drm-misc-fixes for v6.16-rc6 or final:
- Fix nouveau fail on debugfs errors.
- Magic 50 ms to fix nouveau suspend.
- Call rust destructor on drm device release.
- Fix DMA api error handling in tegra/nvdec.
- Fix PVR device reset.
- Habanalabs maintainer update.
- Small memory leak fix when nouveau acpi init fails.
- Do not attempt to bind to any PCI device with AGP capability.
- Make FB's acquire handles on backing object, same as i915/xe already does.
- Fix race in drm_gem_handle_create_tail.
Signed-off-by: Simona Vetter <simona.vetter@ffwll.ch>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/e522cdc7-1787-48f2-97e5-0f94783970ab@linux.intel.com
Driver Changes:
- Clear LMTT page to avoid leaking data from one VF to another
- Align PF queue size to power of 2
- Disable Indirect Ring State to avoid intermittent issues on context
switch: feature is not currently needed, so can be disabled for now.
- Fix compression handling when the BO pages are very fragmented
- Restore display pm on error path
- Fix runtime pm handling in xe devcoredump
- Fix xe_pm_set_vram_threshold() doc
- Recommend new minor versions of GuC firmware
- Drop some workarounds on VF
- Do not use verbose GuC logging by default: it should be only for
debugging
Signed-off-by: Simona Vetter <simona.vetter@ffwll.ch>
From: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/s6jyd24mimbzb4vxtgc5vupvbyqplfep2c6eupue7znnlbhuxy@lmvzexfzhrnn
Currently xe sets the guc log level to a verbose level since it's useful
to debug hangs and general development. However the verbose level may
already be too much and affect performance.
Michal Mrozek did some tests with the L0 compute stack for submission
latency with ULLS disabled. Below are the normalized numbers with log
level 3 (the current default) as baseline for each test:
Test \ Log Level 3 0 1 2
----------------------------------------------------------- ------ ------ ------ ------
BestWalkerNthCommandListSubmission(CmdListCount=2) 1.00 0.63 0.63 0.96
BestWalkerNthSubmission(KernelCount=2) 1.00 0.62 0.63 0.96
BestWalkerNthSubmissionImmediate(KernelCount=2) 1.00 0.58 0.58 0.85
BestWalkerSubmission 1.00 0.62 0.62 0.96
BestWalkerSubmissionImmediate 1.00 0.63 0.62 0.96
BestWalkerSubmissionImmediateMultiCmdlists(cmdlistCount=2) 1.00 0.58 0.58 0.86
BestWalkerSubmissionImmediateMultiCmdlists(cmdlistCount=4) 1.00 0.70 0.70 0.83
BestWalkerSubmissionImmediateMultiCmdlists(cmdlistCount=8) 1.00 0.53 0.52 0.78
Log level 2 is the first "verbose level" for GuC, where the biggest
difference happens. Keep log level 3 for CONFIG_DRM_XE_DEBUG, but switch
to 1, i.e. GUC_LOG_LEVEL_NON_VERBOSE, for "normal" builds.
Cc: Michal Mrozek <michal.mrozek@intel.com>
Cc: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: John Harrison <John.C.Harrison@Intel.com>
Link: https://lore.kernel.org/r/20250613-guc-log-level-v2-1-cb84a63e49fe@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit a37128ba61)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
It turns out that the fixup from vlv_fixup_mipi_sequences() is necessary
for some DSI panel's with version 2 mipi-sequences too.
Specifically the Acer Iconia One 8 A1-840 (not to be confused with the
A1-840FHD which is different) has the following sequences:
BDB block 53 (1284 bytes) - MIPI sequence block:
Sequence block version v2
Panel 0 *
Sequence 2 - MIPI_SEQ_INIT_OTP
GPIO index 9, source 0, set 0 (0x00)
Delay: 50000 us
GPIO index 9, source 0, set 1 (0x01)
Delay: 6000 us
GPIO index 9, source 0, set 0 (0x00)
Delay: 6000 us
GPIO index 9, source 0, set 1 (0x01)
Delay: 25000 us
Send DCS: Port A, VC 0, LP, Type 39, Length 5, Data ff aa 55 a5 80
Send DCS: Port A, VC 0, LP, Type 39, Length 3, Data 6f 11 00
...
