[Intel-gfx] [PATCH v5] drm/i915: Force 2*96 MHz cdclk on glk/cnl when audio power is enabled

Imre Deak imre.deak at intel.com
Wed Sep 12 13:18:12 UTC 2018


Hi Kumar, Takashi,

On Tue, Jun 19, 2018 at 03:01:11PM -0700, Abhay Kumar wrote:
> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> 
> CDCLK has to be at least twice the BLCK regardless of audio. Audio
> driver has to probe using this hook and increase the clock even in
> absence of any display.
> 
> v2: Use atomic refcount for get_power, put_power so that we can
>     call each once(Abhay).
> v3: Reset power well 2 to avoid any transaction on iDisp link
>     during cdclk change(Abhay).
> v4: Remove Power well 2 reset workaround(Ville).
> v5: Remove unwanted Power well 2 register defined in v4(Abhay).
> 
> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> Signed-off-by: Abhay Kumar <abhay.kumar at intel.com>
> ---
>  drivers/gpu/drm/i915/i915_drv.h      |  3 ++
>  drivers/gpu/drm/i915/intel_audio.c   | 67 +++++++++++++++++++++++++++++++++---
>  drivers/gpu/drm/i915/intel_cdclk.c   | 29 +++++-----------
>  drivers/gpu/drm/i915/intel_display.c |  7 +++-
>  drivers/gpu/drm/i915/intel_drv.h     |  2 ++
>  5 files changed, 83 insertions(+), 25 deletions(-)
> 
> diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> index 6104d7115054..a4a386a5db69 100644
> --- a/drivers/gpu/drm/i915/i915_drv.h
> +++ b/drivers/gpu/drm/i915/i915_drv.h
> @@ -1702,6 +1702,7 @@ struct drm_i915_private {
>  	unsigned int hpll_freq;
>  	unsigned int fdi_pll_freq;
>  	unsigned int czclk_freq;
> +	u32 get_put_refcount;
>  
>  	struct {
>  		/*
> @@ -1719,6 +1720,8 @@ struct drm_i915_private {
>  		struct intel_cdclk_state actual;
>  		/* The current hardware cdclk state */
>  		struct intel_cdclk_state hw;
> +
> +		int force_min_cdclk;
>  	} cdclk;
>  
>  	/**
> diff --git a/drivers/gpu/drm/i915/intel_audio.c b/drivers/gpu/drm/i915/intel_audio.c
> index 3ea566f99450..ca8f04c7cbb3 100644
> --- a/drivers/gpu/drm/i915/intel_audio.c
> +++ b/drivers/gpu/drm/i915/intel_audio.c
> @@ -618,7 +618,6 @@ void intel_audio_codec_enable(struct intel_encoder *encoder,
>  
>  	if (!connector->eld[0])
>  		return;
> -
>  	DRM_DEBUG_DRIVER("ELD on [CONNECTOR:%d:%s], [ENCODER:%d:%s]\n",
>  			 connector->base.id,
>  			 connector->name,
> @@ -713,14 +712,74 @@ void intel_init_audio_hooks(struct drm_i915_private *dev_priv)
>  	}
>  }
>  
> +static void glk_force_audio_cdclk(struct drm_i915_private *dev_priv,
> +				bool enable)
> +{
> +	struct drm_modeset_acquire_ctx ctx;
> +	struct drm_atomic_state *state;
> +	int ret;
> +
> +	drm_modeset_acquire_init(&ctx, 0);
> +	state = drm_atomic_state_alloc(&dev_priv->drm);
> +	if (WARN_ON(!state))
> +		return;
> +
> +	state->acquire_ctx = &ctx;
> +
> +retry:
> +	to_intel_atomic_state(state)->modeset = true;
> +	to_intel_atomic_state(state)->cdclk.force_min_cdclk =
> +		enable ? 2 * 96000 : 0;
> +
> +	/*
> +	 * Protects dev_priv->cdclk.force_min_cdclk
> +	 * Need to lock this here in case we have no active pipes
> +	 * and thus wouldn't lock it during the commit otherwise.
> +	 */
> +	ret = drm_modeset_lock(&dev_priv->drm.mode_config.connection_mutex, &ctx);
> +	if (!ret)
> +		ret = drm_atomic_commit(state);

Ville mentioned a potential dead-lock problem here: a normal modeset
enabling an HDMI/DP output with an audio sink will call the
drm_audio_component_audio_ops::pin_eld_notify() callback with the
connection_mutex already held. This callback in turn could call
drm_audio_component_ops::get_power()/put_power() callbacks and
dead-lock at the above drm_modeset_lock() call.

