[why]
dml2 based on num_enabled clock != 2 to do clock ramming to dpm.
apu has 8 levels dispclk/dppclk/dcfclk/fclk, but only 4 levels of memclk.
to avoid mapping dispclk/dppclk to DPM clock,
based on arch review, force dispclk/dppclk num_level as 2.
Reviewed-by: Dillon Varone <dillon.varone@amd.com>
Signed-off-by: Charlene Liu <Charlene.Liu@amd.com>
Signed-off-by: Roman Li <roman.li@amd.com>
Signed-off-by: Chuanyu Tseng <Chuanyu.Tseng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
PASID resue could cause interrupt issue when process
immediately runs into hw state left by previous
process exited with the same PASID, it's possible that
page faults are still pending in the IH ring buffer when
the process exits and frees up its PASID. To prevent the
case, it uses idr cyclic allocator same as kernel pid's.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_device_get_job_timeout_settings() passes a pointer directly
to the global amdgpu_lockup_timeout[] buffer into strsep().
strsep() destructively replaces delimiter characters with '\0'
in-place.
On multi-GPU systems, this function is called once per device.
When a multi-value setting like "0,0,0,-1" is used, the first
GPU's call transforms the global buffer into "0\00\00\0-1". The
second GPU then sees only "0" (terminated at the first '\0'),
parses a single value, hits the single-value fallthrough
(index == 1), and applies timeout=0 to all rings — causing
immediate false job timeouts.
Fix this by copying into a stack-local array before calling
strsep(), so the global module parameter buffer remains intact
across calls. The buffer is AMDGPU_MAX_TIMEOUT_PARAM_LENGTH
(256) bytes, which is safe for the stack.
v2: wrap commit message to 72 columns, add Assisted-by tag.
v3: use stack array with strscpy() instead of kstrdup()/kfree()
to avoid unnecessary heap allocation (Christian).
This patch was developed with assistance from Claude (claude-opus-4-6).
Assisted-by: Claude:claude-opus-4-6
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Look at hqd_queue_priority rather than hqd_pipe_priority.
In practice, it didn't matter as both were always set for
kernel queues, but that will change in the future.
Fixes: 2e216b1e6b ("drm/amdgpu/gfx11: handle priority setup for gfx pipe1")
Reviewed-by：Jesse Zhang <jesse.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Look at hqd_queue_priority rather than hqd_pipe_priority.
In practice, it didn't matter as both were always set for
kernel queues, but that will change in the future.
Fixes: b07d1d73b0 ("drm/amd/amdgpu: Enable high priority gfx queue")
Reviewed-by：Jesse Zhang <jesse.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When generating a devcoredump, amdgpu_discovery_dump() prints the IP
discovery topology.
The function already needs to handle the case where
adev->discovery.ip_top is NULL to avoid a crash.
Currently, the code prints a section header and an additional message
when the topology is unavailable.
However, for platforms where discovery is not used, this section is not
expected to be present. Printing an extra message adds unnecessary
output.
Simplify this by skipping the entire section when ip_top is NULL.
The NULL check is kept to avoid a crash, but no output is generated when
the discovery topology is unavailable.
Cc: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@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>
When a GPU fault or timeout happens, the driver creates a devcoredump
to collect debug information.
During this, amdgpu_devcoredump_format() calls
amdgpu_discovery_dump() to print IP discovery data.
amdgpu_discovery_dump() uses:
adev->discovery.ip_top
and then accesses:
ip_top->die_kset
amdgpu_discovery_dump() uses adev->discovery.ip_top. However,
ip_top may be NULL if the discovery topology was never initialized.
The current code does not check for this before using ip_top. As a
result, when ip_top is NULL, the coredump worker crashes while taking
the spinlock for ip_top->die_kset.
Fix this by checking for a missing ip_top before walking the discovery
topology. If it is unavailable, print a short message in the dump and
return safely.
- If ip_top is NULL, print a message and skip the dump
- Also add the same check in the cleanup path
This makes the coredump and cleanup paths safe even when the
discovery topology is not available.
KASAN trace:
[ 522.228252] [IGT] amd_deadlock: starting subtest amdgpu-deadlock-sdma
[ 522.240681] [IGT] amd_deadlock: starting dynamic subtest amdgpu-deadlock-sdma
...
