ALSA: hda - Retry codec-verbs at errors

The current error-recovery scheme for the codec communication errors
doesn't work always well.  Especially falling back to the
single-command mode causes the fatal problem on many systems.

In this patch, the problematic verb is re-issued again after the error
(even with polling mode) instead of the single-cmd mode.  The
single-cmd mode will be used only when specified via the command
option explicitly, mainly just for testing.

Signed-off-by: Takashi Iwai <tiwai@suse.de>

3 files changed, 24 insertions, 13 deletions

diff --git a/sound/pci/hda/hda_codec.c b/sound/pci/hda/hda_codec.c
index 3d8bf39e6d9..1736ccbebc7 100644
--- a/sound/pci/hda/hda_codec.c
+++ b/sound/pci/hda/hda_codec.c
@@ -174,14 +174,23 @@ unsigned int snd_hda_codec_read(struct hda_codec *codec, hda_nid_t nid,
 				unsigned int verb, unsigned int parm)
 {
 	struct hda_bus *bus = codec->bus;
-	unsigned int res;
+	unsigned int cmd, res;
+	int repeated = 0;
 
-	res = make_codec_cmd(codec, nid, direct, verb, parm);
+	cmd = make_codec_cmd(codec, nid, direct, verb, parm);
 	snd_hda_power_up(codec);
 	mutex_lock(&bus->cmd_mutex);
-	if (!bus->ops.command(bus, res))
+ again:
+	if (!bus->ops.command(bus, cmd)) {
 		res = bus->ops.get_response(bus);
-	else
+		if (res == -1 && bus->rirb_error) {
+			if (repeated++ < 1) {
+				snd_printd(KERN_WARNING "hda_codec: "
+					   "Trying verb 0x%08x again\n", cmd);
+				goto again;
+			}
+		}
+	} else
 		res = (unsigned int)-1;
 	mutex_unlock(&bus->cmd_mutex);
 	snd_hda_power_down(codec);
diff --git a/sound/pci/hda/hda_codec.h b/sound/pci/hda/hda_codec.h
index 2fdecf4b0eb..cd8979c7670 100644
--- a/sound/pci/hda/hda_codec.h
+++ b/sound/pci/hda/hda_codec.h
@@ -623,6 +623,7 @@ struct hda_bus {
 	/* misc op flags */
 	unsigned int needs_damn_long_delay :1;
 	unsigned int shutdown :1;	/* being unloaded */
+	unsigned int rirb_error:1;	/* error in codec communication */
 };
 
 /*
diff --git a/sound/pci/hda/hda_intel.c b/sound/pci/hda/hda_intel.c
index 30829ee920c..803b72098ed 100644
--- a/sound/pci/hda/hda_intel.c
+++ b/sound/pci/hda/hda_intel.c
@@ -604,6 +604,7 @@ static unsigned int azx_rirb_get_response(struct hda_bus *bus)
 		}
 		if (!chip->rirb.cmds) {
 			smp_rmb();
+			bus->rirb_error = 0;
 			return chip->rirb.res; /* the last value */
 		}
 		if (time_after(jiffies, timeout))
@@ -623,8 +624,10 @@ static unsigned int azx_rirb_get_response(struct hda_bus *bus)
 		chip->irq = -1;
 		pci_disable_msi(chip->pci);
 		chip->msi = 0;
-		if (azx_acquire_irq(chip, 1) < 0)
+		if (azx_acquire_irq(chip, 1) < 0) {
+			bus->rirb_error = 1;
 			return -1;
+		}
 		goto again;
 	}
 
@@ -644,14 +647,12 @@ static unsigned int azx_rirb_get_response(struct hda_bus *bus)
 		return -1;
 	}
 
-	snd_printk(KERN_ERR "hda_intel: azx_get_response timeout, "
-		   "switching to single_cmd mode: last cmd=0x%08x\n",
-		   chip->last_cmd);
-	chip->rirb.rp = azx_readb(chip, RIRBWP);
-	chip->rirb.cmds = 0;
-	/* switch to single_cmd mode */
-	chip->single_cmd = 1;
-	azx_free_cmd_io(chip);
+	snd_printk(KERN_ERR "hda_intel: azx_get_response timeout (ERROR): "
+		   "last cmd=0x%08x\n", chip->last_cmd);
+	spin_lock_irq(&chip->reg_lock);
+	chip->rirb.cmds = 0; /* reset the index */
+	bus->rirb_error = 1;
+	spin_unlock_irq(&chip->reg_lock);
 	return -1;
 }