This is a note to let you know that I've just added the patch titled drm/i915/gt: Use the local HWSP offset during submission to the 5.9-stable tree which can be found at: http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary The filename of the patch is: drm-i915-gt-use-the-local-hwsp-offset-during-submission.patch and it can be found in the queue-5.9 subdirectory. If you, or anyone else, feels it should not be added to the stable tree, please let <stable@xxxxxxxxxxxxxxx> know about it. >From 8ce70996f759a37bac92e69ae0addd715227bfd1 Mon Sep 17 00:00:00 2001 From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Date: Thu, 22 Oct 2020 07:41:27 +0100 Subject: drm/i915/gt: Use the local HWSP offset during submission From: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> commit 8ce70996f759a37bac92e69ae0addd715227bfd1 upstream. We wrap the timeline on construction of the next request, but there may still be requests in flight that have not yet finalized the breadcrumb. (The breadcrumb is delayed as we need engine-local offsets, and for the virtual engine that is not known until execution.) As such, by the time we write to the timeline's HWSP offset it may have changed, and we should use the value we preserved in the request instead. Though the window is small and infrequent (at full flow we can expect a timeline's seqno to wrap once every 30 minutes), the impact of writing the old seqno into the new HWSP is severe: the old requests are never completed, and the new requests are completed before they are even submitted. Fixes: ebece7539242 ("drm/i915: Keep timeline HWSP allocated until idle across the system") Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx> Cc: Joonas Lahtinen <joonas.lahtinen@xxxxxxxxxxxxxxx> Cc: <stable@xxxxxxxxxxxxxxx> # v5.2+ Reviewed-by: Mika Kuoppala <mika.kuoppala@xxxxxxxxxxxxxxx> Link: https://patchwork.freedesktop.org/patch/msgid/20201022064127.10159-1-chris@xxxxxxxxxxxxxxxxxx (cherry picked from commit c10f6019d0b2dc8a6a62b55459f3ada5bc4e5e1a) Signed-off-by: Rodrigo Vivi <rodrigo.vivi@xxxxxxxxx> Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> --- drivers/gpu/drm/i915/gt/intel_lrc.c | 27 +++++++++++++++++-------- drivers/gpu/drm/i915/gt/intel_timeline.c | 18 +++++++++------- drivers/gpu/drm/i915/gt/intel_timeline_types.h | 2 + 3 files changed, 31 insertions(+), 16 deletions(-) --- a/drivers/gpu/drm/i915/gt/intel_lrc.c +++ b/drivers/gpu/drm/i915/gt/intel_lrc.c @@ -3539,6 +3539,19 @@ static const struct intel_context_ops ex .destroy = execlists_context_destroy, }; +static u32 hwsp_offset(const struct i915_request *rq) +{ + const struct intel_timeline_cacheline *cl; + + /* Before the request is executed, the timeline/cachline is fixed */ + + cl = rcu_dereference_protected(rq->hwsp_cacheline, 1); + if (cl) + return cl->ggtt_offset; + + return rcu_dereference_protected(rq->timeline, 1)->hwsp_offset; +} + static int gen8_emit_init_breadcrumb(struct i915_request *rq) { u32 *cs; @@ -3561,7 +3574,7 @@ static int gen8_emit_init_breadcrumb(str *cs++ = MI_NOOP; *cs++ = MI_STORE_DWORD_IMM_GEN4 | MI_USE_GGTT; - *cs++ = i915_request_timeline(rq)->hwsp_offset; + *cs++ = hwsp_offset(rq); *cs++ = 0; *cs++ = rq->fence.seqno - 1; @@ -4865,11 +4878,9 @@ gen8_emit_fini_breadcrumb_tail(struct i9 return gen8_emit_wa_tail(request, cs); } -static u32 *emit_xcs_breadcrumb(struct i915_request *request, u32 *cs) +static u32 *emit_xcs_breadcrumb(struct i915_request *rq, u32 *cs) { - u32 addr = i915_request_active_timeline(request)->hwsp_offset; - - return gen8_emit_ggtt_write(cs, request->fence.seqno, addr, 0); + return gen8_emit_ggtt_write(cs, rq->fence.seqno, hwsp_offset(rq), 0); } static u32 *gen8_emit_fini_breadcrumb(struct i915_request *rq, u32 *cs) @@ -4888,7 +4899,7 @@ static u32 *gen8_emit_fini_breadcrumb_rc /* XXX flush+write+CS_STALL all in one upsets gem_concurrent_blt:kbl */ cs = gen8_emit_ggtt_write_rcs(cs, request->fence.seqno, - i915_request_active_timeline(request)->hwsp_offset, + hwsp_offset(request), PIPE_CONTROL_FLUSH_ENABLE | PIPE_CONTROL_CS_STALL); @@ -4900,7 +4911,7 @@ gen11_emit_fini_breadcrumb_rcs(struct i9 { cs = gen8_emit_ggtt_write_rcs(cs, request->fence.seqno, - i915_request_active_timeline(request)->hwsp_offset, + hwsp_offset(request), PIPE_CONTROL_CS_STALL | PIPE_CONTROL_TILE_CACHE_FLUSH | PIPE_CONTROL_RENDER_TARGET_CACHE_FLUSH | @@ -4970,7 +4981,7 @@ gen12_emit_fini_breadcrumb_rcs(struct i9 { cs = gen12_emit_ggtt_write_rcs(cs, request->fence.seqno, - i915_request_active_timeline(request)->hwsp_offset, + hwsp_offset(request), PIPE_CONTROL0_HDC_PIPELINE_FLUSH, PIPE_CONTROL_CS_STALL | PIPE_CONTROL_TILE_CACHE_FLUSH | --- a/drivers/gpu/drm/i915/gt/intel_timeline.c +++ b/drivers/gpu/drm/i915/gt/intel_timeline.c @@ -188,10 +188,14 @@ cacheline_alloc(struct intel_timeline_hw return cl; } -static void cacheline_acquire(struct intel_timeline_cacheline *cl) +static void cacheline_acquire(struct intel_timeline_cacheline *cl, + u32 ggtt_offset) { - if (cl) - i915_active_acquire(&cl->active); + if (!cl) + return; + + cl->ggtt_offset = ggtt_offset; + i915_active_acquire(&cl->active); } static void cacheline_release(struct intel_timeline_cacheline *cl) @@ -332,7 +336,7 @@ int intel_timeline_pin(struct intel_time GT_TRACE(tl->gt, "timeline:%llx using HWSP offset:%x\n", tl->fence_context, tl->hwsp_offset); - cacheline_acquire(tl->hwsp_cacheline); + cacheline_acquire(tl->hwsp_cacheline, tl->hwsp_offset); if (atomic_fetch_inc(&tl->pin_count)) { cacheline_release(tl->hwsp_cacheline); __i915_vma_unpin(tl->hwsp_ggtt); @@ -505,7 +509,7 @@ __intel_timeline_get_seqno(struct intel_ GT_TRACE(tl->gt, "timeline:%llx using HWSP offset:%x\n", tl->fence_context, tl->hwsp_offset); - cacheline_acquire(cl); + cacheline_acquire(cl, tl->hwsp_offset); tl->hwsp_cacheline = cl; *seqno = timeline_advance(tl); @@ -563,9 +567,7 @@ int intel_timeline_read_hwsp(struct i915 if (err) goto out; - *hwsp = i915_ggtt_offset(cl->hwsp->vma) + - ptr_unmask_bits(cl->vaddr, CACHELINE_BITS) * CACHELINE_BYTES; - + *hwsp = cl->ggtt_offset; out: i915_active_release(&cl->active); return err; --- a/drivers/gpu/drm/i915/gt/intel_timeline_types.h +++ b/drivers/gpu/drm/i915/gt/intel_timeline_types.h @@ -94,6 +94,8 @@ struct intel_timeline_cacheline { struct intel_timeline_hwsp *hwsp; void *vaddr; + u32 ggtt_offset; + struct rcu_head rcu; }; Patches currently in stable-queue which might be from chris@xxxxxxxxxxxxxxxxxx are queue-5.9/drm-i915-cancel-outstanding-work-after-disabling-heartbeats-on-an-engine.patch queue-5.9/drm-i915-gem-avoid-implicit-vmap-for-highmem-on-x86-32.patch queue-5.9/drm-i915-gt-use-the-local-hwsp-offset-during-submission.patch queue-5.9/drm-i915-gt-delay-execlist-processing-for-tgl.patch queue-5.9/drm-i915-break-up-error-capture-compression-loops-with-cond_resched.patch queue-5.9/drm-i915-use-the-active-reference-on-the-vma-while-capturing.patch queue-5.9/drm-i915-avoid-mixing-integer-types-during-batch-copies.patch queue-5.9/drm-i915-drop-runtime-pm-assert-from-vgpu-io-accessors.patch queue-5.9/drm-i915-exclude-low-pages-128kib-of-stolen-from-use.patch queue-5.9/drm-i915-gt-initialize-reserved-and-unspecified-mocs-indices.patch queue-5.9/drm-i915-restore-ilk-m-rps-support.patch queue-5.9/drm-i915-gt-undo-forced-context-restores-after-trivial-preemptions.patch queue-5.9/drm-i915-reject-90-270-degree-rotated-initial-fbs.patch queue-5.9/drm-i915-mark-ininitial-fb-obj-as-wt-on-ellc-machines-to-avoid-rcu-lockup-during-fbdev-init.patch queue-5.9/drm-i915-gem-prevent-using-pgprot_writecombine-if-pat-is-not-supported.patch queue-5.9/drm-i915-gem-always-test-execution-status-on-closing-the-context.patch queue-5.9/drm-i915-gt-always-send-a-pulse-down-the-engine-after-disabling-heartbeat.patch