diff options
Diffstat (limited to 'drivers/gpu/drm/i915/gt/selftest_lrc.c')
| -rw-r--r-- | drivers/gpu/drm/i915/gt/selftest_lrc.c | 1842 |
1 files changed, 1662 insertions, 180 deletions
diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index b292f8cbd0bf..6f06ba750a0a 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -68,6 +68,71 @@ static void engine_heartbeat_enable(struct intel_engine_cs *engine, engine->props.heartbeat_interval_ms = saved; } +static int wait_for_submit(struct intel_engine_cs *engine, + struct i915_request *rq, + unsigned long timeout) +{ + timeout += jiffies; + do { + cond_resched(); + intel_engine_flush_submission(engine); + + if (READ_ONCE(engine->execlists.pending[0])) + continue; + + if (i915_request_is_active(rq)) + return 0; + + if (i915_request_started(rq)) /* that was quick! */ + return 0; + } while (time_before(jiffies, timeout)); + + return -ETIME; +} + +static int wait_for_reset(struct intel_engine_cs *engine, + struct i915_request *rq, + unsigned long timeout) +{ + timeout += jiffies; + + do { + cond_resched(); + intel_engine_flush_submission(engine); + + if (READ_ONCE(engine->execlists.pending[0])) + continue; + + if (i915_request_completed(rq)) + break; + + if (READ_ONCE(rq->fence.error)) + break; + } while (time_before(jiffies, timeout)); + + flush_scheduled_work(); + + if (rq->fence.error != -EIO) { + pr_err("%s: hanging request %llx:%lld not reset\n", + engine->name, + rq->fence.context, + rq->fence.seqno); + return -EINVAL; + } + + /* Give the request a jiffie to complete after flushing the worker */ + if (i915_request_wait(rq, 0, + max(0l, (long)(timeout - jiffies)) + 1) < 0) { + pr_err("%s: hanging request %llx:%lld did not complete\n", + engine->name, + rq->fence.context, + rq->fence.seqno); + return -ETIME; + } + + return 0; +} + static int live_sanitycheck(void *arg) { struct intel_gt *gt = arg; @@ -285,6 +350,84 @@ static int live_unlite_preempt(void *arg) return live_unlite_restore(arg, I915_USER_PRIORITY(I915_PRIORITY_MAX)); } +static int live_pin_rewind(void *arg) +{ + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err = 0; + + /* + * We have to be careful not to trust intel_ring too much, for example + * ring->head is updated upon retire which is out of sync with pinning + * the context. Thus we cannot use ring->head to set CTX_RING_HEAD, + * or else we risk writing an older, stale value. + * + * To simulate this, let's apply a bit of deliberate sabotague. + */ + + for_each_engine(engine, gt, id) { + struct intel_context *ce; + struct i915_request *rq; + struct intel_ring *ring; + struct igt_live_test t; + + if (igt_live_test_begin(&t, gt->i915, __func__, engine->name)) { + err = -EIO; + break; + } + + ce = intel_context_create(engine); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + break; + } + + err = intel_context_pin(ce); + if (err) { + intel_context_put(ce); + break; + } + + /* Keep the context awake while we play games */ + err = i915_active_acquire(&ce->active); + if (err) { + intel_context_unpin(ce); + intel_context_put(ce); + break; + } + ring = ce->ring; + + /* Poison the ring, and offset the next request from HEAD */ + memset32(ring->vaddr, STACK_MAGIC, ring->size / sizeof(u32)); + ring->emit = ring->size / 2; + ring->tail = ring->emit; + GEM_BUG_ON(ring->head); + + intel_context_unpin(ce); + + /* Submit a simple nop request */ + GEM_BUG_ON(intel_context_is_pinned(ce)); + rq = intel_context_create_request(ce); + i915_active_release(&ce->active); /* e.g. async retire */ + intel_context_put(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + break; + } + GEM_BUG_ON(!rq->head); + i915_request_add(rq); + + /* Expect not to hang! */ + if (igt_live_test_end(&t)) { + err = -EIO; + break; + } + } + + return err; +} + static int live_hold_reset(void *arg) { struct intel_gt *gt = arg; @@ -386,6 +529,152 @@ out: return err; } +static const char *error_repr(int err) +{ + return err ? "bad" : "good"; +} + +static int live_error_interrupt(void *arg) +{ + static const struct error_phase { + enum { GOOD = 0, BAD = -EIO } error[2]; + } phases[] = { + { { BAD, GOOD } }, + { { BAD, BAD } }, + { { BAD, GOOD } }, + { { GOOD, GOOD } }, /* sentinel */ + }; + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + + /* + * We hook up the CS_MASTER_ERROR_INTERRUPT to have forewarning + * of invalid commands in user batches that will cause a GPU hang. + * This is a faster mechanism than using hangcheck/heartbeats, but + * only detects problems the HW knows about -- it will not warn when + * we kill the HW! + * + * To verify our detection and reset, we throw some invalid commands + * at the HW and wait for the interrupt. + */ + + if (!intel_has_reset_engine(gt)) + return 0; + + for_each_engine(engine, gt, id) { + const struct error_phase *p; + unsigned long heartbeat; + int err = 0; + + engine_heartbeat_disable(engine, &heartbeat); + + for (p = phases; p->error[0] != GOOD; p++) { + struct i915_request *client[ARRAY_SIZE(phases->error)]; + u32 *cs; + int i; + + memset(client, 0, sizeof(*client)); + for (i = 0; i < ARRAY_SIZE(client); i++) { + struct intel_context *ce; + struct i915_request *rq; + + ce = intel_context_create(engine); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + goto out; + } + + rq = intel_context_create_request(ce); + intel_context_put(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto out; + } + + if (rq->engine->emit_init_breadcrumb) { + err = rq->engine->emit_init_breadcrumb(rq); + if (err) { + i915_request_add(rq); + goto out; + } + } + + cs = intel_ring_begin(rq, 2); + if (IS_ERR(cs)) { + i915_request_add(rq); + err = PTR_ERR(cs); + goto out; + } + + if (p->error[i]) { + *cs++ = 0xdeadbeef; + *cs++ = 0xdeadbeef; + } else { + *cs++ = MI_NOOP; + *cs++ = MI_NOOP; + } + + client[i] = i915_request_get(rq); + i915_request_add(rq); + } + + err = wait_for_submit(engine, client[0], HZ / 2); + if (err) { + pr_err("%s: first request did not start within time!