As we use wb slots more dynamically, we need to lock
access to avoid racing on allocation or free.
Reviewed-by: Shaoyun.liu <shaoyunl@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Subtract the VRAM pinned memory when checking for available memory
in amdgpu_amdkfd_reserve_mem_limit function since that memory is not
available for use.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When slice_height is 0, the division by slice_height in the calculation
of the number of slices will cause a division by zero driver crash. This
leaves the kernel in a state that requires a reboot. This patch adds a
check to avoid the division by zero.
The stack trace below is for the 6.8.4 Kernel. I reproduced the issue on
a Z16 Gen 2 Lenovo Thinkpad with a Apple Studio Display monitor
connected via Thunderbolt. The amdgpu driver crashed with this exception
when I rebooted the system with the monitor connected.
kernel: ? die (arch/x86/kernel/dumpstack.c:421 arch/x86/kernel/dumpstack.c:434 arch/x86/kernel/dumpstack.c:447)
kernel: ? do_trap (arch/x86/kernel/traps.c:113 arch/x86/kernel/traps.c:154)
kernel: ? setup_dsc_config (drivers/gpu/drm/amd/amdgpu/../display/dc/dsc/dc_dsc.c:1053) amdgpu
kernel: ? do_error_trap (./arch/x86/include/asm/traps.h:58 arch/x86/kernel/traps.c:175)
kernel: ? setup_dsc_config (drivers/gpu/drm/amd/amdgpu/../display/dc/dsc/dc_dsc.c:1053) amdgpu
kernel: ? exc_divide_error (arch/x86/kernel/traps.c:194 (discriminator 2))
kernel: ? setup_dsc_config (drivers/gpu/drm/amd/amdgpu/../display/dc/dsc/dc_dsc.c:1053) amdgpu
kernel: ? asm_exc_divide_error (./arch/x86/include/asm/idtentry.h:548)
kernel: ? setup_dsc_config (drivers/gpu/drm/amd/amdgpu/../display/dc/dsc/dc_dsc.c:1053) amdgpu
kernel: dc_dsc_compute_config (drivers/gpu/drm/amd/amdgpu/../display/dc/dsc/dc_dsc.c:1109) amdgpu
After applying this patch, the driver no longer crashes when the monitor
is connected and the system is rebooted. I believe this is the same
issue reported for 3113.
Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Jose Fernandez <josef@netflix.com>
Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/3113
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add the prototype to dump ip registers
for all ips of different asics and set
them to NULL for now. Based on the
requirement add a function pointer for
each of them.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit addresses buffer overflow in the smu_v14_0_init_microcode
function. The issue was about the snprintf function writing more bytes
into the fw_name buffer than it can hold.
The line of code is:
snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
Here, snprintf is used to write a formatted string into fw_name. The
format is "amdgpu/%s.bin", where %s is a placeholder for the string
ucode_prefix. The sizeof(fw_name) argument tells snprintf the maximum
number of bytes it can write into fw_name, including the
null-terminating character. In the original code, fw_name is an array of
30 characters.
The string "amdgpu/%s.bin" could be up to 41 bytes long, which exceeds
the 30 bytes allocated for fw_name. This is because %s could be replaced
by ucode_prefix, which can be up to 29 characters long. Adding the 12
characters from "amdgpu/" and ".bin", the total length could be 41
characters.
To address this, the size of ucode_prefix has been reduced to 15
characters. This ensures that the maximum length of the string written into
fw_name does not exceed its capacity.
smu_13/14 etc. don't follow legacy scheme ie., amdgpu_ucode_legacy_naming
Fixes the below with gcc W=1:
drivers/gpu/drm/amd/amdgpu/../pm/swsmu/smu14/smu_v14_0.c: In function ‘smu_v14_0_init_microcode’:
drivers/gpu/drm/amd/amdgpu/../pm/swsmu/smu14/smu_v14_0.c:80:52: warning: ‘%s’ directive output may be truncated writing up to 29 bytes into a region of size 23 [-Wformat-truncation=]
80 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
| ^~ ~~~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../pm/swsmu/smu14/smu_v14_0.c:80:9: note: ‘snprintf’ output between 12 and 41 bytes into a destination of size 30
80 | snprintf(fw_name, sizeof(fw_name), "amdgpu/%s.bin", ucode_prefix);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Fixes: fe6cd91524 ("drm/amd/swsmu: add smu14 ip support")
Cc: Li Ma <li.ma@amd.com>
Cc: Likun Gao <Likun.Gao@amd.com>
Cc: Lijo Lazar <lijo.lazar@amd.com>
Cc: Kenneth Feng <kenneth.feng@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>
Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Here are the corrections needed for the queue ring buffer size
calculation for the following cases:
- Remove the KIQ VM flush ring usage.
- Add the invalidate TLBs packet for gfx10 and gfx11 queue.
- There's no VM flush and PFP sync, so remove the gfx9 real
ring and compute ring buffer usage.