Send DCS: Port A, VC 0, LP, Type 05, Length 1, Data 29
Delay: 120000 us
Sequence 4 - MIPI_SEQ_DISPLAY_OFF
Send DCS: Port A, VC 0, LP, Type 05, Length 1, Data 28
Delay: 105000 us
Send DCS: Port A, VC 0, LP, Type 05, Length 2, Data 10 00
Delay: 10000 us
Sequence 5 - MIPI_SEQ_ASSERT_RESET
Delay: 10000 us
GPIO index 9, source 0, set 0 (0x00)
Notice how there is no MIPI_SEQ_DEASSERT_RESET, instead the deassert
is done at the beginning of MIPI_SEQ_INIT_OTP, which is exactly what
the fixup from vlv_fixup_mipi_sequences() fixes up.
Extend it to also apply to v2 sequences, this fixes the panel not working
on the Acer Iconia One 8 A1-840.
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/14605
Signed-off-by: Hans de Goede <hansg@kernel.org>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Link: https://lore.kernel.org/r/20250703143824.7121-1-hansg@kernel.org
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 11895f3759)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Object creation is a careful dance where we must guarantee that the
object is fully constructed before it is visible to other threads, and
GEM buffer objects are no difference.
Final publishing happens by calling drm_gem_handle_create(). After
that the only allowed thing to do is call drm_gem_object_put() because
a concurrent call to the GEM_CLOSE ioctl with a correctly guessed id
(which is trivial since we have a linear allocator) can already tear
down the object again.
Luckily most drivers get this right, the very few exceptions I've
pinged the relevant maintainers for. Unfortunately we also need
drm_gem_handle_create() when creating additional handles for an
already existing object (e.g. GETFB ioctl or the various bo import
ioctl), and hence we cannot have a drm_gem_handle_create_and_put() as
the only exported function to stop these issues from happening.
Now unfortunately the implementation of drm_gem_handle_create() isn't
living up to standards: It does correctly finishe object
initialization at the global level, and hence is safe against a
concurrent tear down. But it also sets up the file-private aspects of
the handle, and that part goes wrong: We fully register the object in
the drm_file.object_idr before calling drm_vma_node_allow() or
obj->funcs->open, which opens up races against concurrent removal of
that handle in drm_gem_handle_delete().
Fix this with the usual two-stage approach of first reserving the
handle id, and then only registering the object after we've completed
the file-private setup.
Jacek reported this with a testcase of concurrently calling GEM_CLOSE
on a freshly-created object (which also destroys the object), but it
should be possible to hit this with just additional handles created
through import or GETFB without completed destroying the underlying
object with the concurrent GEM_CLOSE ioctl calls.
Note that the close-side of this race was fixed in f6cd7daecf ("drm:
Release driver references to handle before making it available
again"), which means a cool 9 years have passed until someone noticed
that we need to make this symmetry or there's still gaps left :-/
Without the 2-stage close approach we'd still have a race, therefore
that's an integral part of this bugfix.
More importantly, this means we can have NULL pointers behind
allocated id in our drm_file.object_idr. We need to check for that
now:
- drm_gem_handle_delete() checks for ERR_OR_NULL already
- drm_gem.c:object_lookup() also chekcs for NULL
- drm_gem_release() should never be called if there's another thread
still existing that could call into an IOCTL that creates a new
handle, so cannot race. For paranoia I added a NULL check to
drm_gem_object_release_handle() though.
- most drivers (etnaviv, i915, msm) are find because they use
idr_find(), which maps both ENOENT and NULL to NULL.
