[PATCH 0/3] ALSA: hda/hdmi: i915 keepalive fixes
A series with multiple fixes to i915 keepalive (KAE) functionality. First patch fixes issue that is hit on some A750/770 cards: https://gitlab.freedesktop.org/drm/intel/-/issues/7353
The two other improve behaviour especially with certain USB-C docks.
Kai Vehmanen (3): ALSA: hda/hdmi: fix i915 silent stream programming flow ALSA: hda/hdmi: set default audio parameters for KAE silent-stream ALSA: hda/hdmi: fix stream-id config keep-alive for rt suspend
include/sound/hda_codec.h | 1 + sound/pci/hda/hda_codec.c | 3 +- sound/pci/hda/patch_hdmi.c | 114 ++++++++++++++++++++++++++++++++++++- 3 files changed, 114 insertions(+), 4 deletions(-)
base-commit: 81a2da5a10a6eaa6ae16108eed4e74651cc296bf
The i915 display codec may not successfully transition to normal audio streaming mode, if the stream id is programmed while codec is actively transmitting data. This can happen when silent stream is enabled in KAE mode.
Fix the issue by implementing a i915 specific programming flow, where the silent streaming is temporarily stopped, a small delay is applied to ensure display codec becomes idle, and then proceed with reprogramming the stream ID.
Fixes: 15175a4f2bbb ("ALSA: hda/hdmi: add keep-alive support for ADL-P and DG2") Link: https://gitlab.freedesktop.org/drm/intel/-/issues/7353 Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com Tested-by: Rodrigo Vivi rodrigo.vivi@intel.com --- sound/pci/hda/patch_hdmi.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-)
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index 7a40ddfd695a..a0ba24165ae2 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -2879,9 +2879,28 @@ static int i915_hsw_setup_stream(struct hda_codec *codec, hda_nid_t cvt_nid, hda_nid_t pin_nid, int dev_id, u32 stream_tag, int format) { + struct hdmi_spec *spec = codec->spec; + int pin_idx = pin_id_to_pin_index(codec, pin_nid, dev_id); + struct hdmi_spec_per_pin *per_pin = get_pin(spec, pin_idx); + int res; + haswell_verify_D0(codec, cvt_nid, pin_nid); - return hdmi_setup_stream(codec, cvt_nid, pin_nid, dev_id, - stream_tag, format); + + if (spec->silent_stream_type == SILENT_STREAM_KAE && per_pin && per_pin->silent_stream) { + silent_stream_set_kae(codec, per_pin, false); + /* wait for pending transfers in codec to clear */ + usleep_range(100, 200); + } + + res = hdmi_setup_stream(codec, cvt_nid, pin_nid, dev_id, + stream_tag, format); + + if (spec->silent_stream_type == SILENT_STREAM_KAE && per_pin && per_pin->silent_stream) { + usleep_range(100, 200); + silent_stream_set_kae(codec, per_pin, true); + } + + return res; }
/* pin_cvt_fixup ops override for HSW+ and VLV+ */
On 12/8/2022 4:43 PM, Kai Vehmanen wrote:
The i915 display codec may not successfully transition to normal audio streaming mode, if the stream id is programmed while codec is actively transmitting data. This can happen when silent stream is enabled in KAE mode.
Fix the issue by implementing a i915 specific programming flow, where the silent streaming is temporarily stopped, a small delay is applied to ensure display codec becomes idle, and then proceed with reprogramming the stream ID.
Fixes: 15175a4f2bbb ("ALSA: hda/hdmi: add keep-alive support for ADL-P and DG2") Link: https://gitlab.freedesktop.org/drm/intel/-/issues/7353 Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com Tested-by: Rodrigo Vivi rodrigo.vivi@intel.com
sound/pci/hda/patch_hdmi.c | 23 +++++++++++++++++++++-- 1 file changed, 21 insertions(+), 2 deletions(-)
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index 7a40ddfd695a..a0ba24165ae2 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -2879,9 +2879,28 @@ static int i915_hsw_setup_stream(struct hda_codec *codec, hda_nid_t cvt_nid, hda_nid_t pin_nid, int dev_id, u32 stream_tag, int format) {
- struct hdmi_spec *spec = codec->spec;
- int pin_idx = pin_id_to_pin_index(codec, pin_nid, dev_id);
Shouldn't return value from pin_id_to_pin_index() be checked? It seems that it can return -EINVAL.
