On 4/17/2024 1:14 PM, Khatri, Sunil wrote:
>
> On 4/17/2024 1:06 PM, Khatri, Sunil wrote:
>> devcoredump is used to debug gpu hangs/resets. So in normal process
>> when there is a hang due to ring timeout or page fault we are doing a
>> hard reset as soft reset fail in those cases. How are we making sure
>> that the devcoredump is triggered in those cases and captured?
>>
>> Regards
>> Sunil Khatri
>>
>> On 4/17/2024 9:43 AM, Ahmad Rehman wrote:
>>> In passthrough environment, the driver triggers the mode-1 reset on
>>> reload. The reset causes the core dump collection which is delayed task
>>> and prevents driver from unloading until it is completed. Since we do
>>> not need to collect data on "reset on reload" case, we can skip core
>>> dump collection.
>>>
>>> v2: Use the same flag to avoid calling amdgpu_reset_reg_dumps as well.
>>>
>>> Signed-off-by: Ahmad Rehman <Ahmad.Rehman@xxxxxxx>
>>> ---
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 7 +++++--
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 1 +
>>> drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h | 1 +
>>> 3 files changed, 7 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> index 1b2e177bc2d6..c718982cffa8 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> @@ -5357,7 +5357,9 @@ int amdgpu_do_asic_reset(struct list_head
>>> *device_list_handle,
>>> /* Try reset handler method first */
>>> tmp_adev = list_first_entry(device_list_handle, struct
>>> amdgpu_device,
>>> reset_list);
>>> - amdgpu_reset_reg_dumps(tmp_adev);
>>> +
>>> + if (!test_bit(AMDGPU_SKIP_COREDUMP, &reset_context->flags))
>>> + amdgpu_reset_reg_dumps(tmp_adev);
>>> reset_context->reset_device_list = device_list_handle;
>>> r = amdgpu_reset_perform_reset(tmp_adev, reset_context);
>>> @@ -5430,7 +5432,8 @@ int amdgpu_do_asic_reset(struct list_head
>>> *device_list_handle,
>>> vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
>>> - amdgpu_coredump(tmp_adev, vram_lost, reset_context);
>>> + if (!test_bit(AMDGPU_SKIP_COREDUMP,
>>> &reset_context->flags))
>>> + amdgpu_coredump(tmp_adev, vram_lost,
>>> reset_context);
>>> if (vram_lost) {
>>> DRM_INFO("VRAM is lost due to GPU reset!\n");
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>> index 6ea893ad9a36..c512f70b8272 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c
>>> @@ -2481,6 +2481,7 @@ static void
>>> amdgpu_drv_delayed_reset_work_handler(struct work_struct *work)
>>> /* Use a common context, just need to make sure full reset is
>>> done */
>>> set_bit(AMDGPU_SKIP_HW_RESET, &reset_context.flags);
>>> + set_bit(AMDGPU_SKIP_COREDUMP, &reset_context.flags);
> If this is used for guests only can we better have a flag like
> amdgpu_sriov_vf for setting the skip coredump flag ??
>
A reset is not always triggered just because of hang. There are other
cases like we want to do a reset after a suspend/resume cycle so that
the device starts from a clean state. Those are intentionally triggered
by driver. Also, there are case like RAS errors where we reset and that
also really doesn't need a core dump. In all such cases, this flag is
required, and this is one such case (this patch only addresses passthrough).
Thanks,
Lijo
> Regards
> Sunil khatri
>
>>> r = amdgpu_do_asic_reset(&device_list, &reset_context);
>>> if (r) {
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
>>> index 66125d43cf21..b11d190ece53 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_reset.h
>>> @@ -32,6 +32,7 @@ enum AMDGPU_RESET_FLAGS {
>>> AMDGPU_NEED_FULL_RESET = 0,
>>> AMDGPU_SKIP_HW_RESET = 1,
>>> + AMDGPU_SKIP_COREDUMP = 2,
>>> };
>>> struct amdgpu_reset_context {
