Infinite waits for completion of GPU activity have been observed in CI,
mostly inside __i915_active_wait(), triggered by igt@gem_barrier_race or
igt@perf@stress-open-close. Root cause analysis, based of ftrace dumps
generated with a lot of extra trace_printk() calls added to the code,
revealed loops of request dependencies being accidentally built,
preventing the requests from being processed, each waiting for completion
of another one's activity.
After we substitute a new request for a last active one tracked on a
timeline, we set up a dependency of our new request to wait on completion
of current activity of that previous one. While doing that, we must take
care of keeping the old request still in memory until we use its
attributes for setting up that await dependency, or we can happen to set
up the await dependency on an unrelated request that already reuses the
memory previously allocated to the old one, already released. Combined
with perf adding consecutive kernel context remote requests to different
user context timelines, unresolvable loops of await dependencies can be
built, leading do infinite waits.
We obtain a pointer to the previous request to wait upon when we
substitute it with a pointer to our new request in an active tracker,
e.g. in intel_timeline.last_request. In some processing paths we protect
that old request from being freed before we use it by getting a reference
to it under RCU protection, but in others, e.g. __i915_request_commit()
-> __i915_request_add_to_timeline() -> __i915_request_ensure_ordering(),
we don't. But anyway, since the requests' memory is SLAB_FAILSAFE_BY_RCU,
that RCU protection is not sufficient against reuse of memory.
We could protect i915_request's memory from being prematurely reused by
calling its release function via call_rcu() and using rcu_read_lock()
consequently, as proposed in v1. However, that approach leads to
significant (up to 10 times) increase of SLAB utilization by i915_request
SLAB cache. Another potential approach is to take a reference to the
previous active fence.
When updating an active fence tracker, we first lock the new fence,
substitute a pointer of the current active fence with the new one, then we
lock the substituted fence. With this approach, there is a time window
after the substitution and before the lock when the request can be
concurrently released by an interrupt handler and its memory reused, then
we may happen to lock and return a new, unrelated request.
Always get a reference to the current active fence first, before
replacing it with a new one. Having it protected from premature release
and reuse, lock it and then replace with the new one but only if not
yet signalled via a potential concurrent interrupt nor replaced with
another one by a potential concurrent thread, otherwise retry, starting
from getting a reference to the new current one. Adjust users to not
get a reference to the previous active fence themselves and always put the
reference got by __i915_active_fence_set() when no longer needed.
v3: Fix lockdep splat reports and other issues caused by incorrect use of
try_cmpxchg() (use (cmpxchg() != prev) instead)
v2: Protect request's memory by getting a reference to it in favor of
delegating its release to call_rcu() (Chris)
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/8211
Fixes: df9f85d858 ("drm/i915: Serialise i915_active_fence_set() with itself")
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Cc: <stable@vger.kernel.org> # v5.6+
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/20230720093543.832147-2-janusz.krzysztofik@linux.intel.com
On MTL, if the GSC Proxy init flows haven't completed, submissions to the
GSC engine will fail. Those init flows are dependent on the mei's
gsc_proxy component that is loaded in parallel with i915 and a
worker that could potentially start after i915 driver init is done.
That said, all subsytems that access the GSC engine today does check
for such init flow completion before using the GSC engine. However,
selftests currently don't wait on anything before starting.
To fix this, add a waiter function at the start of __run_selftests
that waits for gsc-proxy init flows to complete. Selftests shouldn't
care if the proxy-init failed as that should be flagged elsewhere.
Difference from prior versions:
v7: - Change the fw status to INTEL_UC_FIRMWARE_LOAD_FAIL if the
proxy-init fails so that intel_gsc_uc_fw_proxy_get_status
catches it. (Daniele)
v6: - Add a helper that returns something more than a boolean
so we selftest can stop waiting if proxy-init hadn't
completed but failed (Daniele).
v5: - Move the call to __wait_gsc_proxy_completed from common
__run_selftests dispatcher to the group-level selftest
function (Trvtko).
- change the pr_info to pr_warn if we hit the timeout.
v4: - Remove generalized waiters function table framework (Tvrtko).
- Remove mention of CI-framework-timeout from comments (Tvrtko).
v3: - Rebase to latest drm-tip.
v2: - Based on internal testing, increase the timeout for gsc-proxy
specific case to 8 seconds.
Signed-off-by: Alan Previn <alan.previn.teres.alexis@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230720230126.375566-1-alan.previn.teres.alexis@intel.com
A proposed update to clang's -Wconstant-logical-operand to warn when the
left hand side is a constant shows the following instance in
nsecs_to_jiffies_timeout() when NSEC_PER_SEC is not a multiple of HZ,
such as CONFIG_HZ=300:
drivers/gpu/drm/i915/gem/i915_gem_wait.c:189:24: warning: use of logical '&&' with constant operand [-Wconstant-logical-operand]
189 | if (NSEC_PER_SEC % HZ &&
| ~~~~~~~~~~~~~~~~~ ^
drivers/gpu/drm/i915/gem/i915_gem_wait.c:189:24: note: use '&' for a bitwise operation
189 | if (NSEC_PER_SEC % HZ &&
| ^~
| &
drivers/gpu/drm/i915/gem/i915_gem_wait.c:189:24: note: remove constant to silence this warning
1 warning generated.
