UAPI Changes:
- drm/i915/guc: Use context hints for GT frequency
Allow user to provide a low latency context hint. When set, KMD
sends a hint to GuC which results in special handling for this
context. SLPC will ramp the GT frequency aggressively every time
it switches to this context. The down freq threshold will also be
lower so GuC will ramp down the GT freq for this context more slowly.
We also disable waitboost for this context as that will interfere with
the strategy.
We need to enable the use of SLPC Compute strategy during init, but
it will apply only to contexts that set this bit during context
creation.
Userland can check whether this feature is supported using a new param-
I915_PARAM_HAS_CONTEXT_FREQ_HINT. This flag is true for all guc submission
enabled platforms as they use SLPC for frequency management.
The Mesa usage model for this flag is here -
https://gitlab.freedesktop.org/sushmave/mesa/-/commits/compute_hint
- drm/i915/gt: Enable only one CCS for compute workload
Enable only one CCS engine by default with all the compute sices
allocated to it.
While generating the list of UABI engines to be exposed to the
user, exclude any additional CCS engines beyond the first
instance
***
NOTE: This W/A will make all DG2 SKUs appear like single CCS SKUs by
default to mitigate a hardware bug. All the EUs will still remain
usable, and all the userspace drivers have been confirmed to be able
to dynamically detect the change in number of CCS engines and adjust.
For the smaller percent of applications that get perf benefit from
letting the userspace driver dispatch across all 4 CCS engines we will
be introducing a sysfs control as a later patch to choose 4 CCS each
with 25% EUs (or 50% if 2 CCS).
NOTE: A regression has been reported at
https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/10895
However Andi has been triaging the issue and we're closing in a fix
to the gap in the W/A implementation:
https://lists.freedesktop.org/archives/intel-gfx/2024-April/348747.html
Driver Changes:
- Add new and fix to existing workarounds: Wa_14018575942 (MTL),
Wa_16019325821 (Gen12.70), Wa_14019159160 (MTL), Wa_16015675438,
Wa_14020495402 (Gen12.70) (Tejas, John, Lucas)
- Fix UAF on destroy against retire race and remove two earlier
partial fixes (Janusz)
- Limit the reserved VM space to only the platforms that need it (Andi)
- Reset queue_priority_hint on parking for execlist platforms (Chris)
- Fix gt reset with GuC submission is disabled (Nirmoy)
- Correct capture of EIR register on hang (John)
- Remove usage of the deprecated ida_simple_xx() API
- Refactor confusing __intel_gt_reset() (Nirmoy)
- Fix the fix for GuC reset lock confusion (John)
- Simplify/extend platform check for Wa_14018913170 (John)
- Replace dev_priv with i915 (Andi)
- Add and use gt_to_guc() wrapper (Andi)
- Remove bogus null check (Rodrigo, Dan)
. Selftest improvements (Janusz, Nirmoy, Daniele)
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/ZitVBTvZmityDi7D@jlahtine-mobl.ger.corp.intel.com
Currently intel_gt_reset() kills the GuC and then resets requested
engines. This is problematic because there is a dedicated CSB FIFO
which only GuC can access and if that FIFO fills up, the hardware
will block on the next context switch until there is space that means
the system is effectively hung. If an engine is reset whilst actively
executing a context, a CSB entry will be sent to say that the context
has gone idle. Thus if reset happens on a very busy system then
killing GuC before killing the engines will lead to deadlock because
of filled up CSB FIFO.
To address this issue, the GuC should be killed only after resetting
the requested engines and before calling intel_gt_init_hw().
v2: Improve commit message(John)
Cc: John Harrison <john.c.harrison@intel.com>
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240422201951.633-2-nirmoy.das@intel.com
Linux 6.9-rc5
I've had a persistent msm failure on clang, and the fix is in fixes
so just pull it back to fix that.
Signed-off-by: Dave Airlie <airlied@redhat.com>
The previous fix for the circlular lock splat about the busyness
worker wasn't quite complete. Even though the reset-in-progress flag
is cleared at the start of intel_uc_reset_finish, the entire function
is still inside the reset mutex lock. Not sure why the patch appeared
to fix the issue both locally and in CI. However, it is now back
again.
There is a further complication that the wedge code path within
intel_gt_reset() jumps around so much that it results in nested
reset_prepare/_finish calls. That is, the call sequence is:
intel_gt_reset
| reset_prepare
| __intel_gt_set_wedged
| | reset_prepare
| | reset_finish
| reset_finish
The nested finish means that even if the clear of the in-progress flag
was moved to the end of _finish, it would still be clear for the
entire second call. Surprisingly, this does not seem to be causing any
other problems at present.
As an aside, a wedge on fini does not call the finish functions at
all. The reset_in_progress flag is left set (twice).
