Skip to contentSkip to conversation

ALSA: hda/hdmi: fix failures at PCM open on Intel ICL and later

PATCH Applied Kai Vehmanen 2 participants 4 messages Jul 3, 2020 – Jul 7, 2020 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 ALSA: hda/hdmi: fix failures at PCM open on Intel ICL and later

When HDMI PCM devices are opened in a specific order, with at least one HDMI/DP receiver connected, ALSA PCM open fails to -EBUSY on the connected monitor, on recent Intel platforms (ICL/JSL and newer). While this is not a typical sequence, at least Pulseaudio does this every time when it is started, to discover the available PCMs.

The rootcause is an invalid assumption in hdmi_add_pin(), where the total number of converters is assumed to be known at the time the function is called. On older Intel platforms this held true, but after ICL/JSL, the order how pins and converters are in the subnode list as returned by snd_hda_get_sub_nodes(), was changed. As a result, information for some converters was not stored to per_pin->mux_nids. And this means some pins cannot be connected to all converters, and application instead gets -EBUSY instead at open.

The assumption that converters are always before pins in the subnode list, is not really a valid one. Fix the problem in hdmi_parse_codec() by introducing separate loops for discovering converters and pins.

---
 sound/pci/hda/patch_hdmi.c | 36 +++++++++++++++++++++++-------------
 1 file changed, 23 insertions(+), 13 deletions(-)
sound/pci/hda/patch_hdmi.c +23 -13
index e2b21ef5d7d1..295cc5a989d5 100644
@@ -1804,33 +1804,43 @@ static int hdmi_add_cvt(struct hda_codec *codec, hda_nid_t cvt_nid)
 
 static int hdmi_parse_codec(struct hda_codec *codec)
 {
-	hda_nid_t nid;
+	hda_nid_t start_nid;
+	unsigned int caps;
 	int i, nodes;
 
-	nodes = snd_hda_get_sub_nodes(codec, codec->core.afg, &nid);
-	if (!nid || nodes < 0) {
+	nodes = snd_hda_get_sub_nodes(codec, codec->core.afg, &start_nid);
+	if (!start_nid || nodes < 0) {
 		codec_warn(codec, "HDMI: failed to get afg sub nodes\n");
 		return -EINVAL;
 	}
 
-	for (i = 0; i < nodes; i++, nid++) {
-		unsigned int caps;
-		unsigned int type;
+	/*
+	 * hdmi_add_pin() assumes total amount of converters to
+	 * be known, so first discover all converters
+	 */
+	for (i = 0; i < nodes; i++) {
+		hda_nid_t nid = start_nid + i;
 
 		caps = get_wcaps(codec, nid);
-		type = get_wcaps_type(caps);
 
 		if (!(caps & AC_WCAP_DIGITAL))
 			continue;
 
-		switch (type) {
-		case AC_WID_AUD_OUT:
+		if (get_wcaps_type(caps) == AC_WID_AUD_OUT)
 			hdmi_add_cvt(codec, nid);
-			break;
-		case AC_WID_PIN:
+	}
+
+	/* discover audio pins */
+	for (i = 0; i < nodes; i++) {
+		hda_nid_t nid = start_nid + i;
+
+		caps = get_wcaps(codec, nid);
+
+		if (!(caps & AC_WCAP_DIGITAL))
+			continue;
+
+		if (get_wcaps_type(caps) == AC_WID_PIN)
 			hdmi_add_pin(codec, nid);
-			break;
-		}
 	}
 
 	return 0;
Signature · 1 line
2.27.0
#1 lore raw

PATCH 2/2 ALSA: hda/hdmi: improve debug traces for stream lookups

The HDMI codec driver has two debug traces printed from different functions but with identical message content:

"HDMI: hinfo 000000006a6b84d9 not registered"

Fix this duplication and also add a bit more context in addition to raw object pointer, to help analysis of kernel logs.

---
 sound/pci/hda/patch_hdmi.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
sound/pci/hda/patch_hdmi.c +3 -2
index 295cc5a989d5..41eaa89660c3 100644
@@ -259,7 +259,7 @@ static int hinfo_to_pcm_index(struct hda_codec *codec,
 		if (get_pcm_rec(spec, pcm_idx)->stream == hinfo)
 			return pcm_idx;
 
-	codec_warn(codec, "HDMI: hinfo %p not registered\n", hinfo);
+	codec_warn(codec, "HDMI: hinfo %p not tied to a PCM\n", hinfo);
 	return -EINVAL;
 }
 
@@ -277,7 +277,8 @@ static int hinfo_to_pin_index(struct hda_codec *codec,
 			return pin_idx;
 	}
 
-	codec_dbg(codec, "HDMI: hinfo %p not registered\n", hinfo);
+	codec_dbg(codec, "HDMI: hinfo %p (pcm %d) not registered\n", hinfo,
+		  hinfo_to_pcm_index(codec, hinfo));
 	return -EINVAL;
 }
 
Signature · 1 line
2.27.0
Applied
#2 lore raw

Re: [PATCH 2/2] ALSA: hda/hdmi: improve debug traces for stream lookups

Quoted 4 lines · Kai Vehmanen

The HDMI codec driver has two debug traces printed from different functions but with identical message content:

Quoted 8 lines

"HDMI: hinfo 000000006a6b84d9 not registered"

Fix this duplication and also add a bit more context in addition to raw object pointer, to help analysis of kernel logs.

Thanks, applied now.

Takashi

Applied
#3 lore raw
In reply to Kai Vehmanen
Quoted 20 lines · Kai Vehmanen

When HDMI PCM devices are opened in a specific order, with at least one HDMI/DP receiver connected, ALSA PCM open fails to -EBUSY on the connected monitor, on recent Intel platforms (ICL/JSL and newer). While this is not a typical sequence, at least Pulseaudio does this every time when it is started, to discover the available PCMs.

The rootcause is an invalid assumption in hdmi_add_pin(), where the total number of converters is assumed to be known at the time the function is called. On older Intel platforms this held true, but after ICL/JSL, the order how pins and converters are in the subnode list as returned by snd_hda_get_sub_nodes(), was changed. As a result, information for some converters was not stored to per_pin->mux_nids. And this means some pins cannot be connected to all converters, and application instead gets -EBUSY instead at open.

The assumption that converters are always before pins in the subnode list, is not really a valid one. Fix the problem in hdmi_parse_codec() by introducing separate loops for discovering converters and pins.

Quoted 6 lines

Thanks, applied.

Takashi