- drivers using idr_for_each_entry() should also be fine, because
idr_get_next does filter out NULL entries and continues the
iteration.
- The same holds for drm_show_memory_stats().
v2: Use drm_WARN_ON (Thomas)
Reported-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Tested-by: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Cc: stable@vger.kernel.org
Cc: Jacek Lawrynowicz <jacek.lawrynowicz@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Maxime Ripard <mripard@kernel.org>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Signed-off-by: Simona Vetter <simona.vetter@intel.com>
Signed-off-by: Simona Vetter <simona.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20250707151814.603897-1-simona.vetter@ffwll.ch
There looks to be an issue in our compression handling when the BO pages
are very fragmented, where we choose to skip the identity map and
instead fall back to emitting the PTEs by hand when migrating memory,
such that we can hopefully do more work per blit operation. However in
such a case we need to ensure the src PTEs are correctly tagged with a
compression enabled PAT index on dgpu xe2+, otherwise the copy will
simply treat the src memory as uncompressed, leading to corruption if
the memory was compressed by the user.
To fix this pass along use_comp_pat into emit_pte() on the src side, to
indicate that compression should be considered.
v2 (Jonathan): tweak the commit message
Fixes: 523f191cc0 ("drm/xe/xe_migrate: Handle migration logic for xe2+ dgfx")
Signed-off-by: Matthew Auld <matthew.auld@intel.com>
Cc: Himal Prasad Ghimiray <himal.prasad.ghimiray@intel.com>
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Akshata Jahagirdar <akshata.jahagirdar@intel.com>
Cc: <stable@vger.kernel.org> # v6.12+
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Link: https://lore.kernel.org/r/20250701103949.83116-2-matthew.auld@intel.com
(cherry picked from commit f7a2fd776e)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
This reverts commit fe0154cf82.
Seeing some unexplained random failures during LRC context switches with
indirect ring state enabled. The failures were always there, but the
repro rate increased with the addition of WA BB as a separate BO.
Commit 3a1edef8f4 ("drm/xe: Make WA BB part of LRC BO") helped to
reduce the issues in the context switches, but didn't eliminate them
completely.
Indirect ring state is not required for any current features, so disable
for now until failures can be root caused.
Cc: stable@vger.kernel.org
Fixes: fe0154cf82 ("drm/xe/xe2: Enable Indirect Ring State support for Xe2")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/20250702035846.3178344-1-matthew.brost@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit 03d85ab36b)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Our LMEM buffer objects are not cleared by default on alloc
and during VF provisioning we only setup LMTT PTEs for the
actually provisioned LMEM range. But beyond that valid range
we might leave some stale data that could either point to some
other VFs allocations or even to the PF pages.
Explicitly clear all new LMTT page to avoid the risk that a
malicious VF would try to exploit that gap.
While around add asserts to catch any undesired PTE overwrites
and low-level debug traces to track LMTT PT life-cycle.
Fixes: b1d2040582 ("drm/xe/pf: Introduce Local Memory Translation Table")
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Lukasz Laguna <lukasz.laguna@intel.com>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Piotr Piórkowski <piotr.piorkowski@intel.com>
Link: https://lore.kernel.org/r/20250701220052.1612-1-michal.wajdeczko@intel.com
(cherry picked from commit 3fae6918a3)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
The GPU hard reset sequence calls pm_runtime_force_suspend() and
pm_runtime_force_resume(), which according to their documentation should
only be used during system-wide PM transitions to sleep states.
The main issue though is that depending on some internal runtime PM
state as seen by pm_runtime_force_suspend() (whether the usage count is
<= 1), pm_runtime_force_resume() might not resume the device unless
needed. If that happens, the runtime PM resume callback
pvr_power_device_resume() is not called, the GPU clocks are not
re-enabled, and the kernel crashes on the next attempt to access GPU
registers as part of the power-on sequence.
