public inbox for intel-gfx@lists.freedesktop.org
 help / color / mirror / Atom feed
From: Thomas Daniel <thomas.daniel@intel.com>
To: intel-gfx@lists.freedesktop.org
Subject: [PATCH 32/43] drm/i915/bdw: Avoid non-lite-restore preemptions
Date: Thu, 24 Jul 2014 17:04:40 +0100	[thread overview]
Message-ID: <1406217891-8912-33-git-send-email-thomas.daniel@intel.com> (raw)
In-Reply-To: <1406217891-8912-1-git-send-email-thomas.daniel@intel.com>

From: Oscar Mateo <oscar.mateo@intel.com>

In the current Execlists feeding mechanism, full preemption is not
supported yet: only lite-restores are allowed (this is: the GPU
simply samples a new tail pointer for the context currently in
execution).

But we have identified an scenario in which a full preemption occurs:
1) We submit two contexts for execution (A & B).
2) The GPU finishes with the first one (A), switches to the second one
(B) and informs us.
3) We submit B again (hoping to cause a lite restore) together with C,
but in the time we spend writing to the ELSP, the GPU finishes B.
4) The GPU start executing B again (since we told it so).
5) We receive a B finished interrupt and, mistakenly, we submit C (again)
and D, causing a full preemption of B.

The race is avoided by keeping track of how many times a context has been
submitted to the hardware and by better discriminating the received context
switch interrupts: in the example, when we have submitted B twice, we won´t
submit C and D as soon as we receive the notification that B is completed
because we were expecting to get a LITE_RESTORE and we didn´t, so we know a
second completion will be received shortly.

Without this explicit checking, somehow, the batch buffer execution order
gets messed with. This can be verified with the IGT test I sent together with
the series. I don´t know the exact mechanism by which the pre-emption messes
with the execution order but, since other people is working on the Scheduler
+ Preemption on Execlists, I didn´t try to fix it. In these series, only Lite
Restores are supported (other kind of preemptions WARN).

v2: elsp_submitted belongs in the new intel_ctx_submit_request. Several
rebase changes.

v3: Clarify how the race is avoided, as requested by Daniel.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
---
 drivers/gpu/drm/i915/intel_lrc.c |   28 ++++++++++++++++++++++++----
 drivers/gpu/drm/i915/intel_lrc.h |    2 ++
 2 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index 65f4f26..895dbfc 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -264,6 +264,7 @@ static void execlists_context_unqueue(struct intel_engine_cs *ring)
 		else if (req0->ctx == cursor->ctx) {
 			/* Same ctx: ignore first request, as second request
 			 * will update tail past first request's workload */
+			cursor->elsp_submitted = req0->elsp_submitted;
 			list_del(&req0->execlist_link);
 			queue_work(dev_priv->wq, &req0->work);
 			req0 = cursor;
@@ -273,8 +274,14 @@ static void execlists_context_unqueue(struct intel_engine_cs *ring)
 		}
 	}
 
+	WARN_ON(req1 && req1->elsp_submitted);
+
 	BUG_ON(execlists_submit_context(ring, req0->ctx, req0->tail,
 			req1? req1->ctx : NULL, req1? req1->tail : 0));
+
+	req0->elsp_submitted++;
+	if (req1)
+		req1->elsp_submitted++;
 }
 
 static bool execlists_check_remove_request(struct intel_engine_cs *ring,
@@ -291,9 +298,13 @@ static bool execlists_check_remove_request(struct intel_engine_cs *ring,
 		struct drm_i915_gem_object *ctx_obj =
 				head_req->ctx->engine[ring->id].state;
 		if (intel_execlists_ctx_id(ctx_obj) == request_id) {
-			list_del(&head_req->execlist_link);
-			queue_work(dev_priv->wq, &head_req->work);
-			return true;
+			WARN(head_req->elsp_submitted == 0,
+					"Never submitted head request\n");
+			if (--head_req->elsp_submitted <= 0) {
+				list_del(&head_req->execlist_link);
+				queue_work(dev_priv->wq, &head_req->work);
+				return true;
+			}
 		}
 	}
 