[ 522.952317] Write of size 4 at addr 0000000000000050 by task kworker/u129:5/5434
[ 522.937526] BUG: KASAN: null-ptr-deref in _raw_spin_lock+0x66/0xc0
[ 522.967659] Workqueue: events_unbound amdgpu_devcoredump_deferred_work [amdgpu]
...
[ 522.969445] Call Trace:
[ 522.969508] _raw_spin_lock+0x66/0xc0
[ 522.969518] ? __pfx__raw_spin_lock+0x10/0x10
[ 522.969534] amdgpu_discovery_dump+0x61/0x530 [amdgpu]
[ 522.971346] ? pick_next_task_fair+0x3f6/0x1c60
[ 522.971363] amdgpu_devcoredump_format+0x84f/0x26f0 [amdgpu]
[ 522.973188] ? __pfx_amdgpu_devcoredump_format+0x10/0x10 [amdgpu]
[ 522.975012] ? psi_task_switch+0x2b5/0x9b0
[ 522.975027] ? __pfx___drm_printfn_coredump+0x10/0x10 [drm]
[ 522.975198] ? __pfx___drm_puts_coredump+0x10/0x10 [drm]
[ 522.975366] ? __schedule+0x113c/0x38d0
[ 522.975381] amdgpu_devcoredump_deferred_work+0x4c/0x1f0 [amdgpu]
v2: Updated commit message - Clarified that ip_top is not freed, it can
just be NULL if discovery was not initialized. (Christian/Lijo)
v3: Removed the extra drm_warn() for sysfs init failure as sysfs already
reports errors. (Christian)
Fixes: e81eff80aa ("drm/amdgpu: include ip discovery data in devcoredump")
Cc: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@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>
Starting with commit 17ce8a6907 ("drm/amd/display: Add dsc pre-validation in
atomic check"), amdgpu resets the CRTC state mode_changed flag to false when
recomputing the DSC configuration results in no timing change for a particular
stream.
However, this is incorrect in scenarios where a change in MST/DSC configuration
happens in the same KMS commit as another (unrelated) mode change. For example,
the integrated panel of a laptop may be configured differently (e.g., HDR
enabled/disabled) depending on whether external screens are attached. In this
case, plugging in external DP-MST screens may result in the mode_changed flag
being dropped incorrectly for the integrated panel if its DSC configuration
did not change during precomputation in pre_validate_dsc().
At this point, however, dm_update_crtc_state() has already created new streams
for CRTCs with DSC-independent mode changes. In turn,
amdgpu_dm_commit_streams() will never release the old stream, resulting in a
memory leak. amdgpu_dm_atomic_commit_tail() will never acquire a reference to
the new stream either, which manifests as a use-after-free when the stream gets
disabled later on:
BUG: KASAN: use-after-free in dc_stream_release+0x25/0x90 [amdgpu]
Write of size 4 at addr ffff88813d836524 by task kworker/9:9/29977
Workqueue: events drm_mode_rmfb_work_fn
Call Trace:
<TASK>
dump_stack_lvl+0x6e/0xa0
print_address_description.constprop.0+0x88/0x320
? dc_stream_release+0x25/0x90 [amdgpu]
print_report+0xfc/0x1ff
? srso_alias_return_thunk+0x5/0xfbef5
? __virt_addr_valid+0x225/0x4e0
? dc_stream_release+0x25/0x90 [amdgpu]
kasan_report+0xe1/0x180
? dc_stream_release+0x25/0x90 [amdgpu]
kasan_check_range+0x125/0x200
dc_stream_release+0x25/0x90 [amdgpu]
dc_state_destruct+0x14d/0x5c0 [amdgpu]
dc_state_release.part.0+0x4e/0x130 [amdgpu]
dm_atomic_destroy_state+0x3f/0x70 [amdgpu]
drm_atomic_state_default_clear+0x8ee/0xf30
? drm_mode_object_put.part.0+0xb1/0x130
__drm_atomic_state_free+0x15c/0x2d0
atomic_remove_fb+0x67e/0x980
Since there is no reliable way of figuring out whether a CRTC has unrelated
mode changes pending at the time of DSC validation, remember the value of the
mode_changed flag from before the point where a CRTC was marked as potentially
affected by a change in DSC configuration. Reset the mode_changed flag to this
earlier value instead in pre_validate_dsc().
