[PATCH v1 2/2] drm/amdgpu: adding device coredump support

[Sharma, Shashank]

On 5/24/2022 8:42 AM, Somalapuram, Amaranath wrote:
> 
> On 5/20/2022 7:52 PM, Sharma, Shashank wrote:
>>
>>
>> On 5/20/2022 3:49 PM, Somalapuram Amaranath wrote:
>>> Added device coredump information:
>>> - Kernel version
>>> - Module
>>> - Time
>>> - VRAM status
>>> - Guilty process name and PID
>>> - GPU register dumps
>>>
>>> Signed-off-by: Somalapuram Amaranath <Amaranath.Somalapuram at amd.com>
>>> ---
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  3 ++
>>>   drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 59 ++++++++++++++++++++++
>>>   2 files changed, 62 insertions(+)
>>>
>>> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>>> index c79d9992b113..f28d9c563f74 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
>>> @@ -1044,6 +1044,9 @@ struct amdgpu_device {
>>>       uint32_t                        *reset_dump_reg_list;
>>>       uint32_t            *reset_dump_reg_value;
>>>       int                             num_regs;
>>> +    struct amdgpu_task_info         reset_context_task_info;
>>> +    bool                            reset_context_vram_lost;
>>
>> How about drop the 'context' from name and just reset_task_info and 
>> reset_vram_lost ?
> OK.
>>
>>> +    struct timespec64 reset_time;
>>>         struct amdgpu_reset_domain    *reset_domain;
>>>   diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c 
>>> b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> index 963c897a76e6..f9b710e741a7 100644
>>> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
>>> @@ -32,6 +32,8 @@
>>>   #include <linux/slab.h>
>>>   #include <linux/iommu.h>
>>>   #include <linux/pci.h>
>>> +#include <linux/devcoredump.h>
>>> +#include <generated/utsrelease.h>
>>>     #include <drm/drm_atomic_helper.h>
>>>   #include <drm/drm_probe_helper.h>
>>> @@ -4733,6 +4735,55 @@ static int amdgpu_reset_reg_dumps(struct 
>>> amdgpu_device *adev)
>>>       return 0;
>>>   }
>>>   +#ifdef CONFIG_DEV_COREDUMP
>>> +static ssize_t amdgpu_devcoredump_read(char *buffer, loff_t offset,
>>> +        size_t count, void *data, size_t datalen)
>>> +{
>>> +    struct drm_printer p;
>>> +    struct amdgpu_device *adev = data;
>>> +    struct drm_print_iterator iter;
>>> +    int i;
>>> +
>>
>> A NULL check for 'buffer' here could prevent a segfault later.
>>
> Agreed.
>>> +    iter.data = buffer;
>>> +    iter.offset = 0;
>>> +    iter.start = offset;
>>> +    iter.remain = count;
>>> +
>>> +    p = drm_coredump_printer(&iter);
>>> +
>>> +    drm_printf(&p, "**** AMDGPU Device Coredump ****\n");
>>> +    drm_printf(&p, "kernel: " UTS_RELEASE "\n");
>>> +    drm_printf(&p, "module: " KBUILD_MODNAME "\n");
>>> +    drm_printf(&p, "time: %lld.%09ld\n", adev->reset_time.tv_sec, 
>>> adev->reset_time.tv_nsec);
>>> +    if (adev->reset_context_task_info.pid)
>>> +        drm_printf(&p, "process_name: %s PID: %d\n",
>>> + adev->reset_context_task_info.process_name,
>>> + adev->reset_context_task_info.pid);
>> Please fix the alignment of print variables.
>>
> I will cross check this.
>>> +
>>> +    if (adev->reset_context_vram_lost)
>>> +        drm_printf(&p, "VRAM is lost due to GPU reset!\n");
>>> +    if (adev->num_regs) {
>>> +        drm_printf(&p, "AMDGPU register dumps:\nOffset: Value:\n");
>>> +
>>> +        for (i = 0; i < adev->num_regs; i++)
>>> +            drm_printf(&p, "0x%08x: 0x%08x\n",
>>> +                    adev->reset_dump_reg_list[i],
>>> +                    adev->reset_dump_reg_value[i]);
>>> +    }
>>> +
>>> +    return count - iter.remain;
>>> +}
>>> +
>>> +static void amdgpu_reset_capture_coredumpm(struct amdgpu_device *adev)
>>> +{
>>> +    struct drm_device *dev = adev_to_drm(adev);
>>> +
>>> +    ktime_get_ts64(&adev->reset_time);
>>> +    dev_coredumpm(dev->dev, THIS_MODULE, adev, 0, GFP_KERNEL,
>>> +            amdgpu_devcoredump_read, NULL);
>> instead of registering NULL as free function, I would prefer you to 
>> have a dummy no_op free function registered, which we can consume if 
>> something changes.
> you mean something like this (function without any code):
> staticvoidamdgpu_devcoredump_free(void*data)
> {
> }

Yes, precisely.

>>> +}
>>> +#endif
>>> +
>>>   int amdgpu_do_asic_reset(struct list_head *device_list_handle,
>>>                struct amdgpu_reset_context *reset_context)
>>>   {
>>> @@ -4817,6 +4868,14 @@ int amdgpu_do_asic_reset(struct list_head 
>>> *device_list_handle,
>>>                       goto out;
>>>                     vram_lost = amdgpu_device_check_vram_lost(tmp_adev);
>>> +#ifdef CONFIG_DEV_COREDUMP
>>> +                tmp_adev->reset_context_vram_lost = vram_lost;
>>> +                tmp_adev->reset_context_task_info.pid = 0;
>> why is the PID hardcoded to 0 ?
> in case of reset context reset_context->job->vm  is null (possibility 
> that reset can be non VM related).
> If we don't set tmp_adev->reset_context_task_info.pid = 0, it will show 
> previous reset valid PID.
> 

But when the VM is not NULL, are we updating this PID somewhere ? I did 
not see that happening in this series.

- Shashank

> 
> Regards,
> S.Amarnath
>>> +                if (reset_context->job && reset_context->job->vm)
>>> +                    tmp_adev->reset_context_task_info =
>>> + reset_context->job->vm->task_info;
>>> +                amdgpu_reset_capture_coredumpm(tmp_adev);
>>> +#endif
>>>                   if (vram_lost) {
>>>                       DRM_INFO("VRAM is lost due to GPU reset!\n");
>>>
>> - Shashank
>>                      amdgpu_inc_vram_lost(tmp_adev);