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

["Sharma, Shashank" <shashank.sharma@amd.com>]

On 5/24/2022 2:10 PM, Somalapuram, Amaranath wrote:
> 
> On 5/24/2022 3:23 PM, Sharma, Shashank wrote:
>>
>>
>> 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@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.
> This is the only place where PID get updated.
> For example sequence of operation like:
> 1st reset:
> -valid VM and tmp_adev->reset_context_task_info.pid is set to some valid 
> PID
> 2nd reset:
> -invalid VM
> -tmp_adev context will remain same (adev context will be same after 
> successful  GPU reset sequence).
> -tmp_adev->reset_context_task_info.pid will be holding 1st reset PID value
> 
> if we don't set it 0, it can give wrong PID.
>

I get your point, and agree, that when the vm is invalid, PID is invalid 
so we should send pid=0. But where are you handling the case when VM is 
valid and pid is valid ?

- Shashank

> Regards,
> S.Amarnath
>>
>> - 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);