, ensuring that its accesses are
* appropriately bounded, and recording how many texture accesses are
* made and where so that we can do relocations for them in the
* uniform stream.
*/
#include "vc4_drv.h"
#include "vc4_qpu_defines.h"
#define LIVE_REG_COUNT (32 + 32 + 4)
struct vc4_shader_validation_state {
/* Current IP being validated. */
uint32_t ip;
/* IP at the end of the BO, do not read shader[max_ip] */
uint32_t max_ip;
uint64_t *shader;
struct vc4_texture_sample_info tmu_setup[2];
int tmu_write_count[2];
/* For registers that were last written to by a MIN instruction with
* one argument being a uniform, the address of the uniform.
* Otherwise, ~0.
*
* This is used for the validation of direct address memory reads.
*/
uint32_t live_min_clamp_offsets[LIVE_REG_COUNT];
bool live_max_clamp_regs[LIVE_REG_COUNT];
uint32_t live_immediates[LIVE_REG_COUNT];
/* Bitfield of which IPs are used as branch targets.
*
* Used for validation that the uniform stream is updated at the right
* points and clearing the texturing/clamping state.
*/
unsigned long *branch_targets;
/* Set when entering a basic block, and cleared when the uniform
* address update is found. This is used to make sure that we don't
* read uniforms when the address is undefined.
*/
bool needs_uniform_address_update;
/* Set when we find a backwards branch. If the branch is backwards,
* the taraget is probably doing an address reset to read uniforms,
* and so we need to be sure that a uniforms address is present in the
* stream, even if the shader didn't need to read uniforms in later
* basic blocks.
*/
bool needs_uniform_address_for_loop;
/* Set when we find an instruction writing the top half of the
* register files. If we allowed writing the unusable regs in
* a threaded shader, then the other shader running on our
* QPU's clamp validation would be invalid.
*/
bool all_registers_used;
};
static uint32_t
waddr_to_live_reg_index(uint32_t waddr, bool is_b)
{
if (waddr < 32) {
if (is_b)
return 32 + waddr;
else
return waddr;
} else if (waddr <= QPU_W_ACC3) {
return 64 + waddr - QPU_W_ACC0;
} else {
return ~0;
}
}
static uint32_t
raddr_add_a_to_live_reg_index(uint64_t inst)
{
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
uint32_t add_a = QPU_GET_FIELD(inst, QPU_ADD_A);
uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
if (add_a == QPU_MUX_A)
return raddr_a;
else if (add_a == QPU_MUX_B && sig != QPU_SIG_SMALL_IMM)
return 32 + raddr_b;
else if (add_a <= QPU_MUX_R3)
return 64 + add_a;
else
return ~0;
}
static bool
live_reg_is_upper_half(uint32_t lri)
{
return (lri >= 16 && lri < 32) ||
(lri >= 32 + 16 && lri < 32 + 32);
}
static bool
is_tmu_submit(uint32_t waddr)
{
return (waddr == QPU_W_TMU0_S ||
waddr == QPU_W_TMU1_S);
}
static bool
is_tmu_write(uint32_t waddr)
{
return (waddr >= QPU_W_TMU0_S &&
waddr <= QPU_W_TMU1_B);
}
static bool
record_texture_sample(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state,
int tmu)
{
uint32_t s = validated_shader->num_texture_samples;
int i;
struct vc4_texture_sample_info *temp_samples;
temp_samples = krealloc(validated_shader->texture_samples,
(s + 1) * sizeof(*temp_samples),
GFP_KERNEL);
if (!temp_samples)
return false;
memcpy(&temp_samples[s],
&validation_state->tmu_setup[tmu],
sizeof(*temp_samples));
validated_shader->num_texture_samples = s + 1;
validated_shader->texture_samples = temp_samples;
for (i = 0; i < 4; i++)
validation_state->tmu_setup[tmu].p_offset[i] = ~0;
return true;
}
static bool
check_tmu_write(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state,
bool is_mul)
{
uint64_t inst = validation_state->shader[validation_state->ip];
uint32_t waddr = (is_mul ?
