From: John.C.Harrison@Intel.com
To: Intel-GFX@Lists.FreeDesktop.Org
Cc: John Harrison <John.C.Harrison@Intel.com>,
DRI-Devel@Lists.FreeDesktop.Org
Subject: [PATCH 4/4] drm/i915: Improve long running OCL w/a for GuC submission
Date: Fri, 10 Dec 2021 22:58:59 -0800 [thread overview]
Message-ID: <20211211065859.2248188-5-John.C.Harrison@Intel.com> (raw)
In-Reply-To: <20211211065859.2248188-1-John.C.Harrison@Intel.com>
From: John Harrison <John.C.Harrison@Intel.com>
A workaround was added to the driver to allow OpenCL workloads to run
'forever' by disabling pre-emption on the RCS engine for Gen12.
It is not totally unbound as the heartbeat will kick in eventually
and cause a reset of the hung engine.
However, this does not work well in GuC submission mode. In GuC mode,
the pre-emption timeout is how GuC detects hung contexts and triggers
a per engine reset. Thus, disabling the timeout means also losing all
per engine reset ability. A full GT reset will still occur when the
heartbeat finally expires, but that is a much more destructive and
undesirable mechanism.
The purpose of the workaround is actually to give OpenCL tasks longer
to reach a pre-emption point after a pre-emption request has been
issued. This is necessary because Gen12 does not support mid-thread
pre-emption and OpenCL can have long running threads.
So, rather than disabling the timeout completely, just set it to a
'long' value. Likewise, bump the heartbeat interval. That gives the
OpenCL thread sufficient time to reach a pre-emption point without
being killed off either by the GuC or by the heartbeat.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
---
drivers/gpu/drm/i915/gt/intel_engine_cs.c | 42 +++++++++++++++++++++--
1 file changed, 39 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_engine_cs.c b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
index 352254e001b4..26af8d60fe2b 100644
--- a/drivers/gpu/drm/i915/gt/intel_engine_cs.c
+++ b/drivers/gpu/drm/i915/gt/intel_engine_cs.c
@@ -382,9 +382,45 @@ static int intel_engine_setup(struct intel_gt *gt, enum intel_engine_id id,
engine->props.timeslice_duration_ms =
CONFIG_DRM_I915_TIMESLICE_DURATION;
- /* Override to uninterruptible for OpenCL workloads. */
- if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS)
- engine->props.preempt_timeout_ms = 0;
+ /*
+ * Mid-thread pre-emption is not available in Gen12. Unfortunately,
+ * some OpenCL workloads run quite long threads. That means they get
+ * reset due to not pre-empting in a timely manner.
+ * The execlist solution was to disable pre-emption completely.
+ * However, pre-emption timeouts are the way GuC detects hung contexts
+ * and triggers engine resets. Thus, without pre-emption, there is no
+ * per engine reset. And full GT reset is much more intrusive. So keep
+ * the timeout for GuC submission platforms and just bump it to be
+ * much larger. Also bump the heartbeat timeout to match, otherwise
+ * the heartbeat can expire before the pre-emption can timeout and
+ * thus trigger a full GT reset anyway.
+ */
+ if (GRAPHICS_VER(i915) == 12 && engine->class == RENDER_CLASS) {
+ if (intel_uc_wants_guc_submission(>->uc)) {
+ const unsigned long min_preempt = 7500;
+ const unsigned long min_beat = 5000;
+
+ if (engine->props.preempt_timeout_ms &&
+ engine->props.preempt_timeout_ms < min_preempt) {
+ drm_info(>->i915->drm, "Bumping pre-emption timeout from %ld to %ld on %s to allow slow compute pre-emption\n",
+ engine->props.preempt_timeout_ms,
+ min_preempt, engine->name);
+
+ engine->props.preempt_timeout_ms = min_preempt;
+ }
+
+ if (engine->props.heartbeat_interval_ms &&
+ engine->props.heartbeat_interval_ms < min_beat) {
+ drm_info(>->i915->drm, "Bumping heartbeat interval from %ld to %ld on %s to allow slow compute pre-emption\n",
+ engine->props.heartbeat_interval_ms,
+ min_beat, engine->name);
+
+ engine->props.heartbeat_interval_ms = min_beat;
+ }
+ } else {
+ engine->props.preempt_timeout_ms = 0;
+ }
+ }
engine->defaults = engine->props; /* never to change again */
--
2.25.1
next prev parent reply other threads:[~2021-12-11 6:59 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-12-11 6:58 [PATCH 0/4] More fixes/tweaks to GuC support John.C.Harrison
2021-12-11 6:58 ` [PATCH 1/4] drm/i915/guc: Speed up GuC log dumps John.C.Harrison
2021-12-20 14:43 ` Daniele Ceraolo Spurio
2021-12-11 6:58 ` [PATCH 2/4] drm/i915/guc: Increase GuC log size for CONFIG_DEBUG_GEM John.C.Harrison
2021-12-11 6:58 ` [PATCH 3/4] drm/i915/guc: Flag an error if an engine reset fails John.C.Harrison
2021-12-16 1:16 ` [Intel-gfx] " Daniele Ceraolo Spurio
2021-12-11 6:58 ` John.C.Harrison [this message]
2021-12-16 1:11 ` [Intel-gfx] [PATCH 4/4] drm/i915: Improve long running OCL w/a for GuC submission Daniele Ceraolo Spurio
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=20211211065859.2248188-5-John.C.Harrison@Intel.com \
--to=john.c.harrison@intel.com \
--cc=DRI-Devel@Lists.FreeDesktop.Org \
--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;
as well as URLs for NNTP newsgroup(s).