All of lore.kernel.org
 help / color / mirror / Atom feed
From: John.C.Harrison@Intel.com
To: Intel-GFX@Lists.FreeDesktop.Org
Subject: [PATCH v4 09/38] drm/i915: Disable hardware semaphores when GPU scheduler is enabled
Date: Mon, 11 Jan 2016 18:42:38 +0000	[thread overview]
Message-ID: <1452537788-32136-10-git-send-email-John.C.Harrison@Intel.com> (raw)
In-Reply-To: <1452537788-32136-1-git-send-email-John.C.Harrison@Intel.com>

From: John Harrison <John.C.Harrison@Intel.com>

Hardware sempahores require seqno values to be continuously
incrementing. However, the scheduler's reordering of batch buffers
means that the seqno values going through the hardware could be out of
order. Thus semaphores can not be used.

On the other hand, the scheduler superceeds the need for hardware
semaphores anyway. Having one ring stall waiting for something to
complete on another ring is inefficient if that ring could be working
on some other, independent task. This is what the scheduler is meant
to do - keep the hardware as busy as possible by reordering batch
buffers to avoid dependency stalls.

v4: Downgraded a BUG_ON to WARN_ON as the latter is preferred.

For: VIZ-1587
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
---
 drivers/gpu/drm/i915/i915_drv.c         | 9 +++++++++
 drivers/gpu/drm/i915/i915_scheduler.c   | 7 +++++++
 drivers/gpu/drm/i915/i915_scheduler.h   | 1 +
 drivers/gpu/drm/i915/intel_ringbuffer.c | 4 ++++
 4 files changed, 21 insertions(+)

diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c
index 858d58c..e6be8f5 100644
--- a/drivers/gpu/drm/i915/i915_drv.c
+++ b/drivers/gpu/drm/i915/i915_drv.c
@@ -34,6 +34,7 @@
 #include "i915_drv.h"
 #include "i915_trace.h"
 #include "intel_drv.h"
+#include "i915_scheduler.h"
 
 #include <linux/console.h>
 #include <linux/module.h>
@@ -581,6 +582,14 @@ void intel_detect_pch(struct drm_device *dev)
 
 bool i915_semaphore_is_enabled(struct drm_device *dev)
 {
+	/* Hardware semaphores are not compatible with the scheduler due to the
+	 * seqno values being potentially out of order. However, semaphores are
+	 * also not required as the scheduler will handle interring dependencies
+	 * and try do so in a way that does not cause dead time on the hardware.
+	 */
+	if (i915_scheduler_is_enabled(dev))
+		return false;
+
 	if (INTEL_INFO(dev)->gen < 6)
 		return false;
 
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 8cb9063..fc48955 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -38,6 +38,13 @@ static int         i915_scheduler_priority_bump(struct i915_scheduler *scheduler
 						struct i915_scheduler_queue_entry *target,
 						uint32_t bump);
 
+bool i915_scheduler_is_enabled(struct drm_device *dev)
+{
+	struct drm_i915_private *dev_priv = dev->dev_private;
+
+	return dev_priv->scheduler != NULL;
+}
+
 int i915_scheduler_init(struct drm_device *dev)
 {
 	struct drm_i915_private *dev_priv = dev->dev_private;
diff --git a/drivers/gpu/drm/i915/i915_scheduler.h b/drivers/gpu/drm/i915/i915_scheduler.h
index 00dc7f3..2d50d83 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.h
+++ b/drivers/gpu/drm/i915/i915_scheduler.h
@@ -84,6 +84,7 @@ enum {
 	i915_sf_submitting          = (1 << 1),
 };
 
+bool        i915_scheduler_is_enabled(struct drm_device *dev);
 int         i915_scheduler_init(struct drm_device *dev);
 int         i915_scheduler_queue_execbuffer(struct i915_scheduler_queue_entry *qe);
 bool        i915_scheduler_notify_request(struct drm_i915_gem_request *req);
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 1dec252..a93bbce 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -32,6 +32,7 @@
 #include <drm/i915_drm.h>
 #include "i915_trace.h"
 #include "intel_drv.h"
+#include "i915_scheduler.h"
 
 bool
 intel_ring_initialized(struct intel_engine_cs *ring)
@@ -1411,6 +1412,9 @@ gen6_ring_sync(struct drm_i915_gem_request *waiter_req,
 	u32 wait_mbox = signaller->semaphore.mbox.wait[waiter->id];
 	int ret;
 
