The adev->pm.mutx is already held at the beginning of
amdgpu_dpm_compute_clocks/amdgpu_dpm_enable_uvd/amdgpu_dpm_enable_vce.
But on their calling path, amdgpu_display_bandwidth_update will be
called and thus its sub functions amdgpu_dpm_get_sclk/mclk. They
will then try to acquire the same adev->pm.mutex and deadlock will
occur.
By placing amdgpu_display_bandwidth_update outside of adev->pm.mutex
protection(considering logically they do not need such protection) and
restructuring the call flow accordingly, we can eliminate the deadlock
issue. This comes with no real logics change.
Fixes: 3712e7a494 ("drm/amd/pm: unified lock protections in amdgpu_dpm.c")
Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
Reported-by: Arthur Marsh <arthur.marsh@internode.on.net>
Link: https://lore.kernel.org/all/9e689fea-6c69-f4b0-8dee-32c4cf7d8f9c@molgen.mpg.de/
BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1957
Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The fatal error handler is independent from general ras interrupt
handler since there is no related IH ring.
Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add support for general RAS poison consumption handler.
v2: remove callback function for poison consumption.
Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Prepare for the implementation of poison consumption handler.
v2: separate umc handler from poison creation.
Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
- In the latest version of the header, there is a variable name change.
This should not cause any backward compatibility since the variable is
at the same offset in the struct.
Signed-off-by: Bokun Zhang <Bokun.Zhang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When dcn20_clk_src_construct() fails, we need to release clk_src.
Fixes: 6f4e6361c3 ("drm/amd/display: Add Renoir resource (v2)")
Signed-off-by: Miaoqian Lin <linmq006@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We normally runtime suspend when there are displays attached if they
are in the DPMS off state, however, if something wakes the GPU
we send a hotplug event on resume (in case any displays were connected
while the GPU was in suspend) which can cause userspace to light
up the displays again soon after they were turned off.
Prior to
commit 087451f372 ("drm/amdgpu: use generic fb helpers instead of setting up AMD own's."),
the driver took a runtime pm reference when the fbdev emulation was
enabled because we didn't implement proper shadowing support for
vram access when the device was off so the device never runtime
suspended when there was a console bound. Once that commit landed,
we now utilize the core fb helper implementation which properly
handles the emulation, so runtime pm now suspends in cases where it did
not before. Ultimately, we need to sort out why runtime suspend in not
working in this case for some users, but this should restore similar
behavior to before.
v2: move check into runtime_suspend
v3: wake ups -> wakeups in comment, retain pm_runtime behavior in
runtime_idle callback
Fixes: 087451f372 ("drm/amdgpu: use generic fb helpers instead of setting up AMD own's.")
Link: https://lore.kernel.org/r/20220403132322.51c90903@darkstar.example.org/
Tested-by: Michele Ballabio <ballabio.m@gmail.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit 36bf93216e.
It causes SVM regressions on Vega10 with XNACK-ON. Just revert it
at the moment.
./kfdtest --gtest_filter=KFDSVMRangeTest.MigratePolicyTest
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Reviewed-by: Philip Yang<Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
v1:
Add debugfs support to load/unload/invoke TA in runtime.
v2:
1. Update some variables to static.
2. Use PAGE_ALIGN to calculate shared buf size directly.
3. Remove fp check.
4. Update debugfs from read to write.
Signed-off-by: John Clements <john.clements@amd.com>
Signed-off-by: Candice Li <candice.li@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
dqm->gws_queue_count and pdd->qpd.mapped_gws_queue need to be updated
each time the queue gets evicted.
Fixes: b8020b0304 ("drm/amdkfd: Enable over-subscription with >1 GWS queue")
Signed-off-by: David Yat Sin <david.yatsin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Smatch reports this issue
virtual_link_hwss.c:32:6: warning: symbol
'virtual_setup_stream_attribute' was not declared.