@@ -326,7 +337,16 @@ void intel_execlists_handle_ctx_events(struct intel_engine_cs *ring)
 		status_id = I915_READ(RING_CONTEXT_STATUS_BUF(ring) +
 				(read_pointer % 6) * 8 + 4);
 
-		if (status & GEN8_CTX_STATUS_COMPLETE) {
+		if (status & GEN8_CTX_STATUS_PREEMPTED) {
+			if (status & GEN8_CTX_STATUS_LITE_RESTORE) {
+				if (execlists_check_remove_request(ring, status_id))
+					WARN(1, "Lite Restored request removed from queue\n");
+			} else
+				WARN(1, "Preemption without Lite Restore\n");
+		}
+
+		 if ((status & GEN8_CTX_STATUS_ACTIVE_IDLE) ||
+		     (status & GEN8_CTX_STATUS_ELEMENT_SWITCH)) {
 			if (execlists_check_remove_request(ring, status_id))
 				submit_contexts++;
 		}
diff --git a/drivers/gpu/drm/i915/intel_lrc.h b/drivers/gpu/drm/i915/intel_lrc.h
index 2e8929f..074b44f 100644
--- a/drivers/gpu/drm/i915/intel_lrc.h
+++ b/drivers/gpu/drm/i915/intel_lrc.h
@@ -67,6 +67,8 @@ struct intel_ctx_submit_request {
 
 	struct list_head execlist_link;
 	struct work_struct work;
+
+	int elsp_submitted;
 };
 
 void intel_execlists_handle_ctx_events(struct intel_engine_cs *ring);
-- 
1.7.9.5

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/intel-gfx

  parent reply	other threads:[~2014-07-24 16:05 UTC|newest]