QPU_GET_FIELD(inst, QPU_WADDR_MUL) :
QPU_GET_FIELD(inst, QPU_WADDR_ADD));
uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
int tmu = waddr > QPU_W_TMU0_B;
bool submit = is_tmu_submit(waddr);
bool is_direct = submit && validation_state->tmu_write_count[tmu] == 0;
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
if (is_direct) {
uint32_t add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
uint32_t clamp_reg, clamp_offset;
if (sig == QPU_SIG_SMALL_IMM) {
DRM_ERROR("direct TMU read used small immediate\n");
return false;
}
/* Make sure that this texture load is an add of the base
* address of the UBO to a clamped offset within the UBO.
*/
if (is_mul ||
QPU_GET_FIELD(inst, QPU_OP_ADD) != QPU_A_ADD) {
DRM_ERROR("direct TMU load wasn't an add\n");
return false;
}
/* We assert that the clamped address is the first
* argument, and the UBO base address is the second argument.
* This is arbitrary, but simpler than supporting flipping the
* two either way.
*/
clamp_reg = raddr_add_a_to_live_reg_index(inst);
if (clamp_reg == ~0) {
DRM_ERROR("direct TMU load wasn't clamped\n");
return false;
}
clamp_offset = validation_state->live_min_clamp_offsets[clamp_reg];
if (clamp_offset == ~0) {
DRM_ERROR("direct TMU load wasn't clamped\n");
return false;
}
/* Store the clamp value's offset in p1 (see reloc_tex() in
* vc4_validate.c).
*/
validation_state->tmu_setup[tmu].p_offset[1] =
clamp_offset;
if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
!(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF)) {
DRM_ERROR("direct TMU load didn't add to a uniform\n");
return false;
}
validation_state->tmu_setup[tmu].is_direct = true;
} else {
if (raddr_a == QPU_R_UNIF || (sig != QPU_SIG_SMALL_IMM &&
raddr_b == QPU_R_UNIF)) {
DRM_ERROR("uniform read in the same instruction as "
"texture setup.\n");
return false;
}
}
if (validation_state->tmu_write_count[tmu] >= 4) {
DRM_ERROR("TMU%d got too many parameters before dispatch\n",
tmu);
return false;
}
validation_state->tmu_setup[tmu].p_offset[validation_state->tmu_write_count[tmu]] =
validated_shader->uniforms_size;
validation_state->tmu_write_count[tmu]++;
/* Since direct uses a RADDR uniform reference, it will get counted in
* check_instruction_reads()
*/
if (!is_direct) {
if (validation_state->needs_uniform_address_update) {
DRM_ERROR("Texturing with undefined uniform address\n");
return false;
}
validated_shader->uniforms_size += 4;
}
if (submit) {
if (!record_texture_sample(validated_shader,
validation_state, tmu)) {
return false;
}
validation_state->tmu_write_count[tmu] = 0;
}
return true;
}
static bool require_uniform_address_uniform(struct vc4_validated_shader_info *validated_shader)
{
uint32_t o = validated_shader->num_uniform_addr_offsets;
uint32_t num_uniforms = validated_shader->uniforms_size / 4;
validated_shader->uniform_addr_offsets =
krealloc(validated_shader->uniform_addr_offsets,
(o + 1) *
sizeof(*validated_shader->uniform_addr_offsets),
GFP_KERNEL);
if (!validated_shader->uniform_addr_offsets)
return false;
validated_shader->uniform_addr_offsets[o] = num_uniforms;
validated_shader->num_uniform_addr_offsets++;
return true;
}
static bool
validate_uniform_address_write(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state,
bool is_mul)
{
uint64_t inst = validation_state->shader[validation_state->ip];
u32 add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
u32 raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
u32 raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
u32 add_lri = raddr_add_a_to_live_reg_index(inst);
/* We want our reset to be pointing at whatever uniform follows the
* uniforms base address.
