The KFD pre_reset should be called before reset been executed, it will
hold the lock to prevent other rocm process to sent the packlage to hiq
during host execute the real reset on the HW
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There was a change(below) target for such issue:
d82e2c249c ("drm/amdgpu: Fix crash on device remove/driver unload")
But the fix for VI ASICs was missing there. This is a supplement for
that.
Fixes: d82e2c249c ("drm/amdgpu: Fix crash on device remove/driver unload")
Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Properly handle SI DC support when CONFIG_DRM_AMD_DC_SI is not
set.
Fixes: f7f12b2582 ("drm/amdgpu: default to true in amdgpu_device_asic_has_dc_support")
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If a kfd_bo was shared (e.g. a dmabuf export), the original kfd_bo may be
freed when the amdgpu_bo still lives on. Free the kfd_bo struct in the
release_notify callback then the amdgpu_bo is freed.
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-By: Ramesh Errabolu <Ramesh.Errabolu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
As part of the ib padding process, accessing the RLC_SPM_* register may
trigger gfx hang. Since gfxoff may be already kicked during the whole period.
To address that, we manually toggle gfx on/off around the RLC_SPM_*
register access.
This can resolve the gfx hang issue observed on running Talos with RDP launched
in parallel.
Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
csb bo is not unpinned in gfx 9. It will lead to pin_count leak on
driver unload.
[How]
Call bo_free_kernel corresponding to bo_create_kernel in
gfx_rlc_init_csb. This will also unify the code path with other gfx
versions.
Signed-off-by: YuBiao Wang <YuBiao.Wang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
For Vega10, disabling gart of gfxhub could mess up KIQ and PSP
under sriov mode, and lead to DMAR on host side.
[How]
Skip writing GMC registers under sriov.
Signed-off-by: YuBiao Wang <YuBiao.Wang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
BOs need to be reserved before they are added or removed, so ensure that
they are reserved during kfd_mem_attach and kfd_mem_detach
Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The huge page functionality in TTM does not work safely because PUD and
PMD entries do not have a special bit.
get_user_pages_fast() considers any page that passed pmd_huge() as
usable:
if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd) ||
pmd_devmap(pmd))) {
And vmf_insert_pfn_pmd_prot() unconditionally sets
entry = pmd_mkhuge(pfn_t_pmd(pfn, prot));
eg on x86 the page will be _PAGE_PRESENT | PAGE_PSE.
As such gup_huge_pmd() will try to deref a struct page:
head = try_grab_compound_head(pmd_page(orig), refs, flags);
and thus crash.
Thomas further notices that the drivers are not expecting the struct page
to be used by anything - in particular the refcount incr above will cause
them to malfunction.
Thus everything about this is not able to fully work correctly considering
GUP_fast. Delete it entirely. It can return someday along with a proper
PMD/PUD_SPECIAL bit in the page table itself to gate GUP_fast.
Fixes: 314b6580ad ("drm/ttm, drm/vmwgfx: Support huge TTM pagefaults")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Reviewed-by: Thomas Hellström <thomas.helllstrom@linux.intel.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
[danvet: Update subject per Thomas' &Christian's review]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/0-v2-a44694790652+4ac-ttm_pmd_jgg@nvidia.com
This is more useful when talking to the SMU team to have the information
in this format, save one less step to manually do it.
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Aldebaran has different register mask definitions for
regiter MC_VM_XGMI_LFB_CNTL. Use the correct masks
to interpret fields of this register.
Signed-off-by: Oak Zeng <Oak.Zeng@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
In advance tdr mode, the real bad job will be resubmitted twice, while
in drm_sched_resubmit_jobs_ext, there's a dma_fence_put, so the bad job
is put one more time than other jobs.
[How]
Adding dma_fence_get before resbumit job in
amdgpu_device_recheck_guilty_jobs and put the fence for normal jobs
Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Remove GPRs init for ALDEBARAN in gpu reset temporarily, will add the init once the
algorithm is stable.
v2: Only remove GPRs init in gpu reset.
v3: Suspend needs it, only skip it in gpu reset.