So instead of trying to cancel the worker anywhere at all in the reset
path, just add a cancel to intel_guc_submission_fini instead. Note
that it is not a problem if the worker is still active during a reset.
Either it will run before the reset path starts locking things and
will simply block the reset code for a tiny amount of time. Or it will
run after the locks have been acquired and will early exit due to the
try-lock.
Also, do not use the reset-in-progress flag to decide whether a
synchronous cancel is safe (from a lockdep perspective) or not.
Instead, use the actual reset mutex state (both the genuine one and
the custom rolled BACKOFF one).
Fixes: 0e00a8814e ("drm/i915/guc: Avoid circular locking issue on busyness flush")
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Cc: Zhanjun Dong <zhanjun.dong@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: Madhumitha Tolakanahalli Pradeep <madhumitha.tolakanahalli.pradeep@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Cc: Dnyaneshwar Bhadane <dnyaneshwar.bhadane@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240329235306.1559639-1-John.C.Harrison@Intel.com
(cherry picked from commit 3563d85531)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
The previous fix for the circlular lock splat about the busyness
worker wasn't quite complete. Even though the reset-in-progress flag
is cleared at the start of intel_uc_reset_finish, the entire function
is still inside the reset mutex lock. Not sure why the patch appeared
to fix the issue both locally and in CI. However, it is now back
again.
There is a further complication that the wedge code path within
intel_gt_reset() jumps around so much that it results in nested
reset_prepare/_finish calls. That is, the call sequence is:
intel_gt_reset
| reset_prepare
| __intel_gt_set_wedged
| | reset_prepare
| | reset_finish
| reset_finish
The nested finish means that even if the clear of the in-progress flag
was moved to the end of _finish, it would still be clear for the
entire second call. Surprisingly, this does not seem to be causing any
other problems at present.
As an aside, a wedge on fini does not call the finish functions at
all. The reset_in_progress flag is left set (twice).
So instead of trying to cancel the worker anywhere at all in the reset
path, just add a cancel to intel_guc_submission_fini instead. Note
that it is not a problem if the worker is still active during a reset.
Either it will run before the reset path starts locking things and
will simply block the reset code for a tiny amount of time. Or it will
run after the locks have been acquired and will early exit due to the
try-lock.
Also, do not use the reset-in-progress flag to decide whether a
synchronous cancel is safe (from a lockdep perspective) or not.
Instead, use the actual reset mutex state (both the genuine one and
the custom rolled BACKOFF one).
Fixes: 0e00a8814e ("drm/i915/guc: Avoid circular locking issue on busyness flush")
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Cc: Zhanjun Dong <zhanjun.dong@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Cc: Andi Shyti <andi.shyti@linux.intel.com>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Nirmoy Das <nirmoy.das@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Cc: Andrzej Hajda <andrzej.hajda@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: Prathap Kumar Valsan <prathap.kumar.valsan@intel.com>
Cc: Alan Previn <alan.previn.teres.alexis@intel.com>
Cc: Madhumitha Tolakanahalli Pradeep <madhumitha.tolakanahalli.pradeep@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Cc: Dnyaneshwar Bhadane <dnyaneshwar.bhadane@intel.com>
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240329235306.1559639-1-John.C.Harrison@Intel.com
Allow user to provide a low latency context hint. When set, KMD
sends a hint to GuC which results in special handling for this
context. SLPC will ramp the GT frequency aggressively every time
it switches to this context. The down freq threshold will also be
lower so GuC will ramp down the GT freq for this context more slowly.
We also disable waitboost for this context as that will interfere with
the strategy.
We need to enable the use of SLPC Compute strategy during init, but
it will apply only to contexts that set this bit during context
creation.
Userland can check whether this feature is supported using a new param-
I915_PARAM_HAS_CONTEXT_FREQ_HINT. This flag is true for all guc submission
enabled platforms as they use SLPC for frequency management.
The Mesa usage model for this flag is here -
https://gitlab.freedesktop.org/sushmave/mesa/-/commits/compute_hint
v2: Rename flags as per review suggestions (Rodrigo, Tvrtko).
Also, use flag bits in intel_context as it allows finer control for
toggling per engine if needed (Tvrtko).
v3: Minor review comments (Tvrtko)
v4: Update comment (Sushma)
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@linux.intel.com>
Cc: Sushma Venkatesh Reddy <sushma.venkatesh.reddy@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Acked-by: Ivan Briano <ivan.briano@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240306012759.204938-1-vinay.belgaumkar@intel.com
While trying to reproduce some other issues reported by CI for i915
hangcheck live selftest, I found them hidden behind timeout failures
reported by igt_hang_sanitycheck -- the very first hangcheck test case
executed.