+	/* Arithmetic on sequence numbers is unreliable with a scheduler. */
+	WARN_ON(i915_scheduler_is_enabled(signaller->dev));
+
 	/* Throughout all of the GEM code, seqno passed implies our current
 	 * seqno is >= the last seqno executed. However for hardware the
 	 * comparison is strictly greater than.
-- 
1.9.1

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

  parent reply	other threads:[~2016-01-11 18:43 UTC|newest]

Thread overview: 143+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-11-23 11:38 [PATCH 00/39] GPU scheduler for i915 driver John.C.Harrison
2015-11-23 11:38 ` [PATCH 01/39] drm/i915: Add total count to context status debugfs output John.C.Harrison
2016-01-08  9:50   ` Joonas Lahtinen
2015-11-23 11:38 ` [PATCH 02/39] drm/i915: Updating assorted register and status page definitions John.C.Harrison
2016-01-08 12:26   ` Joonas Lahtinen
2016-01-11  7:47     ` Daniel Vetter
2015-11-23 11:38 ` [PATCH 03/39] drm/i915: Explicit power enable during deferred context initialisation John.C.Harrison
2016-01-08 12:35   ` Joonas Lahtinen
2015-11-23 11:38 ` [PATCH 04/39] drm/i915: Prelude to splitting i915_gem_do_execbuffer in two John.C.Harrison
2015-11-23 11:39 ` [PATCH 05/39] drm/i915: Split i915_dem_do_execbuffer() in half John.C.Harrison
2015-12-11 13:15   ` [PATCH 05/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 06/39] drm/i915: Re-instate request->uniq because it is extremely useful John.C.Harrison
2015-11-23 11:39 ` [PATCH 07/39] drm/i915: Start of GPU scheduler John.C.Harrison
2015-12-11 13:16   ` [PATCH 08/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 08/39] drm/i915: Prepare retire_requests to handle out-of-order seqnos John.C.Harrison
2015-11-23 11:39 ` [PATCH 09/39] drm/i915: Disable hardware semaphores when GPU scheduler is enabled John.C.Harrison
2015-11-23 11:39 ` [PATCH 10/39] drm/i915: Force MMIO flips when scheduler enabled John.C.Harrison
2015-11-23 11:39 ` [PATCH 11/39] drm/i915: Added scheduler hook when closing DRM file handles John.C.Harrison
2015-12-11 13:19   ` [PATCH 12/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 12/39] drm/i915: Added scheduler hook into i915_gem_request_notify() John.C.Harrison
2015-11-23 11:39 ` [PATCH 13/39] drm/i915: Added deferred work handler for scheduler John.C.Harrison
2015-11-23 11:39 ` [PATCH 14/39] drm/i915: Redirect execbuffer_final() via scheduler John.C.Harrison
2015-11-23 11:39 ` [PATCH 15/39] drm/i915: Keep the reserved space mechanism happy John.C.Harrison
2015-12-11 13:19   ` [PATCH 16/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 16/39] drm/i915: Added tracking/locking of batch buffer objects John.C.Harrison
2015-12-11 13:19   ` [PATCH 17/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 17/39] drm/i915: Hook scheduler node clean up into retire requests John.C.Harrison
2015-12-11 13:19   ` [PATCH 18/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 18/39] drm/i915: Added scheduler support to __wait_request() calls John.C.Harrison
2015-12-11 13:20   ` [PATCH 19/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 19/39] drm/i915: Added scheduler support to page fault handler John.C.Harrison
2015-11-23 11:39 ` [PATCH 20/39] drm/i915: Added scheduler flush calls to ring throttle and idle functions John.C.Harrison
2015-12-11 13:20   ` [PATCH 21/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 21/39] drm/i915: Added a module parameter for allowing scheduler overrides John.C.Harrison
2015-11-23 11:39 ` [PATCH 22/39] drm/i915: Support for 'unflushed' ring idle John.C.Harrison
2015-11-23 11:39 ` [PATCH 23/39] drm/i915: Defer seqno allocation until actual hardware submission time John.C.Harrison
2015-12-11 13:20   ` [PATCH 24/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 24/39] drm/i915: Added immediate submission override to scheduler John.C.Harrison
2015-11-23 11:39 ` [PATCH 25/39] drm/i915: Add sync wait support " John.C.Harrison
2015-11-23 11:39 ` [PATCH 26/39] drm/i915: Connecting execbuff fences " John.C.Harrison
2015-11-23 11:39 ` [PATCH 27/39] drm/i915: Added trace points " John.C.Harrison