Thread overview: 137+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-07-24 16:04 [PATCH 00/43] Execlists v5 Thomas Daniel
2014-07-24 16:04 ` [PATCH 01/43] drm/i915: Reorder the actual workload submission so that args checking is done earlier Thomas Daniel
2014-07-25  8:30   ` Daniel Vetter
2014-07-25  9:16     ` Chris Wilson
2014-07-24 16:04 ` [PATCH 02/43] drm/i915/bdw: New source and header file for LRs, LRCs and Execlists Thomas Daniel
2014-07-24 16:04 ` [PATCH 03/43] drm/i915/bdw: Macro for LRCs and module option for Execlists Thomas Daniel
2014-08-11 13:57   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 04/43] drm/i915/bdw: Initialization for Logical Ring Contexts Thomas Daniel
2014-08-11 14:03   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 05/43] drm/i915/bdw: Introduce one context backing object per engine Thomas Daniel
2014-08-11 13:59   ` [PATCH] drm/i915: WARN if module opt sanitization goes out of order Daniel Vetter
2014-08-11 14:28     ` Damien Lespiau
2014-07-24 16:04 ` [PATCH 06/43] drm/i915/bdw: A bit more advanced LR context alloc/free Thomas Daniel
2014-07-24 16:04 ` [PATCH 07/43] drm/i915/bdw: Allocate ringbuffers for Logical Ring Contexts Thomas Daniel
2014-07-24 16:04 ` [PATCH 08/43] drm/i915/bdw: Add a context and an engine pointers to the ringbuffer Thomas Daniel
2014-08-11 14:14   ` Daniel Vetter
2014-08-11 14:20     ` Daniel Vetter
2014-08-13 13:34       ` Daniel, Thomas
2014-08-13 15:16         ` Daniel Vetter
2014-08-14 15:09           ` Daniel, Thomas
2014-08-14 15:32             ` Daniel Vetter
2014-08-14 15:37               ` Daniel Vetter
2014-08-14 15:56                 ` Daniel, Thomas
2014-08-14 16:19                   ` Daniel Vetter
2014-08-14 16:27                     ` [PATCH] drm/i915: Add temporary ring->ctx backpointer Daniel Vetter
2014-08-14 16:33                       ` Daniel, Thomas
2014-07-24 16:04 ` [PATCH 09/43] drm/i915/bdw: Populate LR contexts (somewhat) Thomas Daniel
2014-07-24 16:04 ` [PATCH 10/43] drm/i915/bdw: Deferred creation of user-created LRCs Thomas Daniel
2014-08-11 14:25   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 11/43] drm/i915/bdw: Render moot context reset and switch with Execlists Thomas Daniel
2014-08-11 14:30   ` Daniel Vetter
2014-08-15 10:22     ` Daniel, Thomas
2014-08-15 15:39       ` Daniel Vetter
2014-08-20 15:29   ` [PATCH] " Thomas Daniel
2014-08-20 15:36     ` Chris Wilson
2014-08-25 20:39       ` Daniel Vetter
2014-08-25 22:01         ` Scot Doyle
2014-08-26  5:59         ` Chris Wilson
2014-08-26 13:54           ` Siluvery, Arun
2014-08-26 14:11             ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 12/43] drm/i915/bdw: Don't write PDP in the legacy way when using LRCs Thomas Daniel
2014-08-01 13:46   ` Damien Lespiau
2014-08-07 12:17   ` Thomas Daniel
2014-08-08 15:59     ` Damien Lespiau
2014-08-11 14:32     ` Daniel Vetter
2014-08-15 11:01     ` [PATCH] " Thomas Daniel
2014-07-24 16:04 ` [PATCH 13/43] drm/i915: Abstract the legacy workload submission mechanism away Thomas Daniel
2014-08-11 14:36   ` Daniel Vetter
2014-08-11 14:39     ` Daniel Vetter
2014-08-11 14:39   ` Daniel Vetter
2014-08-11 15:02   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 14/43] drm/i915/bdw: Skeleton for the new logical rings submission path Thomas Daniel
2014-07-24 16:04 ` [PATCH 15/43] drm/i915/bdw: Generic logical ring init and cleanup Thomas Daniel
2014-08-11 15:01   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 16/43] drm/i915/bdw: GEN-specific logical ring init Thomas Daniel
2014-08-11 15:04   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 17/43] drm/i915/bdw: GEN-specific logical ring set/get seqno Thomas Daniel
2014-08-11 15:05   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 18/43] drm/i915/bdw: New logical ring submission mechanism Thomas Daniel
2014-08-11 20:40   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 19/43] drm/i915/bdw: GEN-specific logical ring emit request Thomas Daniel
2014-07-24 16:04 ` [PATCH 20/43] drm/i915/bdw: GEN-specific logical ring emit flush Thomas Daniel
2014-07-24 16:04 ` [PATCH 21/43] drm/i915/bdw: Emission of requests with logical rings Thomas Daniel
2014-08-11 20:56   ` Daniel Vetter
2014-08-13 13:34     ` Daniel, Thomas
2014-08-13 15:25       ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 22/43] drm/i915/bdw: Ring idle and stop " Thomas Daniel