*/
u32 expected_offset = validated_shader->uniforms_size + 4;
/* We only support absolute uniform address changes, and we
* require that they be in the current basic block before any
* of its uniform reads.
*
* One could potentially emit more efficient QPU code, by
* noticing that (say) an if statement does uniform control
* flow for all threads and that the if reads the same number
* of uniforms on each side. However, this scheme is easy to
* validate so it's all we allow for now.
*/
switch (QPU_GET_FIELD(inst, QPU_SIG)) {
case QPU_SIG_NONE:
case QPU_SIG_SCOREBOARD_UNLOCK:
case QPU_SIG_COLOR_LOAD:
case QPU_SIG_LOAD_TMU0:
case QPU_SIG_LOAD_TMU1:
break;
default:
DRM_ERROR("uniforms address change must be "
"normal math\n");
return false;
}
if (is_mul || QPU_GET_FIELD(inst, QPU_OP_ADD) != QPU_A_ADD) {
DRM_ERROR("Uniform address reset must be an ADD.\n");
return false;
}
if (QPU_GET_FIELD(inst, QPU_COND_ADD) != QPU_COND_ALWAYS) {
DRM_ERROR("Uniform address reset must be unconditional.\n");
return false;
}
if (QPU_GET_FIELD(inst, QPU_PACK) != QPU_PACK_A_NOP &&
!(inst & QPU_PM)) {
DRM_ERROR("No packing allowed on uniforms reset\n");
return false;
}
if (add_lri == -1) {
DRM_ERROR("First argument of uniform address write must be "
"an immediate value.\n");
return false;
}
if (validation_state->live_immediates[add_lri] != expected_offset) {
DRM_ERROR("Resetting uniforms with offset %db instead of %db\n",
validation_state->live_immediates[add_lri],
expected_offset);
return false;
}
if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
!(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF)) {
DRM_ERROR("Second argument of uniform address write must be "
"a uniform.\n");
return false;
}
validation_state->needs_uniform_address_update = false;
validation_state->needs_uniform_address_for_loop = false;
return require_uniform_address_uniform(validated_shader);
}
static bool
check_reg_write(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state,
bool is_mul)
{
uint64_t inst = validation_state->shader[validation_state->ip];
uint32_t waddr = (is_mul ?
QPU_GET_FIELD(inst, QPU_WADDR_MUL) :
QPU_GET_FIELD(inst, QPU_WADDR_ADD));
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
bool ws = inst & QPU_WS;
bool is_b = is_mul ^ ws;
u32 lri = waddr_to_live_reg_index(waddr, is_b);
if (lri != -1) {
uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
uint32_t cond_mul = QPU_GET_FIELD(inst, QPU_COND_MUL);
if (sig == QPU_SIG_LOAD_IMM &&
QPU_GET_FIELD(inst, QPU_PACK) == QPU_PACK_A_NOP &&
((is_mul && cond_mul == QPU_COND_ALWAYS) ||
(!is_mul && cond_add == QPU_COND_ALWAYS))) {
validation_state->live_immediates[lri] =
QPU_GET_FIELD(inst, QPU_LOAD_IMM);
} else {
validation_state->live_immediates[lri] = ~0;
}
if (live_reg_is_upper_half(lri))
validation_state->all_registers_used = true;
}
switch (waddr) {
case QPU_W_UNIFORMS_ADDRESS:
if (is_b) {
DRM_ERROR("relative uniforms address change "
"unsupported\n");
return false;
}
return validate_uniform_address_write(validated_shader,
validation_state,
is_mul);
case QPU_W_TLB_COLOR_MS:
case QPU_W_TLB_COLOR_ALL:
case QPU_W_TLB_Z:
/* These only interact with the tile buffer, not main memory,
* so they're safe.
