Make sure to not release the vm root bo after vm validation
and to make that happen we moved the restore function within
amdgpu_userq_vm_validate function.
Also update the function name to reflect the intent.
Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Zhu Lingshan <lingshan.zhu@amd.com>
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This function is not called from anywhere anymore and
every implementation was bogus.
Some implementations checked busy flags of the IP blocks,
which are not really indicative of whether the block is
hung and needs to be reset. For example the blocks
could be busy just normally executing submissions,
and not need to be reset.
Other implementations checked IB tests, which is actually
more useful, but could still just indicate that an IP block
is executing submissions normally.
It is also unnecessary because the GPU recovery code path
already knows which ring is hung so we know exactly what
we need to reset.
Just delete check_soft_reset() entirely.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This was basically dead code, not used or called from anywhere.
Now that DC is the default display driver for all ASICs,
it is unlikely that anyone wants to develop this further.
Display hang related work should be focused on DC.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
We should only reset the memory controller during ASIC reset
and only when it's absolutely necessary. Otherwise, resetting
the memory controller typically just breaks everything and
on dGPUs may also clear the contents of VRAM (it's unclear if
it really does, but it's likely).
Specifically for GMC 8, the memory controller is reset as part
of the ASIC reset and otherwise should be left alone.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Soft reset means resetting IP blocks individually using
a hardware interconnect (SRBM or GRBM) without assistance
from firmware.
Soft reset is a useful tool for implementing GPU recovery,
eg. it is already successfully used for SDMA queue resets.
It should be used by a GPU recovery method instead of
being called directly from the ASIC reset code path.
Currently, this is only used on Carrizo and Stoney,
but doesn't work well and fails on those chips.
A subsequent commit will add a working GFX8 recovery
implementation after the cleanups.
Note that this commit only cleans up the ASIC reset path,
which also unblocks more opportunities for cleanup for
the various IP blocks. Those will be done in subsequent commits.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Soft recovery is not the same as soft reset:
* Soft recovery attempts to resolve a GPU hang by sending a
command to terminate shaders.
* Soft reset completely re-initializes an entire device IP block,
which may affect multiple rings and jobs at the same time.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If the user has disabled kernel queues, then make all vmids
available to HWS.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For SOC24 ASICs (RDNA4 / Navi 4x dGPUs) re-enabling PM features fails if an
S3 suspend got aborted, the same issue already handled for SOC21 and SOC15:
commit df3c7dc5c5 ("drm/amdgpu: Reset dGPU if suspend got aborted")
commit 38e8ca3e4b ("amdgpu/soc15: enable asic reset for dGPU in case of suspend abort")
The aborted resume fails with:
amdgpu: SMU: No response msg_reg: 6 resp_reg: 0
amdgpu: Failed to enable requested dpm features!
amdgpu: resume of IP block <smu> failed -62
Apply the same workaround for soc24: detect the aborted-suspend state at
resume via the sign-of-life register and reset the device before re-init.
This is a workaround till a proper solution is finalized.
Fixes: 98b912c50e ("drm/amdgpu: Add soc24 common ip block (v2)")
Signed-off-by: Jakob Linke <jakob@linke.cx>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add amdgpu_atpx_buffer_validate() to check that the returned ACPI
buffer is of type ACPI_TYPE_BUFFER, is large enough to hold the u16
size field, and that the BIOS-reported size does not exceed the actual
allocation length or fall below the minimum required by the caller.
Use it in VERIFY_INTERFACE and GET_PX_PARAMETERS callers.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add a min_size parameter to amdgpu_atif_call() to validate that the
returned ACPI buffer is of type ACPI_TYPE_BUFFER, holds at least a u16
size field, does not claim more data than was actually returned, and
meets the minimum size required by the calling function. Each caller
passes its required minimum via sizeof() or offsetof() of the expected
output struct and drops its own size check.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use common helper function to get pptable from firmware binary in
SMUv15.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use common helper function to get pptable from firmware binary in
SMUv14.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use common helper function to get pptable from firmware binary in
SMUv13.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use common helper function to get pptable from firmware binary in
SMUv11.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
PPTables could be embedded in firmware binaries with v2.0 or v2.1
format. Add a common helper to get pptable from firmware binaries.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Assisted-by: Claude Sonnet (Cursor AI)
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Abort driver load when num_mem_partitions is zero since operation is
unreliable without valid memory partition info. Skip absent resources
in soc_v1_0_get_xcp_res_info() to avoid divide-by-zero on firmware-
reported zero instance counts.
v2: Remove redundant checks (Lijo)
v3: Return error instead when num_mem_partitions is zero (Lijo)
Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
In aqua_vanjaram_get_xcp_res_info(), max_res[i] can be zero.
When res_lt_xcp is true the code divides num_xcp by max_res[i],
causing a divide fault.
Skip the loop body for absent resources.
v2: Remove redundant checks (Lijo)
Signed-off-by: Asad Kamal <asad.kamal@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Add a helper function to extract long values passed in a string. The
string may have values of multiple parameters separated by space char.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
down_read/up_read adev->reset_domain semaphore should be placed around
remove queue.
v2: remove the empty function, recover_bad_queue_mes to avoid compile
error on rhel
Fixes: f401a2633e ("drm/amdgpu: Remove faulty queue before resume")
Signed-off-by: Amber Lin <Amber.Lin@amd.com>
Reviewed-by: Jesse Zhang <jesse.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Check if a valid buffer object is returned after ATRM call. Also, match
the buffer length against requested size before copying.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Removing the output parameter from a few functions should result in more
readable code and also enables us to save some lines.
v2: fix build (Alex)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
IDR is deprecated so let's replace it with xarray.