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>
Skip GPRs init in specific condition since current GPRs init algorithm only works for some CU settings.
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>
amdgpu_fence_driver_sw_fini() should be executed before
amdgpu_device_ip_fini(), otherwise fence driver resource
won't be properly freed as adev->rings have been tore down.
Fixes: 72c8c97b15 ("drm/amdgpu: Split amdgpu_device_fini into early and late")
Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently, all kfd BOs use same destruction routine. But pinned
BOs are not unpinned properly. Separate them from general routine.
v2 (Felix):
Add safeguard to prevent user space from freeing signal BO.
Kunmap signal BO in the event of setting event page error.
Just kunmap signal BO to avoid duplicating the code.
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>
The userptr can be unmapped by application and still registered to
driver, restore userptr work return user pages will get -EFAULT bad
address error. Pretend this error as succeed. GPU access this userptr
will have VM fault later, it is better than application soft hangs with
stalled user mode queues.
v2: squash in warning fix (Alex)
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When a GPU hits the bad_page_threshold, it will not be initialized by
the amdgpu driver. This means that the table cannot be cleared, nor can
information gathering be performed (getting serial number, BDF, etc).
If the bad_page_threshold kernel parameter is set to -2,
continue to initialize the GPU, while printing a warning to dmesg that
this action has been done
v2: squash in Luben's fix to restore RAS info reporting
Cc: Luben Tuikov <luben.tuikov@amd.com>
Cc: Mukul Joshi <Mukul.Joshi@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
dmesg doesn't warn when the number of bad pages approaches the
threshold for page retirement. WARN when the number of bad pages
is at 90% or greater for easier checks and planning, instead of waiting
until the GPU is full of bad pages.
Cc: Luben Tuikov <luben.tuikov@amd.com>
Cc: Mukul Joshi <Mukul.Joshi@amd.com>
Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Luben Tuikov <luben.tuikov@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The extended bits were not available for use on vega20 and
presumably arcturus as well.
Fixes: a0f9f85466 ("drm/amdgpu/nbio7.4: don't use GPU_HDP_FLUSH bit 12")
Reviewed-and-tested-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
No need to use the id variable, just use the constant
plus instance offset directly.
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Roughly the same code was present in all VCN versions.
Consolidate it into a single function.
v2: use AMDGPU_UCODE_ID_VCN + i, check if num_inst >= 2
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: James Zhu <James.Zhu@amd.com>
PSP firmware will be responsible for applying the GRBM CAM remapping in
the production. And the GRBM_CAM_INDEX / GRBM_CAM_DATA registers will be
protected by PSP under security policy. So remove it according to the
new security policy.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
No longer used since IP enumeration is driven by the IP
discovery table now.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
No longer used since IP enumeration is now driven by
amdgpu IP discovery code.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
No longer used since IP enumeration is now driven by
amdgpu IP discovery code.
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Check was incorrectly converted to IP version checking.
Fixes: 4b0ad84254 ("drm/amdgpu/gfx10: convert to IP version checking")
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
show() must not use snprintf() when formatting the value to be
returned to user space.
Fix the following coccicheck warning:
drivers/gpu/drm/amd/amdgpu/amdgpu_ucode.c:427:
WARNING: use scnprintf or sprintf.
Signed-off-by: Qing Wang <wangqing@vivo.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
B0 internal rev_id is 0x01, B1 internal rev_id is 0x02.
The external rev_id for B0 and B1 is 0x20.
The original expression is not suitable for B1.
v2: squash in fix for display code (Alex)
Signed-off-by: Aaron Liu <aaron.liu@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Output a warning message if RAS TA returns UNSUPPORTED_ERROR_INJ status.
v2: implement it in psp_ras_ta_check_status function.
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>
[Why]
drm_irq_uninstall is called in irq_fini_hw so that irq is disabled in sw
stage. SMU (and maybe other IP blocks) fini_hw will call irq_put for
cleanup and the whole cleanup process will be skipped because of
drm->irq_enable = false.
[How]
Move ip_fini_early before irq_fini_hw.
Signed-off-by: YuBiao Wang <YuBiao.Wang@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
