- 24 Oct, 2019 2 commits
-
-
Chris Wilson authored
Make trebly sure that all possible callbacks and their delayed brethren are complete before asserting that the i915_active should be idle after flushing all barriers. Signed-off-by:
Chris Wilson <chris@chris-wilson.co.uk> Reviewed-by:
Mika Kuoppala <mika.kuoppala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191023235359.27132-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
When setting up the system to perform the atomic reset, we need to serialise with any ongoing interrupt tasklet or else: <0> [472.951428] i915_sel-4442 0d..1 466527056us : __i915_request_submit: rcs0 fence 11659:2, current 0 <0> [472.951554] i915_sel-4442 0d..1 466527059us : __execlists_submission_tasklet: rcs0: queue_priority_hint:-2147483648, submit:yes <0> [472.951681] i915_sel-4442 0d..1 466527061us : trace_ports: rcs0: submit { 11659:2, 0:0 } <0> [472.951805] i915_sel-4442 0.... 466527114us : __igt_atomic_reset_engine: i915_reset_engine(rcs0:active) under hardirq <0> [472.951932] i915_sel-4442 0d... 466527115us : intel_engine_reset: rcs0 flags=11d <0> [472.952056] i915_sel-4442 0d... 466527117us : execlists_reset_prepare: rcs0: depth<-1 <0> [472.952179] i915_sel-4442 0d... 466527119us : intel_engine_stop_cs: rcs0 <0> [472.952305] <idle>-0 1..s1 466527119us : process_csb: rcs0 cs-irq head=3, tail=4 <0> [472.952431] i915_sel-4442 0d... 466527122us : __intel_gt_reset: engine_mask=1 <0> [472.952557] <idle>-0 1..s1 466527124us : process_csb: rcs0 csb[4]: status=0x00000001:0x00000000 <0> [472.952683] <idle>-0 1..s1 466527130us : trace_ports: rcs0: promote { 11659:2*, 0:0 } <0> [472.952808] i915_sel-4442 0d... 466527131us : execlists_reset: rcs0 <0> [472.952933] i915_sel-4442 0d..1 466527133us : process_csb: rcs0 cs-irq head=3, tail=4 <0> [472.953059] i915_sel-4442 0d..1 466527134us : process_csb: rcs0 csb[4]: status=0x00000001:0x00000000 <0> [472.953185] i915_sel-4442 0d..1 466527136us : trace_ports: rcs0: preempted { 11659:2*, 0:0 } <0> [472.953310] i915_sel-4442 0d..1 466527150us : assert_pending_valid: Nothing pending for promotion! <0> [472.953436] i915_sel-4442 0d..1 466527158us : process_csb: process_csb:1930 GEM_BUG_ON(!assert_pending_valid(execlists, "promote")) We have the same CSB events being seen by process_csb() on two different processors. One being issued by the reset in the test, the other by the interrupt; this scenario is supposed to be prevented by flushing the interrupt tasklet with tasklet_disable() before we enter the atomic reset. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112069Signed-off-by:
-
- 23 Oct, 2019 12 commits
-
-
Zhenyu Wang authored
As early workload scan and shadow happens in execlist mmio handler, which has already taken vgpu_lock. So remove extra lock taking here. Fixes: 952f89f0 ("drm/i915/gvt: Wean off struct_mutex") Cc: Chris Wilson <chris@chris-wilson.co.uk> Signed-off-by:
Zhenyu Wang <zhenyuw@linux.intel.com> Reviewed-by:
-
Chris Wilson authored
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:
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
-
Chris Wilson authored
Normally, we rely on our hangcheck to prevent persistent batches from hogging the GPU. However, if the user disables hangcheck, this mechanism breaks down. Despite our insistence that this is unsafe, the users are equally insistent that they want to use endless batches and will disable the hangcheck mechanism. We are looking at replacing hangcheck, in the next patch, with a softer mechanism, that sends a pulse down the engine to check if it is well. We can use the same preemptive pulse to flush an active context off the GPU upon context close, preventing resources being lost and unkillable requests remaining on the GPU after process termination. Testcase: igt/gem_ctx_exec/basic-nohangcheck Signed-off-by:
-
Chris Wilson authored
On schedule-out (CS completion) of a banned context, scrub the context image so that we do not replay the active payload. The intent is that we skip banned payloads on request submission so that the timeline advancement continues on in the background. However, if we are returning to a preempted request, i915_request_skip() is ineffective and instead we need to patch up the context image so that it continues from the start of the next request. v2: Fixup cancellation so that we only scrub the payload of the active request and do not short-circuit the breadcrumbs (which might cause other contexts to execute out of order). v3: Grammar pass Signed-off-by:
-
Chris Wilson authored
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 authored
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:
-
Robert M. Fosha authored
GuC enable logging H2G action definition changed some time ago from 0xE000 to 0x40. All current GuC FW blobs use this definition, so fix the action definition in driver to match. Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by:
Robert M. Fosha <robert.m.fosha@intel.com> Reviewed-by:
Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Signed-off-by:
-
Robert M. Fosha authored
Creating and opening the GuC log relay file enables and starts the relay potentially before the caller is ready to consume logs. Change the behavior so that relay starts only on an explicit call to the write function (with a value of '1'). Other values flush the log relay as before. v2: Style changes and fix typos. Add guc_log_relay_stop() function. (Daniele) Cc: Matthew Brost <matthew.brost@intel.com> Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com> Cc: Michal Wajdeczko <michal.wajdeczko@intel.com> Signed-off-by:
Matthew Brost <matthew.brost@intel.com> Signed-off-by:
-
Imre Deak authored
Atm we don't detect a PCH with PCI ID 0xA3C1 which showed up now on a CML platform. We don't have the official assignment of the PCH PCI IDs, but this looks like a CNP which was already used on CML platforms. Let's add the new ID->PCH type mapping accordingly. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112051Reported-and-tested-by:
Cyrus <cyrus.lien@canonical.com> Cc: Cyrus <cyrus.lien@canonical.com> Signed-off-by:
Imre Deak <imre.deak@intel.com> Reviewed-by:
José Roberto de Souza <jose.souza@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191022095155.30991-1-imre.deak@intel.com
-
Jani Nikula authored
DSC isn't DP specific, so remove the dp_ prefix from the crtc state member name. Also moving the member under the dsc sub-struct gives us enough context to allow shortening the name to just config. No functional changes. Cc: Manasi Navare <manasi.d.navare@intel.com> Reviewed-by:
Manasi Navare <manasi.d.navare@intel.com> Signed-off-by:
Jani Nikula <jani.nikula@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191022133414.8293-2-jani.nikula@intel.com
-
Jani Nikula authored
Reduce verbosity in code by renaming dsc_params member of crtc state to simply dsc. There is enough context for this to be clear. No functional changes. Cc: Manasi Navare <manasi.d.navare@intel.com> Reviewed-by:
-
Chris Wilson authored
A lock once taken must be released again. Fixes: c31c9e82 ("drm/i915/selftests: Teach switch_to_context() to use the context") Signed-off-by:
-
- 22 Oct, 2019 20 commits
-
-
Chris Wilson authored
If retirement is running on another thread, we may inspect the status of the i915_active before its retirement callback is complete. As we expect it to be running synchronously, we can wait for any callback to complete by acquiring the i915_active.mutex. Signed-off-by:
Andi Shyti <andi.shyti@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191022112111.9317-1-chris@chris-wilson.co.uk
-
Chris Wilson authored
Forcewake is the speciality of the GT, so it is natural to run the intel_uncore_forcewake tests over the GT. So pass intel_gt as the parameter to our selftests. Signed-off-by:
-
Chris Wilson authored
The context details which engines to use, so use the ctx->engines[] to generate the requests to cause the context switch. Signed-off-by:
-
Chris Wilson authored
Again we wish to operate on the engines, which are owned by the intel_gt. As such it is easier, and much more consistent, to pass the intel_gt parameter. v2: Unexport i915_gem_load_power_context() Signed-off-by:
-
Chris Wilson authored
Separate each object class into a separate lock type to avoid lockdep cross-contamination between paths (i.e. userptr!). Signed-off-by:
Matthew Auld <matthew.auld@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191022144501.26486-1-chris@chris-wilson.co.uk
-
James Ausmus authored
Even the AML devices that behave like CFLs can be paired with an SPT PCH. Allow this to happen without blowing up dmesg. BSpec: 33665 Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112013 Cc: Quanxian Wang <quanxian.wang@intel.com> Cc: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by:
James Ausmus <james.ausmus@intel.com> Reviewed-by:
Lyude Paul <lyude@redhat.com> Signed-off-by:
Matt Roper <matthew.d.roper@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20191017194203.9645-1-james.ausmus@intel.com
-
Chris Wilson authored
The actual conditions are that we know the GPU is not accessing the context, and we hold a pin on the context image to allow CPU access. We used a fake lock on ce->pin_mutex so that we could try and use lockdep to assert that access is serialised, but the various different hardirq/softirq contexts where we need to *fake* holding the pin_mutex are causing more trouble. Still it would be nice if we did have a way to reassure ourselves that the direct update to the context image is serialised with GPU execution. In the meantime, stop lockdep complaining about false irq inversions. Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=111923Signed-off-by:
-
Chris Wilson authored
Loop over all engines, issuing a request for the object on each in order to make sure we leave no stone unturned when creating an active ref. The purpose is to make sure that we can reap a zombie object (one that is only alive due to an active reference on the GPU) no matter where that active reference emanates from. Signed-off-by:
-
Tvrtko Ursulin authored
Contexts are not testing physical engines so it makes sense to use the uabi iterator. Signed-off-by:
-
Tvrtko Ursulin authored
Frees up two call sites from passing i915 to for_each_engine. Signed-off-by:
-
Tvrtko Ursulin authored
Just freeing up two more call sites from passing in i915 to for_each_engine. Signed-off-by:
-
Tvrtko Ursulin authored
Convert the test code to work directly on what it needs rather than going through the top-level i915. This enables another natural usage for for_each_engine(.., gt, ..). Signed-off-by:
-
Tvrtko Ursulin authored
Just compartmentalizes code a bit more. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Tvrtko Ursulin authored
Engines belong to the GT so make it indicative in the API. Signed-off-by:
-
Chris Wilson authored
Add the backpointer to ppgtt and i915->gt so that we can traverse across the device hierarchy. Reported-by:
-
- 21 Oct, 2019 6 commits
-
-
Chris Wilson authored
With the last user, i915_vma_parked(), retired, there are no more users of the per-gt pm notifications and we can remove the unused infrastructure. Signed-off-by:
-
Chris Wilson authored
Currently even though i915_vma_parked() operates on a per-gt struct, it is called from a global pm notify. This oddity was only because the long term plan is to decouple the vma cache from the pm notification, but right now the oddity stands out like a sore thumb! Suggested-by:
-
Chris Wilson authored
To flush idle barriers, and even inflight requests, we want to send a preemptive 'pulse' along an engine. We use a no-op request along the pinned kernel_context at high priority so that it should run or else kick off the stuck requests. We can use this to ensure idle barriers are immediately flushed, as part of a context cancellation mechanism, or as part of a heartbeat mechanism to detect and reset a stuck GPU. Signed-off-by:
-
Chris Wilson authored
i915_active must track over any engine, so expand the selftest to iterate over all uabi engines. Signed-off-by:
-
Ville Syrjälä authored
On ILK+ the documented min hdisplay is 64, min hblank is 32, and min vblank is 5. On earlier platforms min hblank is also 32, and min vblank is 3. Make sure the mode satisfies those limits. There are further limits for HDMI and pfit use cases, but we'll check for those in a more specific location. Signed-off-by:
Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20190718144340.1114-2-ville.syrjala@linux.intel.comReviewed-by:
-
Chris Wilson authored
If we change the priority of the active context, then it has no impact on the decision of whether to preempt the active context -- we don't preempt the context with itself. In this situation, we elide the tasklet rescheduling and should *not* be marking up the queue_priority_hint as that may mask a later submission where we decide we don't have to kick the tasklet as a higher priority submission is pending (spoiler alert, it was not). Signed-off-by:
-
