From patchwork Sun Oct 6 16:49:59 2019 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Chris Wilson X-Patchwork-Id: 11176529 Return-Path: Received: from mail.kernel.org (pdx-korg-mail-1.web.codeaurora.org [172.30.200.123]) by pdx-korg-patchwork-2.web.codeaurora.org (Postfix) with ESMTP id 338681709 for ; Sun, 6 Oct 2019 16:50:22 +0000 (UTC) Received: from gabe.freedesktop.org (gabe.freedesktop.org [131.252.210.177]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by mail.kernel.org (Postfix) with ESMTPS id 1BFC520835 for ; Sun, 6 Oct 2019 16:50:22 +0000 (UTC) DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org 1BFC520835 Authentication-Results: mail.kernel.org; dmarc=none (p=none dis=none) header.from=chris-wilson.co.uk Authentication-Results: mail.kernel.org; spf=none smtp.mailfrom=intel-gfx-bounces@lists.freedesktop.org Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 6E3BB89AC0; Sun, 6 Oct 2019 16:50:18 +0000 (UTC) X-Original-To: intel-gfx@lists.freedesktop.org Delivered-To: intel-gfx@lists.freedesktop.org Received: from fireflyinternet.com (mail.fireflyinternet.com [109.228.58.192]) by gabe.freedesktop.org (Postfix) with ESMTPS id 6C39C6E3E5 for ; Sun, 6 Oct 2019 16:50:16 +0000 (UTC) X-Default-Received-SPF: pass (skip=forwardok (res=PASS)) x-ip-name=78.156.65.138; Received: from haswell.alporthouse.com (unverified [78.156.65.138]) by fireflyinternet.com (Firefly Internet (M1)) with ESMTP id 18742847-1500050 for multiple; Sun, 06 Oct 2019 17:50:05 +0100 From: Chris Wilson To: intel-gfx@lists.freedesktop.org Date: Sun, 6 Oct 2019 17:49:59 +0100 Message-Id: <20191006165002.30312-9-chris@chris-wilson.co.uk> X-Mailer: git-send-email 2.23.0 In-Reply-To: <20191006165002.30312-1-chris@chris-wilson.co.uk> References: <20191006165002.30312-1-chris@chris-wilson.co.uk> MIME-Version: 1.0 Subject: [Intel-gfx] [PATCH 09/12] drm/i915/execlists: Cancel banned contexts on schedule-out X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.23 Precedence: list List-Id: Intel graphics driver community testing & development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" On completion of a banned context, scrub the context image so that we do not replay the active payload. The intent is that we skip banned payloads on request submission so that the timeline advancement continues on in the background. However, if we are returning to a preempted request, i915_request_skip() is ineffective and instead we need to patch up the context image so that it continues from the start of the next request. Signed-off-by: Chris Wilson --- drivers/gpu/drm/i915/gt/intel_lrc.c | 58 ++++++ drivers/gpu/drm/i915/gt/selftest_lrc.c | 273 +++++++++++++++++++++++++ 2 files changed, 331 insertions(+) diff --git a/drivers/gpu/drm/i915/gt/intel_lrc.c b/drivers/gpu/drm/i915/gt/intel_lrc.c index 177b0f6874dd..9cfc436634d2 100644 --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -234,6 +234,9 @@ static void execlists_init_reg_state(u32 *reg_state, const struct intel_engine_cs *engine, const struct intel_ring *ring, bool close); +static void +__execlists_update_reg_state(const struct intel_context *ce, + const struct intel_engine_cs *engine); static void __context_pin_acquire(struct intel_context *ce) { @@ -1022,6 +1025,58 @@ static void kick_siblings(struct i915_request *rq, struct intel_context *ce) tasklet_schedule(&ve->base.execlists.tasklet); } +static void +mark_complete(struct i915_request *rq, struct intel_engine_cs *engine) +{ + const struct intel_timeline * const tl = rcu_dereference(rq->timeline); + + *(u32 *)tl->hwsp_seqno = rq->fence.seqno; + GEM_BUG_ON(!i915_request_completed(rq)); + + list_for_each_entry_from_reverse(rq, &tl->requests, link) { + if (i915_request_signaled(rq)) + break; + + mark_eio(rq); + } + + intel_engine_queue_breadcrumbs(engine); +} + +static void cancel_active(struct i915_request *rq, + struct intel_engine_cs *engine) +{ + struct intel_context * const ce = rq->hw_context; + u32 *regs = ce->lrc_reg_state; + + if (i915_request_completed(rq)) + return; + + GEM_TRACE("%s(%s): { rq=%llx:%lld }\n", + __func__, engine->name, rq->fence.context, rq->fence.seqno); + __context_pin_acquire(ce); + + /* Scrub the context image to prevent replaying the previous batch */ + memcpy(regs, /* skip restoring the vanilla PPHWSP */ + engine->pinned_default_state + LRC_STATE_PN * PAGE_SIZE, + engine->context_size - PAGE_SIZE); + execlists_init_reg_state(regs, ce, engine, ce->ring, false); + + /* Ring will be advanced on retire; here we need to reset the context */ + ce->ring->head = intel_ring_wrap(ce->ring, rq->wa_tail); + __execlists_update_reg_state(ce, engine); + + /* We've switched away, so this should be a no-op, but intent matters */ + ce->lrc_desc |= CTX_DESC_FORCE_RESTORE; + + /* Let everyone know that the request may now be retired */ + rcu_read_lock(); + mark_complete(rq, engine); + rcu_read_unlock(); + + __context_pin_release(ce); +} + static inline void __execlists_schedule_out(struct i915_request *rq, struct intel_engine_cs * const engine) @@ -1032,6 +1087,9 @@ __execlists_schedule_out(struct i915_request *rq, execlists_context_status_change(rq, INTEL_CONTEXT_SCHEDULE_OUT); intel_gt_pm_put(engine->gt); + if (unlikely(i915_gem_context_is_banned(ce->gem_context))) + cancel_active(rq, engine); + /* * If this is part of a virtual engine, its next request may * have been blocked waiting for access to the active context. diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index 198cf2f754f4..1703130ef0ef 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -7,6 +7,7 @@ #include #include "gem/i915_gem_pm.h" +#include "gt/intel_engine_heartbeat.h" #include "gt/intel_reset.h" #include "i915_selftest.h" @@ -986,6 +987,277 @@ static int live_nopreempt(void *arg) goto err_client_b; } +struct live_preempt_cancel { + struct intel_engine_cs *engine; + struct preempt_client a, b; +}; + +static int __cancel_active0(struct live_preempt_cancel *arg) +{ + struct i915_request *rq; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP0 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq)) + return PTR_ERR(rq); + + i915_request_get(rq); + i915_request_add(rq); + if (!igt_wait_for_spinner(&arg->a.spin, rq)) { + err = -EIO; + goto out; + } + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + 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; + goto out; + } + +out: + i915_request_put(rq); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_active1(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[2] = {}; + struct igt_live_test t; + int err; + + /* Preempt cancel of ELSP1 */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_NOOP); /* no preemption */ + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = spinner_create_request(&arg->b.spin, + arg->b.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->b.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + igt_spinner_end(&arg->a.spin); + if (i915_request_wait(rq[1], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != 0) { + pr_err("Normal inflight0 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != -EIO) { + pr_err("Cancelled inflight1 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int __cancel_queued(struct live_preempt_cancel *arg) +{ + struct i915_request *rq[3] = {}; + struct igt_live_test t; + int err; + + /* Full ELSP and one in the wings */ + GEM_TRACE("%s(%s)\n", __func__, arg->engine->name); + + if (igt_live_test_begin(&t, arg->engine->i915, + __func__, arg->engine->name)) + return -EIO; + + clear_bit(CONTEXT_BANNED, &arg->a.ctx->flags); + rq[0] = spinner_create_request(&arg->a.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[0])) + return PTR_ERR(rq[0]); + + i915_request_get(rq[0]); + i915_request_add(rq[0]); + if (!igt_wait_for_spinner(&arg->a.spin, rq[0])) { + err = -EIO; + goto out; + } + + clear_bit(CONTEXT_BANNED, &arg->b.ctx->flags); + rq[1] = igt_request_alloc(arg->b.ctx, arg->engine); + if (IS_ERR(rq[1])) { + err = PTR_ERR(rq[1]); + goto out; + } + + i915_request_get(rq[1]); + err = i915_request_await_dma_fence(rq[1], &rq[0]->fence); + i915_request_add(rq[1]); + if (err) + goto out; + + rq[2] = spinner_create_request(&arg->b.spin, + arg->a.ctx, arg->engine, + MI_ARB_CHECK); + if (IS_ERR(rq[2])) { + err = PTR_ERR(rq[2]); + goto out; + } + + i915_request_get(rq[2]); + err = i915_request_await_dma_fence(rq[2], &rq[1]->fence); + i915_request_add(rq[2]); + if (err) + goto out; + + i915_gem_context_set_banned(arg->a.ctx); + err = intel_engine_pulse(arg->engine); + if (err) + goto out; + + if (i915_request_wait(rq[2], 0, HZ / 5) < 0) { + err = -EIO; + goto out; + } + + if (rq[0]->fence.error != -EIO) { + pr_err("Cancelled inflight0 request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + + if (rq[1]->fence.error != 0) { + pr_err("Normal inflight1 request did not complete\n"); + err = -EINVAL; + goto out; + } + + if (rq[2]->fence.error != -EIO) { + pr_err("Cancelled queued request did not report -EIO\n"); + err = -EINVAL; + goto out; + } + +out: + i915_request_put(rq[2]); + i915_request_put(rq[1]); + i915_request_put(rq[0]); + if (igt_live_test_end(&t)) + err = -EIO; + return err; +} + +static int live_preempt_cancel(void *arg) +{ + struct drm_i915_private *i915 = arg; + struct live_preempt_cancel data; + enum intel_engine_id id; + int err = -ENOMEM; + + /* + * To cancel an inflight context, we need to first remove it from the + * GPU. That sounds like preemption! Plus a little bit of bookkeeping. + */ + + if (!HAS_LOGICAL_RING_PREEMPTION(i915)) + return 0; + + if (preempt_client_init(i915, &data.a)) + return -ENOMEM; + if (preempt_client_init(i915, &data.b)) + goto err_client_a; + + for_each_engine(data.engine, i915, id) { + if (!intel_engine_has_preemption(data.engine)) + continue; + + err = __cancel_active0(&data); + if (err) + goto err_wedged; + + err = __cancel_active1(&data); + if (err) + goto err_wedged; + + err = __cancel_queued(&data); + if (err) + goto err_wedged; + } + + err = 0; +err_client_b: + preempt_client_fini(&data.b); +err_client_a: + preempt_client_fini(&data.a); + return err; + +err_wedged: + GEM_TRACE_DUMP(); + igt_spinner_end(&data.b.spin); + igt_spinner_end(&data.a.spin); + intel_gt_set_wedged(&i915->gt); + goto err_client_b; +} + static int live_suppress_self_preempt(void *arg) { struct drm_i915_private *i915 = arg; @@ -2270,6 +2542,7 @@ int intel_execlists_live_selftests(struct drm_i915_private *i915) SUBTEST(live_preempt), SUBTEST(live_late_preempt), SUBTEST(live_nopreempt), + SUBTEST(live_preempt_cancel), SUBTEST(live_suppress_self_preempt), SUBTEST(live_suppress_wait_preempt), SUBTEST(live_chain_preempt),