Previously, the driver's internal wedged.mode state was updated without
verifying whether the corresponding engine reset policy update in GuC
succeeded. This could leave the driver reporting a wedged.mode state
that doesn't match the actual reset behavior programmed in GuC.
With this change, the reset policy is updated first, and the driver's
wedged.mode state is modified only if the policy update succeeds on all
available GTs.
This patch also introduces two functional improvements:
- The policy is sent to GuC only when a change is required. An update
is needed only when entering or leaving XE_WEDGED_MODE_UPON_ANY_HANG,
because only in that case the reset policy changes. For example,
switching between XE_WEDGED_MODE_UPON_CRITICAL_ERROR and
XE_WEDGED_MODE_NEVER doesn't affect the reset policy, so there is no
need to send the same value to GuC.
- An inconsistent_reset flag is added to track cases where reset policy
update succeeds only on a subset of GTs. If such inconsistency is
detected, future wedged mode configuration will force a retry of the
reset policy update to restore a consistent state across all GTs.
Fixes: 6b8ef44cc0 ("drm/xe: Introduce the wedged_mode debugfs")
Signed-off-by: Lukasz Laguna <lukasz.laguna@intel.com>
Link: https://patch.msgid.link/20260107174741.29163-3-lukasz.laguna@intel.com
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
This reverts commit 22a36e660d once,
which was merged twice due to an incorrect backmerge resolution.
Fixes: ce0478b02e ("Merge tag 'v6.18-rc6' into drm-next")
Signed-off-by: Peter Colberg <pcolberg@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In commit 9ce4aef9a5 ("drm/gpuvm: take GEM lock inside
drm_gpuvm_bo_obtain_prealloc()") we update
drm_gpuvm_bo_obtain_prealloc() to take locks internally, which means
that it's only usable in immediate mode.
In this commit, we notice that drm_gpuvm_bo_obtain() requires you to use
staged mode. This means that we now have one variant of obtain for each
mode you might use gpuvm in.
To reflect this information, we add a warning about using it in
immediate mode, and to make the distinction clearer we rename the method
with a _locked() suffix so that it's clear that it requires the caller
to take the locks.
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Alice Ryhl <aliceryhl@google.com>
Link: https://patch.msgid.link/20260108-gpuvm-rust-v2-2-dbd014005a0b@google.com
[ Slightly reword commit message to refer to commit 9ce4aef9a5
("drm/gpuvm: take GEM lock inside drm_gpuvm_bo_obtain_prealloc()").
- Danilo ]
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
When calling drm_gpuvm_bo_obtain_prealloc() and using immediate mode,
this may result in a call to ops->vm_bo_free(vm_bo) while holding the
GEMs gpuva mutex. This is a problem if ops->vm_bo_free(vm_bo) performs
any operations that are not safe in the fence signalling critical path,
and it turns out that Panthor (the only current user of the method)
calls drm_gem_shmem_unpin() which takes a resv lock internally.
This constitutes both a violation of signalling safety and lock
inversion. To fix this, we modify the method to internally take the GEMs
gpuva mutex so that the mutex can be unlocked before freeing the
preallocated vm_bo.
Note that this modification introduces a requirement that the driver
uses immediate mode to call drm_gpuvm_bo_obtain_prealloc() as it would
otherwise take the wrong lock.
Fixes: 63e919a316 ("panthor: use drm_gpuva_unlink_defer()")
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Signed-off-by: Alice Ryhl <aliceryhl@google.com>
Link: https://patch.msgid.link/20260108-gpuvm-rust-v2-1-dbd014005a0b@google.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
When an eGPU is unplugged the KFD topology should also be destroyed
for that GPU. This never happens because the fini_sw callbacks never
get to run. Run them manually before calling amdgpu_device_ip_fini_early()
when a device has already been disconnected.
This location is intentionally chosen to make sure that the kfd locking
refcount doesn't get incremented unintentionally.
Cc: kent.russell@amd.com
Closes: https://community.frame.work/t/amd-egpu-on-linux/8691/33
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Replace legacy DRM_ERROR()/DRM_INFO() logging in the UVD v1.0 code
with drm_err() and drm_info() helpers that take a struct drm_device.
Using drm_* logging provides proper device context in dmesg, which is
important for systems with multiple DRM devices, and aligns the radeon
driver with current DRM logging practices.
No functional change intended.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Mukesh Ogare <mukeshogare871@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In SRIOV, guest drivers no longer setup/destory
VMR starting from mp0 v11_0_7.
In bare-metal, if boot-time TMR is enabled, some
generation (e.g., mp0 v13_0_x) don’t need runtime
TMR allocation but still require SETUP_TMR command
with tmr address 0 for backward compatibility.
some newer generations require neither SETUP_TMR nor
DESTROY_TMR and will return errors if they are sent.
Driver relies on boot_time_tmr and autoload_supported
to handle these cases correctly.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Likun Gao <Likun.Gao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add helper amdgpu_gtt_mgr_alloc/free_entries, define
GART_ENTRY_WITHOUT_BO_COLOR color for GART node not allocated with GTT
bo, then amdgpu_gtt_mgr_recover skip those mm_node.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When driver not support atomic, fb using plane->fb rather than
plane->state->fb.
