The purpose of halt_if_hws_hang is to preserve GPU state for driver
debugging when queue preemption fails. Issuing per-queue reset may
kill wavefronts which caused the preemption failure.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Reviewed-by: Jonathan Kim <Jonathan.Kim@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org # 6.12.x
[why]
Updating the cursor enablement register can be a slow operation and accumulates
when high polling rate cursors cause frequent updates asynchronously to the
cursor position.
[how]
Since the cursor enable bit is cached there is no need to update the
enablement register if there is no change to it. This removes the
read-modify-write from the cursor position programming path in HUBP and
DPP, leaving only the register writes.
Cc: Mario Limonciello <mario.limonciello@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Reviewed-by: Sung Lee <sung.lee@amd.com>
Signed-off-by: Aric Cyr <Aric.Cyr@amd.com>
Signed-off-by: Wayne Lin <wayne.lin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
When HUBP is power gated, the SW state can get out of sync with the
hardware state causing cursor to not be programmed correctly.
[How]
Similar to DPP, add a HUBP reset function which is called wherever
HUBP is initialized or powergated. This function will clear the cursor
position and attribute cache allowing for proper programming when the
HUBP is brought back up.
Cc: Mario Limonciello <mario.limonciello@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Reviewed-by: Sung Lee <sung.lee@amd.com>
Signed-off-by: Aric Cyr <Aric.Cyr@amd.com>
Signed-off-by: Wayne Lin <wayne.lin@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
MES internal will check CP_MES_MSCRATCH_LO/HI register to set scratch
data location during ucode start, driver side need to start the MES
one by one with different setting for each pipe
Signed-off-by: Shaoyun Liu <shaoyun.liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Effectively amdgpu.gttsize gets set to ~1/2 of RAM, but that's controlled
by what the TTM page limit is set to. Clarify the kdoc.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If the parent is NULL, adev->pdev is used to retrieve the PCIe speed and
width, ensuring that the function can still determine these
capabilities from the device itself.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c:6193 amdgpu_device_gpu_bandwidth()
error: we previously assumed 'parent' could be null (see line 6180)
drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
6170 static void amdgpu_device_gpu_bandwidth(struct amdgpu_device *adev,
6171 enum pci_bus_speed *speed,
6172 enum pcie_link_width *width)
6173 {
6174 struct pci_dev *parent = adev->pdev;
6175
6176 if (!speed || !width)
6177 return;
6178
6179 parent = pci_upstream_bridge(parent);
6180 if (parent && parent->vendor == PCI_VENDOR_ID_ATI) {
^^^^^^
If parent is NULL
6181 /* use the upstream/downstream switches internal to dGPU */
6182 *speed = pcie_get_speed_cap(parent);
6183 *width = pcie_get_width_cap(parent);
6184 while ((parent = pci_upstream_bridge(parent))) {
6185 if (parent->vendor == PCI_VENDOR_ID_ATI) {
6186 /* use the upstream/downstream switches internal to dGPU */
6187 *speed = pcie_get_speed_cap(parent);
6188 *width = pcie_get_width_cap(parent);
6189 }
6190 }
6191 } else {
6192 /* use the device itself */
--> 6193 *speed = pcie_get_speed_cap(parent);
^^^^^^ Then we are toasted here.
6194 *width = pcie_get_width_cap(parent);
6195 }
6196 }
Fixes: 757e8b951c ("drm/amdgpu: cache gpu pcie link width")
Cc: Christian König <christian.koenig@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The function amdgpu_dm_crtc_mem_type_changed was dereferencing pointers
returned by drm_atomic_get_plane_state without checking for errors. This
could lead to undefined behavior if the function returns an error pointer.
This commit adds checks using IS_ERR to ensure that new_plane_state and
old_plane_state are valid before dereferencing them.