Replace calls to pm_runtime_force_suspend() and
pm_runtime_force_resume() with direct calls to the driver's runtime PM
callbacks, pvr_power_device_suspend() and pvr_power_device_resume(),
to ensure clocks are re-enabled and avoid the kernel crash.
Fixes: cc1aeedb98 ("drm/imagination: Implement firmware infrastructure and META FW support")
Signed-off-by: Alessio Belle <alessio.belle@imgtec.com>
Reviewed-by: Matt Coster <matt.coster@imgtec.com>
Link: https://lore.kernel.org/r/20250624-fix-kernel-crash-gpu-hard-reset-v1-1-6d24810d72a6@imgtec.com
Cc: stable@vger.kernel.org
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Driver Changes:
- Fix chunking the PTE updates and overflowing the maximum number of
dwords with with MI_STORE_DATA_IMM (Jia Yao)
- Move WA BB to the LRC BO to mitigate hangs on context switch (Matthew
Brost)
- Fix frequency/flush WAs for BMG (Vinay / Lucas)
- Fix kconfig prompt title and description (Lucas)
- Do not require kunit (Harry Austen / Lucas)
- Extend 14018094691 WA to BMG (Daniele)
- Fix wedging the device on signal (Matthew Brost)
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://lore.kernel.org/r/o5662wz6nrlf6xt5sjgxq5oe6qoujefzywuwblm3m626hreifv@foqayqydd6ig
Fixups
- Fixed raw pointer leakage and unsafe behavior in printk()
. Switch from %pK to %p for pointer formatting, as %p is now safer
and prevents issues like raw pointer leakage and acquiring sleeping
locks in atomic contexts.
- Fixed kernel panic during boot
. A NULL pointer dereference issue occasionally occurred
when the vblank interrupt handler was called before
the DRM driver was fully initialized during boot.
So this patch fixes the issue by adding a check in the interrupt handler
to ensure the DRM driver is properly initialized.
- Fixed a lockup issue on Samsung Peach-Pit/Pi Chromebooks
. The issue occurred after commit c9b1150a68 changed
the call order of CRTC enable/disable and bridge pre_enable/post_disable
methods, causing fimd_dp_clock_enable() to be called
before the FIMD device was activated. To fix this,
runtime PM guards were added to fimd_dp_clock_enable()
to ensure proper operation even when CRTC is not enabled.
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Inki Dae <inki.dae@samsung.com>
Link: https://lore.kernel.org/r/20250629083554.28628-1-inki.dae@samsung.com
This fixes a bunch of command hangs after runtime suspend/resume.
This fixes a regression caused by code movement in the commit below,
the commit seems to just change timings enough to cause this to happen
now, and adding the sleep seems to avoid it.
I've spent some time trying to root cause it to no great avail,
it seems like a bug on the firmware side, but it could be a bug
in our rpc handling that I can't find.
Either way, we should land the workaround to fix the problem,
while we continue to work out the root cause.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Cc: Ben Skeggs <bskeggs@nvidia.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Fixes: c21b039715 ("drm/nouveau/gsp: add hals for fbsr.suspend/resume()")
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
Link: https://lore.kernel.org/r/20250702232707.175679-1-airlied@gmail.com
If CONFIG_DEBUG_FS is enabled, nouveau_drm_init() returns an error if it
fails to create the "nouveau" directory in debugfs. One case where that
will happen is when debugfs access is restricted by
CONFIG_DEBUG_FS_ALLOW_NONE or by the boot parameter debugfs=off, which
cause the debugfs APIs to return -EPERM.
So just ignore errors from debugfs. Note that nouveau_debugfs_root may
be an error now, but that is a standard pattern for debugfs. From
include/linux/debugfs.h:
"NOTE: it's expected that most callers should _ignore_ the errors
returned by this function. Other debugfs functions handle the fact that
the "dentry" passed to them could be an error and they don't crash in
that case. Drivers should generally work fine even if debugfs fails to
init anyway."