\n", + engine->name); + err = -ETIME; + goto out; + } + + for (i = 0; i < ARRAY_SIZE(client); i++) { + if (i915_request_wait(client[i], 0, HZ / 5) < 0) + pr_debug("%s: %s request incomplete!\n", + engine->name, + error_repr(p->error[i])); + + if (!i915_request_started(client[i])) { + pr_debug("%s: %s request not stated!\n", + engine->name, + error_repr(p->error[i])); + err = -ETIME; + goto out; + } + + /* Kick the tasklet to process the error */ + intel_engine_flush_submission(engine); + if (client[i]->fence.error != p->error[i]) { + pr_err("%s: %s request completed with wrong error code: %d\n", + engine->name, + error_repr(p->error[i]), + client[i]->fence.error); + err = -EINVAL; + goto out; + } + } + +out: + for (i = 0; i < ARRAY_SIZE(client); i++) + if (client[i]) + i915_request_put(client[i]); + if (err) { + pr_err("%s: failed at phase[%zd] { %d, %d }\n", + engine->name, p - phases, + p->error[0], p->error[1]); + break; + } + } + + engine_heartbeat_enable(engine, heartbeat); + if (err) { + intel_gt_set_wedged(gt); + return err; + } + } + + return 0; +} + static int emit_semaphore_chain(struct i915_request *rq, struct i915_vma *vma, int idx) { @@ -580,6 +869,10 @@ static int live_timeslice_preempt(void *arg) if (err) goto err_map; + err = i915_vma_sync(vma); + if (err) + goto err_pin; + for_each_prime_number_from(count, 1, 16) { struct intel_engine_cs *engine; enum intel_engine_id id; @@ -614,33 +907,227 @@ err_obj: return err; } -static struct i915_request *nop_request(struct intel_engine_cs *engine) +static struct i915_request * +create_rewinder(struct intel_context *ce, + struct i915_request *wait, + void *slot, int idx) { + const u32 offset = + i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(slot); struct i915_request *rq; + u32 *cs; + int err; - rq = intel_engine_create_kernel_request(engine); + rq = intel_context_create_request(ce); if (IS_ERR(rq)) return rq; + if (wait) { + err = i915_request_await_dma_fence(rq, &wait->fence); + if (err) + goto err; + } + + cs = intel_ring_begin(rq, 10); + if (IS_ERR(cs)) { + err = PTR_ERR(cs); + goto err; + } + + *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE; + *cs++ = MI_NOOP; + + *cs++ = MI_SEMAPHORE_WAIT | + MI_SEMAPHORE_GLOBAL_GTT | + MI_SEMAPHORE_POLL | + MI_SEMAPHORE_SAD_NEQ_SDD; + *cs++ = 0; + *cs++ = offset; + *cs++ = 0; + + *cs++ = MI_STORE_REGISTER_MEM_GEN8 | MI_USE_GGTT; + *cs++ = i915_mmio_reg_offset(RING_TIMESTAMP(rq->engine->mmio_base)); + *cs++ = offset + idx * sizeof(u32); + *cs++ = 0; + + intel_ring_advance(rq, cs); + + rq->sched.attr.priority = I915_PRIORITY_MASK; + err = 0; +err: i915_request_get(rq); i915_request_add(rq); + if (err) { + i915_request_put(rq); + return ERR_PTR(err); + } return rq; } -static int wait_for_submit(struct intel_engine_cs *engine, - struct i915_request *rq, - unsigned long timeout) +static int live_timeslice_rewind(void *arg) { - timeout += jiffies; - do { - cond_resched(); + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + + /* + * The usual presumption on timeslice expiration is that we replace + * the active context with another. However, given a chain of + * dependencies we may end up with replacing the context with itself, + * but only a few of those requests, forcing us to rewind the + * RING_TAIL of the original request. + */ + if (!IS_ACTIVE(CONFIG_DRM_I915_TIMESLICE_DURATION)) + return 0; + + for_each_engine(engine, gt, id) { + enum { A1, A2, B1 }; + enum { X = 1, Y, Z }; + struct i915_request *rq[3] = {}; + struct intel_context *ce; + unsigned long heartbeat; + unsigned long timeslice; + int i, err = 0; + u32 *slot; + + if (!intel_engine_has_timeslices(engine)) + continue; + + /* + * A:rq1 -- semaphore wait, timestamp X + * A:rq2 -- write timestamp Y + * + * B:rq1 [await A:rq1] -- write timestamp Z + * + * Force timeslice, release semaphore. + * + * Expect execution/evaluation order XZY + */ + + engine_heartbeat_disable(engine, &heartbeat); + timeslice = xchg(&engine->props.timeslice_duration_ms, 1); + + slot = memset32(engine->status_page.addr + 1000, 0, 4); + + ce = intel_context_create(engine); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + goto err; + } + + rq[0] = create_rewinder(ce, NULL, slot, 1); + if (IS_ERR(rq[0])) { + intel_context_put(ce); + goto err; + } + + rq[1] = create_rewinder(ce, NULL, slot, 2); + intel_context_put(ce); + if (IS_ERR(rq[1])) + goto err; + + err = wait_for_submit(engine, rq[1], HZ / 2); + if (err) { + pr_err("%s: failed to submit first context\n", + engine->name); + goto err; + } + + ce = intel_context_create(engine); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + goto err; + } + + rq[2] = create_rewinder(ce, rq[0], slot, 3); + intel_context_put(ce); + if (IS_ERR(rq[2])) + goto err; + + err = wait_for_submit(engine, rq[2], HZ / 2); + if (err) { + pr_err("%s: failed to submit second context\n", + engine->name); + goto err; + } + GEM_BUG_ON(!timer_pending(&engine->execlists.timer)); + + /* ELSP[] = { { A:rq1, A:rq2 }, { B:rq1 } } */ + GEM_BUG_ON(!i915_request_is_active(rq[A1])); + GEM_BUG_ON(!i915_request_is_active(rq[A2])); + GEM_BUG_ON(!i915_request_is_active(rq[B1])); + + /* Wait for the timeslice to kick in */ + del_timer(&engine->execlists.timer); + tasklet_hi_schedule(&engine->execlists.tasklet); intel_engine_flush_submission(engine); - if (i915_request_is_active(rq)) - return 0; - } while (time_before(jiffies, timeout)); - return -ETIME; + /* -> ELSP[] = { { A:rq1 }, { B:rq1 } } */ + GEM_BUG_ON(!i915_request_is_active(rq[A1])); + GEM_BUG_ON(!i915_request_is_active(rq[B1])); + GEM_BUG_ON(i915_request_is_active(rq[A2])); + + /* Release the hounds! */ + slot[0] = 1; + wmb(); /* "pairs" with GPU; paranoid kick of internal CPU$ */ + + for (i = 1; i <= 3; i++) { + unsigned long timeout = jiffies + HZ / 2; + + while (!READ_ONCE(slot[i]) && + time_before(jiffies, timeout)) + ; + + if (!time_before(jiffies, timeout)) { + pr_err("%s: rq[%d] timed out\n", + engine->name, i - 1); + err = -ETIME; + goto err; + } + + pr_debug("%s: slot[%d]:%x\n", engine->name, i, slot[i]); + } + + /* XZY: XZ < XY */ + if (slot[Z] - slot[X] >= slot[Y] - slot[X]) { + pr_err("%s: timeslicing did not run context B [%u] before A [%u]!