Fixes the below:
drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c:11486 amdgpu_dm_crtc_mem_type_changed()
error: 'new_plane_state' dereferencing possible ERR_PTR()
drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm.c
11475 static bool amdgpu_dm_crtc_mem_type_changed(struct drm_device *dev,
11476 struct drm_atomic_state *state,
11477 struct drm_crtc_state *crtc_state)
11478 {
11479 struct drm_plane *plane;
11480 struct drm_plane_state *new_plane_state, *old_plane_state;
11481
11482 drm_for_each_plane_mask(plane, dev, crtc_state->plane_mask) {
11483 new_plane_state = drm_atomic_get_plane_state(state, plane);
11484 old_plane_state = drm_atomic_get_plane_state(state, plane);
^^^^^^^^^^^^^^^^^^^^^^^^^^ These functions can fail.
11485
--> 11486 if (old_plane_state->fb && new_plane_state->fb &&
11487 get_mem_type(old_plane_state->fb) != get_mem_type(new_plane_state->fb))
11488 return true;
11489 }
11490
11491 return false;
11492 }
Fixes: 4caacd1671 ("drm/amd/display: Do not elevate mem_type change to full update")
Cc: Leo Li <sunpeng.li@amd.com>
Cc: Tom Chung <chiahsuan.chung@amd.com>
Cc: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Cc: Roman Li <roman.li@amd.com>
Cc: Alex Hung <alex.hung@amd.com>
Cc: Aurabindo Pillai <aurabindo.pillai@amd.com>
Cc: Harry Wentland <harry.wentland@amd.com>
Cc: Hamza Mahfooz <hamza.mahfooz@amd.com>
Reported-by: Dan Carpenter <dan.carpenter@linaro.org>
Signed-off-by: Srinivasan Shanmugam <srinivasan.shanmugam@amd.com>
Reviewed-by: Roman Li <roman.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
commit 26c95e838e ("drm/amdgpu: set the VM pointer to NULL in
amdgpu_job_prepare") set job->vm as NULL if there is no fence. It will
cause emit switch buffer be skippen if job->vm set as NULL.
Check job rather than vm could solve this problem.
Fixes: 26c95e838e ("drm/amdgpu: set the VM pointer to NULL in amdgpu_job_prepare")
Signed-off-by: Lin.Cao <lincao12@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the initialization and usage of SMU v13.0.6 capability values. Use
caps_set/clear functions to set/clear capability.
Also, fix SET_UCLK_MAX capability on APUs, it is supported on APUs.
Fixes: e9b86b841b ("drm/amd/pm: Add capability flags for SMU v13.0.6")
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch refactors the firmware version checks in `smu_v13_0_6_reset_sdma`
to support multiple SMU programs with different firmware version thresholds.
V2: return -EOPNOTSUPP for unspported pmfw
Suggested-by: Lazar Lijo <Lijo.Lazar@amd.com>
Signed-off-by: Jesse Zhang <jesse.zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add capability flags for SMU v13.0.6 variants. Initialize the flags
based on firmware support. As there are multiple IP versions maintained,
it is more manageable with one time initialization of caps flags based
on IP version and firmware feature support.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When compiling allmodconfig (CONFIG_WERROR=y) with clang-19, see the
following errors:
.../display/dc/dml2/display_mode_core.c:6268:13: warning: stack frame size (3128) exceeds limit (3072) in 'dml_prefetch_check' [-Wframe-larger-than]
.../display/dc/dml2/dml21/src/dml2_core/dml2_core_dcn4_calcs.c:7236:13: warning: stack frame size (3256) exceeds limit (3072) in 'dml_core_mode_support' [-Wframe-larger-than]
Mark static functions called by dml_prefetch_check() and
dml_core_mode_support() noinline_for_stack to avoid them become huge
functions and thus exceed the frame size limit.
A way to reproduce:
$ git checkout next-20250107
$ mkdir build_dir
$ export PATH=/tmp/llvm-19.1.6-x86_64/bin:$PATH
$ make LLVM=1 O=build_dir allmodconfig
$ make LLVM=1 O=build_dir drivers/gpu/drm/ -j
The way how it chose static functions to mark:
[0] Unset CONFIG_WERROR in build_dir/.config.
To get display_mode_core.o without errors.
[1] Get a function list called by dml_prefetch_check().
$ sed -n '6268,6711p' drivers/gpu/drm/amd/display/dc/dml2/display_mode_core.c \
| sed -n -r 's/.*\W(\w+)\(.*/\1/p' | sort -u >/tmp/syms
[2] Get the non-inline function list.
