drm/i915/selftest: Check that GPR are restored across noa_wait

Perf implements a GPU delay (noa_wait) by looping until the CS timestamp
has passed a certain point. This use MI_MATH and the general purpose
registers of the user's context, and since it is clobbering the user
state it must carefully save and restore the user's data around the
noa_wait. We can verify this by loading some values in the GPR that we
know will be clobbered by the noa_wait, and then inspecting the GPR after
the noa_wait completes and confirming that they have been restored.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200709224504.11345-2-chris@chris-wilson.co.uk
This commit is contained in:
Chris Wilson 2020-07-09 23:45:04 +01:00
parent e43ff99c8d
commit ed2690a9ca

View File

@ -280,11 +280,142 @@ static int live_noa_delay(void *arg)
return err;
}
static int live_noa_gpr(void *arg)
{
struct drm_i915_private *i915 = arg;
struct i915_perf_stream *stream;
struct intel_context *ce;
struct i915_request *rq;
u32 *cs, *store;
void *scratch;
u32 gpr0;
int err;
int i;
/* Check that the delay does not clobber user context state (GPR) */
stream = test_stream(&i915->perf);
if (!stream)
return -ENOMEM;
gpr0 = i915_mmio_reg_offset(GEN8_RING_CS_GPR(stream->engine->mmio_base, 0));
ce = intel_context_create(stream->engine);
if (IS_ERR(ce)) {
err = PTR_ERR(ce);
goto out;
}
/* Poison the ce->vm so we detect writes not to the GGTT gt->scratch */
scratch = kmap(ce->vm->scratch[0].base.page);
memset(scratch, POISON_FREE, PAGE_SIZE);
rq = intel_context_create_request(ce);
if (IS_ERR(rq)) {
err = PTR_ERR(rq);
goto out_ce;
}
i915_request_get(rq);
if (rq->engine->emit_init_breadcrumb) {
err = rq->engine->emit_init_breadcrumb(rq);
if (err) {
i915_request_add(rq);
goto out_rq;
}
}
/* Fill the 16 qword [32 dword] GPR with a known unlikely value */
cs = intel_ring_begin(rq, 2 * 32 + 2);
if (IS_ERR(cs)) {
i915_request_add(rq);
goto out_rq;
}
*cs++ = MI_LOAD_REGISTER_IMM(32);
for (i = 0; i < 32; i++) {
*cs++ = gpr0 + i * sizeof(u32);
*cs++ = STACK_MAGIC;
}
*cs++ = MI_NOOP;
intel_ring_advance(rq, cs);
/* Execute the GPU delay */
err = rq->engine->emit_bb_start(rq,
i915_ggtt_offset(stream->noa_wait), 0,
I915_DISPATCH_SECURE);
if (err) {
i915_request_add(rq);
goto out_rq;
}
/* Read the GPR back, using the pinned global HWSP for convenience */
store = memset32(rq->engine->status_page.addr + 512, 0, 32);
for (i = 0; i < 32; i++) {
u32 cmd;
cs = intel_ring_begin(rq, 4);
if (IS_ERR(cs)) {
i915_request_add(rq);
goto out_rq;
}
cmd = MI_STORE_REGISTER_MEM;
if (INTEL_GEN(i915) >= 8)
cmd++;
cmd |= MI_USE_GGTT;
*cs++ = cmd;
*cs++ = gpr0 + i * sizeof(u32);
*cs++ = i915_ggtt_offset(rq->engine->status_page.vma) +
offset_in_page(store) +
i * sizeof(u32);
*cs++ = 0;
intel_ring_advance(rq, cs);
}
i915_request_add(rq);
if (i915_request_wait(rq, I915_WAIT_INTERRUPTIBLE, HZ / 2) < 0) {
pr_err("noa_wait timed out\n");
intel_gt_set_wedged(stream->engine->gt);
err = -EIO;
goto out_rq;
}
/* Verify that the GPR contain our expected values */
for (i = 0; i < 32; i++) {
if (store[i] == STACK_MAGIC)
continue;
pr_err("GPR[%d] lost, found:%08x, expected:%08x!\n",
i, store[i], STACK_MAGIC);
err = -EINVAL;
}
/* Verify that the user's scratch page was not used for GPR storage */
if (memchr_inv(scratch, POISON_FREE, PAGE_SIZE)) {
pr_err("Scratch page overwritten!\n");
igt_hexdump(scratch, 4096);
err = -EINVAL;
}
out_rq:
i915_request_put(rq);
out_ce:
kunmap(ce->vm->scratch[0].base.page);
intel_context_put(ce);
out:
stream_destroy(stream);
return err;
}
int i915_perf_live_selftests(struct drm_i915_private *i915)
{
static const struct i915_subtest tests[] = {
SUBTEST(live_sanitycheck),
SUBTEST(live_noa_delay),
SUBTEST(live_noa_gpr),
};
struct i915_perf *perf = &i915->perf;
int err;