- struct hdmi_spec_per_pin *per_pin = get_pin(spec, pin_idx);
- int res;
- haswell_verify_D0(codec, cvt_nid, pin_nid);
- return hdmi_setup_stream(codec, cvt_nid, pin_nid, dev_id,
stream_tag, format);
if (spec->silent_stream_type == SILENT_STREAM_KAE && per_pin && per_pin->silent_stream) {
silent_stream_set_kae(codec, per_pin, false);
/* wait for pending transfers in codec to clear */
usleep_range(100, 200);
}
res = hdmi_setup_stream(codec, cvt_nid, pin_nid, dev_id,
stream_tag, format);
if (spec->silent_stream_type == SILENT_STREAM_KAE && per_pin && per_pin->silent_stream) {
usleep_range(100, 200);
silent_stream_set_kae(codec, per_pin, true);
}
return res; }
/* pin_cvt_fixup ops override for HSW+ and VLV+ */
Hi,
On Thu, 8 Dec 2022, Amadeusz Sławiński wrote:
On 12/8/2022 4:43 PM, Kai Vehmanen wrote:
@@ -2879,9 +2879,28 @@ static int i915_hsw_setup_stream(struct hda_codec *codec, hda_nid_t cvt_nid, hda_nid_t pin_nid, int dev_id, u32 stream_tag, int format) {
- struct hdmi_spec *spec = codec->spec;
- int pin_idx = pin_id_to_pin_index(codec, pin_nid, dev_id);
Shouldn't return value from pin_id_to_pin_index() be checked? It seems that it can return -EINVAL.
that's a good point. I think we are safe with current code as setup_stream ops is only called from generic_hdmi_playback_pcm_prepare() and spec->ops.setup_stream() there is only called with a valid pin. But this leaves room for future errors, and passing negative index to get_pin() is pretty bad. Let me send a V2 later today.
Thanks for the review!
Br, Kai
If the stream-id is zero, the keep-alive (KAE) will only ensure clock is generated, but no audio samples are sent over display link. This happens before first real audio stream is played out to a newly connected receiver.
Reuse the code in silent_stream_enable() to set up stream parameters to sane defaults values, also when using the newer keep-alive flow.
Fixes: 15175a4f2bbb ("ALSA: hda/hdmi: add keep-alive support for ADL-P and DG2") Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com Tested-by: Rodrigo Vivi rodrigo.vivi@intel.com --- sound/pci/hda/patch_hdmi.c | 1 + 1 file changed, 1 insertion(+)
diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index a0ba24165ae2..1618419647aa 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -1738,6 +1738,7 @@ static void silent_stream_enable(struct hda_codec *codec,
switch (spec->silent_stream_type) { case SILENT_STREAM_KAE: + silent_stream_enable_i915(codec, per_pin); silent_stream_set_kae(codec, per_pin, true); break; case SILENT_STREAM_I915:
When the new style KAE keep-alive implementation is used on compatible Intel hardware, the clocks are maintained when codec is in D3. The generic code in hda_cleanup_all_streams() can however interfere with generation of audio samples in this mode, by setting the stream and channel ids to zero.
To get full benefit of the keepalive, set the new no_stream_clean_at_suspend quirk bit on affected Intel hardware. When this bit is set, stream cleanup is skipped in hda_call_codec_suspend().
Special handling is needed for the case when system goes to suspend. The stream id programming can be lost in this case. This will also cause codec->cvt_setups to be out of sync. Handle this by implementing custom suspend/resume handlers. If keep-alive is active for any converter, set the quirk flags no_stream_clean_at_suspend and forced_resume. Upon resume, keepalive programming is restored if needed.
