Skip to contentSkip to conversation

xhci: unratelimited "buffer overrun event" warning floods log (~7000/s) with USB S/PDIF capture device

PATCH Alex Kononov 2 participants 2 messages Nov 21

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Good day Mathias,

With a USB S/PDIF TX/RX adapter (Savitech, 262a:9123, UAC2, high speed), the xHCI driver logs an isochronous buffer overrun warning for nearly every microframe while the adapter's S/PDIF input has no signal:

xhci_hcd 0000:00:14.0: WARN: buffer overrun event for slot 2 ep 4 on endpoint

Rate measured: ~433,000 messages per minute (~7,200/s). systemd-journald sits at 100% of one core and reports "Missed N kernel messages".

Setup:
- Intel N100 (PCH xHCI 0000:00:14.0), Debian 13
- Kernel: 6.12.117+deb13-amd64 (same code in mainline 7.4-rc3)
- Capture stream open continuously (CamillaDSP), 48 kHz S16_LE, EP 0x82 IN isochronous, adaptive, 125 us interval
- TV feeding the S/PDIF input goes idle -> no input signal -> flood. With a valid input signal the count is 0.

The device is clearly misbehaving (it sends more data than requested when the S/PDIF receiver is unlocked; it also reports its clock source as valid while unlocked). The driver handles the events correctly and audio recovers fine. The problem is only the logging: the warning in handle_tx_event() for COMP_ISOCH_BUFFER_OVERRUN is not rate-limited, so a single misbehaving isoc device can generate thousands of lines per second.

Locally I changed that xhci_warn() to dev_warn_ratelimited() and the rate dropped to ~100/min with no functional change.
Would a patch rate-limiting this message (or demoting it to xhci_dbg) be acceptable?

drivers/usb/host/xhci-ring.c +0 -0
@@ handle_tx_event()
 	case COMP_ISOCH_BUFFER_OVERRUN:
-		xhci_warn(xhci,
-			  "WARN: buffer overrun event for slot %u ep %u on

endpoint",
- slot_id, ep_index);
+ dev_warn_ratelimited(xhci_to_hcd(xhci)->self.controller,
+ "WARN: buffer overrun event for slot %u
ep %u on endpoint\n",
+ slot_id, ep_index);
break;
or

case COMP_ISOCH_BUFFER_OVERRUN:
- xhci_warn(xhci,
- "WARN: buffer overrun event for slot %u ep %u on
endpoint",
- slot_id, ep_index);
+ xhci_dbg(xhci, "Buffer overrun event for slot %u ep %u\n",
+ slot_id, ep_index);
break;

Endpoint descriptor:

Endpoint 0x82 (capture), from lsusb -v:
altset 1 (S16_LE): Isochronous, Adaptive, wMaxPacketSize 104, bInterval 1 altset 2 (S24_3LE): Isochronous, Adaptive, wMaxPacketSize 156, bInterval 1

/proc/asound/card0/stream0 while flooding:
Capture: Status: Running, Interface 3, Altset 1, Packet Size = 36, Momentary freq = 48000 Hz

Thank you,
Alex.

#1 lore raw

Hi,

Quoted 22 lines · Alex Kononov

Good day Mathias,

With a USB S/PDIF TX/RX adapter (Savitech, 262a:9123, UAC2, high speed), the xHCI driver logs an isochronous buffer overrun warning for nearly every microframe while the adapter's S/PDIF input has no signal:

xhci_hcd 0000:00:14.0: WARN: buffer overrun event for slot 2 ep 4 on endpoint

Rate measured: ~433,000 messages per minute (~7,200/s). systemd-journald sits at 100% of one core and reports "Missed N kernel messages".

Setup:
- Intel N100 (PCH xHCI 0000:00:14.0), Debian 13
- Kernel: 6.12.117+deb13-amd64 (same code in mainline 7.4-rc3)
- Capture stream open continuously (CamillaDSP), 48 kHz S16_LE, EP 0x82 IN isochronous, adaptive, 125 us interval
- TV feeding the S/PDIF input goes idle -> no input signal -> flood. With a valid input signal the count is 0.

Quoted 3 lines

The device is clearly misbehaving (it sends more data than requested when the S/PDIF receiver is unlocked; it also reports its clock source as valid while unlocked).

Either the device is buggy or snd_usb_audio driver submits wrong URBs, but that's a sound subsystem issue.

Quoted 9 lines

The driver handles the events correctly and audio recovers fine. The problem is only the logging: the warning in handle_tx_event() for COMP_ISOCH_BUFFER_OVERRUN is not rate-limited, so a single misbehaving isoc device can generate thousands of lines per second.

Locally I changed that xhci_warn() to dev_warn_ratelimited() and the rate dropped to ~100/min with no functional change.
Would a patch rate-limiting this message (or demoting it to xhci_dbg) be acceptable?

Not Mathias, but I think it should be xhci_dbg() like all the other conditions which are device-triggerable rather than driver bugs. People can still see this as -EOVERFLOW status in iso_frame_desc[].

Quoted 7 lines
drivers/usb/host/xhci-ring.c +0 -0

@@ handle_tx_event()

drivers/usb/host/xhci-ring.c +0 -2
 	case COMP_ISOCH_BUFFER_OVERRUN:
-		xhci_warn(xhci,
-			  "WARN: buffer overrun event for slot %u ep %u on

endpoint",

When submitting, please make sure to avoid line breaks like this.

Regards,
Michal