Feb 22 19:49:06 DUT1394ACMR kernel: calling mei_gsc_driver_init+0x0/0xff0 [mei_gsc] @ 121074
Feb 22 19:49:06 DUT1394ACMR kernel: i915 0000:03:00.0: [drm] DRM_I915_DEBUG enabled
Feb 22 19:49:06 DUT1394ACMR kernel: i915 0000:03:00.0: [drm] Cannot find any crtc or sizes
Feb 22 19:49:06 DUT1394ACMR kernel: probe of i915.mei-gsc.768 returned 0 after 1475 usecs
Feb 22 19:49:06 DUT1394ACMR kernel: probe of i915.mei-gscfi.768 returned 0 after 1441 usecs
Feb 22 19:49:06 DUT1394ACMR kernel: initcall mei_gsc_driver_init+0x0/0xff0 [mei_gsc] returned 0 after 3010 usecs
Feb 22 19:49:06 DUT1394ACMR kernel: i915 0000:03:00.0: [drm] DRM_I915_DEBUG_GEM enabled
Feb 22 19:49:06 DUT1394ACMR kernel: i915 0000:03:00.0: [drm] DRM_I915_DEBUG_RUNTIME_PM enabled
Feb 22 19:49:06 DUT1394ACMR kernel: i915: Performing live selftests with st_random_seed=0x4c26c048 st_timeout=500
Feb 22 19:49:07 DUT1394ACMR kernel: i915: Running hangcheck
Feb 22 19:49:07 DUT1394ACMR kernel: calling mei_hdcp_driver_init+0x0/0xff0 [mei_hdcp] @ 121074
Feb 22 19:49:07 DUT1394ACMR kernel: i915: Running intel_hangcheck_live_selftests/igt_hang_sanitycheck
Feb 22 19:49:07 DUT1394ACMR kernel: probe of 0000:00:16.0-b638ab7e-94e2-4ea2-a552-d1c54b627f04 returned 0 after 1398 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: probe of i915.mei-gsc.768-b638ab7e-94e2-4ea2-a552-d1c54b627f04 returned 0 after 97 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: initcall mei_hdcp_driver_init+0x0/0xff0 [mei_hdcp] returned 0 after 101960 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: calling mei_pxp_driver_init+0x0/0xff0 [mei_pxp] @ 121094
Feb 22 19:49:07 DUT1394ACMR kernel: probe of 0000:00:16.0-fbf6fcf1-96cf-4e2e-a6a6-1bab8cbe36b1 returned 0 after 435 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: mei_pxp i915.mei-gsc.768-fbf6fcf1-96cf-4e2e-a6a6-1bab8cbe36b1: bound 0000:03:00.0 (ops i915_pxp_tee_component_ops [i915])
Feb 22 19:49:07 DUT1394ACMR kernel: 100ms wait for request failed on rcs0, err=-62
Feb 22 19:49:07 DUT1394ACMR kernel: probe of i915.mei-gsc.768-fbf6fcf1-96cf-4e2e-a6a6-1bab8cbe36b1 returned 0 after 158425 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: initcall mei_pxp_driver_init+0x0/0xff0 [mei_pxp] returned 0 after 224159 usecs
Feb 22 19:49:07 DUT1394ACMR kernel: i915/intel_hangcheck_live_selftests: igt_hang_sanitycheck failed with error -5
Feb 22 19:49:07 DUT1394ACMR kernel: i915: probe of 0000:03:00.0 failed with error -5
Those request waits, once timed out after 100ms, have never been
confirmed to still persist over another 100ms, always being able to
complete within the originally requested wait time doubled.
Taking into account potentially significant additional concurrent workload
generated by new auxiliary drivers that didn't exist before and now are
loaded in parallel with the i915 module also when loaded in selftest mode,
relax our expectations on time consumed by the sanity check request before
it completes.
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240228152500.38267-2-janusz.krzysztofik@linux.intel.com
The EIR register (0x20B0) was being included in the engine class list
for render and compute as the absolute register address. However, it
is actually a ring register available on all engines at an offset of
(base) + 0xB0. As it was included as an RCS engine but with the
absolute address, GuC was adding on another 0x2000 and coming out at
an invalid location. Thus it would reject the register and complain
about only managing a partial capture.
So update the list to use the RING_EIR version of the register and
include it for all engines.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240223203204.1533410-1-John.C.Harrison@Intel.com
The context persistence code does things like send super high priority
heartbeat pulses to ensure any leaked context can still be pre-empted
and thus isn't a total denial of service but only a minor denial of
service. Unfortunately, it wasn't bothering to restart the heartbeat
worker with a fresh timeout. Thus, if a persistent context happened to
be closed just before the heartbeat was going to go ping anyway then
the forced pulse would get a negligble execution time. And as the
forced pulse is super high priority, the worker thread's next step is
a reset. Which means a potentially innocent system randomly goes boom
when attempting to close a context. So, force a re-schedule of the
worker thread with the appropriate timeout.
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240110210216.4125092-1-John.C.Harrison@Intel.com