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/5004
Fixes: 17ce8a6907 ("drm/amd/display: Add dsc pre-validation in atomic check")
Signed-off-by: Yussuf Khalil <dev@pp3345.net>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
ras_umc_log_pending_bad_bank() walks through a list of pending ECC
bad-bank entries. These entries are saved when a bad-bank error cannot
be processed immediately, for example during a GPU reset.
Later, this function iterates over the pending list and retries logging
each bad-bank error. If logging succeeds, the entry is removed from the
list and the memory for that node is freed.
The loop uses list_for_each_entry_safe(), which already guarantees that
ecc_node points to a valid list entry while the loop body is executing.
Checking "ecc_node &&" inside the loop is therefore unnecessary and
redundant.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/../ras/rascore/ras_umc.c:225 ras_umc_log_pending_bad_bank() warn: variable dereferenced before check 'ecc_node' (see line 223)
Fixes: 7a3f9c0992 ("drm/amd/ras: Add umc common ras functions")
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: YiPeng Chai <YiPeng.Chai@amd.com>
Cc: Tao Zhou <tao.zhou1@amd.com>
Cc: Hawking Zhang <Hawking.Zhang@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: YiPeng Chai <YiPeng.Chai@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When the memory allocated by userspace isn't sufficient for all the
fences then just wait on them instead of returning an error.
v2: use correct variable as pointed out by Sunil
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_sched_ioctl() currently uses WARN(1, ...) when userspace passes
an out-of-range context priority value. WARN(1, ...) is unconditional
and produces a full stack trace, which is disproportionate for a simple
input validation failure -- the invalid value is already rejected with
-EINVAL on the next line.
Replace WARN(1, ...) with DRM_ERROR() to log the invalid value at an
appropriate level without generating a stack dump. The -EINVAL return
to userspace is unchanged.
No functional change for well-formed userspace callers.
v2:
- Reworked commit message to focus on appropriate log level for
parameter validation
- Clarified that -EINVAL behavior is preserved (Vitaly)
v3: completely drop that warning.
Invalid parameters should never clutter the system log. (Christian)
Reviewed-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Jesse Zhang <jesse.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
With this change we're adding NV12 and P010 twice to reported
formats on a primary plane, which causes us to hit an assert
in Weston.
This reverts commit 63fff55131.
Fixes: 63fff55131 ("drm/amd/display: Add NV12/P010 formats to primary plane")
Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_virt_init_critical_region() stores init_hdr_offset as u64.
The subsequent check for init_hdr_offset < 0 is therefore always false.
Drop the unreachable validation and rely on the existing
check_add_overflow() and VRAM end bounds check for offset validation.
This resolves the Smatch warning about comparing an unsigned value
against zero.
drivers/gpu/drm/amd/amdgpu/amdgpu_virt.c:953 amdgpu_virt_init_critical_region() warn: unsigned 'init_hdr_offset' is never less than zero.
Fixes: 07009df649 ("drm/amdgpu: Introduce SRIOV critical regions v2 during VF init")
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Ellen Pan <yunru.pan@amd.com>
Cc: Lijo Lazar <lijo.lazar@amd.com>
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: Bokun Zhang <bokun.zhang@amd.com>
Reviewed-by: Ellen Pan <yunru.pan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_userq_hang_detect_work() retrieves the queue pointer using
container_of() from the embedded work item.
Since the work structure is part of struct amdgpu_usermode_queue,
the returned queue pointer cannot be NULL in normal execution.
Remove the redundant !queue check and keep the validation for
queue->userq_mgr.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c:159 amdgpu_userq_hang_detect_work() warn: can 'queue' even be NULL?
Fixes: 290f46cf57 ("drm/amdgpu: Implement user queue reset functionality")
Cc: Jesse Zhang <Jesse.Zhang@amd.com>
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Acked-by: Jesse Zhang <jesse.zhang@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lockdep was complaining about a number of issues here. Especially lock
inversion between syncobj, dma_resv and copying things into userspace.