*/
return true;
case QPU_W_TMU0_S:
case QPU_W_TMU0_T:
case QPU_W_TMU0_R:
case QPU_W_TMU0_B:
case QPU_W_TMU1_S:
case QPU_W_TMU1_T:
case QPU_W_TMU1_R:
case QPU_W_TMU1_B:
return check_tmu_write(validated_shader, validation_state,
is_mul);
case QPU_W_HOST_INT:
case QPU_W_TMU_NOSWAP:
case QPU_W_TLB_ALPHA_MASK:
case QPU_W_MUTEX_RELEASE:
/* XXX: I haven't thought about these, so don't support them
* for now.
*/
DRM_ERROR("Unsupported waddr %d\n", waddr);
return false;
case QPU_W_VPM_ADDR:
DRM_ERROR("General VPM DMA unsupported\n");
return false;
case QPU_W_VPM:
case QPU_W_VPMVCD_SETUP:
/* We allow VPM setup in general, even including VPM DMA
* configuration setup, because the (unsafe) DMA can only be
* triggered by QPU_W_VPM_ADDR writes.
*/
return true;
case QPU_W_TLB_STENCIL_SETUP:
return true;
}
return true;
}
static void
track_live_clamps(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state)
{
uint64_t inst = validation_state->shader[validation_state->ip];
uint32_t op_add = QPU_GET_FIELD(inst, QPU_OP_ADD);
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
uint32_t cond_add = QPU_GET_FIELD(inst, QPU_COND_ADD);
uint32_t add_a = QPU_GET_FIELD(inst, QPU_ADD_A);
uint32_t add_b = QPU_GET_FIELD(inst, QPU_ADD_B);
uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
bool ws = inst & QPU_WS;
uint32_t lri_add_a, lri_add, lri_mul;
bool add_a_is_min_0;
/* Check whether OP_ADD's A argumennt comes from a live MAX(x, 0),
* before we clear previous live state.
*/
lri_add_a = raddr_add_a_to_live_reg_index(inst);
add_a_is_min_0 = (lri_add_a != ~0 &&
validation_state->live_max_clamp_regs[lri_add_a]);
/* Clear live state for registers written by our instruction. */
lri_add = waddr_to_live_reg_index(waddr_add, ws);
lri_mul = waddr_to_live_reg_index(waddr_mul, !ws);
if (lri_mul != ~0) {
validation_state->live_max_clamp_regs[lri_mul] = false;
validation_state->live_min_clamp_offsets[lri_mul] = ~0;
}
if (lri_add != ~0) {
validation_state->live_max_clamp_regs[lri_add] = false;
validation_state->live_min_clamp_offsets[lri_add] = ~0;
} else {
/* Nothing further to do for live tracking, since only ADDs
* generate new live clamp registers.
*/
return;
}
/* Now, handle remaining live clamp tracking for the ADD operation. */
if (cond_add != QPU_COND_ALWAYS)
return;
if (op_add == QPU_A_MAX) {
/* Track live clamps of a value to a minimum of 0 (in either
* arg).
*/
if (sig != QPU_SIG_SMALL_IMM || raddr_b != 0 ||
(add_a != QPU_MUX_B && add_b != QPU_MUX_B)) {
return;
}
validation_state->live_max_clamp_regs[lri_add] = true;
} else if (op_add == QPU_A_MIN) {
/* Track live clamps of a value clamped to a minimum of 0 and
* a maximum of some uniform's offset.