Fixes: 97118a1816 ("drm/nouveau: create module debugfs root")
Cc: stable@vger.kernel.org
Signed-off-by: Aaron Thompson <dev@aaront.org>
Acked-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
Link: https://lore.kernel.org/r/20250703211949.9916-1-dev@aaront.org
A GEM handle can be released while the GEM buffer object is attached
to a DRM framebuffer. This leads to the release of the dma-buf backing
the buffer object, if any. [1] Trying to use the framebuffer in further
mode-setting operations leads to a segmentation fault. Most easily
happens with driver that use shadow planes for vmap-ing the dma-buf
during a page flip. An example is shown below.
[ 156.791968] ------------[ cut here ]------------
[ 156.796830] WARNING: CPU: 2 PID: 2255 at drivers/dma-buf/dma-buf.c:1527 dma_buf_vmap+0x224/0x430
[...]
[ 156.942028] RIP: 0010:dma_buf_vmap+0x224/0x430
[ 157.043420] Call Trace:
[ 157.045898] <TASK>
[ 157.048030] ? show_trace_log_lvl+0x1af/0x2c0
[ 157.052436] ? show_trace_log_lvl+0x1af/0x2c0
[ 157.056836] ? show_trace_log_lvl+0x1af/0x2c0
[ 157.061253] ? drm_gem_shmem_vmap+0x74/0x710
[ 157.065567] ? dma_buf_vmap+0x224/0x430
[ 157.069446] ? __warn.cold+0x58/0xe4
[ 157.073061] ? dma_buf_vmap+0x224/0x430
[ 157.077111] ? report_bug+0x1dd/0x390
[ 157.080842] ? handle_bug+0x5e/0xa0
[ 157.084389] ? exc_invalid_op+0x14/0x50
[ 157.088291] ? asm_exc_invalid_op+0x16/0x20
[ 157.092548] ? dma_buf_vmap+0x224/0x430
[ 157.096663] ? dma_resv_get_singleton+0x6d/0x230
[ 157.101341] ? __pfx_dma_buf_vmap+0x10/0x10
[ 157.105588] ? __pfx_dma_resv_get_singleton+0x10/0x10
[ 157.110697] drm_gem_shmem_vmap+0x74/0x710
[ 157.114866] drm_gem_vmap+0xa9/0x1b0
[ 157.118763] drm_gem_vmap_unlocked+0x46/0xa0
[ 157.123086] drm_gem_fb_vmap+0xab/0x300
[ 157.126979] drm_atomic_helper_prepare_planes.part.0+0x487/0xb10
[ 157.133032] ? lockdep_init_map_type+0x19d/0x880
[ 157.137701] drm_atomic_helper_commit+0x13d/0x2e0
[ 157.142671] ? drm_atomic_nonblocking_commit+0xa0/0x180
[ 157.147988] drm_mode_atomic_ioctl+0x766/0xe40
[...]
[ 157.346424] ---[ end trace 0000000000000000 ]---
Acquiring GEM handles for the framebuffer's GEM buffer objects prevents
this from happening. The framebuffer's cleanup later puts the handle
references.
Commit 1a148af060 ("drm/gem-shmem: Use dma_buf from GEM object
instance") triggers the segmentation fault easily by using the dma-buf
field more widely. The underlying issue with reference counting has
been present before.
v2:
- acquire the handle instead of the BO (Christian)
- fix comment style (Christian)
- drop the Fixes tag (Christian)
- rename err_ gotos
- add missing Link tag
Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://elixir.bootlin.com/linux/v6.15/source/drivers/gpu/drm/drm_gem.c#L241 # [1]
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Cc: Anusha Srivatsa <asrivats@redhat.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Maxime Ripard <mripard@kernel.org>
Cc: Sumit Semwal <sumit.semwal@linaro.org>
Cc: "Christian König" <christian.koenig@amd.com>
Cc: linux-media@vger.kernel.org
Cc: dri-devel@lists.freedesktop.org
Cc: linaro-mm-sig@lists.linaro.org
Cc: <stable@vger.kernel.org>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20250630084001.293053-1-tzimmermann@suse.de
Limit GT max frequency to 2600MHz and wait for frequency to reduce
before proceeding with a transient flush. This is really only needed for
the transient flush: if L2 flush is needed due to 16023588340 then
there's no need to do this additional wait since we are already using
the bigger hammer.