Turn this into an explicit comparison against zero to make the
expression a boolean to make it clear this should be a logical check,
not a bitwise one.
Link: https://reviews.llvm.org/D142609
Signed-off-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Maíra Canal <mairacanal@riseup.net>
Link: https://patchwork.freedesktop.org/patch/msgid/20230718-nsecs_to_jiffies_timeout-constant-logical-operand-v1-2-36ed8fc8faea@kernel.org
MTL onwards Degamma LUT/PRE-CSC LUT precision has been increased from
16 bits to 24 bits. Currently, drm framework only supports LUTs up to 16
bit precision. Until a new uapi comes along to support higher bitdepth,
upscale the values sent from userland to 24 bit before writing into the
HW to continue supporting degamma on MTL.
Add helper function to upscale or downscale lut values. Parameters
'to' and 'from' needs to be less than 32. This should be sufficient
as currently there are no lut values exceeding 32 bit.
v2: (Jani)
- Reuse glk_load_degamma_lut()
- Create a helper function for upscaling values
v3: Fix multi line comment style (Uma)
v4: Remove extra line(Ankit)
Signed-off-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230727123342.4077017-1-chaitanya.kumar.borah@intel.com
Due to a change in the auth flow on MTL, GuC 70.7.0 and newer will only
be able to authenticate HuC 8.5.1 and newer. The plan is to update the 2
binaries synchronously in linux-firmware so that the fw repo always has
a matching pair that works; still, it's better to check in the kernel so
we can print an error message and abort HuC loading if the binaries are
out of sync instead of failing the authentication.
v2: Add clarification comment, fix typo in commit msg, clean up variable
declaration (John)
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Andrzej Hajda <andrzej.hajda@intel.com> #v1
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230717234905.117114-1-daniele.ceraolospurio@intel.com
drm-misc-next for v6.6:
UAPI Changes:
* fbdev:
* Make fbdev userspace interfaces optional; only leaves the
framebuffer console active
* prime:
* Support dma-buf self-import for all drivers automatically: improves
support for many userspace compositors
Cross-subsystem Changes:
* backlight:
* Fix interaction with fbdev in several drivers
* base: Convert struct platform.remove to return void; part of a larger,
tree-wide effort
* dma-buf: Acquire reservation lock for mmap() in exporters; part
of an on-going effort to simplify locking around dma-bufs
* fbdev:
* Use Linux device instead of fbdev device in many places
* Use deferred-I/O helper macros in various drivers
* i2c: Convert struct i2c from .probe_new to .probe; part of a larger,
tree-wide effort
* video:
* Avoid including <linux/screen_info.h>
Core Changes:
* atomic:
* Improve logging
* prime:
* Remove struct drm_driver.gem_prime_mmap plus driver updates: all
drivers now implement this callback with drm_gem_prime_mmap()
* gem:
* Support execution contexts: provides locking over multiple GEM
objects
* ttm:
* Support init_on_free
* Swapout fixes
Driver Changes:
* accel:
* ivpu: MMU updates; Support debugfs
* ast:
* Improve device-model detection
* Cleanups
* bridge:
* dw-hdmi: Improve support for YUV420 bus format
* dw-mipi-dsi: Fix enable/disable of DSI controller
* lt9611uxc: Use MODULE_FIRMWARE()
* ps8640: Remove broken EDID code
* samsung-dsim: Fix command transfer
* tc358764: Handle HS/VS polarity; Use BIT() macro; Various cleanups
* Cleanups
* ingenic:
* Kconfig REGMAP fixes
* loongson:
* Support display controller
* mgag200:
* Minor fixes
* mxsfb:
* Support disabling overlay planes
* nouveau:
* Improve VRAM detection
* Various fixes and cleanups
* panel:
* panel-edp: Support AUO B116XAB01.4
* Support Visionox R66451 plus DT bindings
* Cleanups
* ssd130x:
* Support per-controller default resolution plus DT bindings
* Reduce memory-allocation overhead
* Cleanups
* tidss:
* Support TI AM625 plus DT bindings
* Implement new connector model plus driver updates
* vkms
* Improve write-back support
* Documentation fixes
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
From: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20230713090830.GA23281@linux-uq9g
We are currently having FIFO underruns happening for kms_dsc test case,
problem is that, we check if curreny cdclk is >= pixel rate only if
there is a single VDSC engine enabled(i.e dsc_split=false) however if
we happen to have 2 VDSC engines enabled, we just kinda rely that this
would be automatically enough.
However pixel rate can be even >= than VDSC clock(cdclk) * 2, so in that
case even with 2 VDSC engines enabled, we still need to tweak it up.
So lets compare pixel rate with cdclk * VDSC engine count and
check if it still requires bumping up.
Previously we had to bump up CDCLK many times for similar reasons.
v2: - Use new intel_dsc_get_num_vdsc_instances to determine number of VDSC
engines, instead of slice count(Ankit Nautiyal)
v3: - s/u8/int/ (Jani Nikula)
v4: - Remove slice count mentions(Ankit Nautiyal)
- Use DIV_ROUND_UP in order to make sure that resulting CDCLK would
be always >= than required, after division(Ankit Nautiyal)
Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20230704131758.14024-3-stanislav.lisovskiy@intel.com
