Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider: - [Medium] drm/amdgpu: broken audio pin state and sharing in dce_v*_0_afmt_setmode() - [Medium] drm/amdgpu: dce_v6_0_afmt_setmode() ignores AMDGPU_AUDIO_DISABLE on DP --
commit bd861db0f0aadc54fc085212778f034d398ae2b3 Author: renpanpan <[email protected]> drm/amdgpu: Only enable HDMI audio if display supports it This commit gates the initial and final audio_enable() calls in dce_v6_0_afmt_setmode(), dce_v8_0_afmt_setmode(), and dce_v10_0_afmt_setmode() on whether the connected display reports audio support in its EDID or has audio forced on via AMDGPU_AUDIO_ENABLE. > diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c > b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c > index 7de7adc8f225c..6d5f25ea67c2f 100644 > --- a/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c > +++ b/drivers/gpu/drm/amd/amdgpu/dce_v10_0.c > @@ -1555,9 +1555,11 @@ static void dce_v10_0_afmt_setmode(struct drm_encoder > *encoder, > bpc = amdgpu_crtc->bpc; > } > > - /* disable audio prior to setting up hw */ > + /* disable audio prior to setting up hw, unless the display has no > audio */ > dig->afmt->pin = dce_v10_0_audio_get_pin(adev); > - dce_v10_0_audio_enable(adev, dig->afmt->pin, false); > + if (connector->display_info.has_audio || > + to_amdgpu_connector(connector)->audio == AMDGPU_AUDIO_ENABLE) > + dce_v10_0_audio_enable(adev, dig->afmt->pin, false); > > dce_v10_0_audio_set_dto(encoder, mode->clock); > > @@ -1705,8 +1707,10 @@ static void dce_v10_0_afmt_setmode(struct drm_encoder > *encoder, > WREG32(mmAFMT_RAMP_CONTROL2 + dig->afmt->offset, 0x00000001); > WREG32(mmAFMT_RAMP_CONTROL3 + dig->afmt->offset, 0x00000001); > > - /* enable audio after to setting up hw */ > - dce_v10_0_audio_enable(adev, dig->afmt->pin, true); > + /* Only enable audio if the display supports it or is forced on */ > + if (connector->display_info.has_audio || > + to_amdgpu_connector(connector)->audio == AMDGPU_AUDIO_ENABLE) > + dce_v10_0_audio_enable(adev, dig->afmt->pin, true); > } [Severity: Medium] As reported by Sashiko on v2 [1], gating only the initial audio_enable(adev, dig->afmt->pin, false) and final audio_enable(adev, dig->afmt->pin, true) calls in dce_v10_0_afmt_setmode(), dce_v6_0_afmt_setmode(), and dce_v8_0_afmt_setmode() still leaves two issues with audio pin state and sharing: First, when an already-active encoder transitions from an audio-enabled state to a non-audio state (for example, changing the connector audio property from AMDGPU_AUDIO_ENABLE back to AMDGPU_AUDIO_AUTO on a non-audio HDMI monitor via amdgpu_connector_set_property() -> amdgpu_connector_property_change_mode() -> drm_crtc_helper_set_mode(), or hot-swapping an audio-capable display for a non-audio display on the same active CRTC), drm_crtc_prepare_encoders() does not disable the active encoder, and dce_v10_0_afmt_enable(encoder, true) returns early because dig->afmt->enabled is already true. When dce_v10_0_afmt_setmode() then runs with both connector->display_info.has_audio and audio == AMDGPU_AUDIO_ENABLE evaluating to false, both the initial dce_v10_0_audio_enable(..., false) and final dce_v10_0_audio_enable(..., true) calls are skipped. Doesn't this leave AZALIA_F0_CODEC_PIN_CONTROL_HOT_PLUG_CONTROL__AUDIO_ENABLED_MASK stuck enabled in hardware? Second, dce_v10_0_audio_get_pin(), dce_v8_0_audio_get_pin(), and dce_v6_0_audio_get_pin() do not track per-encoder pin ownership and always return the first connected pin in adev->mode_info.audio.pin[]: drivers/gpu/drm/amd/amdgpu/dce_v10_0.c:dce_v10_0_afmt_setmode() { ... dig->afmt->pin = dce_v10_0_audio_get_pin(adev); if (connector->display_info.has_audio || to_amdgpu_connector(connector)->audio == AMDGPU_AUDIO_ENABLE) dce_v10_0_audio_enable(adev, dig->afmt->pin, false); dce_v10_0_audio_set_dto(encoder, mode->clock); ... dce_v10_0_audio_write_speaker_allocation(encoder); ... dce_v10_0_afmt_audio_select_pin(encoder); dce_v10_0_audio_write_sad_regs(encoder); dce_v10_0_audio_write_latency_fields(encoder, mode); ... } Because dig->afmt->pin is still unconditionally assigned on a non-audio display, dce_v10_0_afmt_setmode() still overwrites the global mmDCCG_AUDIO_DTO_SOURCE and mmDCCG_AUDIO_DTO0_* registers as well as the shared pin's speaker allocation, SAD, and latency registers, and enables audio infoframes and sample packets. Furthermore, when the non-audio encoder is later disabled via dce_v10_0_encoder_disable() -> dce_v10_0_afmt_enable(encoder, false): drivers/gpu/drm/amd/amdgpu/dce_v10_0.c:dce_v10_0_afmt_enable() { ... if (!enable && dig->afmt->pin) { dce_v10_0_audio_enable(adev, dig->afmt->pin, false); dig->afmt->pin = NULL; } ... } won't disabling the non-audio encoder unconditionally call dce_v10_0_audio_enable(adev, dig->afmt->pin, false) and turn off audio on another active display sharing that pin (and similarly in dce_v6_0.c and dce_v8_0.c)? > diff --git a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c > b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c > index 3c1d6901127d1..3add851974b47 100644 > --- a/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c > +++ b/drivers/gpu/drm/amd/amdgpu/dce_v6_0.c > @@ -1724,8 +1724,10 @@ static void dce_v6_0_afmt_setmode(struct drm_encoder > *encoder, > bpc = amdgpu_crtc->bpc; > } > > - /* disable audio before setting up hw */ > - dce_v6_0_audio_enable(adev, dig->afmt->pin, false); > + /* disable audio before setting up hw, unless the display has no audio > */ > + if (connector->display_info.has_audio || > + amdgpu_connector->audio == AMDGPU_AUDIO_ENABLE) > + dce_v6_0_audio_enable(adev, dig->afmt->pin, false); > > dce_v6_0_audio_set_mute(encoder, true); > dce_v6_0_audio_write_speaker_allocation(encoder); > @@ -1748,8 +1750,10 @@ static void dce_v6_0_afmt_setmode(struct drm_encoder > *encoder, > dce_v6_0_audio_dp_enable(encoder, 1); > } > > - /* enable audio after setting up hw */ > - dce_v6_0_audio_enable(adev, dig->afmt->pin, true); > + /* Only enable audio if the display supports it or is forced on */ > + if (connector->display_info.has_audio || > + amdgpu_connector->audio == AMDGPU_AUDIO_ENABLE) > + dce_v6_0_audio_enable(adev, dig->afmt->pin, true); > } [Severity: Medium] For HDMI and DVI connectors, amdgpu_atombios_encoder_get_encoder_mode() returns ATOM_ENCODER_MODE_DVI when amdgpu_connector->audio == AMDGPU_AUDIO_DISABLE, which prevents dce_v6_0_afmt_setmode() from being called. However, for native DisplayPort sinks, amdgpu_atombios_encoder_get_encoder_mode() unconditionally returns ATOM_ENCODER_MODE_DP regardless of amdgpu_connector->audio: drivers/gpu/drm/amd/amdgpu/atombios_encoders.c:amdgpu_atombios_encoder_get_encoder_mode() { ... case DRM_MODE_CONNECTOR_DisplayPort: dig_connector = amdgpu_connector->con_priv; if ((dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_DISPLAYPORT) || (dig_connector->dp_sink_type == CONNECTOR_OBJECT_ID_eDP)) { return ATOM_ENCODER_MODE_DP; } ... } and dce_v6_0_encoder_mode_set() calls dce_v6_0_afmt_setmode() whenever ENCODER_MODE_IS_DP(em) is true. Because dce_v6_0_afmt_setmode() checks connector->display_info.has_audio without checking amdgpu_connector->audio != AMDGPU_AUDIO_DISABLE, won't setting the audio connector property to off (AMDGPU_AUDIO_DISABLE) on an audio-capable DisplayPort monitor still enable audio here? [1] https://lore.kernel.org/r/[email protected] -- Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
