On Wed, 24 Aug 2022, "Borah, Chaitanya Kumar" <chaitanya.kumar.borah@xxxxxxxxx> wrote: > In certain scenarios, we might have to filter out some audio > configuration depending on HW limitation. For example, in > GLK DP port more than 2 channels are not supported for audio. > A monitor provides information of it's supported audio configurations > through SAD (Short Audio Descriptors) which are part of the ELD/EDID. > In this patch we add a bit mask to indicate which SADs are supported. > The bit mask is updated in the function intel_eld_compute_config(). > Based on this bit mask, we prune SADs which are not supported in > the function i915_audio_component_get_eld() before sending out the > data to the audio driver. > > We use a bit mask instead of operating on the eld data directly as > eld data can get updated after intel_eld_compute_config() and before > i915_audio_component_get_eld() is called > > fixes: https://gitlab.freedesktop.org/drm/intel/-/issues/5368 > > Signed-off-by: Borah, Chaitanya Kumar <chaitanya.kumar.borah@xxxxxxxxx> > --- > drivers/gpu/drm/drm_edid.c | 36 ++++++++++ > drivers/gpu/drm/i915/display/intel_audio.c | 65 ++++++++++++++++++- > drivers/gpu/drm/i915/display/intel_audio.h | 3 + > drivers/gpu/drm/i915/display/intel_ddi.c | 2 + > .../drm/i915/display/intel_display_types.h | 9 +++ > include/drm/drm_edid.h | 9 +++ First of all, drm_edid.[ch] and i915 changes need to be separated. > 6 files changed, 123 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/drm_edid.c b/drivers/gpu/drm/drm_edid.c > index 90a5e26eafa8..3495af8d8596 100644 > --- a/drivers/gpu/drm/drm_edid.c > +++ b/drivers/gpu/drm/drm_edid.c > @@ -6988,3 +6988,39 @@ static void _drm_update_tile_info(struct drm_connector *connector, > connector->tile_group = NULL; > } > } > + > +/* > + * drm_sad_to_format - extract format of an SAD > + * @sad: pointer to an sad > + * > + * extract the format of an SAD from byte 0 > + */ > +int drm_sad_to_format(const u8 *sad) > +{ > + return (sad[0] & DRM_ELD_SAD_FORMAT_MASK) >> DRM_ELD_SAD_FORMAT_SHIFT; > +} > +EXPORT_SYMBOL(drm_sad_to_format); > + > +/* > + * drm_sad_to_format - extract channels of an SAD > + * @sad: pointer to an sad > + * > + * extract number of supported channels from byte 0 of SAD > + */ > +int drm_sad_to_channels(const u8 *sad) > +{ > + return (sad[0] & DRM_ELD_SAD_CHANNELS_MASK) + 1; > +} > +EXPORT_SYMBOL(drm_sad_to_channels); > + > +/* > + * drm_sad_to_format - extract supported rates of an SAD > + * @sad: pointer to an sad > + * > + * extract supported rates from byte 1 of SAD > + */ > +int drm_sad_to_rates(const u8 *sad) > +{ > + return sad[1] & DRM_ELD_SAD_RATES_MASK; > +} > +EXPORT_SYMBOL(drm_sad_to_rates); Not enthusiastic about adding a bunch of new functions to inspect the sad here. Makes me think this should be parsed and stored in drm_display_info in drm_edid_to_eld() instead if it's generally useful data. > diff --git a/drivers/gpu/drm/i915/display/intel_audio.c b/drivers/gpu/drm/i915/display/intel_audio.c > index 6c9ee905f132..ac425b652331 100644 > --- a/drivers/gpu/drm/i915/display/intel_audio.c > +++ b/drivers/gpu/drm/i915/display/intel_audio.c > @@ -1235,7 +1235,33 @@ static int i915_audio_component_get_eld(struct device *kdev, int port, > if (*enabled) { > eld = intel_encoder->audio_connector->eld; > ret = drm_eld_size(eld); > - memcpy(buf, eld, min(max_bytes, ret)); > + > + if (intel_encoder->sad_mask_valid) { > + int i; > + u8 *temp_buf; > + u8 *sad; > + > + temp_buf = kzalloc(ret, GFP_KERNEL); > + > + if (!temp_buf) { > + mutex_unlock(&dev_priv->audio.mutex); > + return -ENOMEM; Please no error returns like this; use goto and common exit handling with the unlock. > + } > + > + memcpy(temp_buf, eld, ret); > + > + sad = (u8 *)drm_eld_sad(temp_buf); This may return NULL. And cringe about casting const away. > + > + for (i = 0; i < drm_eld_sad_count(temp_buf); i++, sad += 3) { > + if (!(intel_encoder->supported_sads & (1 << i))) > + memset(&sad[0], 0, 3); Here's the main question I have about the change, really. The ELD (literally EDID-like-data) and SAD are supposed to describe the *sink* capabilities. Why are we filtering the data here, telling the audio controller driver this is what the *sink* supports, when the limitation comes from the *source*? I could just be clueless about how audio works, but semantically this feels the same as modifying the EDID based on what the *source* supports. > + } > + > + memcpy(buf, temp_buf, min(max_bytes, ret)); > + kfree(temp_buf); > + } else { > + memcpy(buf, eld, min(max_bytes, ret)); > + } > } > > mutex_unlock(&dev_priv->audio.mutex); > @@ -1408,3 +1434,40 @@ void intel_audio_deinit(struct drm_i915_private *dev_priv) > else > i915_audio_component_cleanup(dev_priv); > } > + > +void