Changes in Kconfig.profile

d7a8680e - drm/i915: Improve long running compute w/a for GuC submission

Thu, 06 Oct 2022 21:38:13 +0000

drm/i915: Improve long running compute w/a for GuC submissionA workaround was added to the driver to allow compute 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 eventuallyand 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 triggersa per engine reset. Thus, disabling the timeout means also losing allper engine reset ability. A full GT reset will still occur when theheartbeat finally expires, but that is a much more destructive andundesirable mechanism.The purpose of the workaround is actually to give compute tasks longerto reach a pre-emption point after a pre-emption request has beenissued. This is necessary because Gen12 does not support mid-threadpre-emption and compute tasks can have long running threads.So, rather than disabling the timeout completely, just set it to a'long' value.v2: Review feedback from Tvrtko - must hard code the 'long' valueinstead of determining it algorithmically. So make it an extra CONFIGdefinition. Also, remove the execlist centric comment from theexisting pre-emption timeout CONFIG option given that it applies tomore than just execlists.Signed-off-by: John Harrison Reviewed-by: Daniele Ceraolo Spurio Acked-by: Michal Mrozek Acked-by: Tvrtko Ursulin Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-5-John.C.Harrison@Intel.com List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

e8dbb566 - drm/i915: Fail too long user submissions by default

Wed, 24 Mar 2021 12:13:34 +0000

drm/i915: Fail too long user submissions by defaultA new Kconfig option CONFIG_DRM_I915_REQUEST_TIMEOUT is added, defaultingto 20s, and this timeout is applied to all users contexts using thepreviously added watchdog facility.Result of this is that any user submission will simply fail after thistimeout, either causing a reset (for non-preemptable), or incompleteresults.This can have an effect that workloads which used to work fine willsuddenly start failing. Even workloads comprised of short batches but inlong dependency chains can be terminated.And because of lack of agreement on usefulness and safety of fence errorpropagation this partial execution can be invisible to userspace even ifit is "listening" to returned fence status.Another interaction is with hangcheck where care needs to be taken timeoutis not set lower or close to three times the heartbeat interval. Otherwisea hang in any application can cause complete termination of allsubmissions from unrelated clients. Any users modifying the per engineheartbeat intervals therefore need to be aware of this potential denial ofservice to avoid inadvertently enabling it.Given all this I am personally not convinced the scheme is a good idea.Intuitively it feels object importers would be better positioned toenforce the time they are willing to wait for something to complete.v2: * Improved commit message and Kconfig text. * Pull in some helper code from patch which got dropped.v3: * Bump timeout to 20s to see if it helps Tigerlake.Signed-off-by: Tvrtko Ursulin Cc: Daniel Vetter Acked-by: Matthew Auld Signed-off-by: Daniel Vetter Link: https://patchwork.freedesktop.org/patch/msgid/20210324121335.2307063-7-tvrtko.ursulin@linux.intel.com List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

16dc224f - drm/i915: Replace the hardcoded I915_FENCE_TIMEOUT

Sat, 09 May 2020 10:50:21 +0000

drm/i915: Replace the hardcoded I915_FENCE_TIMEOUTExpose the hardcoded timeout for unsignaled foreign fences as a Kconfigoption, primarily to allow brave systems to disable the timeout andsolely rely on correct signaling.Signed-off-by: Chris Wilson Cc: Joonas Lahtinen Acked-by: Michael J. Ruhl Link: https://patchwork.freedesktop.org/patch/msgid/20200509105021.12542-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

07bcfd12 - drm/i915/gen12: Disable preemption timeout

Thu, 12 Mar 2020 11:57:48 +0000

drm/i915/gen12: Disable preemption timeoutAllow super long OpenCL workloads which cannot be preempted withinthe default timeout to run out of the box.v2: * Make it stick out more and apply only to RCS. (Chris)v3: * Mention platform override in kconfig. (Joonas)Signed-off-by: Tvrtko Ursulin Cc: Chris Wilson Cc: Joonas Lahtinen Cc: Michal Mrozek Cc: # v5.6+Acked-by: Chris Wilson Acked-by: Michal Mrozek Link: https://patchwork.freedesktop.org/patch/msgid/20200312115748.29970-1-tvrtko.ursulin@linux.intel.com List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

9a40bddd - drm/i915/gt: Expose heartbeat interval via sysfs

Fri, 28 Feb 2020 13:17:16 +0000

drm/i915/gt: Expose heartbeat interval via sysfsWe monitor the health of the system via periodic heartbeat pulses. Thepulses also provide the opportunity to perform garbage collection.However, we interpret an incomplete pulse (a missed heartbeat) as anindication that the system is no longer responsive, i.e. hung, andperform an engine or full GPU reset. Given that the preemptiongranularity can be very coarse on a system, we let the sysadmin overrideour legacy timeouts which were "optimised" for desktop applications.The heartbeat interval can be adjusted per-engine using, /sys/class/drm/card?/engine/*/heartbeat_interval_msSigned-off-by: Chris Wilson Reviewed-by: Steve Carbonari Tested-by: Steve Carbonari Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-7-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

db3d8338 - drm/i915/gt: Expose preempt reset timeout via sysfs

Fri, 28 Feb 2020 13:17:15 +0000

drm/i915/gt: Expose preempt reset timeout via sysfsAfter initialising a preemption request, we give the current resident asmall amount of time to vacate the GPU. The preemption request is for ahigher priority context and should be immediate to maintain highquality of service (and avoid priority inversion). However, thepreemption granularity of the GPU can be quite coarse and so we need acompromise.The preempt timeout can be adjusted per-engine using, /sys/class/drm/card?/engine/*/preempt_timeout_msand can be disabled by setting it to 0.Signed-off-by: Chris Wilson Reviewed-by: Steve Carbonari Tested-by: Steve Carbonari Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-6-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

72338a1f - drm/i915/gt: Expose reset stop timeout via sysfs

Fri, 28 Feb 2020 13:17:14 +0000

drm/i915/gt: Expose reset stop timeout via sysfsWhen we allow ourselves to sleep before a GPU reset after disablingsubmission, even for a few milliseconds, gives an innocent context theopportunity to clear the GPU before the reset occurs. However, how longto sleep depends on the typical non-preemptible duration (a similarproblem to determining the ideal preempt-reset timeout or even theheartbeat interval). As this seems of a hard policy decision, punt it touserspace.The timeout can be adjusted using /sys/class/drm/card?/engine/*/stop_timeout_msSigned-off-by: Chris Wilson Cc: Joonas Lahtinen Cc: Jon Bloomfield Reviewed-by: Steve Carbonari Tested-by: Steve Carbonari Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-5-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

062444bb - drm/i915/gt: Expose busywait duration to sysfs

Fri, 28 Feb 2020 13:17:13 +0000

drm/i915/gt: Expose busywait duration to sysfsWe busywait on an inflight request (one that is currently executing onHW, and so might complete quickly) prior to setting up an interrupt andsleeping. The trade off is that we keep an expensive CPU core busy inorder to avoid wake up latency: where that trade off should lie is bestleft to the sysadmin.The busywait mechanism can be compiled out with ./scripts/config --set-val DRM_I915_SPIN_REQUEST 0The maximum busywait duration can be adjusted per-engine using, /sys/class/drm/card?/engine/*/ms_busywait_duration_nsSigned-off-by: Chris Wilson Cc: Joonas Lahtinen Reviewed-by: Steve Carbonari Tested-by: Steve Carbonari Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-4-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

1a2695a7 - drm/i915/gt: Expose timeslice duration to sysfs

Fri, 28 Feb 2020 13:17:12 +0000

drm/i915/gt: Expose timeslice duration to sysfsExeclists uses a scheduling quantum (a timeslice) to alternate executionbetween ready-to-run contexts of equal priority. This ensures that allusers (though only if they of equal importance) have the opportunity torun and prevents livelocks where contexts may have implicit ordering dueto userspace semaphores.The timeslicing mechanism can be compiled out with ./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0The timeslice duration can be adjusted per-engine using, /sys/class/drm/card?/engine/*/timeslice_duration_msSigned-off-by: Chris Wilson Cc: Joonas Lahtinen Reviewed-by: Steve Carbonari Tested-by: Steve Carbonari Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-3-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

5766a5ff - drm/i915: Default to a more lenient forced preemption timeout

Mon, 25 Nov 2019 16:27:37 +0000

drm/i915: Default to a more lenient forced preemption timeoutBased on a sampling of a number of benchmarks across platforms, bydefault opt for a much more lenient timeout so that we should notadversely affect existing "good" clients.640ms ought to be enough for anyone.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112169Fixes: 3a7a92aba8fb ("drm/i915/execlists: Force preemption")Signed-off-by: Chris Wilson Cc: Joonas Lahtinen Cc: Eero Tamminen Cc: Dmitry Rogozhkin Reviewed-by: Tvrtko Ursulin Link: https://patchwork.freedesktop.org/patch/msgid/20191125162737.2161069-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

8d15ede5 - drm/i915: Default to a more lenient forced preemption timeout

Mon, 25 Nov 2019 16:27:37 +0000

drm/i915: Default to a more lenient forced preemption timeoutBased on a sampling of a number of benchmarks across platforms, bydefault opt for a much more lenient timeout so that we should notadversely affect existing "good" clients.640ms ought to be enough for anyone.Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112169Fixes: 3a7a92aba8fb ("drm/i915/execlists: Force preemption")Signed-off-by: Chris Wilson Cc: Joonas Lahtinen Cc: Eero Tamminen Cc: Dmitry Rogozhkin Reviewed-by: Tvrtko Ursulin Link: https://patchwork.freedesktop.org/patch/msgid/20191125162737.2161069-1-chris@chris-wilson.co.uk(cherry picked from commit 5766a5ffc6a69595903865518c43636bde0e4ac4)Signed-off-by: Joonas Lahtinen List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

b79029b2 - drm/i915/gt: Make timeslice duration configurable

Tue, 29 Oct 2019 09:16:32 +0000

drm/i915/gt: Make timeslice duration configurableExeclists uses a scheduling quantum (a timeslice) to alternate executionbetween ready-to-run contexts of equal priority. This ensures that allusers (though only if they of equal importance) have the opportunity torun and prevents livelocks where contexts may have implicit ordering dueto userspace semaphores. However, not all workloads necessarily benefitfrom timeslicing and in the extreme some sysadmin may want to disable orreduce the timeslicing granularity.The timeslicing mechanism can be compiled out^W^W disabled (but shouldDCE!) with ./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0Signed-off-by: Chris Wilson Cc: Joonas Lahtinen Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20191029091632.26281-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

058179e7 - drm/i915/gt: Replace hangcheck by heartbeats

Wed, 23 Oct 2019 13:31:08 +0000

drm/i915/gt: Replace hangcheck by heartbeatsReplace sampling the engine state every so often with a periodicheartbeat request to measure the health of an engine. This is coupledwith the forced-preemption to allow long running requests to survive solong as they do not block other users.Signed-off-by: Chris Wilson Cc: Joonas Lahtinen Cc: Tvrtko Ursulin Cc: Jon Bloomfield Reviewed-by: Jon Bloomfield Reviewed-by: Tvrtko Ursulin Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-5-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

3a7a92ab - drm/i915/execlists: Force preemption

Wed, 23 Oct 2019 13:31:05 +0000

drm/i915/execlists: Force preemptionIf the preempted context takes too long to relinquish control, e.g. itis stuck inside a shader with arbitration disabled, evict that contextwith an engine reset. This ensures that preemptions are reasonablyresponsive, providing a tighter QoS for the more important context atthe cost of flagging unresponsive contexts more frequently (i.e. insteadof using an ~10s hangcheck, we now evict at ~100ms). The challenge oflies in picking a timeout that can be reasonably serviced by HW fortypical workloads, balancing the existing clients against the needs forresponsiveness.Note that coupled with timeslicing, this will lead to rapid GPU "hang"detection with multiple active contexts vying for GPU time.The forced preemption mechanism can be compiled out with ./scripts/config --set-val DRM_I915_PREEMPT_TIMEOUT 0Signed-off-by: Chris Wilson Cc: Mika Kuoppala Cc: Tvrtko Ursulin Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-2-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

a8c51ed2 - drm/i915/gt: Try to more gracefully quiesce the system before resets

Wed, 23 Oct 2019 13:31:04 +0000

drm/i915/gt: Try to more gracefully quiesce the system before resetsIf we are doing a normal GPU reset triggered after detecting a longperiod of stalled work, we can take our time and allow the engines toquiesce. Since we've stopped submission to the engine, and if we waitlong enough an innocent context should complete, leaving the engine idle.So by waiting a short amount of time, we should prevent clobbering otherusers when resetting a stuck context.Suggested-by: Joonas Lahtinen Suggested-by: Jon Bloomfield Signed-off-by: Chris Wilson Cc: Mika Kuoppala Cc: Joonas Lahtinen Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

ea60f4bd - drm/i915: Add a label for config DRM_I915_SPIN_REQUEST

Wed, 12 Jun 2019 09:31:11 +0000

drm/i915: Add a label for config DRM_I915_SPIN_REQUESTIf we don't give it a label, it does not appear as a configurationoption.Signed-off-by: Chris Wilson Cc: Mika Kuoppala Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20190612093111.11684-9-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

b27e35ae - drm/i915: Keep user GGTT alive for a minimum of 250ms

Mon, 27 May 2019 11:51:14 +0000

drm/i915: Keep user GGTT alive for a minimum of 250msDo not allow runtime pm autosuspend to remove userspace GGTT mmaps tooquickly. For example, igt sets the autosuspend delay to 0, and so weimmediately attempt to perform runtime suspend upon releasing thewakeref. Unfortunately, that involves tearing down GGTT mmaps as theyrequire an active device.Override the autosuspend for GGTT mmaps, by keeping the wakeref aroundfor 250ms after populating the PTE for a fresh mmap.v2: Prefer refcount_t for its under/overflow error detectionv3: Flush the user runtime autosuspend prior to system system.Signed-off-by: Chris Wilson Cc: Mika Kuoppala Reviewed-by: Mika Kuoppala Link: https://patchwork.freedesktop.org/patch/msgid/20190527115114.13448-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile

7ce99d24 - drm/i915: Expose the busyspin durations for i915_wait_request

Fri, 19 Apr 2019 18:26:25 +0000

drm/i915: Expose the busyspin durations for i915_wait_requestAn interesting discussion regarding "hybrid interrupt polling" for NVMecame to the conclusion that the ideal busyspin before sleeping was halfof the expected request latency (and better if it was already halfwaythrough that request). This suggested that we too should look again atour tradeoff between spinning and waiting. Currently, our spin simplytries to hide the cost of enabling the interrupt, which is good to avoidpenalising nop requests (i.e. test throughput) and not much else.Studying real world workloads suggests that a spin of upto 500us candramatically boost performance, but the suggestion is that this is notfrom avoiding interrupt latency per-se, but from secondary effects ofsleeping such as allowing the CPU reduce cstate and context switch away.In a truly hybrid interrupt polling scheme, we would aim to sleep untiljust before the request completed and then wake up in advance of theinterrupt and do a quick poll to handle completion. This is tricky forourselves at the moment as we are not recording request times, and sincewe allow preemption, our requests are not on as a nicely orderedtimeline as IO. However, the idea is interesting, for it will certainlyhelp us decide when busyspinning is worthwhile.v2: Expose the spin setting via Kconfig options for easier adjustmentand testing.v3: Don't get caught sneaking in a change to the busyspin parameters.v4: Explain more about the "hybrid interrupt polling" scheme that wewant to migrate towards.Suggested-by: Sagar Kamble References: http://events.linuxfoundation.org/sites/events/files/slides/lemoal-nvme-polling-vault-2017-final_0.pdfSigned-off-by: Chris Wilson Cc: Sagar Kamble Cc: Eero Tamminen Cc: Tvrtko Ursulin Cc: Ben Widawsky Cc: Joonas Lahtinen Cc: Michał Winiarski Reviewed-by: Sagar Kamble Link: https://patchwork.freedesktop.org/patch/msgid/20190419182625.11186-1-chris@chris-wilson.co.uk List of files: /linux-6.15/drivers/gpu/drm/i915/Kconfig.profile