Rework the functionality. Split it up into multiple functions,
consistenly use memdup_array_user(), fix the lock inversions and a few
more bugs in error handling.
v2: drop the dma_fence leak fix, turned out that was actually correct,
just not well documented. Apply some more cleanup suggestion from
Tvrtko.
v3: rebase on already done cleanups
v4: add missing dma_fence_put() in error path.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Well that was broken on multiple levels.
First of all a lot of checks were placed at incorrect locations, especially if
the resume worker should run or not.
Then a bunch of code was just mid-layering because of incorrect assignment who
should do what.
And finally comments explaining what happens instead of why.
Just re-write it from scratch, that should at least fix some of the hangs we
are seeing.
Use RCU for the eviction fence pointer in the manager, the spinlock usage was
mostly incorrect as well. Then finally remove all the nonsense checks and
actually add them in the correct locations.
v2: some typo fixes and cleanups suggested by Sunil
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Sunil Khatri <sunil.khatri@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add support for JPEG_5_0_2
v2: comment out RAS for now (Alex)
v3: drop some bringup leftovers (Alex)
Signed-off-by: Sonny Jiang <sonjiang@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add support for VCN_5_0_2
v2: squash in RRMT enable bit fix from Sonny (Alex)
v3: sqaush in doorbell enablement patch (Alex)
v4: drop some bringup leftovers (Alex)
Signed-off-by: Sonny Jiang <sonjiang@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
ras_core_get_utc_second_timestamp() retrieves the current UTC timestamp
(in seconds since the Unix epoch) through a platform-specific RAS system
callback and is used for timestamping RAS error events.
The function checks ras_core in the conditional statement before calling
the sys_fn callback. However, when the condition fails, the function
prints an error message using ras_core->dev.
If ras_core is NULL, this can lead to a potential NULL pointer
dereference when accessing ras_core->dev.
Add an early NULL check for ras_core at the beginning of the function
and return 0 when the pointer is not valid. This prevents the
dereference and makes the control flow clearer.
Fixes: 13c91b5b43 ("drm/amd/ras: Add rascore unified interface function")
Cc: YiPeng Chai <YiPeng.Chai@amd.com>
Cc: Dan Carpenter <dan.carpenter@linaro.org>
Cc: Tao Zhou <tao.zhou1@amd.com>
Cc: Hawking Zhang <Hawking.Zhang@amd.com>
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: YiPeng Chai <YiPeng.Chai@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Newer DCN bandwidth calculations require new definitions.
[How]
Add new fields cpu_id and vram_bit_width for
atom_integrated_system_info_v2_3, and add a memtype for LpDDR5x.
Reviewed-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add NULL pointer checks for dev->type before accessing
dev->type->name in ISP genpd add/remove functions to
prevent kernel crashes.
This regression was introduced in v7.0 as the wakeup sources
are registered using physical device instead of ACPI device.
This led to adding wakeup source device as the first child of
AMDGPU device without initializing dev-type variable, and
resulted in segfault when accessed it in the amdgpu isp driver.
Fixes: 057edc58aa ("ACPI: PM: Register wakeup sources under physical devices")
Suggested-by: Bin Du <Bin.Du@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Pratap Nirujogi <pratap.nirujogi@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The value should never exceed the array size as those
are the only values the hardware is expected to return,
but add checks anyway.
Cc: Benjamin Cheng <benjamin.cheng@amd.com>
Reviewed-by: Benjamin Cheng <benjamin.cheng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The value should never exceed the array size as those
are the only values the hardware is expected to return,
but add checks anyway.
Reviewed-by: Benjamin Cheng <benjamin.cheng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The value should never exceed the array size as those
are the only values the hardware is expected to return,
but add checks anyway.
Reviewed-by: Benjamin Cheng <benjamin.cheng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Charlene Liu
Meenakshikumar Somasundaram
Eric Huang
Ruijing Dong
Alex Deucher
Srinivasan Shanmugam
Christian König
Yussuf Khalil
Hawking Zhang
Jesse.Zhang
Dave Airlie
Lijo Lazar
Harry Wentland
Mangesh Gadre
Sonny Jiang
Asad Kamal
Leo Li
Pratap Nirujogi