*/
if (!add_a_is_min_0)
return;
if (!(add_b == QPU_MUX_A && raddr_a == QPU_R_UNIF) &&
!(add_b == QPU_MUX_B && raddr_b == QPU_R_UNIF &&
sig != QPU_SIG_SMALL_IMM)) {
return;
}
validation_state->live_min_clamp_offsets[lri_add] =
validated_shader->uniforms_size;
}
}
static bool
check_instruction_writes(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state)
{
uint64_t inst = validation_state->shader[validation_state->ip];
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
bool ok;
if (is_tmu_write(waddr_add) && is_tmu_write(waddr_mul)) {
DRM_ERROR("ADD and MUL both set up textures\n");
return false;
}
ok = (check_reg_write(validated_shader, validation_state, false) &&
check_reg_write(validated_shader, validation_state, true));
track_live_clamps(validated_shader, validation_state);
return ok;
}
static bool
check_branch(uint64_t inst,
struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state,
int ip)
{
int32_t branch_imm = QPU_GET_FIELD(inst, QPU_BRANCH_TARGET);
uint32_t waddr_add = QPU_GET_FIELD(inst, QPU_WADDR_ADD);
uint32_t waddr_mul = QPU_GET_FIELD(inst, QPU_WADDR_MUL);
if ((int)branch_imm < 0)
validation_state->needs_uniform_address_for_loop = true;
/* We don't want to have to worry about validation of this, and
* there's no need for it.
*/
if (waddr_add != QPU_W_NOP || waddr_mul != QPU_W_NOP) {
DRM_ERROR("branch instruction at %d wrote a register.\n",
validation_state->ip);
return false;
}
return true;
}
static bool
check_instruction_reads(struct vc4_validated_shader_info *validated_shader,
struct vc4_shader_validation_state *validation_state)
{
uint64_t inst = validation_state->shader[validation_state->ip];
uint32_t raddr_a = QPU_GET_FIELD(inst, QPU_RADDR_A);
uint32_t raddr_b = QPU_GET_FIELD(inst, QPU_RADDR_B);
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
if (raddr_a == QPU_R_UNIF ||
(raddr_b == QPU_R_UNIF && sig != QPU_SIG_SMALL_IMM)) {
/* This can't overflow the uint32_t, because we're reading 8
* bytes of instruction to increment by 4 here, so we'd
* already be OOM.
*/
validated_shader->uniforms_size += 4;
if (validation_state->needs_uniform_address_update) {
DRM_ERROR("Uniform read with undefined uniform "
"address\n");
return false;
}
}
if ((raddr_a >= 16 && raddr_a < 32) ||
(raddr_b >= 16 && raddr_b < 32 && sig != QPU_SIG_SMALL_IMM)) {
validation_state->all_registers_used = true;
}
return true;
}
/* Make sure that all branches are absolute and point within the shader, and
* note their targets for later.
*/
static bool
vc4_validate_branches(struct vc4_shader_validation_state *validation_state)
{
uint32_t max_branch_target = 0;
int ip;
int last_branch = -2;
for (ip = 0; ip < validation_state->max_ip; ip++) {
uint64_t inst = validation_state->shader[ip];
int32_t branch_imm = QPU_GET_FIELD(inst, QPU_BRANCH_TARGET);
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
uint32_t after_delay_ip = ip + 4;
uint32_t branch_target_ip;
if (sig == QPU_SIG_PROG_END) {
/* There are two delay slots after program end is
* signaled that are still executed, then we're
* finished. validation_state->max_ip is the
* instruction after the last valid instruction in the
* program.
*/
validation_state->max_ip = ip + 3;
continue;
}
if (sig != QPU_SIG_BRANCH)
continue;
if (ip - last_branch < 4) {
DRM_ERROR("Branch at %d during delay slots\n", ip);
return false;
}
last_branch = ip;
if (inst & QPU_BRANCH_REG) {
DRM_ERROR("branching from register relative "
"not supported\n");
return false;
}
if (!(inst & QPU_BRANCH_REL)) {
DRM_ERROR("relative branching required\n");
return false;
}
/* The actual branch target is the instruction after the delay
* slots, plus whatever byte offset is in the low 32 bits of
* the instruction. Make sure we're not branching beyond the
* end of the shader object.