\n", + engine->name, + slot[Z] - slot[X], + slot[Y] - slot[X]); + err = -EINVAL; + } + +err: + memset32(&slot[0], -1, 4); + wmb(); + + engine->props.timeslice_duration_ms = timeslice; + engine_heartbeat_enable(engine, heartbeat); + for (i = 0; i < 3; i++) + i915_request_put(rq[i]); + if (igt_flush_test(gt->i915)) + err = -EIO; + if (err) + return err; + } + + return 0; +} + +static struct i915_request *nop_request(struct intel_engine_cs *engine) +{ + struct i915_request *rq; + + rq = intel_engine_create_kernel_request(engine); + if (IS_ERR(rq)) + return rq; + + i915_request_get(rq); + i915_request_add(rq); + + return rq; } static long timeslice_threshold(const struct intel_engine_cs *engine) @@ -688,6 +1175,10 @@ static int live_timeslice_queue(void *arg) if (err) goto err_map; + err = i915_vma_sync(vma); + if (err) + goto err_pin; + for_each_engine(engine, gt, id) { struct i915_sched_attr attr = { .priority = I915_USER_PRIORITY(I915_PRIORITY_MAX), @@ -774,6 +1265,7 @@ err_heartbeat: break; } +err_pin: i915_vma_unpin(vma); err_map: i915_gem_object_unpin_map(obj); @@ -832,6 +1324,10 @@ static int live_busywait_preempt(void *arg) if (err) goto err_map; + err = i915_vma_sync(vma); + if (err) + goto err_vma; + for_each_engine(engine, gt, id) { struct i915_request *lo, *hi; struct igt_live_test t; @@ -1352,14 +1848,9 @@ static int __cancel_active0(struct live_preempt_cancel *arg) if (err) goto out; - if (i915_request_wait(rq, 0, HZ / 5) < 0) { - err = -EIO; - goto out; - } - - if (rq->fence.error != -EIO) { - pr_err("Cancelled inflight0 request did not report -EIO\n"); - err = -EINVAL; + err = wait_for_reset(arg->engine, rq, HZ / 2); + if (err) { + pr_err("Cancelled inflight0 request did not reset\n"); goto out; } @@ -1417,10 +1908,9 @@ static int __cancel_active1(struct live_preempt_cancel *arg) goto out; igt_spinner_end(&arg->a.spin); - if (i915_request_wait(rq[1], 0, HZ / 5) < 0) { - err = -EIO; + err = wait_for_reset(arg->engine, rq[1], HZ / 2); + if (err) goto out; - } if (rq[0]->fence.error != 0) { pr_err("Normal inflight0 request did not complete\n"); @@ -1500,10 +1990,9 @@ static int __cancel_queued(struct live_preempt_cancel *arg) if (err) goto out; - if (i915_request_wait(rq[2], 0, HZ / 5) < 0) { - err = -EIO; + err = wait_for_reset(arg->engine, rq[2], HZ / 2); + if (err) goto out; - } if (rq[0]->fence.error != -EIO) { pr_err("Cancelled inflight0 request did not report -EIO\n"); @@ -1561,14 +2050,9 @@ static int __cancel_hostile(struct live_preempt_cancel *arg) if (err) goto out; - if (i915_request_wait(rq, 0, HZ / 5) < 0) { - err = -EIO; - goto out; - } - - if (rq->fence.error != -EIO) { - pr_err("Cancelled inflight0 request did not report -EIO\n"); - err = -EINVAL; + err = wait_for_reset(arg->engine, rq, HZ / 2); + if (err) { + pr_err("Cancelled inflight0 request did not reset\n"); goto out; } @@ -1656,7 +2140,7 @@ static int live_suppress_self_preempt(void *arg) if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) return 0; - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; /* presume black blox */ if (intel_vgpu_active(gt->i915)) @@ -2279,117 +2763,6 @@ static int live_preempt_gang(void *arg) return 0; } -static int live_preempt_hang(void *arg) -{ - struct intel_gt *gt = arg; - struct i915_gem_context *ctx_hi, *ctx_lo; - struct igt_spinner spin_hi, spin_lo; - struct intel_engine_cs *engine; - enum intel_engine_id id; - int err = -ENOMEM; - - if (!HAS_LOGICAL_RING_PREEMPTION(gt->i915)) - return 0; - - if (!intel_has_reset_engine(gt)) - return 0; - - if (igt_spinner_init(&spin_hi, gt)) - return -ENOMEM; - - if (igt_spinner_init(&spin_lo, gt)) - goto err_spin_hi; - - ctx_hi = kernel_context(gt->i915); - if (!ctx_hi) - goto err_spin_lo; - ctx_hi->sched.priority = - I915_USER_PRIORITY(I915_CONTEXT_MAX_USER_PRIORITY); - - ctx_lo = kernel_context(gt->i915); - if (!ctx_lo) - goto err_ctx_hi; - ctx_lo->sched.priority = - I915_USER_PRIORITY(I915_CONTEXT_MIN_USER_PRIORITY); - - for_each_engine(engine, gt, id) { - struct i915_request *rq; - - if (!intel_engine_has_preemption(engine)) - continue; - - rq = spinner_create_request(&spin_lo, ctx_lo, engine, - MI_ARB_CHECK); - if (IS_ERR(rq)) { - err = PTR_ERR(rq); - goto err_ctx_lo; - } - - i915_request_add(rq); - if (!igt_wait_for_spinner(&spin_lo, rq)) { - GEM_TRACE("lo spinner failed to start\n"); - GEM_TRACE_DUMP(); - intel_gt_set_wedged(gt); - err = -EIO; - goto err_ctx_lo; - } - - rq = spinner_create_request(&spin_hi, ctx_hi, engine, - MI_ARB_CHECK); - if (IS_ERR(rq)) { - igt_spinner_end(&spin_lo); - err = PTR_ERR(rq); - goto err_ctx_lo; - } - - init_completion(&engine->execlists.preempt_hang.completion); - engine->execlists.preempt_hang.inject_hang = true; - - i915_request_add(rq); - - if (!wait_for_completion_timeout(&engine->execlists.preempt_hang.completion, - HZ / 10)) { - pr_err("Preemption did not occur within timeout!"); - GEM_TRACE_DUMP(); - intel_gt_set_wedged(gt); - err = -EIO; - goto err_ctx_lo; - } - - set_bit(I915_RESET_ENGINE + id, >->reset.flags); - intel_engine_reset(engine, NULL); - clear_bit(I915_RESET_ENGINE + id, >->reset.flags); - - engine->execlists.preempt_hang.inject_hang = false; - - if (!igt_wait_for_spinner(&spin_hi, rq)) { - GEM_TRACE("hi spinner failed to start\n"); - GEM_TRACE_DUMP(); - intel_gt_set_wedged(gt); - err = -EIO; - goto err_ctx_lo; - } - - igt_spinner_end(&spin_hi); - igt_spinner_end(&spin_lo); - if (igt_flush_test(gt->i915)) { - err = -EIO; - goto err_ctx_lo; - } - } - - err = 0; -err_ctx_lo: - kernel_context_close(ctx_lo); -err_ctx_hi: - kernel_context_close(ctx_hi); -err_spin_lo: - igt_spinner_fini(&spin_lo); -err_spin_hi: - igt_spinner_fini(&spin_hi); - return err; -} - static int live_preempt_timeout(void *arg) { struct intel_gt *gt = arg; @@ -2882,7 +3255,7 @@ static int live_virtual_engine(void *arg) unsigned int class, inst; int err; - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; for_each_engine(engine, gt, id) { @@ -3015,7 +3388,7 @@ static int live_virtual_mask(void *arg) unsigned int class, inst; int err; - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; for (class = 0; class <= MAX_ENGINE_CLASS; class++) { @@ -3055,6 +3428,10 @@ static int preserved_virtual_engine(struct intel_gt *gt, if (IS_ERR(scratch)) return PTR_ERR(scratch); + err = i915_vma_sync(scratch); + if (err) + goto out_scratch; + ve = intel_execlists_create_virtual(siblings, nsibling); if (IS_ERR(ve)) { err = PTR_ERR(ve); @@ -3153,7 +3530,7 @@ static int live_virtual_preserved(void *arg) * are preserved. */ - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; /* As we use CS_GPR we cannot run before they existed on all engines. */ @@ -3243,15 +3620,21 @@ static int bond_virtual_engine(struct intel_gt *gt, rq[0] = ERR_PTR(-ENOMEM); for_each_engine(master, gt, id) { struct i915_sw_fence fence = {}; + struct intel_context *ce; if (master->class == class) continue; + ce = intel_context_create(master); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + goto out; + } + memset_p((void *)rq, ERR_PTR(-EINVAL), ARRAY_SIZE(rq)); - rq[0] = igt_spinner_create_request(&spin, - master->kernel_context, - MI_NOOP); + rq[0] = igt_spinner_create_request(&spin, ce, MI_NOOP); + intel_context_put(ce); if (IS_ERR(rq[0])) { err = PTR_ERR(rq[0]); goto out; @@ -3377,7 +3760,7 @@ static int live_virtual_bond(void *arg) unsigned int class, inst; int err; - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; for (class = 0; class <= MAX_ENGINE_CLASS; class++) { @@ -3538,7 +3921,7 @@ static int live_virtual_reset(void *arg) * forgotten. */ - if (USES_GUC_SUBMISSION(gt->i915)) + if (intel_uc_uses_guc_submission(>->uc)) return 0; if (!intel_has_reset_engine(gt)) @@ -3571,8 +3954,11 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_sanitycheck), SUBTEST(live_unlite_switch), SUBTEST(live_unlite_preempt), + SUBTEST(live_pin_rewind), SUBTEST(live_hold_reset), + SUBTEST(live_error_interrupt), SUBTEST(live_timeslice_preempt), + SUBTEST(live_timeslice_rewind), SUBTEST(live_timeslice_queue), SUBTEST(live_busywait_preempt), SUBTEST(live_preempt), @@ -3583,7 +3969,6 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt), SUBTEST(live_preempt_gang), - SUBTEST(live_preempt_hang), SUBTEST(live_preempt_timeout), SUBTEST(live_preempt_smoke), SUBTEST(live_virtual_engine), @@ -3631,6 +4016,62 @@ static void hexdump(const void *buf, size_t len) } } +static int emit_semaphore_signal(struct intel_context *ce, void *slot) +{ + const u32 offset = + i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(slot); + struct i915_request *rq; + u32 *cs; + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + cs = intel_ring_begin(rq, 4); + if (IS_ERR(cs)) { + i915_request_add(rq); + return PTR_ERR(cs); + } + + *cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT; + *cs++ = offset; + *cs++ = 0; + *cs++ = 1; + + intel_ring_advance(rq, cs); + + rq->sched.attr.priority = I915_PRIORITY_BARRIER; + i915_request_add(rq); + return 0; +} + +static int context_flush(struct intel_context *ce, long timeout) +{ + struct i915_request *rq; + struct dma_fence *fence; + int err = 0; + + rq = intel_engine_create_kernel_request(ce->engine); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + fence = i915_active_fence_get(&ce->timeline->last_request); + if (fence) { + i915_request_await_dma_fence(rq, fence); + dma_fence_put(fence); + } + + rq = i915_request_get(rq); + i915_request_add(rq); + if (i915_request_wait(rq, 0, timeout) < 0) + err = -ETIME; + i915_request_put(rq); + + rmb(); /* We know the request is written, make sure all state is too! */ + return err; +} + static int live_lrc_layout(void *arg) { struct intel_gt *gt = arg; @@ -3797,6 +4238,11 @@ static int live_lrc_fixed(void *arg) CTX_BB_STATE - 1, "BB_STATE" }, + { + i915_mmio_reg_offset(RING_CTX_TIMESTAMP(engine->mmio_base)), + CTX_TIMESTAMP - 1, + "RING_CTX_TIMESTAMP" + }, { }, }, *t; u32 *hw; @@ -3880,8 +4326,16 @@ static int __live_lrc_state(struct intel_engine_cs *engine, *cs++ = i915_ggtt_offset(scratch) + RING_TAIL_IDX * sizeof(u32); *cs++ = 0; + i915_vma_lock(scratch); + err = i915_request_await_object(rq, scratch->obj, true); + if (!err) + err = i915_vma_move_to_active(scratch, rq, EXEC_OBJECT_WRITE); + i915_vma_unlock(scratch); + i915_request_get(rq); i915_request_add(rq); + if (err) + goto err_rq; intel_engine_flush_submission(engine); expected[RING_TAIL_IDX] = ce->ring->tail; @@ -3947,13 +4401,13 @@ static int live_lrc_state(void *arg) return err; } -static int gpr_make_dirty(struct intel_engine_cs *engine) +static int gpr_make_dirty(struct intel_context *ce) { struct i915_request *rq; u32 *cs; int n; - rq = intel_engine_create_kernel_request(engine); + rq = intel_context_create_request(ce); if (IS_ERR(rq)) return PTR_ERR(rq); @@ -3965,20 +4419,79 @@ static int gpr_make_dirty(struct intel_engine_cs *engine) *cs++ = MI_LOAD_REGISTER_IMM(NUM_GPR_DW); for (n = 0; n < NUM_GPR_DW; n++) { - *cs++ = CS_GPR(engine, n); + *cs++ = CS_GPR(ce->engine, n); *cs++ = STACK_MAGIC; } *cs++ = MI_NOOP; intel_ring_advance(rq, cs); + + rq->sched.attr.priority = I915_PRIORITY_BARRIER; i915_request_add(rq); return 0; } -static int __live_gpr_clear(struct intel_engine_cs *engine, - struct i915_vma *scratch) +static struct i915_request * +__gpr_read(struct intel_context *ce, struct i915_vma *scratch, u32 *slot) { + const u32 offset = + i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(slot); + struct i915_request *rq; + u32 *cs; + int err; + int n; + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) + return rq; + + cs = intel_ring_begin(rq, 6 + 4 * NUM_GPR_DW); + if (IS_ERR(cs)) { + i915_request_add(rq); + return ERR_CAST(cs); + } + + *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE; + *cs++ = MI_NOOP; + + *cs++ = MI_SEMAPHORE_WAIT | + MI_SEMAPHORE_GLOBAL_GTT | + MI_SEMAPHORE_POLL | + MI_SEMAPHORE_SAD_NEQ_SDD; + *cs++ = 0; + *cs++ = offset; + *cs++ = 0; + + for (n = 0; n < NUM_GPR_DW; n++) { + *cs++ = MI_STORE_REGISTER_MEM_GEN8 | MI_USE_GGTT; + *cs++ = CS_GPR(ce->engine, n); + *cs++ = i915_ggtt_offset(scratch) + n * sizeof(u32); + *cs++ = 0; + } + + i915_vma_lock(scratch); + err = i915_request_await_object(rq, scratch->obj, true); + if (!err) + err = i915_vma_move_to_active(scratch, rq, EXEC_OBJECT_WRITE); + i915_vma_unlock(scratch); + + i915_request_get(rq); + i915_request_add(rq); + if (err) { + i915_request_put(rq); + rq = ERR_PTR(err); + } + + return rq; +} + +static int __live_lrc_gpr(struct intel_engine_cs *engine, + struct i915_vma *scratch, + bool preempt) +{ + u32 *slot = memset32(engine->status_page.addr + 1000, 0, 4); struct intel_context *ce; struct i915_request *rq; u32 *cs; @@ -3988,7 +4501,7 @@ static int __live_gpr_clear(struct intel_engine_cs *engine, if (INTEL_GEN(engine->i915) < 9 && engine->class != RENDER_CLASS) return 0; /* GPR only on rcs0 for gen8 */ - err = gpr_make_dirty(engine); + err = gpr_make_dirty(engine->kernel_context); if (err) return err; @@ -3996,28 +4509,28 @@ static int __live_gpr_clear(struct intel_engine_cs *engine, if (IS_ERR(ce)) return PTR_ERR(ce); - rq = intel_context_create_request(ce); + rq = __gpr_read(ce, scratch, slot); if (IS_ERR(rq)) { err = PTR_ERR(rq); goto err_put; } - cs = intel_ring_begin(rq, 4 * NUM_GPR_DW); - if (IS_ERR(cs)) { - err = PTR_ERR(cs); - i915_request_add(rq); - goto err_put; - } + err = wait_for_submit(engine, rq, HZ / 2); + if (err) + goto err_rq; - for (n = 0; n < NUM_GPR_DW; n++) { - *cs++ = MI_STORE_REGISTER_MEM_GEN8 | MI_USE_GGTT; - *cs++ = CS_GPR(engine, n); - *cs++ = i915_ggtt_offset(scratch) + n * sizeof(u32); - *cs++ = 0; - } + if (preempt) { + err = gpr_make_dirty(engine->kernel_context); + if (err) + goto err_rq; - i915_request_get(rq); - i915_request_add(rq); + err = emit_semaphore_signal(engine->kernel_context, slot); + if (err) + goto err_rq; + } else { + slot[0] = 1; + wmb(); + } if (i915_request_wait(rq, 0, HZ / 5) < 0) { err = -ETIME; @@ -4044,13 +4557,15 @@ static int __live_gpr_clear(struct intel_engine_cs *engine, i915_gem_object_unpin_map(scratch->obj); err_rq: + memset32(&slot[0], -1, 4); + wmb(); i915_request_put(rq); err_put: intel_context_put(ce); return err; } -static int live_gpr_clear(void *arg) +static int live_lrc_gpr(void *arg) { struct intel_gt *gt = arg; struct intel_engine_cs *engine; @@ -4068,7 +4583,971 @@ static int live_gpr_clear(void *arg) return PTR_ERR(scratch); for_each_engine(engine, gt, id) { - err = __live_gpr_clear(engine, scratch); + unsigned long heartbeat; + + engine_heartbeat_disable(engine, &heartbeat); + + err = __live_lrc_gpr(engine, scratch, false); + if (err) + goto err; + + err = __live_lrc_gpr(engine, scratch, true); + if (err) + goto err; + +err: + engine_heartbeat_enable(engine, heartbeat); + if (igt_flush_test(gt->i915)) + err = -EIO; + if (err) + break; + } + + i915_vma_unpin_and_release(&scratch, 0); + return err; +} + +static struct i915_request * +create_timestamp(struct intel_context *ce, void *slot, int idx) +{ + const u32 offset = + i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(slot); + struct i915_request *rq; + u32 *cs; + int err; + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) + return rq; + + cs = intel_ring_begin(rq, 10); + if (IS_ERR(cs)) { + err = PTR_ERR(cs); + goto err; + } + + *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE; + *cs++ = MI_NOOP; + + *cs++ = MI_SEMAPHORE_WAIT | + MI_SEMAPHORE_GLOBAL_GTT | + MI_SEMAPHORE_POLL | + MI_SEMAPHORE_SAD_NEQ_SDD; + *cs++ = 0; + *cs++ = offset; + *cs++ = 0; + + *cs++ = MI_STORE_REGISTER_MEM_GEN8 | MI_USE_GGTT; + *cs++ = i915_mmio_reg_offset(RING_CTX_TIMESTAMP(rq->engine->mmio_base)); + *cs++ = offset + idx * sizeof(u32); + *cs++ = 0; + + intel_ring_advance(rq, cs); + + rq->sched.attr.priority = I915_PRIORITY_MASK; + err = 0; +err: + i915_request_get(rq); + i915_request_add(rq); + if (err) { + i915_request_put(rq); + return ERR_PTR(err); + } + + return rq; +} + +struct lrc_timestamp { + struct intel_engine_cs *engine; + struct intel_context *ce[2]; + u32 poison; +}; + +static bool timestamp_advanced(u32 start, u32 end) +{ + return (s32)(end - start) > 0; +} + +static int __lrc_timestamp(const struct lrc_timestamp *arg, bool preempt) +{ + u32 *slot = memset32(arg->engine->status_page.addr + 1000, 0, 4); + struct i915_request *rq; + u32 timestamp; + int err = 0; + + arg->ce[0]->lrc_reg_state[CTX_TIMESTAMP] = arg->poison; + rq = create_timestamp(arg->ce[0], slot, 1); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + err = wait_for_submit(rq->engine, rq, HZ / 2); + if (err) + goto err; + + if (preempt) { + arg->ce[1]->lrc_reg_state[CTX_TIMESTAMP] = 0xdeadbeef; + err = emit_semaphore_signal(arg->ce[1], slot); + if (err) + goto err; + } else { + slot[0] = 1; + wmb(); + } + + /* And wait for switch to kernel (to save our context to memory) */ + err = context_flush(arg->ce[0], HZ / 2); + if (err) + goto err; + + if (!timestamp_advanced(arg->poison, slot[1])) { + pr_err("%s(%s): invalid timestamp on restore, context:%x, request:%x\n", + arg->engine->name, preempt ? "preempt" : "simple", + arg->poison, slot[1]); + err = -EINVAL; + } + + timestamp = READ_ONCE(arg->ce[0]->lrc_reg_state[CTX_TIMESTAMP]); + if (!timestamp_advanced(slot[1], timestamp)) { + pr_err("%s(%s): invalid timestamp on save, request:%x, context:%x\n", + arg->engine->name, preempt ? "preempt" : "simple", + slot[1], timestamp); + err = -EINVAL; + } + +err: + memset32(slot, -1, 4); + i915_request_put(rq); + return err; +} + +static int live_lrc_timestamp(void *arg) +{ + struct lrc_timestamp data = {}; + struct intel_gt *gt = arg; + enum intel_engine_id id; + const u32 poison[] = { + 0, + S32_MAX, + (u32)S32_MAX + 1, + U32_MAX, + }; + + /* + * We want to verify that the timestamp is saved and restore across + * context switches and is monotonic. + * + * So we do this with a little bit of LRC poisoning to check various + * boundary conditions, and see what happens if we preempt the context + * with a second request (carrying more poison into the timestamp). + */ + + for_each_engine(data.engine, gt, id) { + unsigned long heartbeat; + int i, err = 0; + + engine_heartbeat_disable(data.engine, &heartbeat); + + for (i = 0; i < ARRAY_SIZE(data.ce); i++) { + struct intel_context *tmp; + + tmp = intel_context_create(data.engine); + if (IS_ERR(tmp)) { + err = PTR_ERR(tmp); + goto err; + } + + err = intel_context_pin(tmp); + if (err) { + intel_context_put(tmp); + goto err; + } + + data.ce[i] = tmp; + } + + for (i = 0; i < ARRAY_SIZE(poison); i++) { + data.poison = poison[i]; + + err = __lrc_timestamp(&data, false); + if (err) + break; + + err = __lrc_timestamp(&data, true); + if (err) + break; + } + +err: + engine_heartbeat_enable(data.engine, heartbeat); + for (i = 0; i < ARRAY_SIZE(data.ce); i++) { + if (!data.ce[i]) + break; + + intel_context_unpin(data.ce[i]); + intel_context_put(data.ce[i]); + } + + if (igt_flush_test(gt->i915)) + err = -EIO; + if (err) + return err; + } + + return 0; +} + +static struct i915_vma * +create_user_vma(struct i915_address_space *vm, unsigned long size) +{ + struct drm_i915_gem_object *obj; + struct i915_vma *vma; + int err; + + obj = i915_gem_object_create_internal(vm->i915, size); + if (IS_ERR(obj)) + return ERR_CAST(obj); + + vma = i915_vma_instance(obj, vm, NULL); + if (IS_ERR(vma)) { + i915_gem_object_put(obj); + return vma; + } + + err = i915_vma_pin(vma, 0, 0, PIN_USER); + if (err) { + i915_gem_object_put(obj); + return ERR_PTR(err); + } + + return vma; +} + +static struct i915_vma * +store_context(struct intel_context *ce, struct i915_vma *scratch) +{ + struct i915_vma *batch; + u32 dw, x, *cs, *hw; + + batch = create_user_vma(ce->vm, SZ_64K); + if (IS_ERR(batch)) + return batch; + + cs = i915_gem_object_pin_map(batch->obj, I915_MAP_WC); + if (IS_ERR(cs)) { + i915_vma_put(batch); + return ERR_CAST(cs); + } + + x = 0; + dw = 0; + hw = ce->engine->pinned_default_state; + hw += LRC_STATE_PN * PAGE_SIZE / sizeof(*hw); + do { + u32 len = hw[dw] & 0x7f; + + if (hw[dw] == 0) { + dw++; + continue; + } + + if ((hw[dw] & GENMASK(31, 23)) != MI_INSTR(0x22, 0)) { + dw += len + 2; + continue; + } + + dw++; + len = (len + 1) / 2; + while (len--) { + *cs++ = MI_STORE_REGISTER_MEM_GEN8; + *cs++ = hw[dw]; + *cs++ = lower_32_bits(scratch->node.start + x); + *cs++ = upper_32_bits(scratch->node.start + x); + + dw += 2; + x += 4; + } + } while (dw < PAGE_SIZE / sizeof(u32) && + (hw[dw] & ~BIT(0)) != MI_BATCH_BUFFER_END); + + *cs++ = MI_BATCH_BUFFER_END; + + i915_gem_object_flush_map(batch->obj); + i915_gem_object_unpin_map(batch->obj); + + return batch; +} + +static int move_to_active(struct i915_request *rq, + struct i915_vma *vma, + unsigned int flags) +{ + int err; + + i915_vma_lock(vma); + err = i915_request_await_object(rq, vma->obj, flags); + if (!err) + err = i915_vma_move_to_active(vma, rq, flags); + i915_vma_unlock(vma); + + return err; +} + +static struct i915_request * +record_registers(struct intel_context *ce, + struct i915_vma *before, + struct i915_vma *after, + u32 *sema) +{ + struct i915_vma *b_before, *b_after; + struct i915_request *rq; + u32 *cs; + int err; + + b_before = store_context(ce, before); + if (IS_ERR(b_before)) + return ERR_CAST(b_before); + + b_after = store_context(ce, after); + if (IS_ERR(b_after)) { + rq = ERR_CAST(b_after); + goto err_before; + } + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) + goto err_after; + + err = move_to_active(rq, before, EXEC_OBJECT_WRITE); + if (err) + goto err_rq; + + err = move_to_active(rq, b_before, 0); + if (err) + goto err_rq; + + err = move_to_active(rq, after, EXEC_OBJECT_WRITE); + if (err) + goto err_rq; + + err = move_to_active(rq, b_after, 0); + if (err) + goto err_rq; + + cs = intel_ring_begin(rq, 14); + if (IS_ERR(cs)) { + err = PTR_ERR(cs); + goto err_rq; + } + + *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; + *cs++ = MI_BATCH_BUFFER_START_GEN8 | BIT(8); + *cs++ = lower_32_bits(b_before->node.start); + *cs++ = upper_32_bits(b_before->node.start); + + *cs++ = MI_ARB_ON_OFF | MI_ARB_ENABLE; + *cs++ = MI_SEMAPHORE_WAIT | + MI_SEMAPHORE_GLOBAL_GTT | + MI_SEMAPHORE_POLL | + MI_SEMAPHORE_SAD_NEQ_SDD; + *cs++ = 0; + *cs++ = i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(sema); + *cs++ = 0; + *cs++ = MI_NOOP; + + *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; + *cs++ = MI_BATCH_BUFFER_START_GEN8 | BIT(8); + *cs++ = lower_32_bits(b_after->node.start); + *cs++ = upper_32_bits(b_after->node.start); + + intel_ring_advance(rq, cs); + + WRITE_ONCE(*sema, 0); + i915_request_get(rq); + i915_request_add(rq); +err_after: + i915_vma_put(b_after); +err_before: + i915_vma_put(b_before); + return rq; + +err_rq: + i915_request_add(rq); + rq = ERR_PTR(err); + goto err_after; +} + +static struct i915_vma *load_context(struct intel_context *ce, u32 poison) +{ + struct i915_vma *batch; + u32 dw, *cs, *hw; + + batch = create_user_vma(ce->vm, SZ_64K); + if (IS_ERR(batch)) + return batch; + + cs = i915_gem_object_pin_map(batch->obj, I915_MAP_WC); + if (IS_ERR(cs)) { + i915_vma_put(batch); + return