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>
Do VRAM accounting when doing migrations to vram to make sure
there is enough available VRAM and migrating to VRAM doesn't evict
other possible non-unified memory BOs. If migrating to VRAM fails,
driver can fall back to using system memory seamlessly.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During mode-2 reset, pci config space registers are affected at device
side. However, certain platforms have switches which assign virtual BAR
addresses and returns the same even after device is reset. This
affects pci_restore_state() as it doesn't issue another config write, if
the value read is same as the saved value.
Add a workaround to write saved config space values from driver side.
Presently, these switches are in platforms with SMU v13.0.6 SOCs, hence
restrict the workaround only to those.
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>
umsch test needs full GPU functionality(e.g., VM update, TLB flush,
possibly buffer moving under memory pressure) which may be not ready
under these states. Just skip it to avoid potential issues.
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Veerabadhran Gopalakrishnan <Veerabadhran.Gopalakrishnan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Handle the case that the restore worker was already scheduled by another
eviction while the restore was in progress.
Fixes: 9a1c1339ab ("drm/amdkfd: Run restore_workers on freezable WQs")
Signed-off-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Tested-by: Yunxiang Li <Yunxiang.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Here are the corrections needed for the queue ring buffer size
calculation for the following cases:
- Remove the KIQ VM flush ring usage.
- Add the invalidate TLBs packet for gfx10 and gfx11 queue.
- There's no VM flush and PFP sync, so remove the gfx9 real
ring and compute ring buffer usage.
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>
Do VRAM accounting when doing migrations to vram to make sure
there is enough available VRAM and migrating to VRAM doesn't evict
other possible non-unified memory BOs. If migrating to VRAM fails,
driver can fall back to using system memory seamlessly.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During mode-2 reset, pci config space registers are affected at device
side. However, certain platforms have switches which assign virtual BAR
addresses and returns the same even after device is reset. This
affects pci_restore_state() as it doesn't issue another config write, if
the value read is same as the saved value.
Add a workaround to write saved config space values from driver side.
Presently, these switches are in platforms with SMU v13.0.6 SOCs, hence
restrict the workaround only to those.
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>
Make SVM BOs more likely to get evicted than other BOs. These BOs
opportunistically use available VRAM, but can fall back relatively
seamlessly to system memory. It also avoids SVM migrations evicting
other, more important BOs as they will evict other SVM allocations
first.
Signed-off-by: Felix Kuehling <felix.kuehling@amd.com>
Acked-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The vpe dpm settings should be done before firmware is loaded.
Otherwise, the frequency cannot be successfully raised.
Signed-off-by: Peyton Lee <peytolee@amd.com>
Reviewed-by: Lang Yu <lang.yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When page table BOs were evicted but not validated before
updating page tables, VM is still in evicting state,
amdgpu_vm_update_range returns -EBUSY and
restore_process_worker runs into a dead loop.
v2: Split the BO validation and page table update into two
separate loops in amdgpu_amdkfd_restore_process_bos. (Felix)
1.Validate BOs
2.Validate VM (and DMABuf attachments)
3.Update page tables for the BOs validated above
Fixes: 50661eb1a2 ("drm/amdgpu: Auto-validate DMABuf imports in compute VMs")
Signed-off-by: Lang Yu <Lang.Yu@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Handle case that dma_fence_get_rcu_safe returns NULL.
If restore work is already scheduled, only update its timer. The same
work item cannot be queued twice, so undo the extra queue eviction.
Fixes: 9a1c1339ab ("drm/amdkfd: Run restore_workers on freezable WQs")
Signed-off-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Tested-by: Gang BA <Gang.Ba@amd.com>
Reviewed-by: Gang BA <Gang.Ba@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Otherwise we can end up with a frame on unsuspend where color management
is not applied when userspace has not committed themselves.
Fixes re-applying color management on Steam Deck/Gamescope on S3 resume.
Signed-off-by: Joshua Ashton <joshua@froggi.es>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Make SVM BOs more likely to get evicted than other BOs. These BOs
opportunistically use available VRAM, but can fall back relatively
seamlessly to system memory. It also avoids SVM migrations evicting
other, more important BOs as they will evict other SVM allocations
first.
Signed-off-by: Felix Kuehling <felix.kuehling@amd.com>
Acked-by: Mukul Joshi <mukul.joshi@amd.com>
Tested-by: Mukul Joshi <mukul.joshi@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Avoid returning an uninitialized value if we never enter the loop.
This case should never be hit in practice, but returning 0 doesn't
hurt.
The same fix is applied to the 4 places using the same pattern.
v2: - fixed typos in commit message (Alex)
- use "return 0;" before the done label instead of initializing
r to 0
Signed-off-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Makes it easier to review the logs when there are MES
errors.
v2: use dbg for emitted, add helpers for fetching strings
v3: fix missing commas (Harish)
v4: drop command prefixes (Felix)
v5: squash in bounds fix (Jun)
Acked-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed by Shaoyun.liu <Shaoyun.liu@amd.com> (v2)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The vpe dpm settings should be done before firmware is loaded.
Otherwise, the frequency cannot be successfully raised.
Signed-off-by: Peyton Lee <peytolee@amd.com>
Reviewed-by: Lang Yu <lang.yu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
HDP Flush request bit can be kept unique per AID, and doesn't need to be
unique SOC-wide. Assign only bits 10-13 for SDMA v4.4.2.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