Looks like that

   sound/pci/hda/patch_hdmi.c / intel_pin_eld_notify()->
   check_presence_and_report()->
   hdmi_present_sense()

would prevent this, but for instance

   sound/soc/codecs/hdac_hdmi.c / hdac_hdmi_eld_notify_cb()->
   hdac_hdmi_present_sense()->
   hdac_hdmi_jack_report()->
   snd_soc_jack_report()->
   snd_soc_dapm_sync()->
   snd_soc_dapm_sync_unlocked()->
   dapm_power_widgets()->
   dapm_pre_sequence_async()

could call pm_runtime_get_sync() and so I guess also the aops
get_power() hook.

Could someone in your team check if the above can indeed happen?

> +
> +	if (ret == -EDEADLK) {
> +		drm_atomic_state_clear(state);
> +		drm_modeset_backoff(&ctx);
> +		goto retry;
> +	}
> +
> +	WARN_ON(ret);
> +
> +	drm_atomic_state_put(state);
> +
> +	drm_modeset_drop_locks(&ctx);
> +	drm_modeset_acquire_fini(&ctx);
> +}
> +
>  static void i915_audio_component_get_power(struct device *kdev)
>  {
> -	intel_display_power_get(kdev_to_i915(kdev), POWER_DOMAIN_AUDIO);
> +	struct drm_i915_private *dev_priv = kdev_to_i915(kdev);
> +
> +	dev_priv->get_put_refcount++;
> +
> +	/* Force cdclk to 2*BCLK during first time get power call */
> +	if (dev_priv->get_put_refcount == 1)
> +		if (IS_CANNONLAKE(dev_priv) || IS_GEMINILAKE(dev_priv))
> +			glk_force_audio_cdclk(dev_priv, true);
> +
> +	intel_display_power_get(dev_priv, POWER_DOMAIN_AUDIO);

We can't guarantee that the power wells belonging to the audio power
domain are off at this point (since an external output can keep them on).
But according to the tests you did Kumar, this doesn't even seem to be
required, the only requirement is that CDCLK is raised.

So to make the sequence behave the same all the time (and not hide some
problem happening occasionally) and to have it symmetric with the one
in i915_audio_component_put_power() we should get the audio power domain
before bumping CDCLK.