2015-12-11 13:20   ` [PATCH 28/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 28/39] drm/i915: Added scheduler queue throttling by DRM file handle John.C.Harrison
2015-12-11 13:21   ` [PATCH 29/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 29/39] drm/i915: Added debugfs interface to scheduler tuning parameters John.C.Harrison
2015-11-23 11:39 ` [PATCH 30/39] drm/i915: Added debug state dump facilities to scheduler John.C.Harrison
2015-12-11 13:21   ` [PATCH 31/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 31/39] drm/i915: Add early exit to execbuff_final() if insufficient ring space John.C.Harrison
2015-12-11 13:21   ` [PATCH 32/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 32/39] drm/i915: Added scheduler statistic reporting to debugfs John.C.Harrison
2015-12-11 13:21   ` [PATCH 33/40] " John.C.Harrison
2015-11-23 11:39 ` [PATCH 33/39] drm/i915: Added seqno values to scheduler status dump John.C.Harrison
2015-11-23 11:39 ` [PATCH 34/39] drm/i915: Add scheduler support functions for TDR John.C.Harrison
2015-11-23 11:39 ` [PATCH 35/39] drm/i915: GPU priority bumping to prevent starvation John.C.Harrison
2015-11-23 11:39 ` [PATCH 36/39] drm/i915: Scheduler state dump via debugfs John.C.Harrison
2015-11-23 11:39 ` [PATCH 37/39] drm/i915: Enable GPU scheduler by default John.C.Harrison
2015-11-23 11:39 ` [PATCH 38/39] drm/i915: Add scheduling priority to per-context parameters John.C.Harrison
2015-11-23 11:39 ` [PATCH 39/39] drm/i915: Allow scheduler to manage inter-ring object synchronisation John.C.Harrison
2015-12-11 13:16 ` [PATCH 06/40] drm/i915: Cache request pointer in *_submission_final() John.C.Harrison
2015-12-11 13:23 ` [PATCH 00/40] GPU scheduler for i915 driver John.C.Harrison
2016-01-11 18:42 ` [PATCH v4 00/38] " John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 01/38] drm/i915: Add total count to context status debugfs output John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 02/38] drm/i915: Explicit power enable during deferred context initialisation John.C.Harrison
2016-01-12  0:20     ` Chris Wilson
2016-01-12 11:11       ` John Harrison
2016-01-12 11:28         ` Chris Wilson
2016-01-12 11:50           ` John Harrison
2016-01-12 14:04             ` Daniel Vetter
2016-01-12 14:21               ` John Harrison
2016-01-12 15:35                 ` Daniel Vetter
2016-01-12 15:59                   ` Imre Deak
2016-01-12 16:11                     ` Daniel Vetter
2016-01-12 16:59                       ` Chris Wilson
2016-01-11 18:42   ` [PATCH v4 03/38] drm/i915: Prelude to splitting i915_gem_do_execbuffer in two John.C.Harrison
2016-02-04 17:01     ` Jesse Barnes
2016-02-12 16:18       ` John Harrison
2016-01-11 18:42   ` [PATCH v4 04/38] drm/i915: Split i915_dem_do_execbuffer() in half John.C.Harrison
2016-01-11 22:03     ` Chris Wilson
2016-02-04 17:08     ` Jesse Barnes
2016-01-11 18:42   ` [PATCH v4 05/38] drm/i915: Cache request pointer in *_submission_final() John.C.Harrison
2016-02-04 17:09     ` Jesse Barnes
2016-01-11 18:42   ` [PATCH v4 06/38] drm/i915: Re-instate request->uniq because it is extremely useful John.C.Harrison
2016-01-11 22:04     ` Chris Wilson
2016-01-12 11:16       ` John Harrison
2016-01-11 18:42   ` [PATCH v4 07/38] drm/i915: Start of GPU scheduler John.C.Harrison
2016-01-20 13:18     ` Joonas Lahtinen
2016-02-18 14:22       ` John Harrison
2016-02-19 10:13         ` Joonas Lahtinen
2016-01-11 18:42   ` [PATCH v4 08/38] drm/i915: Prepare retire_requests to handle out-of-order seqnos John.C.Harrison
2016-01-11 22:10     ` Chris Wilson
2016-02-04 17:14       ` Jesse Barnes
2016-01-11 18:42   ` John.C.Harrison [this message]
2016-01-11 18:42   ` [PATCH v4 10/38] drm/i915: Force MMIO flips when scheduler enabled John.C.Harrison
2016-01-11 22:16     ` Chris Wilson
2016-01-12 11:19       ` John Harrison
2016-01-12 14:07         ` Daniel Vetter
2016-01-12 21:53           ` Chris Wilson
2016-01-13 12:37             ` John Harrison
2016-01-13 13:14               ` Chris Wilson
2016-01-11 18:42   ` [PATCH v4 11/38] drm/i915: Added scheduler hook when closing DRM file handles John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 12/38] drm/i915: Added scheduler hook into i915_gem_request_notify() John.C.Harrison