Conversion is mostly 1:1 apart from AMDGPU_BO_LIST_OP_UPDATE which was
implemented with idr_replace, and has now been replaced with a sequence of
xa_load and xa_cmpxchg. Should userspace attempt multi-threaded update
operations on the same handle it could theoretically hit a new -ENOENT
path. But I believe this is purely theoretical and still safe.
Also, since we have removed the RCU protection around the handle lookup we
also removed the RCU freeing of the list.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Userspace always uses struct drm_amdgpu_bo_list_in->bo_info_size equal to
sizeof(struct drm_amdgpu_bo_list_entry) and there are no plans to extend
it. Even if the structure is extended at some point, older kernels will
note that they do not support the additional fields by rejecting the new
structure size.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid and HQD resources if kernel queues are not
needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When the RAS core manages the EEPROM, the eeprom_control is never
initialized (amdgpu_ras_init_badpage_info() returns early), so reading
ras/ras_eeprom_table in debugfs printed only a zeroed header and no
records, even though bad-page records exist in the RAS core EEPROM.
Source the table header and records from the RAS core EEPROM
(ras_core->ras_eeprom) in that case, reusing the existing output layout
so the debugfs node keeps the same format. Skip the dump when the
firmware manages the EEPROM, since the records are not stored in the
I2C-backed table then.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid resources if kernel queues are not needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid resources if kernel queues are not needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the kernel to dispatch the soft IH work on other CPUs.
Otherwise it can happen that the soft IH ring fills up
before it actually starts processing anything, which
can easily happen with retry page faults, in which case
the CP repeatedly spams the CPU with a lot of interrupts.
This significantly improves retry page fault handling on
GPUs that don't have the filter CAM and must rely on
software based filtering.
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
SVM ranges use inclusive page indices: prange->last is the last page in
the range. The split-remap logic introduced by commit 448ee45353
("drm/amdkfd: Use huge page size to check split svm range alignment")
uses ALIGN_DOWN(prange->last, 512) to determine whether the original
range can contain a 2MB huge-page mapping.
That aligns the last page itself down. Thus a range ending one page
before the next 2MB boundary is classified as if the final 2MB block did
not exist. When such a range is split inside that final block, the
split head or tail can be left off the remap list even though it was
derived from an original range that may have PMD mappings.
Use prange->last + 1 as the exclusive upper bound when computing the
original range's last 2MB-aligned boundary. Then use the actual split
boundary for the head and tail alignment checks: tail->start for a tail
split, and new_start for a head split. new_start is equivalent to
head->last + 1 and directly names the exclusive end of the split head.
Using head->last for the head-side check can both remap a head that ends
exactly one page before a 2MB boundary and miss a head whose split
boundary is one page after such a boundary. Philip Yang pointed out in
the review of the original change that this condition should use
head->last + 1 or new_start.
Xiaogang Chen identified the inclusive-last cause and posted the
candidate fix in the regression thread. With the culprit change active
and the local revert not applied, the unchanged C/HSA reproducer
completes 10/10 runs with this change on an RX 7600 XT.
Fixes: 448ee45353 ("drm/amdkfd: Use huge page size to check split svm range alignment")
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/4914
Link: https://lore.kernel.org/stable/IA1PR12MB85172F7FE9157C092EDA46A0E3112@IA1PR12MB8517.namprd12.prod.outlook.com/
Link: https://lore.kernel.org/all/32ce2b72-aa16-4202-9f99-92e3cd4408bc@amd.com/
Suggested-by: Xiaogang Chen <xiaogang.chen@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Gerhard Schwanzer <geschw@pm.me>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For Renior APU with gfx9, in some test scenarios with disabling
ring_reset, like accessing an unmapped invalid address, it can
trigger a gpu job timeout event, then driver uses Mode2 reset
to reset GPU, but after Mode2 compute Ring test and IB test fail
randomly. It because the HQDs of MECs are always active before or
after Mode2, that causes MECs use stale HQDs when MECs are unhalted
before driver restore MQDs, and causes CPC and CPF are still stuck
after Mode2, then causes compute Ring and IB tests fail.
So, add sequences to deactivate HQDs of MECs in suspend IP function
of the resetting process.
v2: Move all sequences into a new function gfx_v9_0_cp_mode2_clear_state (Ray Huang)
To check reset Mode2 method in the if condition (Ray Huang)
v3: Move all sequences before Mode2 instead of after Mode2 (Timur Kristóf)
v4: Call amdgpu_gfx_rlc_enter/exit_safe_mode int the begin and end of
gfx_v9_0_deactivate_kcq_hqd (Alex Deucher)
Signed-off-by: Jiqian Chen <Jiqian.Chen@amd.com>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid resources if kernel queues are not needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid resources if kernel queues are not needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Allow the user to disable kernel queues. This can be used
to free up vmid resources if kernel queues are not needed.
Set amdgpu.user_queue=2 to disable kernel queues.
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Alex Deucher
Sunil Khatri
Christian König
Timur Kristóf
Lijo Lazar
Jakob Linke
Asad Kamal
Amber Lin
Tvrtko Ursulin
Xiang Liu
Pierre-Eric Pelloux-Prayer
Prike Liang
Gerhard Schwanzer
Jiqian Chen
YiPeng Chai