Fixes: fe151ed7af ("drm/amdgpu: add generic display panic helper code")
Signed-off-by: Lu Yao <yaolu@kylinos.cn>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix copy&paste error, that should have been an assignment instead of an or,
otherwise MTYPE_UC 0x3 can not be updated to MTYPE_RW 0x1.
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
GPU board and Baseboard temperatures come from system metrics table.
Driver keeps separate metrics table for both. Use the new driver table
structure to represent them.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If cached gpu metrics table is available, return it directly. Also,
deprecate gpu_metrics_table variables as they are no longer used.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The write of cfgdone bits always done at .atomic_flush.
When userspace makes plane zpos changes of two crtc within one commit,
at the .atomic_begin stage, crtcN will never receive the "layer change
cfg done" event of crtcM because crtcM has not yet written "cfgdone".
So only wait when there is pending cfgdone bits to avoid long timeout.
Fixes: 3e89a8c683 ("drm/rockchip: vop2: Fix the update of LAYER/PORT select registers when there are multi display output on rk3588/rk3568")
Signed-off-by: Andy Yan <andy.yan@rock-chips.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patch.msgid.link/20250718064120.8811-2-andyshrk@163.com
According to the implementation of read_poll_timeout_atomic, if the
delay time is 0, it will only use a simple loop based on timeout_us to
decrement the count. Therefore, the final timeout time will differ
significantly from the set timeout time. So, here we set a specific
delay time to ensure that the calculation of the timeout duration
is accurate.
Fixes: 3e89a8c683 ("drm/rockchip: vop2: Fix the update of LAYER/PORT select registers when there are multi display output on rk3588/rk3568")
Signed-off-by: Andy Yan <andy.yan@rock-chips.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patch.msgid.link/20250718064120.8811-1-andyshrk@163.com
The Imagination DRM driver defines pvr_power_put() to pass the return
value of pm_runtime_put() to the caller, but then it never uses the
return value of pvr_power_put().
Modify pvr_power_put() to discard the pm_runtime_put() return value and
change its return type to void.
No intentional functional impact.
This will facilitate a planned change of the pm_runtime_put() return
type to void in the future.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Matt Coster <matt.coster@imgtec.com>
Link: https://patch.msgid.link/8642685.T7Z3S40VBb@rafael.j.wysocki
Device-to-device migration is causing xe_exec_system_allocator --r
*race*no* to intermittently fail with engine resets and a kernel hang on
a page lock. This should work but is clearly buggy somewhere. Disable
device-to-device migration in the interim until the issue can be
root-caused.
The only downside of disabling device-to-device migration is that memory
will bounce through system memory during migration. However, this path
should be rare, as it only occurs when madvise attributes are changed or
atomics are used.
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Fixes: ec265e1f1c ("drm/pagemap: Support source migration over interconnect")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Francois Dugast <francois.dugast@intel.com>
Link: https://patch.msgid.link/20260107182716.2236607-3-matthew.brost@intel.com
Skipping power ungate exposed some scenarios that will fail
like below:
```
amdgpu: Register(0) [regVPEC_QUEUE_RESET_REQ] failed to reach value 0x00000000 != 0x00000001n
amdgpu 0000:c1:00.0: amdgpu: VPE queue reset failed
...
amdgpu: [drm] *ERROR* wait_for_completion_timeout timeout!
```
The underlying s2idle issue that prompted this commit is going to
be fixed in BIOS.
This reverts commit 2a6c826cfe.
This was lost in the 6.19 merge so reapply it.
Fixes: 2a6c826cfe ("drm/amd: Skip power ungate during suspend for VPE")
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reported-by: Konstantin <answer2019@yandex.ru>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220812
Reported-by: Matthew Schwartz <matthew.schwartz@linux.dev>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 3925683515)
During Mode 1 reset, the ASIC undergoes a reset cycle and becomes
temporarily inaccessible via PCIe. Any attempt to access MMIO registers
during this window (e.g., from interrupt handlers or other driver threads)
can result in uncompleted PCIe transactions, leading to NMI panics or
system hangs.
To prevent this, set the `no_hw_access` flag to true immediately after
triggering the reset. This signals other driver components to skip
register accesses while the device is offline.
A memory barrier `smp_mb()` is added to ensure the flag update is
globally visible to all cores before the driver enters the sleep/wait
state.
Signed-off-by: Perry Yuan <perry.yuan@amd.com>
Reviewed-by: Yifan Zhang <yifan1.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 7edb503fe4)
drm-misc-fixes for v6.19-rc5:
pl111:
- Fix error handling in probe
mediatek/atomic/tidss:
- Fix tidss in another way and revert reordering of pre-enable and post-disable operations,
as it breaks other bridge drivers.
nouveau:
- Fix regression from fwsec s/r fix.
pci/vga:
- Fix multiple gpu's being reported a 'boot_display'
fb-helper:
- Fix vblank timeout during suspend/reset
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patch.msgid.link/f50067e6-243a-4ed8-9781-1e4e4fdebc8e@linux.intel.com