>  }
>  
>  static void i915_audio_component_put_power(struct device *kdev)
>  {
> -	intel_display_power_put(kdev_to_i915(kdev), POWER_DOMAIN_AUDIO);
> +	struct drm_i915_private *dev_priv = kdev_to_i915(kdev);
> +
> +	dev_priv->get_put_refcount--;
> +
> +	/* Force required cdclk during last time put power call */
> +	if (dev_priv->get_put_refcount == 0)
> +		if (IS_CANNONLAKE(dev_priv) || IS_GEMINILAKE(dev_priv))
> +			glk_force_audio_cdclk(dev_priv, false);
> +
> +	intel_display_power_put(dev_priv, POWER_DOMAIN_AUDIO);
>  }
>  
>  static void i915_audio_component_codec_wake_override(struct device *kdev,
> @@ -959,7 +1018,7 @@ void i915_audio_component_init(struct drm_i915_private *dev_priv)
>  		/* continue with reduced functionality */
>  		return;
>  	}
> -
> +	dev_priv->get_put_refcount = 0;
>  	dev_priv->audio_component_registered = true;
>  }
>  
> diff --git a/drivers/gpu/drm/i915/intel_cdclk.c b/drivers/gpu/drm/i915/intel_cdclk.c
> index 8ed7bd052e46..0f0aea900ceb 100644
> --- a/drivers/gpu/drm/i915/intel_cdclk.c
> +++ b/drivers/gpu/drm/i915/intel_cdclk.c
> @@ -2153,19 +2153,8 @@ int intel_crtc_compute_min_cdclk(const struct intel_crtc_state *crtc_state)
>  	/*
>  	 * According to BSpec, "The CD clock frequency must be at least twice
>  	 * the frequency of the Azalia BCLK." and BCLK is 96 MHz by default.
> -	 *
> -	 * FIXME: Check the actual, not default, BCLK being used.
> -	 *
> -	 * FIXME: This does not depend on ->has_audio because the higher CDCLK
> -	 * is required for audio probe, also when there are no audio capable
> -	 * displays connected at probe time. This leads to unnecessarily high
> -	 * CDCLK when audio is not required.
> -	 *
> -	 * FIXME: This limit is only applied when there are displays connected
> -	 * at probe time. If we probe without displays, we'll still end up using
> -	 * the platform minimum CDCLK, failing audio probe.
>  	 */
> -	if (INTEL_GEN(dev_priv) >= 9)
> +	if (crtc_state->has_audio && INTEL_GEN(dev_priv) >= 9)
>  		min_cdclk = max(2 * 96000, min_cdclk);
>  
>  	/*
> @@ -2205,7 +2194,7 @@ static int intel_compute_min_cdclk(struct drm_atomic_state *state)
>  		intel_state->min_cdclk[i] = min_cdclk;
>  	}
>  
> -	min_cdclk = 0;
> +	min_cdclk = intel_state->cdclk.force_min_cdclk;
>  	for_each_pipe(dev_priv, pipe)
>  		min_cdclk = max(intel_state->min_cdclk[pipe], min_cdclk);
>  
> @@ -2266,7 +2255,7 @@ static int vlv_modeset_calc_cdclk(struct drm_atomic_state *state)
>  		vlv_calc_voltage_level(dev_priv, cdclk);
>  
>  	if (!intel_state->active_crtcs) {
> -		cdclk = vlv_calc_cdclk(dev_priv, 0);
> +		cdclk = vlv_calc_cdclk(dev_priv, intel_state->cdclk.force_min_cdclk);
>  
>  		intel_state->cdclk.actual.cdclk = cdclk;
>  		intel_state->cdclk.actual.voltage_level =
> @@ -2299,7 +2288,7 @@ static int bdw_modeset_calc_cdclk(struct drm_atomic_state *state)
>  		bdw_calc_voltage_level(cdclk);
>  
>  	if (!intel_state->active_crtcs) {
> -		cdclk = bdw_calc_cdclk(0);
> +		cdclk = bdw_calc_cdclk(intel_state->cdclk.force_min_cdclk);
>  
>  		intel_state->cdclk.actual.cdclk = cdclk;
>  		intel_state->cdclk.actual.voltage_level =
> @@ -2371,7 +2360,7 @@ static int skl_modeset_calc_cdclk(struct drm_atomic_state *state)
>  		skl_calc_voltage_level(cdclk);
>  
>  	if (!intel_state->active_crtcs) {
> -		cdclk = skl_calc_cdclk(0, vco);
> +		cdclk = skl_calc_cdclk(intel_state->cdclk.force_min_cdclk, vco);
>  
>  		intel_state->cdclk.actual.vco = vco;
>  		intel_state->cdclk.actual.cdclk = cdclk;
> @@ -2410,10 +2399,10 @@ static int bxt_modeset_calc_cdclk(struct drm_atomic_state *state)
>  
>  	if (!intel_state->active_crtcs) {
>  		if (IS_GEMINILAKE(dev_priv)) {
> -			cdclk = glk_calc_cdclk(0);
> +			 cdclk = glk_calc_cdclk(intel_state->cdclk.force_min_cdclk);
>  			vco = glk_de_pll_vco(dev_priv, cdclk);
>  		} else {
> -			cdclk = bxt_calc_cdclk(0);
> +			cdclk = bxt_calc_cdclk(intel_state->cdclk.force_min_cdclk);
>  			vco = bxt_de_pll_vco(dev_priv, cdclk);
>  		}
>  
> @@ -2449,7 +2438,7 @@ static int cnl_modeset_calc_cdclk(struct drm_atomic_state *state)
>  		    cnl_compute_min_voltage_level(intel_state));
>  
>  	if (!intel_state->active_crtcs) {
> -		cdclk = cnl_calc_cdclk(0);
> +		cdclk = cnl_calc_cdclk(intel_state->cdclk.force_min_cdclk);
>  		vco = cnl_cdclk_pll_vco(dev_priv, cdclk);
>  
>  		intel_state->cdclk.actual.vco = vco;
> @@ -2482,7 +2471,7 @@ static int icl_modeset_calc_cdclk(struct drm_atomic_state *state)
>  	intel_state->cdclk.logical.cdclk = cdclk;
>  
>  	if (!intel_state->active_crtcs) {
> -		cdclk = icl_calc_cdclk(0, ref);
> +		cdclk = icl_calc_cdclk(intel_state->cdclk.force_min_cdclk, ref);
>  		vco = icl_calc_cdclk_pll_vco(dev_priv, cdclk);
>  
>  		intel_state->cdclk.actual.vco = vco;
> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
> index 17c590b42fd7..3ee1c1f5419d 100644
> --- a/drivers/gpu/drm/i915/intel_display.c
> +++ b/drivers/gpu/drm/i915/intel_display.c
> @@ -12162,6 +12162,10 @@ static int intel_modeset_checks(struct drm_atomic_state *state)
>  		return -EINVAL;
>  	}
>  
> +	/* keep the current setting */
> +	if (!intel_state->modeset)

