Skip to contentSkip to conversation

ALSA: usb-audio: Add retry on -EPROTO from usb_set_interface()

PATCH In review Geoffrey D. Bennett 2 participants 2 messages Apr 16, 2025 – Apr 20, 2025 Sound

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ALSA: usb-audio: Add retry on -EPROTO from usb_set_interface()

During initialisation of Focusrite USB audio interfaces, -EPROTO is sometimes returned from usb_set_interface(), which sometimes prevents the device from working: subsequent usb_set_interface() and uac_clock_source_is_valid() calls fail.

This patch adds up to 5 retries in endpoint_set_interface(), with a delay starting at 5ms and doubling each time. 5 retries was chosen to allow for longer than expected waits for the interface to start responding correctly; in testing, a single 5ms delay was sufficient to fix the issue.

---
 sound/usb/endpoint.c | 7 +++++++
 1 file changed, 7 insertions(+)
sound/usb/endpoint.c +7 -0
index a29f28eb7d0c..f36ec98da460 100644
@@ -926,6 +926,8 @@ static int endpoint_set_interface(struct snd_usb_audio *chip,
 {
 	int altset = set ? ep->altsetting : 0;
 	int err;
+	int retries = 0;
+	const int max_retries = 5;
 
 	if (ep->iface_ref->altset == altset)
 		return 0;
@@ -935,8 +937,13 @@ static int endpoint_set_interface(struct snd_usb_audio *chip,
 
 	usb_audio_dbg(chip, "Setting usb interface %d:%d for EP 0x%x\n",
 		      ep->iface, altset, ep->ep_num);
+retry:
 	err = usb_set_interface(chip->dev, ep->iface, altset);
 	if (err < 0) {
+		if (err == -EPROTO && ++retries <= max_retries) {
+			msleep(5 * (1 << (retries - 1)));
+			goto retry;
+		}
 		usb_audio_err_ratelimited(
 			chip, "%d:%d: usb_set_interface failed (%d)\n",
 			ep->iface, altset, err);
Signature · 1 line
2.45.0
#1 lore raw
Quoted 12 lines · Geoffrey D. Bennett

During initialisation of Focusrite USB audio interfaces, -EPROTO is sometimes returned from usb_set_interface(), which sometimes prevents the device from working: subsequent usb_set_interface() and uac_clock_source_is_valid() calls fail.

This patch adds up to 5 retries in endpoint_set_interface(), with a delay starting at 5ms and doubling each time. 5 retries was chosen to allow for longer than expected waits for the interface to start responding correctly; in testing, a single 5ms delay was sufficient to fix the issue.

Quoted 3 lines

Applied now. Thanks.

Takashi