*/
if (branch_imm % sizeof(inst) != 0) {
DRM_ERROR("branch target not aligned\n");
return false;
}
branch_target_ip = after_delay_ip + (branch_imm >> 3);
if (branch_target_ip >= validation_state->max_ip) {
DRM_ERROR("Branch at %d outside of shader (ip %d/%d)\n",
ip, branch_target_ip,
validation_state->max_ip);
return false;
}
set_bit(branch_target_ip, validation_state->branch_targets);
/* Make sure that the non-branching path is also not outside
* the shader.
*/
if (after_delay_ip >= validation_state->max_ip) {
DRM_ERROR("Branch at %d continues past shader end "
"(%d/%d)\n",
ip, after_delay_ip, validation_state->max_ip);
return false;
}
set_bit(after_delay_ip, validation_state->branch_targets);
max_branch_target = max(max_branch_target, after_delay_ip);
}
if (max_branch_target > validation_state->max_ip - 3) {
DRM_ERROR("Branch landed after QPU_SIG_PROG_END");
return false;
}
return true;
}
/* Resets any known state for the shader, used when we may be branched to from
* multiple locations in the program (or at shader start).
*/
static void
reset_validation_state(struct vc4_shader_validation_state *validation_state)
{
int i;
for (i = 0; i < 8; i++)
validation_state->tmu_setup[i / 4].p_offset[i % 4] = ~0;
for (i = 0; i < LIVE_REG_COUNT; i++) {
validation_state->live_min_clamp_offsets[i] = ~0;
validation_state->live_max_clamp_regs[i] = false;
validation_state->live_immediates[i] = ~0;
}
}
static bool
texturing_in_progress(struct vc4_shader_validation_state *validation_state)
{
return (validation_state->tmu_write_count[0] != 0 ||
validation_state->tmu_write_count[1] != 0);
}
static bool
vc4_handle_branch_target(struct vc4_shader_validation_state *validation_state)
{
uint32_t ip = validation_state->ip;
if (!test_bit(ip, validation_state->branch_targets))
return true;
if (texturing_in_progress(validation_state)) {
DRM_ERROR("Branch target landed during TMU setup\n");
return false;
}
/* Reset our live values tracking, since this instruction may have
* multiple predecessors.
*
* One could potentially do analysis to determine that, for
* example, all predecessors have a live max clamp in the same
* register, but we don't bother with that.
*/
reset_validation_state(validation_state);
/* Since we've entered a basic block from potentially multiple
* predecessors, we need the uniforms address to be updated before any
* unforms are read. We require that after any branch point, the next
* uniform to be loaded is a uniform address offset. That uniform's
* offset will be marked by the uniform address register write
* validation, or a one-off the end-of-program check.
*/
validation_state->needs_uniform_address_update = true;
return true;
}
struct vc4_validated_shader_info *
vc4_validate_shader(struct drm_gem_cma_object *shader_obj)
{
bool found_shader_end = false;
int shader_end_ip = 0;
uint32_t last_thread_switch_ip = -3;
uint32_t ip;
struct vc4_validated_shader_info *validated_shader = NULL;
struct vc4_shader_validation_state validation_state;
memset(&validation_state, 0, sizeof(validation_state));
validation_state.shader = shader_obj->vaddr;
validation_state.max_ip = shader_obj->base.size / sizeof(uint64_t);
reset_validation_state(&validation_state);
validation_state.branch_targets =
kcalloc(BITS_TO_LONGS(validation_state.max_ip),
sizeof(unsigned long), GFP_KERNEL);
if (!validation_state.branch_targets)
goto fail;
validated_shader = kcalloc(1, sizeof(*validated_shader), GFP_KERNEL);
