The kref_put() call uses (void *)kvfree as the release callback, which
is incorrect. kref_put() expects a function with signature
void (*release)(struct kref *), but kvfree has signature
void (*)(const void *). Calling through an incompatible function pointer
is undefined behavior.
The code only worked by accident because ref_count is the first member
of vmw_bo_dirty, making the kref pointer equal to the struct pointer.
Fix this by adding a proper release callback that uses container_of()
to retrieve the containing structure before freeing.
Fixes: c1962742ff ("drm/vmwgfx: Use kref in vmw_bo_dirty")
Signed-off-by: Brad Spengler <brad.spengler@opensrcsec.com>
Signed-off-by: Zack Rusin <zack.rusin@broadcom.com>
Cc: Ian Forbes <ian.forbes@broadcom.com>
Link: https://patch.msgid.link/20260107171236.3573118-1-zack.rusin@broadcom.com
There is an audio channel shift issue with multi channel case - the
channel order is correct for the first run, but the channel order is
shifted for the second run. The fix method is to reset the PAI interface
at the end of playback.
The reset can be handled by PM runtime, so enable PM runtime.
Fixes: 0205fae632 ("drm/bridge: imx: add driver for HDMI TX Parallel Audio Interface")
Signed-off-by: Shengjiu Wang <shengjiu.wang@nxp.com>
Reviewed-by: Liu Ying <victor.liu@nxp.com>
Signed-off-by: Liu Ying <victor.liu@nxp.com>
Link: https://lore.kernel.org/r/20260130080910.3532724-1-shengjiu.wang@nxp.com
The mgag200_bmc_stop_scanout() function is called by the .atomic_disable()
handler for the MGA G200 VGA BMC encoder. This function performs a few
register writes to inform the BMC of an upcoming mode change, and then
polls to wait until the BMC actually stops.
The polling is implemented using a busy loop with udelay() and an iteration
timeout of 300, resulting in the function blocking for 300 milliseconds.
The function gets called ultimately by the output_poll_execute work thread
for the DRM output change polling thread of the mgag200 driver:
kworker/0:0-mm_ 3528 [000] 4555.315364:
ffffffffaa0e25b3 delay_halt.part.0+0x33
ffffffffc03f6188 mgag200_bmc_stop_scanout+0x178
ffffffffc087ae7a disable_outputs+0x12a
ffffffffc087c12a drm_atomic_helper_commit_tail+0x1a
ffffffffc03fa7b6 mgag200_mode_config_helper_atomic_commit_tail+0x26
ffffffffc087c9c1 commit_tail+0x91
ffffffffc087d51b drm_atomic_helper_commit+0x11b
ffffffffc0509694 drm_atomic_commit+0xa4
ffffffffc05105e8 drm_client_modeset_commit_atomic+0x1e8
ffffffffc0510ce6 drm_client_modeset_commit_locked+0x56
ffffffffc0510e24 drm_client_modeset_commit+0x24
ffffffffc088a743 __drm_fb_helper_restore_fbdev_mode_unlocked+0x93
ffffffffc088a683 drm_fb_helper_hotplug_event+0xe3
ffffffffc050f8aa drm_client_dev_hotplug+0x9a
ffffffffc088555a output_poll_execute+0x29a
ffffffffa9b35924 process_one_work+0x194
ffffffffa9b364ee worker_thread+0x2fe
ffffffffa9b3ecad kthread+0xdd
ffffffffa9a08549 ret_from_fork+0x29
On a server running ptp4l with the mgag200 driver loaded, we found that
ptp4l would sometimes get blocked from execution because of this busy
waiting loop.
Every so often, approximately once every 20 minutes -- though with large
variance -- the output_poll_execute() thread would detect some sort of
change that required performing a hotplug event which results in attempting
to stop the BMC scanout, resulting in a 300msec delay on one CPU.
On this system, ptp4l was pinned to a single CPU. When the
output_poll_execute() thread ran on that CPU, it blocked ptp4l from
executing for its 300 millisecond duration.
This resulted in PTP service disruptions such as failure to send a SYNC
message on time, failure to handle ANNOUNCE messages on time, and clock
check warnings from the application. All of this despite the application
being configured with FIFO_RT and a higher priority than the background
workqueue tasks. (However, note that the kernel did not use
CONFIG_PREEMPT...)