Fixes: 15175a4f2bbb ("ALSA: hda/hdmi: add keep-alive support for ADL-P and DG2") Signed-off-by: Kai Vehmanen kai.vehmanen@linux.intel.com Reviewed-by: Pierre-Louis Bossart pierre-louis.bossart@linux.intel.com --- include/sound/hda_codec.h | 1 + sound/pci/hda/hda_codec.c | 3 +- sound/pci/hda/patch_hdmi.c | 90 +++++++++++++++++++++++++++++++++++++- 3 files changed, 92 insertions(+), 2 deletions(-)
diff --git a/include/sound/hda_codec.h b/include/sound/hda_codec.h index 25ec8c181688..eba23daf2c29 100644 --- a/include/sound/hda_codec.h +++ b/include/sound/hda_codec.h @@ -258,6 +258,7 @@ struct hda_codec { unsigned int link_down_at_suspend:1; /* link down at runtime suspend */ unsigned int relaxed_resume:1; /* don't resume forcibly for jack */ unsigned int forced_resume:1; /* forced resume for jack */ + unsigned int no_stream_clean_at_suspend:1; /* do not clean streams at suspend */
#ifdef CONFIG_PM unsigned long power_on_acct; diff --git a/sound/pci/hda/hda_codec.c b/sound/pci/hda/hda_codec.c index b4d1e658c556..edd653ece70d 100644 --- a/sound/pci/hda/hda_codec.c +++ b/sound/pci/hda/hda_codec.c @@ -2886,7 +2886,8 @@ static unsigned int hda_call_codec_suspend(struct hda_codec *codec) snd_hdac_enter_pm(&codec->core); if (codec->patch_ops.suspend) codec->patch_ops.suspend(codec); - hda_cleanup_all_streams(codec); + if (!codec->no_stream_clean_at_suspend) + hda_cleanup_all_streams(codec); state = hda_set_power_state(codec, AC_PWRST_D3); update_power_acct(codec, true); snd_hdac_leave_pm(&codec->core); diff --git a/sound/pci/hda/patch_hdmi.c b/sound/pci/hda/patch_hdmi.c index 1618419647aa..dfae3443f6bd 100644 --- a/sound/pci/hda/patch_hdmi.c +++ b/sound/pci/hda/patch_hdmi.c @@ -2921,6 +2921,88 @@ static void i915_pin_cvt_fixup(struct hda_codec *codec, } }
+#ifdef CONFIG_PM +static int i915_adlp_hdmi_suspend(struct hda_codec *codec) +{ + struct hdmi_spec *spec = codec->spec; + bool silent_streams = false; + int pin_idx, res; + + res = generic_hdmi_suspend(codec); + + for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { + struct hdmi_spec_per_pin *per_pin = get_pin(spec, pin_idx); + + if (per_pin->silent_stream) { + silent_streams = true; + break; + } + } + + if (silent_streams && spec->silent_stream_type == SILENT_STREAM_KAE) { + /* + * stream-id should remain programmed when codec goes + * to runtime suspend + */ + codec->no_stream_clean_at_suspend = 1; + + /* + * the system might go to S3, in which case keep-alive + * must be reprogrammed upon resume + */ + codec->forced_resume = 1; + + codec_dbg(codec, "HDMI: KAE active at suspend\n"); + } else { + codec->no_stream_clean_at_suspend = 0; + codec->forced_resume = 0; + } + + return res; +} + +static int i915_adlp_hdmi_resume(struct hda_codec *codec) +{ + struct hdmi_spec *spec = codec->spec; + int pin_idx, res; + + res = generic_hdmi_resume(codec); + + /* KAE not programmed at suspend, nothing to do here */ + if (!codec->no_stream_clean_at_suspend) + return res; + + for (pin_idx = 0; pin_idx < spec->num_pins; pin_idx++) { + struct hdmi_spec_per_pin *per_pin = get_pin(spec, pin_idx); + + /* + * If system was in suspend with monitor connected, + * the codec setting may have been lost. Re-enable + * keep-alive. + */ + if (per_pin->silent_stream) { + unsigned int param; + + param = snd_hda_codec_read(codec, per_pin->cvt_nid, 0, + AC_VERB_GET_CONV, 0); + if (!param) { + codec_dbg(codec, "HDMI: KAE: restore stream id\n"); + silent_stream_enable_i915(codec, per_pin); + } + + param = snd_hda_codec_read(codec, per_pin->cvt_nid, 0, + AC_VERB_GET_DIGI_CONVERT_1, 0); + if (!(param & (AC_DIG3_KAE << 16))) { + codec_dbg(codec, "HDMI: KAE: restore DIG3_KAE\n"); + silent_stream_set_kae(codec, per_pin, true); + } + } + } + + return res; +} +#endif + /* precondition and allocation for Intel codecs */ static int alloc_intel_hdmi(struct hda_codec *codec) { @@ -3051,8 +3133,14 @@ static int patch_i915_adlp_hdmi(struct hda_codec *codec) if (!res) { spec = codec->spec;
- if (spec->silent_stream_type) + if (spec->silent_stream_type) { spec->silent_stream_type = SILENT_STREAM_KAE; + +#ifdef CONFIG_PM + codec->patch_ops.resume = i915_adlp_hdmi_resume; + codec->patch_ops.suspend = i915_adlp_hdmi_suspend; +#endif + } }
return res;
participants (2)
-
Amadeusz Sławiński
-
Kai Vehmanen