if (!validated_shader)
goto fail;
if (!vc4_validate_branches(&validation_state))
goto fail;
for (ip = 0; ip < validation_state.max_ip; ip++) {
uint64_t inst = validation_state.shader[ip];
uint32_t sig = QPU_GET_FIELD(inst, QPU_SIG);
validation_state.ip = ip;
if (!vc4_handle_branch_target(&validation_state))
goto fail;
if (ip == last_thread_switch_ip + 3) {
/* Reset r0-r3 live clamp data */
int i;
for (i = 64; i < LIVE_REG_COUNT; i++) {
validation_state.live_min_clamp_offsets[i] = ~0;
validation_state.live_max_clamp_regs[i] = false;
validation_state.live_immediates[i] = ~0;
}
}
switch (sig) {
case QPU_SIG_NONE:
case QPU_SIG_WAIT_FOR_SCOREBOARD:
case QPU_SIG_SCOREBOARD_UNLOCK:
case QPU_SIG_COLOR_LOAD:
case QPU_SIG_LOAD_TMU0:
case QPU_SIG_LOAD_TMU1:
case QPU_SIG_PROG_END:
case QPU_SIG_SMALL_IMM:
case QPU_SIG_THREAD_SWITCH:
case QPU_SIG_LAST_THREAD_SWITCH:
if (!check_instruction_writes(validated_shader,
&validation_state)) {
DRM_ERROR("Bad write at ip %d\n", ip);
goto fail;
}
if (!check_instruction_reads(validated_shader,
&validation_state))
goto fail;
if (sig == QPU_SIG_PROG_END) {
found_shader_end = true;
shader_end_ip = ip;
}
if (sig == QPU_SIG_THREAD_SWITCH ||
sig == QPU_SIG_LAST_THREAD_SWITCH) {
validated_shader->is_threaded = true;
if (ip < last_thread_switch_ip + 3) {
DRM_ERROR("Thread switch too soon after "
"last switch at ip %d\n", ip);
goto fail;
}
last_thread_switch_ip = ip;
}
break;
case QPU_SIG_LOAD_IMM:
if (!check_instruction_writes(validated_shader,
&validation_state)) {
DRM_ERROR("Bad LOAD_IMM write at ip %d\n", ip);
goto fail;
}
break;
case QPU_SIG_BRANCH:
if (!check_branch(inst, validated_shader,
&validation_state, ip))
goto fail;
if (ip < last_thread_switch_ip + 3) {
DRM_ERROR("Branch in thread switch at ip %d",
ip);
goto fail;
}
break;
default:
DRM_ERROR("Unsupported QPU signal %d at "
"instruction %d\n", sig, ip);
goto fail;
}
/* There are two delay slots after program end is signaled
* that are still executed, then we're finished.
*/
if (found_shader_end && ip == shader_end_ip + 2)
break;
}
if (ip == validation_state.max_ip) {
DRM_ERROR("shader failed to terminate before "
"shader BO end at %zd\n",
shader_obj->base.size);
goto fail;
}
/* Might corrupt other thread */
if (validated_shader->is_threaded &&
validation_state.all_registers_used) {
DRM_ERROR("Shader uses threading, but uses the upper "
"half of the registers, too\n");
goto fail;
}
/* If we did a backwards branch and we haven't emitted a uniforms
* reset since then, we still need the uniforms stream to have the
* uniforms address available so that the backwards branch can do its
* uniforms reset.
*
* We could potentially prove that the backwards branch doesn't
* contain any uses of uniforms until program exit, but that doesn't
* seem to be worth the trouble.
*/
if (validation_state.needs_uniform_address_for_loop) {
if (!require_uniform_address_uniform(validated_shader))
goto fail;
validated_shader->uniforms_size += 4;
}
/* Again, no chance of integer overflow here because the worst case
* scenario is 8 bytes of uniforms plus handles per 8-byte
* instruction.
*/
validated_shader->uniforms_src_size =
(validated_shader->uniforms_size +
4 * validated_shader->num_texture_samples);
kfree(validation_state.branch_targets);
return validated_shader;
fail:
kfree(validation_state.branch_targets);
if (validated_shader) {
kfree(validated_shader->texture_samples);
kfree(validated_shader);
}
return NULL;
}