It is unclear if the event is due to a faulty VGA connection, another bug,
or actual events causing a change in the connection. At least on the system
under test it is not a one-time event and consistently causes disruption to
the time sensitive applications.
The function has some helpful comments explaining what steps it is
attempting to take. In particular, step 3a and 3b are explained as such:
3a - The third step is to verify if there is an active scan. We are
waiting on a 0 on remhsyncsts (<XSPAREREG<0>.
3b - This step occurs only if the remove is actually scanning. We are
waiting for the end of the frame which is a 1 on remvsyncsts
(<XSPAREREG<1>).
The actual steps 3a and 3b are implemented as while loops with a
non-sleeping udelay(). The first step iterates while the tmp value at
position 0 is *not* set. That is, it keeps iterating as long as the bit is
zero. If the bit is already 0 (because there is no active scan), it will
iterate the entire 300 attempts which wastes 300 milliseconds in total.
This is opposite of what the description claims.
The step 3b logic only executes if we do not iterate over the entire 300
attempts in the first loop. If it does trigger, it is trying to check and
wait for a 1 on the remvsyncsts. However, again the condition is actually
inverted and it will loop as long as the bit is 1, stopping once it hits
zero (rather than the explained attempt to wait until we see a 1).
Worse, both loops are implemented using non-sleeping waits which spin
instead of allowing the scheduler to run other processes. If the kernel is
not configured to allow arbitrary preemption, it will waste valuable CPU
time doing nothing.
There does not appear to be any documentation for the BMC register
interface, beyond what is in the comments here. It seems more probable that
the comment here is correct and the implementation accidentally got
inverted from the intended logic.
Reading through other DRM driver implementations, it does not appear that
the .atomic_enable or .atomic_disable handlers need to delay instead of
sleep. For example, the ast_astdp_encoder_helper_atomic_disable() function
calls ast_dp_set_phy_sleep() which uses msleep(). The "atomic" in the name
is referring to the atomic modesetting support, which is the support to
enable atomic configuration from userspace, and not to the "atomic context"
of the kernel. There is no reason to use udelay() here if a sleep would be
sufficient.
Replace the while loops with a read_poll_timeout() based implementation
that will sleep between iterations, and which stops polling once the
condition is met (instead of looping as long as the condition is met). This
aligns with the commented behavior and avoids blocking on the CPU while
doing nothing.
Note the RREG_DAC is implemented using a statement expression to allow
working properly with the read_poll_timeout family of functions. The other
RREG_<TYPE> macros ought to be cleaned up to have better semantics, and
several places in the mgag200 driver could make use of RREG_DAC or similar
RREG_* macros should likely be cleaned up for better semantics as well, but
that task has been left as a future cleanup for a non-bugfix.
Fixes: 414c453106 ("mgag200: initial g200se driver (v2)")
Suggested-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Jocelyn Falempe <jfalempe@redhat.com>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patch.msgid.link/20260202-jk-mgag200-fix-bad-udelay-v2-1-ce1e9665987d@intel.com
This is just refactoring to allow the lower layers to distinguish
between suspend and runtime suspend.
GSP 570 needs to set a flag with the GPU is going into GCOFF,
this flag taken from the opengpu driver is set whenever runtime
suspend is enterning GCOFF but not for normal suspend paths.
This just refactors the code, a subsequent patch use the information.
Fixes: 53dac06238 ("drm/nouveau/gsp: add support for 570.144")
Cc: <stable@vger.kernel.org>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Tested-by: Lyude Paul <lyude@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patch.msgid.link/20260203052431.2219998-3-airlied@gmail.com
Fix several issues in dw_dp_bind() error handling:
1. Missing return after drm_bridge_attach() failure - the function
continued execution instead of returning an error.
2. Resource leak: drm_dp_aux_register() is not a devm function, so
drm_dp_aux_unregister() must be called on all error paths after
aux registration succeeds. This affects errors from:
- drm_bridge_attach()
- phy_init()
- devm_add_action_or_reset()
- platform_get_irq()
- devm_request_threaded_irq()
3. Bug fix: platform_get_irq() returns the IRQ number or a negative
error code, but the error path was returning ERR_PTR(ret) instead
of ERR_PTR(dp->irq).
Use a goto label for cleanup to ensure consistent error handling.