Objdump won't show the symbols if they are inline functions.
$ make LLVM=1 O=build_dir drivers/gpu/drm/ -j
$ objdump -d build_dir/.../display_mode_core.o | \
./scripts/checkstack.pl x86_64 0 | \
grep -f /tmp/syms | cut -d' ' -f2- >/tmp/orig
[3] Get the full function list.
Append "-fno-inline" to `CFLAGS_.../display_mode_core.o` in
drivers/gpu/drm/amd/display/dc/dml2/Makefile.
$ make LLVM=1 O=build_dir drivers/gpu/drm/ -j
$ objdump -d build_dir/.../display_mode_core.o | \
./scripts/checkstack.pl x86_64 0 | \
grep -f /tmp/syms | cut -d' ' -f2- >/tmp/noinline
[4] Get the inline function list.
If a symbol only in /tmp/noinline but not in /tmp/orig, it is a good
candidate to mark noinline.
$ diff /tmp/orig /tmp/noinline
Chosen functions and their stack sizes:
CalculateBandwidthAvailableForImmediateFlip [display_mode_core.o]:144
CalculateExtraLatency [display_mode_core.o]:176
CalculateTWait [display_mode_core.o]:64
CalculateVActiveBandwithSupport [display_mode_core.o]:112
set_calculate_prefetch_schedule_params [display_mode_core.o]:48
CheckGlobalPrefetchAdmissibility [dml2_core_dcn4_calcs.o]:544
calculate_bandwidth_available [dml2_core_dcn4_calcs.o]:320
calculate_vactive_det_fill_latency [dml2_core_dcn4_calcs.o]:272
CalculateDCFCLKDeepSleep [dml2_core_dcn4_calcs.o]:208
CalculateODMMode [dml2_core_dcn4_calcs.o]:208
CalculateOutputLink [dml2_core_dcn4_calcs.o]:176
Signed-off-by: Tzung-Bi Shih <tzungbi@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If user shader issues S_SETVSKIP then this state will persist when
executing the trap handler, causing vector instructions to be
skipped.
VSKIP state is already saved/restored through the MODE register.
Signed-off-by: Jay Cornwall <jay.cornwall@amd.com>
Reviewed-by: Lancelot Six <lancelot.six@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
'add ip block' causes a confusion if the blocks are disabled later with
ip_block_mask. Instead change to 'detected' and also add device context.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Context empty interrupt is enabled for SDMA 4.4.2. Add a handler for
context empty interrupt so that it is disposed of fast, and not
propagated to KFD layer.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Some monitors flicker when subvp is enabled which maybe related to
an uncommon timing they use. To isolate such issues, add a debug
option to help isolate this the issue for debugging.
Signed-off-by: Aurabindo Pillai <aurabindo.pillai@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For partial migrate from ram to vram, the migrate->cpages is not
equal to migrate->npages, should use migrate->npages to check all needed
migrate pages which could be copied or not.
And only need to set those pages could be migrated to migrate->dst[i], or
the migrate_vma_pages will migrate the wrong pages based on the migrate->dst[i].
v2:
Add mpages to break the loop earlier.
v3:
Uses MIGRATE_PFN_MIGRATE to identify whether page could be migrated.
v4:
Correct the error part.
Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: Philip Yang<Philip.Yang@amd.com>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit enables the cleaner shader feature for GFX12.0 and GFX12.0.1
GPUs. The cleaner shader is important for clearing GPU resources such as
Local Data Share (LDS), Vector General Purpose Registers (VGPRs), and
Scalar General Purpose Registers (SGPRs) between workloads.
- This feature ensures that GPU resources are reset between workloads,
preventing data leaks and ensuring accurate computation.
By enabling the cleaner shader, this update enhances the security and
reliability of GPU operations on GFX12.0 hardware.
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: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We sometimes have people trying to use debugging options in production
environments.
Mark options only meant to be used for debugging as unsafe so that the
kernel is tainted when they are used.
Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Felix Kuehling <felix.kuehling@amd.com>
Acked-by: Simona Vetter <simona.vetter@ffwll.ch>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This version brings along following fixes:
- Add some feature for secure display
- Add replay desync error count tracking and reset
- Update chip_cap defines and usage
- Remove unnecessary eDP power down
- Fix some stuttering/corruption issue on PSR panel
- Cleanup and refactoring DML2.1
Acked-by: Wayne Lin <wayne.lin@amd.com>
Reviewed-by: Martin Leung <martin.leung@amd.com>
Signed-off-by: Ryan Seto <ryanseto@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[WHY]
We see unstable DP LL 4.2.1.3 test result with dpia pre-train. It is
because the outbox interrupt mechanism can not handle HPD
immediately and require some improvement.
[HOW]
1. not enable link if hpd_pending is true.
2. abort pre-train if training failed and hpd_pending is true.
3. check if 2 lane supported when it is alt mode
Reviewed-by: Wenjing Liu <wenjing.liu@amd.com>
Reviewed-by: Meenakshikumar Somasundaram <meenakshikumar.somasundaram@amd.com>
Signed-off-by: Peichen Huang <PeiChen.Huang@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
The functions read_ono_state are no longer in use and have been identified
as redundant.
Removing them helps streamline the codebase and improve maintainability by
eliminating unnecessary code.
[How]
These unused functions were removed from Hwss module, ensuring that no
functionality is affected, and the code is simplified.
Reviewed-by: Martin Leung <martin.leung@amd.com>
Signed-off-by: Karthi Kandasamy <karthi.kandasamy@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
There should not be any need to revalidate bandwidth on memory placement
change, since the fb is expected to be pinned to DCN-accessable memory
before scanout. For APU it's DRAM, and DGPU, it's VRAM. However, async
flips + memory type change needs to be rejected.
[How]
Do not set lock_and_validation_needed on mem_type change. Instead,
reject an async_flip request if the crtc's buffer(s) changed mem_type.
This may fix stuttering/corruption experienced with PSR SU and PSR1
panels, if the compositor allocates fbs in both VRAM carveout and GTT
and flips between them.
Fixes: a7c0cad0dc ("drm/amd/display: ensure async flips are only accepted for fast updates")
Reviewed-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Outside of a modeset/link configuration change, we should not have to
wait for the panel to exit PSR. Depending on the panel and it's state,
it may take multiple frames for it to exit PSR. Therefore, waiting in
all scenarios may cause perceived stuttering, especially in combination
with faster vblank shutdown.
[How]
PSR1 disable is hooked up to the vblank enable event, and vice versa. In
case of vblank enable, do not wait for panel to exit PSR, but still wait
in all other cases.
We also avoid a call to unnecessarily change power_opts on disable -
this ends up sending another command to dmcub fw.
When testing against IGT, some crc tests like kms_plane_alpha_blend and
amd_hotplug were failing due to CRC timeouts. This was found to be
caused by the early return before HW has fully exited PSR1. Fix this by
first making sure we grab a vblank reference, then waiting for panel to
exit PSR1, before programming hw for CRC generation.
Fixes: 58a261bfc9 ("drm/amd/display: use a more lax vblank enable policy for older ASICs")
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3743
Reviewed-by: Tom Chung <chiahsuan.chung@amd.com>
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[why]
When first time of link training is fail,
eDP would be powered down and
would not be powered up for next retry link training.
It causes that all of retry link linking would be fail.
[how]
We has extracted both power up and down sequence from
enable/disable link output function before DCN32.
We remov eDP power down in dcn32_disable_link_output().
Reviewed-by: Charlene Liu <charlene.liu@amd.com>
Signed-off-by: Yiling Chen <yi-ling.chen2@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Revert commit 284f141f5c ("drm/amd/display: Enable urgent latency adjustments for DCN35")
[Why & How]
Urgent latency increase caused 2.8K OLED monitor caused it to
block this panel support P0.
Reverting this change does not reintroduce the netflix corruption issue
which it fixed.
Fixes: 284f141f5c ("drm/amd/display: Enable urgent latency adjustments for DCN35")
Reviewed-by: Charlene Liu <charlene.liu@amd.com>
Signed-off-by: Nicholas Susanto <Nicholas.Susanto@amd.com>
Signed-off-by: Tom Chung <chiahsuan.chung@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