Should it be static?
virtual_setup_stream_attribute is only used in
virtual_link_hwss.c, but the other functions in the
file are declared in the header file and used elsewhere.
For consistency, add the virtual_setup_stream_attribute
decl to virtual_link_hwss.h.
Signed-off-by: Tom Rix <trix@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In function si_parse_power_table(), array adev->pm.dpm.ps and its member
is allocated. If the allocation of each member fails, the array itself
is freed and returned with an error code. However, the array is later
freed again in si_dpm_fini() function which is called when the function
returns an error.
This leads to potential double free of the array adev->pm.dpm.ps, as
well as leak of its array members, since the members are not freed in
the allocation function and the array is not nulled when freed.
In addition adev->pm.dpm.num_ps, which keeps track of the allocated
array member, is not updated until the member allocation is
successfully finished, this could also lead to either use after free,
or uninitialized variable access in si_dpm_fini().
Fix this by postponing the free of the array until si_dpm_fini() and
increment adev->pm.dpm.num_ps everytime the array member is allocated.
Signed-off-by: Keita Suzuki <keitasuzuki.park@sslab.ics.keio.ac.jp>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It's a local function, let's make it static.
AGD: remove prototype in dcn10_hubp.h
Signed-off-by: Tales Lelo da Aparecida <tales.aparecida@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Clarify the smu_cmn_send_smc_msg_with_param documentation to mention two
cases exist where messages are silently dropped with no error returned.
These cases occur in unusual situations where either:
1. the message type is not allowed to a virtual GPU, or
2. a PCI recovery is underway and the HW is not yet in sync with the SW
For more details see
commit 4ea5081c82 ("drm/amd/powerplay: enable SMC message filter")
commit bf36b52e78 ("drm/amdgpu: Avoid accessing HW when suspending SW state")
(v2)
Reworked with suggestions from Luben & Paul
(v3)
Updated wording as per Luben's feedback
Corrected error stating all messages denied on virtual GPU
(each GPU has mask of which messages are allowed)
Signed-off-by: Darren Powell <darren.powell@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The idea is from
commit a50fe70780 ("drm/amdkfd: Only apply heavy-weight TLB flush on Aldebaran")
and
commit f61c40c075 ("drm/amdkfd: enable heavy-weight TLB flush on Arcturus").
At the moment, heavy-weight TLB could cause problems on ASICs except
Aldebaran and Arcturus.
A simple hipMallocManaged/hipFree program could trigger this issue.
[ 97.787657] amdgpu 0000:01:00.0: amdgpu: wait for kiq fence error: 0.
[ 106.868758] amdgpu: qcm fence wait loop timeout expired
[ 106.868966] amdgpu: The cp might be in an unrecoverable state due to an unsuccessful queues preemption
[ 106.869203] amdgpu: Failed to evict process queues
[ 106.869261] amdgpu: Failed to quiesce KFD
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
To make kfd_flush_tlb_after_unmap visible in kfd_svm.c,
move it into kfd_priv.h. And change it to an inline function.
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drm/drm-next has a build fix for the NewVision NV3052C panel
(drivers/gpu/drm/panel/panel-newvision-nv3052c.c), which needs to be
merged back to drm-misc-next, as it was failing to build there.
Signed-off-by: Paul Cercueil <paul@crapouillou.net>
[why]
These static variables save the RLC Scratch registers address.
When we install multiple GPUs (for example: XGMI setting) and
multiple GPUs call the function at same time. The RLC Scratch
registers address are changed each other. Then it caused
reading/writing from/to wrong GPU.
[how]
Removed the static from the variables. The variables are
on the stack.
Fixes: 5d447e2967 ("drm/amdgpu: add helper for rlcg indirect reg access")
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Gavin Wan <Gavin.Wan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add the waiters to the wait queue during initialization, while holding the
event spinlock. Otherwise the waiter will not get activated if the event
signals before being added to the wait queue.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Philip Yang<Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This reverts commit 863fa85e6a.