Fixes: 86eecc3a9c ("drm/bridge: synopsys: Add DW DPTX Controller support library")
Cc: stable@vger.kernel.org
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Louis Chauvet <louis.chauvet@bootlin.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260102155553.13243-1-osama.abdelkader@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
drm-misc-fixes for v6.19-rc6:
vmwgfx:
- Fix hw regression from refactoring cursor handling on v10 'hardware'
- Fix warnings in destructor by merging the 2 release functions
- kernel doc fix
- error handling in vmw_compat_shader_add()
rockchip:
- fix vop2 polling
- fix regression waiting for cfgdone without config change
- fix warning when enabling encoder
core:
- take gem lock when preallocating in gpuvm.
- add single byte read fallback to dp for broken usb-c adapters
- remove duplicate drm_sysfb declarations
gud:
- Fix oops on usb disconnect
Simple panel:
- Re-add fallback when connector is not set to fix regressions
- Set correct type in DataImage SCF0700C48GGU18
nouveau:
- locking fixes for cursor handling.
Signed-off-by: Simona Vetter <simona.vetter@ffwll.ch>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patch.msgid.link/ce0acfe2-9c1a-42b7-8782-f1e7f34b8544@linux.intel.com
[Why&How]
Right now, the HDMI HPD filter is enabled by default at 1500ms.
We want to disable it by default, as most modern displays with HDMI do
not require it for DPMS mode.
The HPD can instead be enabled as a driver parameter with a custom delay
value in ms (up to 5000ms).
Fixes: c918e75e1e ("drm/amd/display: Add an HPD filter for HDMI")
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4859
Signed-off-by: Ivan Lipski <ivan.lipski@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 6a681cd903)
The user mode queue keeps a pointer to the most recent fence in
userq->last_fence. This pointer holds an extra dma_fence reference.
When the queue is destroyed, we free the fence driver and its xarray,
but we forgot to drop the last_fence reference.
Because of the missing dma_fence_put(), the last fence object can stay
alive when the driver unloads. This leaves an allocated object in the
amdgpu_userq_fence slab cache and triggers
This is visible during driver unload as:
BUG amdgpu_userq_fence: Objects remaining on __kmem_cache_shutdown()
kmem_cache_destroy amdgpu_userq_fence: Slab cache still has objects
Call Trace:
kmem_cache_destroy
amdgpu_userq_fence_slab_fini
amdgpu_exit
__do_sys_delete_module
Fix this by putting userq->last_fence and clearing the pointer during
amdgpu_userq_fence_driver_free().
This makes sure the fence reference is released and the slab cache is
empty when the module exits.
v2: Update to only release userq->last_fence with dma_fence_put()
(Christian)
Fixes: edc762a51c ("drm/amdgpu/userq: move some code around")
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 8e051e38a8)
This reverts commit 820b3d376e.
It’s better to validate VM TLB flushes in the flush‑TLB backend
rather than in the generic VM layer.
Reverting this patch depends on
commit fa7c231fc2b0 ("drm/amdgpu: validate the flush_gpu_tlb_pasid()")
being present in the tree.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 9163fe4d79)
Internal backlight levels are initialised from ACPI but the values
are sometimes out of sync with the levels in effect until there has
been a read from hardware (eg triggered by reading from sysfs).
This means that the first drm_commit can cause the levels to be set
to a different value than the actual starting one, which results in
a sudden change in brightness.
This path shows the problem (when the values are out of sync):
amdgpu_dm_atomic_commit_tail()
-> amdgpu_dm_commit_streams()
-> amdgpu_dm_backlight_set_level(..., dm->brightness[n])
This patch calls the backlight ops get_brightness explicitly
at the end of backlight registration to make sure dm->brightness[n]
is in sync with the actual hardware levels.
Fixes: 2fe87f54ab ("drm/amd/display: Set default brightness according to ACPI")
Signed-off-by: Vivek Das Mohapatra <vivek@collabora.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 318b1c36d8)
Cc: stable@vger.kernel.org
[Why]
The PSR message was moved in commit 4321742c39 ("drm/amd/display:
Move PSR support message into amdgpu_dm"). This message however shows
for every single link without showing which link is which. This can
send a confusing message to the user.
[How]
Add link name into the message.
Fixes: 4321742c39 ("drm/amd/display: Move PSR support message into amdgpu_dm")
Reviewed-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Matthew Stewart <matthew.stewart2@amd.com>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 99f77f6229)
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>
(cherry picked from commit 38a0f4cf8c)
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>
(cherry picked from commit 6a23e7b433)
Cc: stable@vger.kernel.org
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>
(cherry picked from commit 2f2a72de67)
For a while, I've been seeing a strange issue where some (usually not all)
of the display DMA channels will suddenly hang, particularly when there is
a visible cursor on the screen that is being frequently updated, and
especially when said cursor happens to go between two screens. While this
brings back lovely memories of fixing Intel Skylake bugs, I would quite
like to fix it :).
It turns out the problem that's happening here is that we're managing to
reach nv50_head_flush_set() in our atomic commit path without actually
holding nv50_disp->mutex. This means that cursor updates happening in
parallel (along with any other atomic updates that need to use the core
channel) will race with eachother, which eventually causes us to corrupt
the pushbuffer - leading to a plethora of various GSP errors, usually:
nouveau 0000:c1:00.0: gsp: Xid:56 CMDre 00000000 00000218 00102680 00000004 00800003
nouveau 0000:c1:00.0: gsp: Xid:56 CMDre 00000000 0000021c 00040509 00000004 00000001
nouveau 0000:c1:00.0: gsp: Xid:56 CMDre 00000000 00000000 00000000 00000001 00000001
The reason this is happening is because generally we check whether we need
to set nv50_atom->lock_core at the end of nv50_head_atomic_check().
However, curs507a_prepare is called from the fb_prepare callback, which
happens after the atomic check phase. As a result, this can lead to commits
that both touch the core channel but also don't grab nv50_disp->mutex.
So, fix this by making sure that we set nv50_atom->lock_core in
cus507a_prepare().
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Lyude Paul <lyude@redhat.com>
Fixes: 1590700d94 ("drm/nouveau/kms/nv50-: split each resource type into their own source files")
Cc: <stable@vger.kernel.org> # v4.18+
Link: https://patch.msgid.link/20251219215344.170852-2-lyude@redhat.com
The switch from devm_kzalloc() + drm_panel_init() to
devm_drm_panel_alloc() introduced a regression.
Several panel descriptors do not set connector_type. For those panels,
panel_simple_probe() used to compute a connector type (currently DPI as a
fallback) and pass that value to drm_panel_init(). After the conversion
to devm_drm_panel_alloc(), the call unconditionally used
desc->connector_type instead, ignoring the computed fallback and
potentially passing DRM_MODE_CONNECTOR_Unknown, which
drm_panel_bridge_add() does not allow.
Move the connector_type validation / fallback logic before the
devm_drm_panel_alloc() call and pass the computed connector_type to
devm_drm_panel_alloc(), so panels without an explicit connector_type
once again get the DPI default.
Signed-off-by: Ludovic Desroches <ludovic.desroches@microchip.com>
Fixes: de04bb0089 ("drm/panel/panel-simple: Use the new allocation in place of devm_kzalloc()")
Cc: stable@vger.kernel.org
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://lore.kernel.org/stable/20251126-lcd_panel_connector_type_fix-v2-1-c15835d1f7cb%40microchip.com
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patch.msgid.link/20251218-lcd_panel_connector_type_fix-v3-1-ddcea6d8d7ef@microchip.com
The connector type for the DataImage SCF0700C48GGU18 panel is missing and
devm_drm_panel_bridge_add() requires connector type to be set. This leads
to a warning and a backtrace in the kernel log and panel does not work:
"
WARNING: CPU: 3 PID: 38 at drivers/gpu/drm/bridge/panel.c:379 devm_drm_of_get_bridge+0xac/0xb8
"
The warning is triggered by a check for valid connector type in
devm_drm_panel_bridge_add(). If there is no valid connector type
set for a panel, the warning is printed and panel is not added.
Fill in the missing connector type to fix the warning and make
the panel operational once again.
Cc: stable@vger.kernel.org
Fixes: 97ceb1fb08 ("drm/panel: simple: Add support for DataImage SCF0700C48GGU18")
Signed-off-by: Marek Vasut <marex@nabladev.com>
Reviewed-by: Neil Armstrong <neil.armstrong@linaro.org>
Signed-off-by: Neil Armstrong <neil.armstrong@linaro.org>
Link: https://patch.msgid.link/20260110152750.73848-1-marex@nabladev.com
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>
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