2014-07-24 16:04 ` [PATCH 23/43] drm/i915/bdw: Interrupts " Thomas Daniel
2014-08-11 21:02   ` Daniel Vetter
2014-08-11 21:08   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 24/43] drm/i915/bdw: GEN-specific logical ring emit batchbuffer start Thomas Daniel
2014-08-11 21:09   ` Daniel Vetter
2014-08-11 21:12     ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 25/43] drm/i915/bdw: Workload submission mechanism for Execlists Thomas Daniel
2014-08-11 20:30   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 26/43] drm/i915/bdw: Always use MMIO flips with Execlists Thomas Daniel
2014-08-11 20:34   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 27/43] drm/i915/bdw: Render state init for Execlists Thomas Daniel
2014-08-11 21:25   ` Daniel Vetter
2014-08-13 15:07     ` Daniel, Thomas
2014-08-13 15:30       ` Daniel Vetter
2014-08-14 20:00         ` Daniel Vetter
2014-08-15  8:43           ` Daniel, Thomas
2014-08-20 15:55           ` Daniel, Thomas
2014-08-25 20:55             ` Daniel Vetter
2014-08-21 10:40   ` [PATCH] " Thomas Daniel
2014-08-28  9:40     ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 28/43] drm/i915/bdw: Implement context switching (somewhat) Thomas Daniel
2014-08-11 21:29   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 29/43] drm/i915/bdw: Write the tail pointer, LRC style Thomas Daniel
2014-08-01 14:33   ` Damien Lespiau
2014-08-11 21:30   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 30/43] drm/i915/bdw: Two-stage execlist submit process Thomas Daniel
2014-08-14 20:05   ` Daniel Vetter
2014-08-14 20:10   ` Daniel Vetter
2014-08-15  8:51     ` Daniel, Thomas
2014-08-15  9:38       ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 31/43] drm/i915/bdw: Handle context switch events Thomas Daniel
2014-08-14 20:13   ` Daniel Vetter
2014-08-14 20:17   ` Daniel Vetter
2014-08-14 20:28   ` Daniel Vetter
2014-08-14 20:37   ` Daniel Vetter
2014-07-24 16:04 ` Thomas Daniel [this message]
2014-08-14 20:31   ` [PATCH 32/43] drm/i915/bdw: Avoid non-lite-restore preemptions Daniel Vetter
2014-07-24 16:04 ` [PATCH 33/43] drm/i915/bdw: Help out the ctx switch interrupt handler Thomas Daniel
2014-08-14 20:43   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 34/43] drm/i915/bdw: Make sure gpu reset still works with Execlists Thomas Daniel
2014-08-01 14:42   ` Damien Lespiau
2014-08-06  9:26     ` Daniel, Thomas
2014-08-01 14:46   ` Damien Lespiau
2014-08-06  9:28     ` Daniel, Thomas
2014-07-24 16:04 ` [PATCH 35/43] drm/i915/bdw: Make sure error capture keeps working " Thomas Daniel
2014-08-15 12:14   ` Daniel Vetter
2014-08-21 10:57     ` Daniel, Thomas
2014-08-25 21:00       ` Daniel Vetter
2014-08-25 21:29       ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 36/43] drm/i915/bdw: Disable semaphores for Execlists Thomas Daniel
2014-07-24 16:04 ` [PATCH 37/43] drm/i915/bdw: Display execlists info in debugfs Thomas Daniel
2014-08-01 14:54   ` Damien Lespiau
2014-08-07 12:23   ` Thomas Daniel
2014-08-08 16:02     ` Damien Lespiau
2014-07-24 16:04 ` [PATCH 38/43] drm/i915/bdw: Display context backing obj & ringbuffer " Thomas Daniel
2014-07-24 16:04 ` [PATCH 39/43] drm/i915/bdw: Print context state " Thomas Daniel
2014-08-01 15:54   ` Damien Lespiau
2014-08-07 12:24   ` Thomas Daniel
2014-08-08 15:57     ` Damien Lespiau
2014-07-24 16:04 ` [PATCH 40/43] drm/i915/bdw: Document Logical Rings, LR contexts and Execlists Thomas Daniel
2014-08-15 12:42   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 41/43] drm/i915/bdw: Enable Logical Ring Contexts (hence, Execlists) Thomas Daniel
2014-08-18  8:33   ` Jani Nikula
2014-08-18 14:52     ` Daniel, Thomas
2014-07-24 16:04 ` [PATCH 42/43] drm/i915/bdw: Pin the context backing objects to GGTT on-demand Thomas Daniel
2014-08-15 13:03   ` Daniel Vetter
2014-07-24 16:04 ` [PATCH 43/43] drm/i915/bdw: Pin the ringbuffer backing object " Thomas Daniel
2014-07-25  8:35 ` [PATCH 00/43] Execlists v5 Daniel Vetter
2014-08-01 16:09 ` Damien Lespiau
2014-08-01 16:29   ` Jesse Barnes

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1406217891-8912-33-git-send-email-thomas.daniel@intel.com \
    --to=thomas.daniel@intel.com \
    --cc=intel-gfx@lists.freedesktop.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox