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

A 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 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 compute 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 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' value
instead of determining it algorithmically. So make it an extra CONFIG
definition. Also, remove the execlist centric comment from the
existing pre-emption timeout CONFIG option given that it applies to
more than just execlists.

Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Acked-by: Michal Mrozek <michal.mrozek@intel.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221006213813.1563435-5-John.C.Harrison@Intel.com

drm/i915: Fail too long user submissions by default

A new Kconfig option CONFIG_DRM_I915_REQUEST_TIMEOUT is added, defaulting
to 20s, and this timeout is applied to all users contexts using the
previously added watchdog facility.

Result of this is that any user submission will simply fail after this
timeout, either causing a reset (for non-preemptable), or incomplete
results.

This can have an effect that workloads which used to work fine will
suddenly start failing. Even workloads comprised of short batches but in
long dependency chains can be terminated.

And because of lack of agreement on usefulness and safety of fence error
propagation this partial execution can be invisible to userspace even if
it is "listening" to returned fence status.

Another interaction is with hangcheck where care needs to be taken timeout
is not set lower or close to three times the heartbeat interval. Otherwise
a hang in any application can cause complete termination of all
submissions from unrelated clients. Any users modifying the per engine
heartbeat intervals therefore need to be aware of this potential denial of
service 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 to
enforce 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 <tvrtko.ursulin@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Acked-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20210324121335.2307063-7-tvrtko.ursulin@linux.intel.com

drm/i915: Replace the hardcoded I915_FENCE_TIMEOUT

Expose the hardcoded timeout for unsignaled foreign fences as a Kconfig
option, primarily to allow brave systems to disable the timeout and
solely rely on correct signaling.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Acked-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200509105021.12542-1-chris@chris-wilson.co.uk

drm/i915/gen12: Disable preemption timeout

Allow super long OpenCL workloads which cannot be preempted within
the 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 <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Michal Mrozek <michal.mrozek@intel.com>
Cc: <stable@vger.kernel.org> # v5.6+
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Michal Mrozek <Michal.mrozek@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200312115748.29970-1-tvrtko.ursulin@linux.intel.com

drm/i915/gt: Expose heartbeat interval via sysfs

We monitor the health of the system via periodic heartbeat pulses. The
pulses also provide the opportunity to perform garbage collection.
However, we interpret an incomplete pulse (a missed heartbeat) as an
indication that the system is no longer responsive, i.e. hung, and
perform an engine or full GPU reset. Given that the preemption
granularity can be very coarse on a system, we let the sysadmin override
our legacy timeouts which were "optimised" for desktop applications.

The heartbeat interval can be adjusted per-engine using,

/sys/class/drm/card?/engine/*/heartbeat_interval_ms

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-7-chris@chris-wilson.co.uk

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

After initialising a preemption request, we give the current resident a
small amount of time to vacate the GPU. The preemption request is for a
higher priority context and should be immediate to maintain high
quality of service (and avoid priority inversion). However, the
preemption granularity of the GPU can be quite coarse and so we need a
compromise.

The preempt timeout can be adjusted per-engine using,

/sys/class/drm/card?/engine/*/preempt_timeout_ms

and can be disabled by setting it to 0.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-6-chris@chris-wilson.co.uk

drm/i915/gt: Expose reset stop timeout via sysfs

When we allow ourselves to sleep before a GPU reset after disabling
submission, even for a few milliseconds, gives an innocent context the
opportunity to clear the GPU before the reset occurs. However, how long
to sleep depends on the typical non-preemptible duration (a similar
problem to determining the ideal preempt-reset timeout or even the
heartbeat interval). As this seems of a hard policy decision, punt it to
userspace.

The timeout can be adjusted using

/sys/class/drm/card?/engine/*/stop_timeout_ms

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Jon Bloomfield <jon.bloomfield@intel.com>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-5-chris@chris-wilson.co.uk

drm/i915/gt: Expose busywait duration to sysfs

We busywait on an inflight request (one that is currently executing on
HW, and so might complete quickly) prior to setting up an interrupt and
sleeping. The trade off is that we keep an expensive CPU core busy in
order to avoid wake up latency: where that trade off should lie is best
left to the sysadmin.

The busywait mechanism can be compiled out with

./scripts/config --set-val DRM_I915_SPIN_REQUEST 0

The maximum busywait duration can be adjusted per-engine using,

/sys/class/drm/card?/engine/*/ms_busywait_duration_ns

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-4-chris@chris-wilson.co.uk

drm/i915/gt: Expose timeslice duration to sysfs

Execlists uses a scheduling quantum (a timeslice) to alternate execution
between ready-to-run contexts of equal priority. This ensures that all
users (though only if they of equal importance) have the opportunity to
run and prevents livelocks where contexts may have implicit ordering due
to userspace semaphores.

The timeslicing mechanism can be compiled out with

./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0

The timeslice duration can be adjusted per-engine using,

/sys/class/drm/card?/engine/*/timeslice_duration_ms

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Steve Carbonari <steven.carbonari@intel.com>
Tested-by: Steve Carbonari <steven.carbonari@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200228131716.3243616-3-chris@chris-wilson.co.uk

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

Based on a sampling of a number of benchmarks across platforms, by
default opt for a much more lenient timeout so that we should not
adversely affect existing "good" clients.