Ville, I'd use here a dedicated flag instead.

> +		intel_state->cdclk.force_min_cdclk = dev_priv->cdclk.force_min_cdclk;
> +
>  	intel_state->modeset = true;
>  	intel_state->active_crtcs = dev_priv->active_crtcs;
>  	intel_state->cdclk.logical = dev_priv->cdclk.logical;
> @@ -12257,7 +12261,7 @@ static int intel_atomic_check(struct drm_device *dev,
>  	struct drm_crtc *crtc;
>  	struct drm_crtc_state *old_crtc_state, *crtc_state;
>  	int ret, i;
> -	bool any_ms = false;
> +	bool any_ms = intel_state->modeset;
>  
>  	/* Catch I915_MODE_FLAG_INHERITED */
>  	for_each_oldnew_crtc_in_state(state, crtc, old_crtc_state,
> @@ -12805,6 +12809,7 @@ static int intel_atomic_commit(struct drm_device *dev,
>  		dev_priv->active_crtcs = intel_state->active_crtcs;
>  		dev_priv->cdclk.logical = intel_state->cdclk.logical;
>  		dev_priv->cdclk.actual = intel_state->cdclk.actual;
> +		dev_priv->cdclk.force_min_cdclk = intel_state->cdclk.force_min_cdclk;
>  	}
>  
>  	drm_atomic_state_get(state);
> diff --git a/drivers/gpu/drm/i915/intel_drv.h b/drivers/gpu/drm/i915/intel_drv.h
> index 8641583842be..0da17ad056ec 100644
> --- a/drivers/gpu/drm/i915/intel_drv.h
> +++ b/drivers/gpu/drm/i915/intel_drv.h
> @@ -459,6 +459,8 @@ struct intel_atomic_state {
>  		 * state only when all crtc's are DPMS off.
>  		 */
>  		struct intel_cdclk_state actual;
> +
> +		int force_min_cdclk;
>  	} cdclk;
>  
>  	bool dpll_set, modeset;
> -- 
> 2.7.4
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> https://lists.freedesktop.org/mailman/listinfo/intel-gfx


More information about the Intel-gfx mailing list