From patchwork Fri Dec 2 12:28:44 2022 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Matthew Auld X-Patchwork-Id: 13062708 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org 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 smtp.lore.kernel.org (Postfix) with ESMTPS id 08E22C4321E for ; Fri, 2 Dec 2022 12:30:36 +0000 (UTC) Received: from gabe.freedesktop.org (localhost [127.0.0.1]) by gabe.freedesktop.org (Postfix) with ESMTP id 96E7E10E18B; Fri, 2 Dec 2022 12:30:35 +0000 (UTC) Received: from mga05.intel.com (mga05.intel.com [192.55.52.43]) by gabe.freedesktop.org (Postfix) with ESMTPS id 7018E10E188 for ; Fri, 2 Dec 2022 12:30:29 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1669984229; x=1701520229; h=from:to:cc:subject:date:message-id:in-reply-to: references:mime-version:content-transfer-encoding; bh=C3TneCwm7mR+l93sj8QjCTjqXXgO2AsvmUePhfNUGX8=; b=gmxtj9e0o9BetuVH57A1KwcZPBM4E+h1ryKiHzZ0wi//X6ZqKFlfoBy+ sBNFTahESQLFxXNa89EJE+QQTV2hjLkKknorr1o5UWBbH2Zlp7ZJiWc4d S0u9dPKTG7L56keatOaLLBSU0xr6gW+mxyVjrhuE4hcLml1QpbexztE4o 7eN735XeJfJ/+VASpOOnXlWfhfrdxhheB5j23fLdQtA82a+9KvGfKoTjs VaY3ajSSVFwzwvxhocTjHixSL8c81WIzkxfyjxLoru96bhN7WusXgxDUj f0S9q4BzF4de9vKp0AkQXDRlOkWGagVYDS3jga+9RXFdK9t1MF9j8k1+t g==; X-IronPort-AV: E=McAfee;i="6500,9779,10548"; a="402223818" X-IronPort-AV: E=Sophos;i="5.96,212,1665471600"; d="scan'208";a="402223818" Received: from fmsmga003.fm.intel.com ([10.253.24.29]) by fmsmga105.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2022 04:30:29 -0800 X-IronPort-AV: E=McAfee;i="6500,9779,10548"; a="733797925" X-IronPort-AV: E=Sophos;i="5.96,212,1665471600"; d="scan'208";a="733797925" Received: from ctfarrel-mobl.ger.corp.intel.com (HELO mwauld-desk1.intel.com) ([10.252.18.53]) by fmsmga003-auth.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 02 Dec 2022 04:30:27 -0800 From: Matthew Auld To: intel-gfx@lists.freedesktop.org Date: Fri, 2 Dec 2022 12:28:44 +0000 Message-Id: <20221202122844.428006-3-matthew.auld@intel.com> X-Mailer: git-send-email 2.38.1 In-Reply-To: <20221202122844.428006-1-matthew.auld@intel.com> References: <20221202122844.428006-1-matthew.auld@intel.com> MIME-Version: 1.0 Subject: [Intel-gfx] [PATCH v6 3/3] drm/i915/selftests: exercise emit_pte() with nearly full ring X-BeenThere: intel-gfx@lists.freedesktop.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: Intel graphics driver community testing & development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Cc: Andrzej Hajda , Chris Wilson , Nirmoy Das Errors-To: intel-gfx-bounces@lists.freedesktop.org Sender: "Intel-gfx" Simple regression test to check that we don't trample the rq->reserved_space when returning from emit_pte(), if the ring is nearly full. v2: Make spinner_kill() static v3: Reduce the ring size further, which should mean we need to execute less noops; hopefully this appeases bsw. Also add some debug logging. v4: Fix the min request construction to account for reserved_space + I915_EMIT_PTE_NUM_DWORDS v5: Use a simple on-stack timer to kill the spinner instead of kthread (Chris) References: https://gitlab.freedesktop.org/drm/intel/-/issues/7535 References: https://gitlab.freedesktop.org/drm/intel/-/issues/6889 Signed-off-by: Matthew Auld Cc: Chris Wilson Cc: Andi Shyti Cc: Andrzej Hajda Cc: Nirmoy Das Reviewed-by: Andi Shyti --- drivers/gpu/drm/i915/gt/intel_migrate.c | 6 +- drivers/gpu/drm/i915/gt/selftest_migrate.c | 145 +++++++++++++++++++++ 2 files changed, 149 insertions(+), 2 deletions(-) diff --git a/drivers/gpu/drm/i915/gt/intel_migrate.c b/drivers/gpu/drm/i915/gt/intel_migrate.c index b783f6f740c8..e08a739b7091 100644 --- a/drivers/gpu/drm/i915/gt/intel_migrate.c +++ b/drivers/gpu/drm/i915/gt/intel_migrate.c @@ -352,6 +352,8 @@ static int max_pte_pkt_size(struct i915_request *rq, int pkt) return pkt; } +#define I915_EMIT_PTE_NUM_DWORDS 6 + static int emit_pte(struct i915_request *rq, struct sgt_dma *it, enum i915_cache_level cache_level, @@ -393,7 +395,7 @@ static int emit_pte(struct i915_request *rq, offset += (u64)rq->engine->instance << 32; - cs = intel_ring_begin(rq, 6); + cs = intel_ring_begin(rq, I915_EMIT_PTE_NUM_DWORDS); if (IS_ERR(cs)) return PTR_ERR(cs); @@ -416,7 +418,7 @@ static int emit_pte(struct i915_request *rq, intel_ring_advance(rq, cs); intel_ring_update_space(ring); - cs = intel_ring_begin(rq, 6); + cs = intel_ring_begin(rq, I915_EMIT_PTE_NUM_DWORDS); if (IS_ERR(cs)) return PTR_ERR(cs); diff --git a/drivers/gpu/drm/i915/gt/selftest_migrate.c b/drivers/gpu/drm/i915/gt/selftest_migrate.c index 1eab025ac002..e677f2da093d 100644 --- a/drivers/gpu/drm/i915/gt/selftest_migrate.c +++ b/drivers/gpu/drm/i915/gt/selftest_migrate.c @@ -8,6 +8,7 @@ #include "gem/i915_gem_internal.h" #include "gem/i915_gem_lmem.h" +#include "selftests/igt_spinner.h" #include "selftests/i915_random.h" static const unsigned int sizes[] = { @@ -529,6 +530,149 @@ static int live_migrate_clear(void *arg) return 0; } +struct spinner_timer { + struct timer_list timer; + struct igt_spinner spin; +}; + +static void spinner_kill(struct timer_list *timer) +{ + struct spinner_timer *st = from_timer(st, timer, timer); + + igt_spinner_end(&st->spin); + pr_info("%s\n", __func__); +} + +static int live_emit_pte_full_ring(void *arg) +{ + struct intel_gt *gt = arg; + struct intel_migrate *migrate = >->migrate; + struct drm_i915_private *i915 = migrate->context->engine->i915; + struct drm_i915_gem_object *obj; + struct intel_context *ce; + struct i915_request *rq, *prev; + struct spinner_timer st; + struct sgt_dma it; + int len, sz, err; + u32 *cs; + + /* + * Simple regression test to check that we don't trample the + * rq->reserved_space when returning from emit_pte(), if the ring is + * nearly full. + */ + + if (igt_spinner_init(&st.spin, to_gt(i915))) + return -ENOMEM; + + obj = i915_gem_object_create_internal(i915, 2 * PAGE_SIZE); + if (IS_ERR(obj)) { + err = PTR_ERR(obj); + goto out_spinner; + } + + err = i915_gem_object_pin_pages_unlocked(obj); + if (err) + goto out_obj; + + ce = intel_migrate_create_context(migrate); + if (IS_ERR(ce)) { + err = PTR_ERR(ce); + goto out_obj; + } + + ce->ring_size = SZ_4K; /* Not too big */ + + err = intel_context_pin(ce); + if (err) + goto out_put; + + rq = igt_spinner_create_request(&st.spin, ce, MI_ARB_CHECK); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto out_unpin; + } + + i915_request_add(rq); + if (!igt_wait_for_spinner(&st.spin, rq)) { + err = -EIO; + goto out_unpin; + } + + /* + * Fill the rest of the ring leaving I915_EMIT_PTE_NUM_DWORDS + + * ring->reserved_space at the end. To actually emit the PTEs we require + * slightly more than I915_EMIT_PTE_NUM_DWORDS, since our object size is + * greater than PAGE_SIZE. The correct behaviour is to wait for more + * ring space in emit_pte(), otherwise we trample on the reserved_space + * resulting in crashes when later submitting the rq. + */ + + prev = NULL; + do { + if (prev) + i915_request_add(rq); + + rq = i915_request_create(ce); + if (IS_ERR(rq)) { + err = PTR_ERR(rq); + goto out_unpin; + } + + sz = (rq->ring->space - rq->reserved_space) / sizeof(u32) - + I915_EMIT_PTE_NUM_DWORDS; + sz = min_t(u32, sz, (SZ_1K - rq->reserved_space) / sizeof(u32) - + I915_EMIT_PTE_NUM_DWORDS); + cs = intel_ring_begin(rq, sz); + if (IS_ERR(cs)) { + err = PTR_ERR(cs); + goto out_rq; + } + + memset32(cs, MI_NOOP, sz); + cs += sz; + intel_ring_advance(rq, cs); + + pr_info("%s emit=%u sz=%d\n", __func__, rq->ring->emit, sz); + + prev = rq; + } while (rq->ring->space > (rq->reserved_space + + I915_EMIT_PTE_NUM_DWORDS * sizeof(u32))); + + timer_setup_on_stack(&st.timer, spinner_kill, 0); + mod_timer(&st.timer, jiffies + 2 * HZ); + + /* + * This should wait for the spinner to be killed, otherwise we should go + * down in flames when doing i915_request_add(). + */ + pr_info("%s emite_pte ring space=%u\n", __func__, rq->ring->space); + it = sg_sgt(obj->mm.pages->sgl); + len = emit_pte(rq, &it, obj->cache_level, false, 0, CHUNK_SZ); + if (!len) { + err = -EINVAL; + goto out_rq; + } + if (len < 0) { + err = len; + goto out_rq; + } + +out_rq: + i915_request_add(rq); /* GEM_BUG_ON(rq->reserved_space > ring->space)? */ + del_timer_sync(&st.timer); + destroy_timer_on_stack(&st.timer); +out_unpin: + intel_context_unpin(ce); +out_put: + intel_context_put(ce); +out_obj: + i915_gem_object_put(obj); +out_spinner: + igt_spinner_fini(&st.spin); + return err; +} + struct threaded_migrate { struct intel_migrate *migrate; struct task_struct *tsk; @@ -651,6 +795,7 @@ int intel_migrate_live_selftests(struct drm_i915_private *i915) static const struct i915_subtest tests[] = { SUBTEST(live_migrate_copy), SUBTEST(live_migrate_clear), + SUBTEST(live_emit_pte_full_ring), SUBTEST(thread_migrate_copy), SUBTEST(thread_migrate_clear), SUBTEST(thread_global_copy),