v2: Use generic names, ensure user set max frequency requests wait
for flush to complete (Rodrigo)
v3:
- User requests wait via wait_var_event_timeout (Lucas)
- Close races on flush + user requests (Lucas)
- Fix xe_guc_pc_remove_flush_freq_limit() being called on last gt
rather than root gt (Lucas)
v4:
- Only apply the freq reducing part if a TDF is needed: L2 flush trumps
the need for waiting a lower frequency
Fixes: aaa08078e7 ("drm/xe/bmg: Apply Wa_22019338487")
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Vinay Belgaumkar <vinay.belgaumkar@intel.com>
Link: https://lore.kernel.org/r/20250618-wa-22019338487-v5-4-b888388477f2@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit deea6a7d6d)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
pc_set_mert_freq_cap() currently lock()/unlock() the mutex multiple times
to stash the current frequencies. It's not a problem since
xe_guc_pc_restore_stashed_freq() is guaranteed to be called only later
in the init sequence. However, now that we have _locked() variants for
this functions, use them and avoid potential issues when called from
other places or using the same pattern.
While at it, prefer and early return for the WA check to reduce
indentation.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://lore.kernel.org/r/20250618-wa-22019338487-v5-2-b888388477f2@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit d878c97daa)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
There are places in which the getters/setters are called one after the
other causing a multiple lock()/unlock(). These are not currently a
problem since they are all happening from the same thread, but there's a
race possibility as calls are added outside of the early init when the
max/min and stashed values need to be correlated.
Add the _locked() variants to prepare for that.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://lore.kernel.org/r/20250618-wa-22019338487-v5-1-b888388477f2@intel.com
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
(cherry picked from commit 1beae9aa2b)
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
MEI GSC interrupt comes from i915. It has top half and bottom half.
Top half is called from i915 interrupt handler. It should be in
irq disabled context.
With RT kernel, by default i915 IRQ handler is in threaded IRQ. MEI GSC
top half might be in threaded IRQ context. generic_handle_irq_safe API
could be called from either IRQ or process context, it disables local
IRQ then calls MEI GSC interrupt top half.
This change fixes A380/A770 GPU boot hang issue with RT kernel.
Fixes: 1e3dc1d862 ("drm/i915/gsc: add gsc as a mei auxiliary device")
Tested-by: Furong Zhou <furong.zhou@intel.com>
Suggested-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Junxiao Chang <junxiao.chang@intel.com>
Link: https://lore.kernel.org/r/20250425151108.643649-1-junxiao.chang@intel.com
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit dccf655f69)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
The following error has been reported sporadically by CI when a test
unbinds the i915 driver on a ring submission platform:
<4> [239.330153] ------------[ cut here ]------------
<4> [239.330166] i915 0000:00:02.0: [drm] drm_WARN_ON(dev_priv->mm.shrink_count)
<4> [239.330196] WARNING: CPU: 1 PID: 18570 at drivers/gpu/drm/i915/i915_gem.c:1309 i915_gem_cleanup_early+0x13e/0x150 [i915]
...
<4> [239.330640] RIP: 0010:i915_gem_cleanup_early+0x13e/0x150 [i915]
...