intel_eld_compute_config(struct intel_encoder *encoder, > + const struct intel_crtc_state *pipe_config, > + const struct drm_connector_state *conn_state) It should probably be named intel_audio_compute_config(). > +{ > + struct drm_connector *connector = conn_state->connector; > + u8 *eld = connector->eld; > + const u8 *sad = drm_eld_sad(eld); > + struct drm_i915_private *dev_priv = to_i915(encoder->base.dev); Should be named i915 for all new code. > + u32 *supported_sads = &encoder->supported_sads; > + int i, channels; > + > + mutex_lock(&dev_priv->audio.mutex); > + > + /* reset information about supported sads to default */ > + *supported_sads = 0; > + encoder->sad_mask_valid = false; > + > + /* Currently filtering SADs is necessary only for GLK due to it's > + * hardware limitations. However, this function can be scaled > + * to any scenario where display driver has to filter out certain > + * SADs before exposing them to the audio driver. > + */ > + if (IS_GEMINILAKE(dev_priv) && > + intel_crtc_has_type(pipe_config, INTEL_OUTPUT_DP)) { > + encoder->sad_mask_valid = true; > + > + for (i = 0; i < drm_eld_sad_count(eld); i++, sad += 3) { > + channels = drm_sad_to_channels(sad); > + > + if (channels <= 2) > + (*supported_sads) |= 1 << i; > + } > + } It's wrong to modify the encoder members in compute config. Just imagine compute config failing after this; you've already changed the encoder and you can't get it back. > + drm_dbg_kms(&dev_priv->drm, "supported_sads 0x%x\n", *supported_sads); > + mutex_unlock(&dev_priv->audio.mutex); The indentation is off in the function. BR, Jani. > +} > diff --git a/drivers/gpu/drm/i915/display/intel_audio.h b/drivers/gpu/drm/i915/display/intel_audio.h > index 63b22131dc45..17c468a9e07e 100644 > --- a/drivers/gpu/drm/i915/display/intel_audio.h > +++ b/drivers/gpu/drm/i915/display/intel_audio.h > @@ -22,5 +22,8 @@ void intel_audio_cdclk_change_pre(struct drm_i915_private *dev_priv); > void intel_audio_cdclk_change_post(struct drm_i915_private *dev_priv); > void intel_audio_init(struct drm_i915_private *dev_priv); > void intel_audio_deinit(struct drm_i915_private *dev_priv); > +void intel_eld_compute_config(struct intel_encoder *encoder, > + const struct intel_crtc_state *pipe_config, > + const struct drm_connector_state *conn_state); > > #endif /* __INTEL_AUDIO_H__ */ > diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c b/drivers/gpu/drm/i915/display/intel_ddi.c > index 6c43a5124cb8..e7807500c88d 100644 > --- a/drivers/gpu/drm/i915/display/intel_ddi.c > +++ b/drivers/gpu/drm/i915/display/intel_ddi.c > @@ -3678,6 +3678,8 @@ static int intel_ddi_compute_config(struct intel_encoder *encoder, > > intel_ddi_compute_min_voltage_level(dev_priv, pipe_config); > > + intel_eld_compute_config(encoder, pipe_config, conn_state); > + > return 0; > } > > diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h b/drivers/gpu/drm/i915/display/intel_display_types.h > index 0da9b208d56e..5b6a694ff0cc 100644 > --- a/drivers/gpu/drm/i915/display/intel_display_types.h > +++ b/drivers/gpu/drm/i915/display/intel_display_types.h > @@ -265,6 +265,15 @@ struct intel_encoder { > /* for communication with audio component; protected by av_mutex */ > const struct drm_connector *audio_connector; > > + /* bitmask to track supported SADs for current audio connector. > + * According to HDA spec Rev 1.0a, ELD SAD limit is 15. Using > + * a 32 bit mask for future proofing; protected by av_mutex > + */ > + u32 supported_sads; > + > + /* indicates if the supported_sads is a valid bitmask; protected by av_mutex */ > + bool sad_mask_valid; > + > /* VBT information for this encoder (may be NULL for older platforms) */ > const struct intel_bios_encoder_data *devdata; > }; > diff --git a/include/drm/drm_edid.h b/include/drm/drm_edid.h > index 2181977ae683..363f4487cdd6 100644 > --- a/include/drm/drm_edid.h > +++ b/include/drm/drm_edid.h > @@ -318,6 +318,11 @@ struct detailed_timing { > > #define DRM_ELD_CEA_SAD(mnl, sad) (20 + (mnl) + 3 * (sad)) > > +#define DRM_ELD_SAD_FORMAT_MASK 0x78 > +#define DRM_ELD_SAD_FORMAT_SHIFT 3 > +#define DRM_ELD_SAD_CHANNELS_MASK 0x7 > +#define DRM_ELD_SAD_RATES_MASK 0x7f > + > struct edid { > u8 header[8]; > /* Vendor & product info */ > @@ -378,6 +383,10 @@ int drm_edid_to_speaker_allocation(const struct edid *edid, u8 **sadb); > int drm_av_sync_delay(struct drm_connector *connector, > const struct drm_display_mode *mode); > > +int drm_sad_to_format(const u8 *sad); > +int drm_sad_to_channels(const u8 *sad); > +int drm_sad_to_rates(const u8 *sad); > + > #ifdef CONFIG_DRM_LOAD_EDID_FIRMWARE > struct edid *drm_load_edid_firmware(struct drm_connector *connector); > int __drm_set_edid_firmware_path(const char *path); -- Jani Nikula, Intel Open Source Graphics Center
