[Intel-gfx] [PATCH 3/6] drm/i915: Exit PSR before do a aux transaction in gen < 9

[Pandiyan, Dhinakaran]

On Fri, 2018-02-23 at 18:03 -0800, José Roberto de Souza wrote:
> As gen < 9 hardware don't have the aux ch mutex, we need to exit PSR
> and wait until it is back to inactive state before do any aux ch
> transaction.
> 

I wonder if we need this for CHV/VLV since the HW does not send PSR aux
transactions on it's own.

> Signed-off-by: José Roberto de Souza <jose.souza at intel.com>
> ---
>  drivers/gpu/drm/i915/intel_dp.c  |  8 +++++++-
>  drivers/gpu/drm/i915/intel_drv.h |  9 +++++++++
>  drivers/gpu/drm/i915/intel_psr.c | 42 ++++++++++++++++++++++++++++++++++------
>  3 files changed, 52 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> index 7be2fec51651..dacdd98bbb2e 100644
> --- a/drivers/gpu/drm/i915/intel_dp.c
> +++ b/drivers/gpu/drm/i915/intel_dp.c
> @@ -1088,8 +1088,14 @@ static bool intel_dp_aux_ch_trylock(struct intel_dp *intel_dp)
>  			to_i915(intel_dig_port->base.base.dev);
>  	i915_reg_t ch_mutex;
>  
> -	if (!intel_dp->aux_ch_mutex_reg)
> +	if (!intel_dp->aux_ch_mutex_reg) {
> +		/* As gen < 9 hardware don't have the aux ch mutex, we need to
> +		 * exit PSR and wait until it is back to inactive state before
> +		 * do any aux ch transaction
> +		 */
> +		intel_psr_exit(intel_dp, true);
>  		return true;
> +	}
>  
>  	ch_mutex = intel_dp->aux_ch_mutex_reg(intel_dp);
>  	I915_WRITE(ch_mutex, DP_AUX_CH_MUTEX_ENABLE);
> diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
> index 267cc6c5a89f..7adcd5955d1b 100644
> --- a/drivers/gpu/drm/i915/intel_drv.h
> +++ b/drivers/gpu/drm/i915/intel_drv.h
> @@ -1881,6 +1881,15 @@ void intel_psr_single_frame_update(struct drm_i915_private *dev_priv,
>  				   unsigned frontbuffer_bits);
>  void intel_psr_compute_config(struct intel_dp *intel_dp,
>  			      struct intel_crtc_state *crtc_state);
> +/**
> + * Exit PSR in the given DisplayPort.
> + * @intel_dp: DisplayPort which PSR should be exit if running
> + * @wait_exit: if true it will wait until PSR have changed to inactive state,
> + * otherwise there is not wait.
> + *
> + * It will also schedule a work to try to active PSR again.
> + */
> +void intel_psr_exit(struct intel_dp *intel_dp, bool wait_exit);
>  
>  /* intel_runtime_pm.c */
>  int intel_power_domains_init(struct drm_i915_private *);
> diff --git a/drivers/gpu/drm/i915/intel_psr.c b/drivers/gpu/drm/i915/intel_psr.c
> index e8c32c3afb0e..0b889c85e8da 100644
> --- a/drivers/gpu/drm/i915/intel_psr.c
> +++ b/drivers/gpu/drm/i915/intel_psr.c
> @@ -56,6 +56,8 @@
>  #include "intel_drv.h"
>  #include "i915_drv.h"
>  
> +#define PSR_ACTIVE_DELAY_MSEC 100
> +
>  static bool vlv_is_psr_active_on_pipe(struct drm_device *dev, int pipe)
>  {
>  	struct drm_i915_private *dev_priv = to_i915(dev);
> @@ -486,6 +488,16 @@ static void hsw_psr_enable_source(struct intel_dp *intel_dp,
>  	}
>  }
>  
> +static void intel_psr_active_schedule(struct i915_psr *psr,
> +				      unsigned int msec_delay)
> +{
> +	if (psr->active || psr->busy_frontbuffer_bits)
> +		return;
> +
> +	if (!work_busy(&psr->work.work))
> +		schedule_delayed_work(&psr->work, msecs_to_jiffies(msec_delay));

This change will conflict with
https://patchwork.freedesktop.org/series/38199/


> +}
> +
>  /**
>   * intel_psr_enable - Enable PSR
>   * @intel_dp: Intel DP
> @@ -534,8 +546,9 @@ void intel_psr_enable(struct intel_dp *intel_dp,
>  		 *     - On HSW/BDW we get a recoverable frozen screen until
>  		 *       next exit-activate sequence.
>  		 */
> -		schedule_delayed_work(&dev_priv->psr.work,
> -				      msecs_to_jiffies(intel_dp->panel_power_cycle_delay * 5));
> +		intel_psr_active_schedule(&dev_priv->psr,
> +					  intel_dp->panel_power_cycle_delay
> +					  * 5);
>  	}
>  
>  unlock:
> @@ -886,10 +899,7 @@ void intel_psr_flush(struct drm_i915_private *dev_priv,
>  	if (frontbuffer_bits)
>  		dev_priv->psr.exit(dev_priv->psr.enabled, false);
>  
> -	if (!dev_priv->psr.active && !dev_priv->psr.busy_frontbuffer_bits)
> -		if (!work_busy(&dev_priv->psr.work.work))
> -			schedule_delayed_work(&dev_priv->psr.work,
> -					      msecs_to_jiffies(100));
> +	intel_psr_active_schedule(&dev_priv->psr, PSR_ACTIVE_DELAY_MSEC);
>  	mutex_unlock(&dev_priv->psr.lock);
>  }
>  
> @@ -955,3 +965,23 @@ void intel_psr_init(struct drm_i915_private *dev_priv)
>  		dev_priv->psr.exit = hsw_psr_exit;
>  	}
>  }
> +
> +void intel_psr_exit(struct intel_dp *intel_dp, bool wait_exit)
> +{
> +	struct intel_digital_port *intel_dig_port = dp_to_dig_port(intel_dp);
> +	struct drm_device *dev = intel_dig_port->base.base.dev;
> +	struct drm_i915_private *dev_priv = to_i915(dev);
> +
> +	if (!HAS_PSR(dev_priv))
> +		return;
> +
> +	mutex_lock(&dev_priv->psr.lock);
> +
> +	if (dev_priv->psr.enabled != intel_dp)
> +		goto out;
> +
> +	dev_priv->psr.exit(intel_dp, wait_exit);
> +	intel_psr_active_schedule(&dev_priv->psr, PSR_ACTIVE_DELAY_MSEC);

psr_exit scheduling an activate looks wrong and it is also exactly what
psr_flush does.

I suppose we could call psr_invalidate at the beginning of the aux
transaction and psr_flush at the end. But the question is what arguments
do we pass to those functions.

Or we could just call psr_exit at the start (without scheduling
activate) and psr_activate at the end.


> +out:
> +	mutex_unlock(&dev_priv->psr.lock);
> +}