While we were testing DCN3.1 with a hub, we noticed that only one of 2
connected displays lights up when using some specific display
resolution. In summary, this was the setup:
1. Displays:
* Sharp LQ156M1JW26 (eDP): 1080@240
* BENQ SW320 (DP): 4k@60
* BENQ EX3203R (DP): 4k@60
2. Hub: Club3D CSV-7300
3. ASIC: DCN3.1
After bisecting this issue, we figured out the commit mentioned above
introduced this issue. We are investigating why this patch introduced
this regression, but we need to revert it for now.
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Mark Broadworth <Mark.Broadworth@amd.com>
Cc: Michael Strauss <michael.strauss@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
VM might already be freed when amdgpu_vm_tlb_seq_cb() is called.
We see the calltrace below.
Fix it by keeping the last flush fence around and wait for it to signal
BUG kmalloc-4k (Not tainted): Poison overwritten
0xffff9c88630414e8-0xffff9c88630414e8 @offset=5352. First byte 0x6c
instead of 0x6b Allocated in amdgpu_driver_open_kms+0x9d/0x360 [amdgpu]
age=44 cpu=0 pid=2343
__slab_alloc.isra.0+0x4f/0x90
kmem_cache_alloc_trace+0x6b8/0x7a0
amdgpu_driver_open_kms+0x9d/0x360 [amdgpu]
drm_file_alloc+0x222/0x3e0 [drm]
drm_open+0x11d/0x410 [drm]
Freed in amdgpu_driver_postclose_kms+0x3e9/0x550 [amdgpu] age=22 cpu=1
pid=2485
kfree+0x4a2/0x580
amdgpu_driver_postclose_kms+0x3e9/0x550 [amdgpu]
drm_file_free+0x24e/0x3c0 [drm]
drm_close_helper.isra.0+0x90/0xb0 [drm]
drm_release+0x97/0x1a0 [drm]
__fput+0xb6/0x280
____fput+0xe/0x10
task_work_run+0x64/0xb0
Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If lookup_event_by_id() returns a NULL "ev" pointer then the
spin_lock(&ev->lock) will crash. This was detected by Smatch:
drivers/gpu/drm/amd/amdgpu/../amdkfd/kfd_events.c:644 kfd_set_event()
error: we previously assumed 'ev' could be null (see line 639)
Fixes: 5273e82c5f ("drm/amdkfd: Improve concurrency of event handling")
Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Enabling gfxoff quirk results in perfectly usable graphical user
interface on MacBook Pro (15-inch, 2019) with Radeon Pro Vega 20 4 GB.
Without the quirk, X server is completely unusable as every few seconds
there is gpu reset due to ring gfx timeout.
Signed-off-by: Tomasz Moń <desowin@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
DP/HDMI audio on AMD PRO VII stops working after S3:
[ 149.450391] amdgpu 0000:63:00.0: amdgpu: MODE1 reset
[ 149.450395] amdgpu 0000:63:00.0: amdgpu: GPU mode1 reset
[ 149.450494] amdgpu 0000:63:00.0: amdgpu: GPU psp mode1 reset
[ 149.983693] snd_hda_intel 0000:63:00.1: refused to change power state from D0 to D3hot
[ 150.003439] amdgpu 0000:63:00.0: refused to change power state from D0 to D3hot
...
[ 155.432975] snd_hda_intel 0000:63:00.1: CORB reset timeout#2, CORBRP = 65535
The offending commit is daf8de0874 ("drm/amdgpu: always reset the asic in
suspend (v2)"). Commit 34452ac303 ("drm/amdgpu: don't use BACO for
reset in S3 ") doesn't help, so the issue is something different.
Assuming that to make HDA resume to D0 fully realized, it needs to be
successfully put to D3 first. And this guesswork proves working, by
moving amdgpu_asic_reset() to noirq callback, so it's called after HDA
function is in D3.