<4> [239.330942] Call Trace:
<4> [239.330944] <TASK>
<4> [239.330949] i915_driver_late_release+0x2b/0xa0 [i915]
<4> [239.331202] i915_driver_release+0x86/0xa0 [i915]
<4> [239.331482] devm_drm_dev_init_release+0x61/0x90
<4> [239.331494] devm_action_release+0x15/0x30
<4> [239.331504] release_nodes+0x3d/0x120
<4> [239.331517] devres_release_all+0x96/0xd0
<4> [239.331533] device_unbind_cleanup+0x12/0x80
<4> [239.331543] device_release_driver_internal+0x23a/0x280
<4> [239.331550] ? bus_find_device+0xa5/0xe0
<4> [239.331563] device_driver_detach+0x14/0x20
...
<4> [357.719679] ---[ end trace 0000000000000000 ]---
If the test also unloads the i915 module then that's followed with:
<3> [357.787478] =============================================================================
<3> [357.788006] BUG i915_vma (Tainted: G U W N ): Objects remaining on __kmem_cache_shutdown()
<3> [357.788031] -----------------------------------------------------------------------------
<3> [357.788204] Object 0xffff888109e7f480 @offset=29824
<3> [357.788670] Allocated in i915_vma_instance+0xee/0xc10 [i915] age=292729 cpu=4 pid=2244
<4> [357.788994] i915_vma_instance+0xee/0xc10 [i915]
<4> [357.789290] init_status_page+0x7b/0x420 [i915]
<4> [357.789532] intel_engines_init+0x1d8/0x980 [i915]
<4> [357.789772] intel_gt_init+0x175/0x450 [i915]
<4> [357.790014] i915_gem_init+0x113/0x340 [i915]
<4> [357.790281] i915_driver_probe+0x847/0xed0 [i915]
<4> [357.790504] i915_pci_probe+0xe6/0x220 [i915]
...
Closer analysis of CI results history has revealed a dependency of the
error on a few IGT tests, namely:
- igt@api_intel_allocator@fork-simple-stress-signal,
- igt@api_intel_allocator@two-level-inception-interruptible,
- igt@gem_linear_blits@interruptible,
- igt@prime_mmap_coherency@ioctl-errors,
which invisibly trigger the issue, then exhibited with first driver unbind
attempt.
All of the above tests perform actions which are actively interrupted with
signals. Further debugging has allowed to narrow that scope down to
DRM_IOCTL_I915_GEM_EXECBUFFER2, and ring_context_alloc(), specific to ring
submission, in particular.
If successful then that function, or its execlists or GuC submission
equivalent, is supposed to be called only once per GEM context engine,
followed by raise of a flag that prevents the function from being called
again. The function is expected to unwind its internal errors itself, so
it may be safely called once more after it returns an error.
In case of ring submission, the function first gets a reference to the
engine's legacy timeline and then allocates a VMA. If the VMA allocation
fails, e.g. when i915_vma_instance() called from inside is interrupted
with a signal, then ring_context_alloc() fails, leaving the timeline held
referenced. On next I915_GEM_EXECBUFFER2 IOCTL, another reference to the
timeline is got, and only that last one is put on successful completion.
As a consequence, the legacy timeline, with its underlying engine status
page's VMA object, is still held and not released on driver unbind.
Get the legacy timeline only after successful allocation of the context
engine's VMA.
v2: Add a note on other submission methods (Krzysztof Karas):
Both execlists and GuC submission use lrc_alloc() which seems free
from a similar issue.
Fixes: 75d0a7f31e ("drm/i915: Lift timeline into intel_context")
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/12061
Cc: Chris Wilson <chris.p.wilson@linux.intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Krzysztof Karas <krzysztof.karas@intel.com>
Reviewed-by: Sebastian Brzezinka <sebastian.brzezinka@intel.com>
Reviewed-by: Krzysztof Niemiec <krzysztof.niemiec@intel.com>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Reviewed-by: Nitin Gote <nitin.r.gote@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Signed-off-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://lore.kernel.org/r/20250611104352.1014011-2-janusz.krzysztofik@linux.intel.com
(cherry picked from commit cc43422b3c)
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
[WHY]
Rounding error sometimes occurs when the refresh rate is equal to a panel's
max refresh rate, causing HDMI compliance failures.