ERR_CAST(cs); + } + + dw = 0; + hw = ce->engine->pinned_default_state; + hw += LRC_STATE_PN * PAGE_SIZE / sizeof(*hw); + do { + u32 len = hw[dw] & 0x7f; + + if (hw[dw] == 0) { + dw++; + continue; + } + + if ((hw[dw] & GENMASK(31, 23)) != MI_INSTR(0x22, 0)) { + dw += len + 2; + continue; + } + + dw++; + len = (len + 1) / 2; + *cs++ = MI_LOAD_REGISTER_IMM(len); + while (len--) { + *cs++ = hw[dw]; + *cs++ = poison; + dw += 2; + } + } while (dw < PAGE_SIZE / sizeof(u32) && + (hw[dw] & ~BIT(0)) != MI_BATCH_BUFFER_END); + + *cs++ = MI_BATCH_BUFFER_END; + + i915_gem_object_flush_map(batch->obj); + i915_gem_object_unpin_map(batch->obj); + + return batch; +} + +static int poison_registers(struct intel_context *ce, u32 poison, u32 *sema) +{ + struct i915_request *rq; + struct i915_vma *batch; + u32 *cs; + int err; + + batch = load_context(ce, poison); + if (IS_ERR(batch)) + return PTR_ERR(batch); + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_batch; + } + + err = move_to_active(rq, batch, 0); + if (err) + goto err_rq; + + cs = intel_ring_begin(rq, 8); + if (IS_ERR(cs)) { + err = PTR_ERR(cs); + goto err_rq; + } + + *cs++ = MI_ARB_ON_OFF | MI_ARB_DISABLE; + *cs++ = MI_BATCH_BUFFER_START_GEN8 | BIT(8); + *cs++ = lower_32_bits(batch->node.start); + *cs++ = upper_32_bits(batch->node.start); + + *cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT; + *cs++ = i915_ggtt_offset(ce->engine->status_page.vma) + + offset_in_page(sema); + *cs++ = 0; + *cs++ = 1; + + intel_ring_advance(rq, cs); + + rq->sched.attr.priority = I915_PRIORITY_BARRIER; +err_rq: + i915_request_add(rq); +err_batch: + i915_vma_put(batch); + return err; +} + +static bool is_moving(u32 a, u32 b) +{ + return a != b; +} + +static int compare_isolation(struct intel_engine_cs *engine, + struct i915_vma *ref[2], + struct i915_vma *result[2], + struct intel_context *ce, + u32 poison) +{ + u32 x, dw, *hw, *lrc; + u32 *A[2], *B[2]; + int err = 0; + + A[0] = i915_gem_object_pin_map(ref[0]->obj, I915_MAP_WC); + if (IS_ERR(A[0])) + return PTR_ERR(A[0]); + + A[1] = i915_gem_object_pin_map(ref[1]->obj, I915_MAP_WC); + if (IS_ERR(A[1])) { + err = PTR_ERR(A[1]); + goto err_A0; + } + + B[0] = i915_gem_object_pin_map(result[0]->obj, I915_MAP_WC); + if (IS_ERR(B[0])) { + err = PTR_ERR(B[0]); + goto err_A1; + } + + B[1] = i915_gem_object_pin_map(result[1]->obj, I915_MAP_WC); + if (IS_ERR(B[1])) { + err = PTR_ERR(B[1]); + goto err_B0; + } + + lrc = i915_gem_object_pin_map(ce->state->obj, + i915_coherent_map_type(engine->i915)); + if (IS_ERR(lrc)) { + err = PTR_ERR(lrc); + goto err_B1; + } + lrc += LRC_STATE_PN * PAGE_SIZE / sizeof(*hw); + + x = 0; + dw = 0; + hw = engine->pinned_default_state; + hw += LRC_STATE_PN * PAGE_SIZE / sizeof(*hw); + do { + u32 len = hw[dw] & 0x7f; + + if (hw[dw] == 0) { + dw++; + continue; + } + + if ((hw[dw] & GENMASK(31, 23)) != MI_INSTR(0x22, 0)) { + dw += len + 2; + continue; + } + + dw++; + len = (len + 1) / 2; + while (len--) { + if (!is_moving(A[0][x], A[1][x]) && + (A[0][x] != B[0][x] || A[1][x] != B[1][x])) { + switch (hw[dw] & 4095) { + case 0x30: /* RING_HEAD */ + case 0x34: /* RING_TAIL */ + break; + + default: + pr_err("%s[%d]: Mismatch for register %4x, default %08x, reference %08x, result (%08x, %08x), poison %08x, context %08x\n", + engine->name, dw, + hw[dw], hw[dw + 1], + A[0][x], B[0][x], B[1][x], + poison, lrc[dw + 1]); + err = -EINVAL; + break; + } + } + dw += 2; + x++; + } + } while (dw < PAGE_SIZE / sizeof(u32) && + (hw[dw] & ~BIT(0)) != MI_BATCH_BUFFER_END); + + i915_gem_object_unpin_map(ce->state->obj); +err_B1: + i915_gem_object_unpin_map(result[1]->obj); +err_B0: + i915_gem_object_unpin_map(result[0]->obj); +err_A1: + i915_gem_object_unpin_map(ref[1]->obj); +err_A0: + i915_gem_object_unpin_map(ref[0]->obj); + return err; +} + +static int __lrc_isolation(struct intel_engine_cs *engine, u32 poison) +{ + u32 *sema = memset32(engine->status_page.addr + 1000, 0, 1); + struct i915_vma *ref[2], *result[2]; + struct intel_context *A, *B; + struct i915_request *rq; + int err; + + A = intel_context_create(engine); + if (IS_ERR(A)) + return PTR_ERR(A); + + B = intel_context_create(engine); + if (IS_ERR(B)) { + err = PTR_ERR(B); + goto err_A; + } + + ref[0] = create_user_vma(A->vm, SZ_64K); + if (IS_ERR(ref[0])) { + err = PTR_ERR(ref[0]); + goto err_B; + } + + ref[1] = create_user_vma(A->vm, SZ_64K); + if (IS_ERR(ref[1])) { + err = PTR_ERR(ref[1]); + goto err_ref0; + } + + rq = record_registers(A, ref[0], ref[1], sema); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_ref1; + } + + WRITE_ONCE(*sema, 1); + wmb(); + + if (i915_request_wait(rq, 0, HZ / 2) < 0) { + i915_request_put(rq); + err = -ETIME; + goto err_ref1; + } + i915_request_put(rq); + + result[0] = create_user_vma(A->vm, SZ_64K); + if (IS_ERR(result[0])) { + err = PTR_ERR(result[0]); + goto err_ref1; + } + + result[1] = create_user_vma(A->vm, SZ_64K); + if (IS_ERR(result[1])) { + err = PTR_ERR(result[1]); + goto err_result0; + } + + rq = record_registers(A, result[0], result[1], sema); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_result1; + } + + err = poison_registers(B, poison, sema); + if (err) { + WRITE_ONCE(*sema, -1); + i915_request_put(rq); + goto err_result1; + } + + if (i915_request_wait(rq, 0, HZ / 2) < 0) { + i915_request_put(rq); + err = -ETIME; + goto err_result1; + } + i915_request_put(rq); + + err = compare_isolation(engine, ref, result, A, poison); + +err_result1: + i915_vma_put(result[1]); +err_result0: + i915_vma_put(result[0]); +err_ref1: + i915_vma_put(ref[1]); +err_ref0: + i915_vma_put(ref[0]); +err_B: + intel_context_put(B); +err_A: + intel_context_put(A); + return err; +} + +static bool skip_isolation(const struct intel_engine_cs *engine) +{ + if (engine->class == COPY_ENGINE_CLASS && INTEL_GEN(engine->i915) == 9) + return true; + + if (engine->class == RENDER_CLASS && INTEL_GEN(engine->i915) == 11) + return true; + + return false; +} + +static int live_lrc_isolation(void *arg) +{ + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + const u32 poison[] = { + STACK_MAGIC, + 0x3a3a3a3a, + 0x5c5c5c5c, + 0xffffffff, + 0xffff0000, + }; + + /* + * Our goal is try and verify that per-context state cannot be + * tampered with by another non-privileged client. + * + * We take the list of context registers from the LRI in the default + * context image and attempt to modify that list from a remote context. + */ + + for_each_engine(engine, gt, id) { + int err = 0; + int i; + + /* Just don't even ask */ + if (!IS_ENABLED(CONFIG_DRM_I915_SELFTEST_BROKEN) && + skip_isolation(engine)) + continue; + + intel_engine_pm_get(engine); + if (engine->pinned_default_state) { + for (i = 0; i < ARRAY_SIZE(poison); i++) { + err = __lrc_isolation(engine, poison[i]); + if (err) + break; + + err = __lrc_isolation(engine, ~poison[i]); + if (err) + break; + } + } + intel_engine_pm_put(engine); + if (igt_flush_test(gt->i915)) + err = -EIO; + if (err) + return err; + } + + return 0; +} + +static void garbage_reset(struct intel_engine_cs *engine, + struct i915_request *rq) +{ + const unsigned int bit = I915_RESET_ENGINE + engine->id; + unsigned long *lock = &engine->gt->reset.flags; + + if (test_and_set_bit(bit, lock)) + return; + + tasklet_disable(&engine->execlists.tasklet); + + if (!rq->fence.error) + intel_engine_reset(engine, NULL); + + tasklet_enable(&engine->execlists.tasklet); + clear_and_wake_up_bit(bit, lock); +} + +static struct i915_request *garbage(struct intel_context *ce, + struct rnd_state *prng) +{ + struct i915_request *rq; + int err; + + err = intel_context_pin(ce); + if (err) + return ERR_PTR(err); + + prandom_bytes_state(prng, + ce->lrc_reg_state, + ce->engine->context_size - + LRC_STATE_PN * PAGE_SIZE); + + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_unpin; + } + + i915_request_get(rq); + i915_request_add(rq); + return rq; + +err_unpin: + intel_context_unpin(ce); + return ERR_PTR(err); +} + +static int __lrc_garbage(struct intel_engine_cs *engine, struct rnd_state *prng) +{ + struct intel_context *ce; + struct i915_request *hang; + int err = 0; + + ce = intel_context_create(engine); + if (IS_ERR(ce)) + return PTR_ERR(ce); + + hang = garbage(ce, prng); + if (IS_ERR(hang)) { + err = PTR_ERR(hang); + goto err_ce; + } + + if (wait_for_submit(engine, hang, HZ / 2)) { + i915_request_put(hang); + err = -ETIME; + goto err_ce; + } + + intel_context_set_banned(ce); + garbage_reset(engine, hang); + + intel_engine_flush_submission(engine); + if (!hang->fence.error) { + i915_request_put(hang); + pr_err("%s: corrupted context was not reset\n", + engine->name); + err = -EINVAL; + goto err_ce; + } + + if (i915_request_wait(hang, 0, HZ / 2) < 0) { + pr_err("%s: corrupted context did not recover\n", + engine->name); + i915_request_put(hang); + err = -EIO; + goto err_ce; + } + i915_request_put(hang); + +err_ce: + intel_context_put(ce); + return err; +} + +static int live_lrc_garbage(void *arg) +{ + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + + /* + * Verify that we can recover if one context state is completely + * corrupted. + */ + + if (!IS_ENABLED(CONFIG_DRM_I915_SELFTEST_BROKEN)) + return 0; + + for_each_engine(engine, gt, id) { + I915_RND_STATE(prng); + int err = 0, i; + + if (!intel_has_reset_engine(engine->gt)) + continue; + + intel_engine_pm_get(engine); + for (i = 0; i < 3; i++) { + err = __lrc_garbage(engine, &prng); + if (err) + break; + } + intel_engine_pm_put(engine); + + if (igt_flush_test(gt->i915)) + err = -EIO; + if (err) + return err; + } + + return 0; +} + +static int __live_pphwsp_runtime(struct intel_engine_cs *engine) +{ + struct intel_context *ce; + struct i915_request *rq; + IGT_TIMEOUT(end_time); + int err; + + ce = intel_context_create(engine); + if (IS_ERR(ce)) + return PTR_ERR(ce); + + ce->runtime.num_underflow = 0; + ce->runtime.max_underflow = 0; + + do { + unsigned int loop = 1024; + + while (loop) { + rq = intel_context_create_request(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto err_rq; + } + + if (--loop == 0) + i915_request_get(rq); + + i915_request_add(rq); + } + + if (__igt_timeout(end_time, NULL)) + break; + + i915_request_put(rq); + } while (1); + + err = i915_request_wait(rq, 0, HZ / 5); + if (err < 0) { + pr_err("%s: request not completed!\n", engine->name); + goto err_wait; + } + + igt_flush_test(engine->i915); + + pr_info("%s: pphwsp runtime %lluns, average %lluns\n", + engine->name, + intel_context_get_total_runtime_ns(ce), + intel_context_get_avg_runtime_ns(ce)); + + err = 0; + if (ce->runtime.num_underflow) { + pr_err("%s: pphwsp underflow %u time(s), max %u cycles!\n", + engine->name, + ce->runtime.num_underflow, + ce->runtime.max_underflow); + GEM_TRACE_DUMP(); + err = -EOVERFLOW; + } + +err_wait: + i915_request_put(rq); +err_rq: + intel_context_put(ce); + return err; +} + +static int live_pphwsp_runtime(void *arg) +{ + struct intel_gt *gt = arg; + struct intel_engine_cs *engine; + enum intel_engine_id id; + int err = 0; + + /* + * Check that cumulative context runtime as stored in the pphwsp[16] + * is monotonic. + */ + + for_each_engine(engine, gt, id) { + err = __live_pphwsp_runtime(engine); if (err) break; } @@ -4076,7 +5555,6 @@ static int live_gpr_clear(void *arg) if (igt_flush_test(gt->i915)) err = -EIO; - i915_vma_unpin_and_release(&scratch, 0); return err; } @@ -4086,7 +5564,11 @@ int intel_lrc_live_selftests(struct drm_i915_private *i915) SUBTEST(live_lrc_layout), SUBTEST(live_lrc_fixed), SUBTEST(live_lrc_state), - SUBTEST(live_gpr_clear), + SUBTEST(live_lrc_gpr), + SUBTEST(live_lrc_isolation), + SUBTEST(live_lrc_timestamp), + SUBTEST(live_lrc_garbage), + SUBTEST(live_pphwsp_runtime), }; if (!HAS_LOGICAL_RING_CONTEXTS(i915)) |