One Vega10 SR-IOV VF, the FW address returned by PSP should be
set into the init table, while not the original BO mc address.
otherwise, UVD and VCE IB test will fail under Vega10 SR-IOV
reference:
commit bfcea52042 ("drm/amdgpu:change VEGA booting with firmware loaded by PSP")
commit aa5873dca4 ("drm/amdgpu: Change VCE booting with firmware loaded by PSP")
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SR-IOV host side will send IDH_QUERY_ALIVE to guest VM to check
if this guest VM is still alive (not destroyed). The only thing
guest KMD need to do is to send ACK back to host.
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_vm_make_compute is used to turn a GFX VM into a compute VM,
the prerequisite is this VM is clean. Let's check if some page tables
are already filled , while not check if some mapping is already made.
Signed-off-by: Trigger Huang <Trigger.Huang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Also reject TDRs if another one already running.
v2:
Stop all schedulers across device and entire XGMI hive before
force signaling HW fences.
Avoid passing job_signaled to helper fnctions to keep all the decision
making about skipping HW reset in one place.
v3:
Fix SW sched. hang after non HW reset. sched.hw_rq_count has to be balanced
against it's decrement in drm_sched_stop in non HW reset case.
v4: rebase
v5: Revert v3 as we do it now in sceduler code.
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1555599624-12285-6-git-send-email-andrey.grodzovsky@amd.com
amdgpu_sriov_vf would return 0x0 or 0x4 to indicate if sriov.
but F32_POLL_ENABLE need 0x0 or 0x1 to determine if enabled.
set 0x4 into F32_POLL_ENABLE would make SDMA0_GFX_RB_WPTR_POLL_CNTL not working.
Signed-off-by: Wentao Lou <Wentao.Lou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
smu.ppt_funcs is only initialized for ASICs supporting SW SMU.
On a Hawaii ASIC, attempting to access the udev attribute
ATTRS{power_dpm_state} will cause a null pointer deref in
amdgpu_get_dpm_state() because of this.
Fix by checking if SW SMU is supported first.
Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Under SRIOV, we need disable DRIVER_ATOMIC.
Otherwise, it will trigger WARN_ON at drm_universal_plane_init.
Signed-off-by: Yintian Tao <yttao@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_bo_destroy had a bug by calling amdgpu_bo_unref outside mutex_lock.
If amdgpu_device_recover_vram executed between amdgpu_bo_unref and list_del_init,
it would get NULL of shadow->parent, then caused Call Trace and GPU reset failed.
Signed-off-by: Wentao Lou <Wentao.Lou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Method of getting firmware version is the same across ASICs, so remove
them from ASIC-specific files and create one in amdgpu_amdkfd.c. This new
created get_fw_version simply reads fw_version from adev->gfx than parsing
the ucode header.
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
shadow was added into shadow_list by amdgpu_bo_create_shadow.
meanwhile, shadow->tbo.mem was not fully configured.
tbo.mem would be fully configured by amdgpu_vm_sdma_map_table until calling amdgpu_vm_clear_bo.
If sriov TDR occurred between amdgpu_bo_create_shadow and amdgpu_vm_sdma_map_table,
amdgpu_device_recover_vram would deal with shadow without tbo.mem.start.
Signed-off-by: Wentao Lou <Wentao.Lou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The extra ; in the macro definition creates an empty statement
preventing any variable declarations from occuring after
any use of to_dm_plane_state(...).
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Under vega10 virtualuzation, smu ip block will not be added.
Therefore, we need add pp clk query and force dpm level function
at amdgpu_virt_ops to support the feature.
v2: add get_pp_clk existence check and use kzalloc to allocate buf
v3: return -ENOMEM for allocation failure and correct the coding style
Signed-off-by: Yintian Tao <yttao@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
XGMI Memory sharing will be disbaled by default for security reason after
boot up, it depends on driver to enable the memory sharing
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
IP initialize ras in late_init, because of the BUGs of PSP or any
other components, driver receives unexpected irqs. It is ok to add such
check anyway.
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Many parts of the whole SW stack can program the ras enablement state
during the boot. Now we handle that case by adding one function which
check the ras flags and choose different code path.
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_bo_restore_shadow would assign zero to r if succeeded.
r would remain zero if there is only one node in shadow_list.
current code would always return failure when r <= 0.
restart the timeout for each wait was a rather problematic bug as well.
The value of tmo SHOULD be changed, otherwise we wait tmo jiffies on each loop.
Signed-off-by: Wentao Lou <Wentao.Lou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In VRR mode, proper vblank/pageflip timestamps can only be computed
after the display scanout position has left front-porch. Therefore
delay calls to drm_crtc_handle_vblank(), and thereby calls to
drm_update_vblank_count() and pageflip event delivery, to after the
end of front-porch when in VRR mode.
We add a new vupdate irq, which triggers at the end of the vupdate
interval, ie. at the end of vblank, and calls the core vblank handler
function. The new irq handler is not executed in standard non-VRR
mode, so vblank handling for fixed refresh rate mode is identical
to the past implementation.
v2: Implement feedback by Nicholas and Paul Menzel.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Nicholas Kazlauskas <Nicholas.Kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For throttling to work correctly, we always need a baseline vblank
count last_flip_vblank that increments at start of front-porch.
This is the case for drm_crtc_vblank_count() in non-VRR mode, where
the vblank irq fires at start of front-porch and triggers DRM core
vblank handling, but it is no longer the case in VRR mode, where
core vblank handling is done later, after end of front-porch.
Therefore drm_crtc_vblank_count() is no longer useful for this.
We also can't use drm_crtc_accurate_vblank_count(), as that would
screw up vblank timestamps in VRR mode when called in front-porch.
To solve this, use the cooked hardware vblank counter returned by
amdgpu_get_vblank_counter_kms() instead, as that one is cooked to
always increment at start of front-porch, independent of when
vblank related irq's fire.
This patch allows vblank irq handling to happen anywhere within
vblank of even after it, without a negative impact on flip
throttling, so followup patches can shift the vblank core
handling trigger point wherever they need it.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Nicholas Kazlauskas <Nicholas.Kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
According to HW engineer, they prefer the TMR address be "naturally aligned", e.g. the start address
must be an integer divide of TME size.
Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Fix the issue about TDR-2 will have "fallback timer expired on ring sdma1".
It is because the wrong number of irq types setting.
Signed-off-by: Emily Deng <Emily.Deng@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu:
- Switch to HMM for userptr (reverted until HMM fixes land)
- New experimental SMU 11 replacement for powerplay for vega20 (not enabled by default)
- Initial RAS support for vega20
- BACO support for vega12
- BACO fixes for vega20
- Rework IH handling for page fault and retry interrupts
- Cleanly split CPU and GPU paths for GPUVM updates
- Powerplay fixes
- XGMI fixes
- Rework how DC interacts with atomic for planes
- Clean up and simplify DC/Powerplay interfaces
- Misc cleanups and bug fixes
amdkfd:
- Switch to HMM for userptr (reverted until HMM fixes land)
- Add initial RAS support
- MQD fixes
ttm:
- Unify DRM_FILE_PAGE_OFFSET handling
- Account for kernel allocations in kernel zone only
- Misc cleanups
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexdeucher@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190402170820.22197-1-alexander.deucher@amd.com