[HOW]
Added a case so that we round up to avoid v_total_min to be below a panel's
minimum bound.
Reviewed-by: Jun Lei <jun.lei@amd.com>
Signed-off-by: Harold Sun <Harold.Sun@amd.com>
Signed-off-by: Ray Wu <ray.wu@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit fe7645d22b)
patch dd64956685 ("drm/amdgpu: Remove duplicated "context still
alive" check") removed ctx put, which will cause amdgpu_ctx_fini()
cannot be called and then cause some finished fence that added by
amdgpu_ctx_add_fence() cannot be released and cause memleak.
Fixes: dd64956685 ("drm/amdgpu: Remove duplicated "context still alive" check")
Signed-off-by: Lin.Cao <lincao12@amd.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8cf66089e2)
Cc: stable@vger.kernel.org
If the process is exiting, the mmput inside mmu notifier callback from
compactd or fork or numa balancing could release the last reference
of mm struct to call exit_mmap and free_pgtable, this triggers deadlock
with below backtrace.
The deadlock will leak kfd process as mmu notifier release is not called
and cause VRAM leaking.
The fix is to take mm reference mmget_non_zero when adding prange to the
deferred list to pair with mmput in deferred list work.
If prange split and add into pchild list, the pchild work_item.mm is not
used, so remove the mm parameter from svm_range_unmap_split and
svm_range_add_child.
The backtrace of hung task:
INFO: task python:348105 blocked for more than 64512 seconds.
Call Trace:
__schedule+0x1c3/0x550
schedule+0x46/0xb0
rwsem_down_write_slowpath+0x24b/0x4c0
unlink_anon_vmas+0xb1/0x1c0
free_pgtables+0xa9/0x130
exit_mmap+0xbc/0x1a0
mmput+0x5a/0x140
svm_range_cpu_invalidate_pagetables+0x2b/0x40 [amdgpu]
mn_itree_invalidate+0x72/0xc0
__mmu_notifier_invalidate_range_start+0x48/0x60
try_to_unmap_one+0x10fa/0x1400
rmap_walk_anon+0x196/0x460
try_to_unmap+0xbb/0x210
migrate_page_unmap+0x54d/0x7e0
migrate_pages_batch+0x1c3/0xae0
migrate_pages_sync+0x98/0x240
migrate_pages+0x25c/0x520
compact_zone+0x29d/0x590
compact_zone_order+0xb6/0xf0
try_to_compact_pages+0xbe/0x220
__alloc_pages_direct_compact+0x96/0x1a0
__alloc_pages_slowpath+0x410/0x930
__alloc_pages_nodemask+0x3a9/0x3e0
do_huge_pmd_anonymous_page+0xd7/0x3e0
__handle_mm_fault+0x5e3/0x5f0
handle_mm_fault+0xf7/0x2e0
hmm_vma_fault.isra.0+0x4d/0xa0
walk_pmd_range.isra.0+0xa8/0x310
walk_pud_range+0x167/0x240
walk_pgd_range+0x55/0x100
__walk_page_range+0x87/0x90
walk_page_range+0xf6/0x160
hmm_range_fault+0x4f/0x90
amdgpu_hmm_range_get_pages+0x123/0x230 [amdgpu]
amdgpu_ttm_tt_get_user_pages+0xb1/0x150 [amdgpu]
init_user_pages+0xb1/0x2a0 [amdgpu]
amdgpu_amdkfd_gpuvm_alloc_memory_of_gpu+0x543/0x7d0 [amdgpu]
kfd_ioctl_alloc_memory_of_gpu+0x24c/0x4e0 [amdgpu]
kfd_ioctl+0x29d/0x500 [amdgpu]
Fixes: fa582c6f36 ("drm/amdkfd: Use mmget_not_zero in MMU notifier")
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit a29e067bd3)
Cc: stable@vger.kernel.org