640ms ought to be enough for anyone.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112169
Fixes: 3a7a92aba8fb ("drm/i915/execlists: Force preemption")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Eero Tamminen <eero.t.tamminen@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
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 <joonas.lahtinen@linux.intel.com>

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

Based on a sampling of a number of benchmarks across platforms, by
default opt for a much more lenient timeout so that we should not
adversely affect existing "good" clients.

640ms ought to be enough for anyone.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112169
Fixes: 3a7a92aba8fb ("drm/i915/execlists: Force preemption")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Eero Tamminen <eero.t.tamminen@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191125162737.2161069-1-chris@chris-wilson.co.uk

drm/i915/gt: Make timeslice duration configurable

Execlists uses a scheduling quantum (a timeslice) to alternate execution
between ready-to-run contexts of equal priority. This ensures that all
users (though only if they of equal importance) have the opportunity to
run and prevents livelocks where contexts may have implicit ordering due
to userspace semaphores. However, not all workloads necessarily benefit
from timeslicing and in the extreme some sysadmin may want to disable or
reduce the timeslicing granularity.

The timeslicing mechanism can be compiled out^W^W disabled (but should
DCE!) with

./scripts/config --set-val DRM_I915_TIMESLICE_DURATION 0

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191029091632.26281-1-chris@chris-wilson.co.uk

drm/i915/gt: Replace hangcheck by heartbeats

Replace sampling the engine state every so often with a periodic
heartbeat request to measure the health of an engine. This is coupled
with the forced-preemption to allow long running requests to survive so
long as they do not block other users.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Jon Bloomfield <jon.bloomfield@intel.com>
Reviewed-by: Jon Bloomfield <jon.bloomfield@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-5-chris@chris-wilson.co.uk

drm/i915/execlists: Force preemption

If the preempted context takes too long to relinquish control, e.g. it
is stuck inside a shader with arbitration disabled, evict that context
with an engine reset. This ensures that preemptions are reasonably
responsive, providing a tighter QoS for the more important context at
the cost of flagging unresponsive contexts more frequently (i.e. instead
of using an ~10s hangcheck, we now evict at ~100ms). The challenge of
lies in picking a timeout that can be reasonably serviced by HW for
typical workloads, balancing the existing clients against the needs for
responsiveness.

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 0

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-2-chris@chris-wilson.co.uk

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

If we are doing a normal GPU reset triggered after detecting a long
period of stalled work, we can take our time and allow the engines to
quiesce. Since we've stopped submission to the engine, and if we wait
long enough an innocent context should complete, leaving the engine idle.
So by waiting a short amount of time, we should prevent clobbering other
users when resetting a stuck context.

Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Suggested-by: Jon Bloomfield <jon.bloomfield@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191023133108.21401-1-chris@chris-wilson.co.uk

drm/i915: Add a label for config DRM_I915_SPIN_REQUEST

If we don't give it a label, it does not appear as a configuration
option.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190612093111.11684-9-chris@chris-wilson.co.uk

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

Do not allow runtime pm autosuspend to remove userspace GGTT mmaps too
quickly. For example, igt sets the autosuspend delay to 0, and so we
immediately attempt to perform runtime suspend upon releasing the
wakeref. Unfortunately, that involves tearing down GGTT mmaps as they
require an active device.

Override the autosuspend for GGTT mmaps, by keeping the wakeref around
for 250ms after populating the PTE for a fresh mmap.

v2: Prefer refcount_t for its under/overflow error detection
v3: Flush the user runtime autosuspend prior to system system.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190527115114.13448-1-chris@chris-wilson.co.uk

drm/i915: Expose the busyspin durations for i915_wait_request

An interesting discussion regarding "hybrid interrupt polling" for NVMe
came to the conclusion that the ideal busyspin before sleeping was half
of the expected request latency (and better if it was already halfway
through that request). This suggested that we too should look again at
our tradeoff between spinning and waiting. Currently, our spin simply
tries to hide the cost of enabling the interrupt, which is good to avoid
penalising nop requests (i.e. test throughput) and not much else.
Studying real world workloads suggests that a spin of upto 500us can
dramatically boost performance, but the suggestion is that this is not
from avoiding interrupt latency per-se, but from secondary effects of
sleeping such as allowing the CPU reduce cstate and context switch away.

In a truly hybrid interrupt polling scheme, we would aim to sleep until
just before the request completed and then wake up in advance of the
interrupt and do a quick poll to handle completion. This is tricky for
ourselves at the moment as we are not recording request times, and since
we allow preemption, our requests are not on as a nicely ordered
timeline as IO. However, the idea is interesting, for it will certainly
help us decide when busyspinning is worthwhile.

v2: Expose the spin setting via Kconfig options for easier adjustment
and testing.
v3: Don't get caught sneaking in a change to the busyspin parameters.
v4: Explain more about the "hybrid interrupt polling" scheme that we
want to migrate towards.

Suggested-by: Sagar Kamble <sagar.a.kamble@intel.com>
References: http://events.linuxfoundation.org/sites/events/files/slides/lemoal-nvme-polling-vault-2017-final_0.pdf
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Sagar Kamble <sagar.a.kamble@intel.com>
Cc: Eero Tamminen <eero.t.tamminen@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Sagar Kamble <sagar.a.kamble@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419182625.11186-1-chris@chris-wilson.co.uk