[PATCH v2 2/4] drm/amdgpu: Put enable gfx off feature to a delay thread

[Felix Kuehling]

NACK.

See inline ...


On 2018-07-30 05:23 AM, Rex Zhu wrote:
> delay to enable gfx off feature to avoid gfx on/off frequently
> suggested by Alex and Evan.
>
> Signed-off-by: Rex Zhu <Rex.Zhu at amd.com>
> ---
>  drivers/gpu/drm/amd/amdgpu/amdgpu.h        |  2 ++
>  drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 15 +++++++++++++++
>  drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c    |  8 ++++++--
>  3 files changed, 23 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> index 188bc53..ff16689 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h
> @@ -992,6 +992,8 @@ struct amdgpu_gfx {
>  	bool                            gfx_off_state; /* true: enabled, false: disabled */
>  	struct mutex                    gfx_off_mutex;
>  	uint32_t                        gfx_off_req_count; /* default 1, enable gfx off: dec 1, disable gfx off: add 1 */
> +	struct delayed_work             gfx_off_delay_work;
> +
>  	/* pipe reservation */
>  	struct mutex			pipe_reserve_mutex;
>  	DECLARE_BITMAP			(pipe_reserve_bitmap, AMDGPU_MAX_COMPUTE_QUEUES);
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> index 122653f..f6a6fb5 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_device.c
> @@ -1925,6 +1925,19 @@ static void amdgpu_device_ip_late_init_func_handler(struct work_struct *work)
>  		DRM_ERROR("ib ring test failed (%d).\n", r);
>  }
>  
> +static void amdgpu_device_delay_enable_gfx_off(struct work_struct *work)
> +{
> +	struct amdgpu_device *adev =
> +		container_of(work, struct amdgpu_device, gfx.gfx_off_delay_work.work);
> +
> +	mutex_lock(&adev->gfx.gfx_off_mutex);
> +	if (!adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) {
> +		if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true))
> +			adev->gfx.gfx_off_state = true;
> +	}
> +	mutex_unlock(&adev->gfx.gfx_off_mutex);
> +}
> +
>  /**
>   * amdgpu_device_ip_suspend_phase1 - run suspend for hardware IPs (phase 1)
>   *
> @@ -2394,6 +2407,8 @@ int amdgpu_device_init(struct amdgpu_device *adev,
>  
>  	INIT_DELAYED_WORK(&adev->late_init_work,
>  			  amdgpu_device_ip_late_init_func_handler);
> +	INIT_DELAYED_WORK(&adev->gfx.gfx_off_delay_work,
> +			  amdgpu_device_delay_enable_gfx_off);
>  
>  	adev->gfx.gfx_off_req_count = 1;
>  	adev->pm.ac_power = power_supply_is_system_supplied() > 0 ? true : false;
> diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
> index 1cdb264..11d4d9f 100644
> --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
> +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.c
> @@ -26,6 +26,9 @@
>  #include "amdgpu.h"
>  #include "amdgpu_gfx.h"
>  
> +/* 0.5 second timeout */
> +#define GFX_OFF_DELAY_ENABLE         msecs_to_jiffies(500)
> +
>  /*
>   * GPU scratch registers helpers function.
>   */
> @@ -360,6 +363,7 @@ void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
>  	if (!adev->powerplay.pp_funcs->set_powergating_by_smu)
>  		return;
>  
> +
>  	mutex_lock(&adev->gfx.gfx_off_mutex);
>  
>  	if (!enable)
> @@ -368,11 +372,11 @@ void amdgpu_gfx_off_ctrl(struct amdgpu_device *adev, bool enable)
>  		adev->gfx.gfx_off_req_count--;
>  
>  	if (enable && !adev->gfx.gfx_off_state && !adev->gfx.gfx_off_req_count) {
> -		if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, true))
> -			adev->gfx.gfx_off_state = true;
> +		schedule_delayed_work(&adev->gfx.gfx_off_delay_work, GFX_OFF_DELAY_ENABLE);
>  	} else if (!enable && adev->gfx.gfx_off_state) {

There may still be delayed work pending to enable GFXOFF. You need to
use cancel_delayed_work_sync to cancel it and make sure GFXOFF doesn't
get enabled again while your caller still needs it disabled.

That may also wait for a worker function that has already started.
Therefore you need to cancel before taking the gfx_off_mutex, otherwise
you can deadlock with the worker function trying to take the lock as well.

Regards,
  Felix

>  		if (!amdgpu_dpm_set_powergating_by_smu(adev, AMD_IP_BLOCK_TYPE_GFX, false))
>  			adev->gfx.gfx_off_state = false;
>  	}
> +
>  	mutex_unlock(&adev->gfx.gfx_off_mutex);
>  }