Fixes: daf8de0874 ("drm/amdgpu: always reset the asic in suspend (v2)")
Signed-off-by: Kai-Heng Feng <kai.heng.feng@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
"Pre-multiplied" is the default pixel blend mode for KMS/DRM, as
documented in supported_modes of drm_plane_create_blend_mode_property():
https://cgit.freedesktop.org/drm/drm-misc/tree/drivers/gpu/drm/drm_blend.c
In this mode, both 'pixel alpha' and 'plane alpha' participate in the
calculation, as described by the pixel blend mode formula in KMS/DRM
documentation:
out.rgb = plane_alpha * fg.rgb +
(1 - (plane_alpha * fg.alpha)) * bg.rgb
Considering the blend config mechanisms we have in the driver so far,
the alpha mode that better fits this blend mode is the
_PER_PIXEL_ALPHA_COMBINED_GLOBAL_GAIN, where the value for global_gain
is the plane alpha (global_alpha).
With this change, alpha property stops to be ignored. It also addresses
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1734
v2:
* keep the 8-bit value for global_alpha_value (Nicholas)
* correct the logical ordering for combined global gain (Nicholas)
* apply to dcn10 too (Nicholas)
Signed-off-by: Melissa Wen <mwen@igalia.com>
Tested-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Tested-by: Simon Ser <contact@emersion.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Currently, the IO-links to the device being removed from topology,
are not cleared. As a result, there would be dangling links left in
the KFD topology. This patch aims to fix the following:
1. Cleanup all IO links to the device being removed.
2. Ensure that node numbering in sysfs and nodes proximity domain
values are consistent after the device is removed:
a. Adding a device and removing a GPU device are made mutually
exclusive.
b. The global proximity domain counter is no longer required to be
an atomic counter. A normal 32-bit counter can be used instead.
3. Update generation_count to let user-mode know that topology has
changed due to device removal.
CC: Shuotao Xu <shuotaoxu@microsoft.com>
Reviewed-by: Shuotao Xu <shuotaoxu@microsoft.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
creating a generic helper for AMD specific PSR-SU sink validation.
Moving the function to the power module to reference it across all
OS.
[how]
- drop PSRSU specific sink validation helper and move to power
module by reading PSR version and other PSR caps
- call the new helper from linux DM (amdgpu_dm_psr)
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Acked-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: David Zhang <dingchen.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Title: DC Patches Apri 6, 2022
This DC patchset brings improvements in multiple areas. In summary, we highlight:
*Disabling Z10 on DCN31
*Fix issue breaking 32bit Linux build
*Fix inconsistent timestamp type
*Add DCN30 support FEC init
*Fix crash on setting VRR with no display connected
*Disable FEC if DSC not supported for EDP
*Add odm seamless boot support
*Select correct DTO source
*Power down hardware if timer not trigger
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Signed-off-by: Aric Cyr <aric.cyr@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The synchronize_rcu call in destroy_events can take several ms, which
noticeably slows down applications destroying many events. Use kfree_rcu
to free the event structure asynchronously and eliminate the
synchronize_rcu call in the user thread.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why] within dc link detecion, dp link training will be
executed for external sst dp. for debug purpose, we may
need skip dp link training.
[How] expose dc debug option to skip_detection_link_training
to debugfs
Reviewed-by: Roman Li <Roman.Li@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Signed-off-by: hersen wu <hersenxs.wu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[WHY]
An unsigned int timestamp variable is assigned with an unsigned
long long value. Also, the assignment directly converts the
tick value to us without using built-in get elapsed time function.
[HOW]
Cast the assigned value correctly and also use built-in function
to get the timestamp in the unit we want.
v2: squash in 64 bit division fix
Reviewed-by: Aric Cyr <Aric.Cyr@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Signed-off-by: Angus Wang <Angus.Wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