2016-01-11 22:14     ` Chris Wilson
2016-01-12 11:25       ` John Harrison
2016-01-11 18:42   ` [PATCH v4 13/38] drm/i915: Added deferred work handler for scheduler John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 14/38] drm/i915: Redirect execbuffer_final() via scheduler John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 15/38] drm/i915: Keep the reserved space mechanism happy John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 16/38] drm/i915: Added tracking/locking of batch buffer objects John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 17/38] drm/i915: Hook scheduler node clean up into retire requests John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 18/38] drm/i915: Added scheduler support to __wait_request() calls John.C.Harrison
2016-01-11 23:14     ` Chris Wilson
2016-01-12 11:28       ` John Harrison
2016-01-11 18:42   ` [PATCH v4 19/38] drm/i915: Added scheduler support to page fault handler John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 20/38] drm/i915: Added scheduler flush calls to ring throttle and idle functions John.C.Harrison
2016-01-11 22:20     ` Chris Wilson
2016-01-11 18:42   ` [PATCH v4 21/38] drm/i915: Added a module parameter for allowing scheduler overrides John.C.Harrison
2016-01-11 22:24     ` Chris Wilson
2016-01-12 11:34       ` John Harrison
2016-01-12 11:55         ` Chris Wilson
2016-01-11 18:42   ` [PATCH v4 22/38] drm/i915: Support for 'unflushed' ring idle John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 23/38] drm/i915: Defer seqno allocation until actual hardware submission time John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 24/38] drm/i915: Added immediate submission override to scheduler John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 25/38] drm/i915: Added trace points " John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 26/38] drm/i915: Added scheduler queue throttling by DRM file handle John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 27/38] drm/i915: Added debugfs interface to scheduler tuning parameters John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 28/38] drm/i915: Added debug state dump facilities to scheduler John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 29/38] drm/i915: Add early exit to execbuff_final() if insufficient ring space John.C.Harrison
2016-01-11 18:42   ` [PATCH v4 30/38] drm/i915: Added scheduler statistic reporting to debugfs John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 31/38] drm/i915: Added seqno values to scheduler status dump John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 32/38] drm/i915: Add scheduler support functions for TDR John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 33/38] drm/i915: GPU priority bumping to prevent starvation John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 34/38] drm/i915: Scheduler state dump via debugfs John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 35/38] drm/i915: Enable GPU scheduler by default John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 36/38] drm/i915: Add scheduling priority to per-context parameters John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 37/38] drm/i915: Add support for retro-actively banning batch buffers John.C.Harrison
2016-01-11 18:43   ` [PATCH v4 38/38] drm/i915: Allow scheduler to manage inter-ring object synchronisation John.C.Harrison
2016-01-11 22:07     ` Chris Wilson
2016-01-12 11:38       ` John Harrison
2016-01-11 18:43   ` [PATCH] igt/gem_ctx_param_basic: Updated to support scheduler priority interface John.C.Harrison
2016-01-11 23:52   ` [PATCH v4 00/38] GPU scheduler for i915 driver Chris Wilson
2016-01-12  4:37   ` Tian, Kevin
2016-01-12 11:43     ` John Harrison
2016-01-12 13:49       ` Dave Gordon
2016-01-13  2:33         ` Tian, Kevin

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=1452537788-32136-10-git-send-email-John.C.Harrison@Intel.com \
    --to=john.c.harrison@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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.