Re: [PATCH v3] drm/amdkfd: Ignored various return code

[Felix Kuehling <felix.kuehling@amd.com>]

On 2025-11-13 13:31, Andrew Martin wrote:
> The return code of a non void function should not be ignored. In cases
> where we do not care, the code needs to suppress it.
>
> Signed-off-by: Andrew Martin <andrew.martin@amd.com>
> ---
>   drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c | 13 ++++++++-----
>   drivers/gpu/drm/amd/amdkfd/kfd_chardev.c   |  3 ++-
>   2 files changed, 10 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> index 644f79f3c9af..e4438fca6283 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
> @@ -313,8 +313,10 @@ int amdgpu_amdkfd_post_reset(struct amdgpu_device *adev)
>   void amdgpu_amdkfd_gpu_reset(struct amdgpu_device *adev)
>   {
>   	if (amdgpu_device_should_recover_gpu(adev))
> -		amdgpu_reset_domain_schedule(adev->reset_domain,
> -					     &adev->kfd.reset_work);
> +		WARN_ONCE(!amdgpu_reset_domain_schedule(adev->reset_domain,
> +							&adev->kfd.reset_work),
> +							"Failed to queue work! at %s",
> +							 __func__);

The return value is not an error. It only indicates that reset work was 
already queued.


>   }
>   
>   int amdgpu_amdkfd_alloc_gtt_mem(struct amdgpu_device *adev, size_t size,
> @@ -715,9 +717,10 @@ void amdgpu_amdkfd_set_compute_idle(struct amdgpu_device *adev, bool idle)
>   		if (gfx_block != NULL)
>   			gfx_block->version->funcs->set_powergating_state((void *)gfx_block, state);
>   	}
> -	amdgpu_dpm_switch_power_profile(adev,
> -					PP_SMC_POWER_PROFILE_COMPUTE,
> -					!idle);
> +	WARN_ONCE(!amdgpu_dpm_switch_power_profile(adev, PP_SMC_POWER_PROFILE_COMPUTE, !idle),
> +		  "(%s) failed to disable video power profile mode",
> +		  __func__);

I think 0 means success here. But I agree with Philip that we can 
probably ignore errors here. I believe some chips just don't support 
power profiles and return an error here.


> +
>   }
>   
>   bool amdgpu_amdkfd_is_kfd_vmid(struct amdgpu_device *adev, u32 vmid)
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> index 22925df6a791..025609a4abcf 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
> @@ -2780,7 +2780,8 @@ static int runtime_enable(struct kfd_process *p, uint64_t r_debug,
>   		 * saved in MES.
>   		 */
>   		if (pdd->dev->kfd->shared_resources.enable_mes)
> -			kfd_dbg_set_mes_debug_mode(pdd, !kfd_dbg_has_cwsr_workaround(pdd->dev));
> +			(void)kfd_dbg_set_mes_debug_mode(pdd,
> +							 !kfd_dbg_has_cwsr_workaround(pdd->dev));

In this case we could actually handle the error and return an error up 
to the caller.

Regards,
   Felix


>   	}
>   
>   	p->runtime_info.runtime_state = DEBUG_RUNTIME_STATE_ENABLED;