Skip to contentSkip to conversation

for USB guys - strange things in dmesg

DISCUSSION Pete Zaitcev 2 participants 3 messages Jan 8, 2005 – Jan 11, 2005 1 message not in the archive USB

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw
Quoted 6 lines · Michal Semler

when inserting my Bluetooth dongle into USB port, I get into dmesg this:
Bluetooth: RFCOMM ver 1.4
Bluetooth: RFCOMM socket layer initialized
Bluetooth: RFCOMM TTY layer initialized drivers/usb/input/hid-core.c: input irq status -84 received

Quoted 8 lines

[.... LONG flood of the same messages ....]

drivers/usb/input/hid-core.c: input irq status -84 received
usb 3-1: USB disconnect, address 3
drivers/usb/input/hid-core.c: input irq status -84 received drivers/usb/input/hid-core.c: can't resubmit intr, 0000:00:1d.2-1/input1, status -19 usb 3-1: new full speed USB device using uhci_hcd and address 4
Bluetooth: HCI USB driver ver 2.7

What happens here is that the device disconnects itself during or after it's initialized.

Once the HC hardware detects the disconnect, future URBs will end with -84 error. However, the HID driver does not do anything about it. It continues to attempt to resubmit until the khubd does its processing and enters its disconnect method. In extreme cases, it is possible to have this submit-and-error-and-repeat loop to monopolize the CPU and prevent khubd from working ever, thus effectively locking up the box. Fortunately, in 2.6 kernel we standardized error codes, and thus drivers like hid can rely on -84 meaning a disconnect and not something else. In such case, hid has to stop resubmitting before its disconnect method is executed.

This is relevant to all drivers which submit interrupt URBs. One driver which does it correctly is mct_u232 (surprisingly enough), so the code can be taken from there.

-- Pete

#2 lore raw
Quoted 20 lines · Pete Zaitcev
Quoted 6 lines · Michal Semler

when inserting my Bluetooth dongle into USB port, I get into dmesg this:
Bluetooth: RFCOMM ver 1.4
Bluetooth: RFCOMM socket layer initialized
Bluetooth: RFCOMM TTY layer initialized drivers/usb/input/hid-core.c: input irq status -84 received

Quoted 8 lines

[.... LONG flood of the same messages ....]

drivers/usb/input/hid-core.c: input irq status -84 received
usb 3-1: USB disconnect, address 3
drivers/usb/input/hid-core.c: input irq status -84 received drivers/usb/input/hid-core.c: can't resubmit intr, 0000:00:1d.2-1/input1, status -19 usb 3-1: new full speed USB device using uhci_hcd and address 4
Bluetooth: HCI USB driver ver 2.7

What happens here is that the device disconnects itself during or after it's initialized.

Quoted 10 lines

Once the HC hardware detects the disconnect, future URBs will end with -84 error. However, the HID driver does not do anything about it. It continues to attempt to resubmit until the khubd does its processing and enters its disconnect method. In extreme cases, it is possible to have this submit-and-error-and-repeat loop to monopolize the CPU and prevent khubd from working ever, thus effectively locking up the box. Fortunately, in 2.6 kernel we standardized error codes, and thus drivers like hid can rely on -84 meaning a disconnect and not something else. In such case, hid has to stop resubmitting before its disconnect method is executed.

-84 is, according to documentation, -EILSEQ, and means "a CRC error", that shouldn't normally happen and "indicates hardware problems such as bad devices (including firmware) or cables".

I'm getting -84 on disconnect now, too, which suggests that uhci-hcd might be using incorrect error code here.

Looking at uhci-hcd, it reports -EILSEQ in case there is a timeout on the bus talking to the device. This can happen due to a disconnect, but the HC should notice the disconnect I believe and return the right error code.

Disconnects (-ESHUTDOWN) are already handled by hid-core.c

Quoted 3 lines

This is relevant to all drivers which submit interrupt URBs. One driver which does it correctly is mct_u232 (surprisingly enough), so the code can be taken from there.

Signature · 2 lines
Vojtech Pavlik
SuSE Labs, SuSE CR
#3 lore raw
In reply to Pete Zaitcev
Quoted 20 lines · Pete Zaitcev
Quoted 6 lines · Michal Semler

when inserting my Bluetooth dongle into USB port, I get into dmesg this:
Bluetooth: RFCOMM ver 1.4
Bluetooth: RFCOMM socket layer initialized
Bluetooth: RFCOMM TTY layer initialized drivers/usb/input/hid-core.c: input irq status -84 received

Quoted 8 lines

[.... LONG flood of the same messages ....]

drivers/usb/input/hid-core.c: input irq status -84 received
usb 3-1: USB disconnect, address 3
drivers/usb/input/hid-core.c: input irq status -84 received drivers/usb/input/hid-core.c: can't resubmit intr, 0000:00:1d.2-1/input1, status -19 usb 3-1: new full speed USB device using uhci_hcd and address 4
Bluetooth: HCI USB driver ver 2.7

What happens here is that the device disconnects itself during or after it's initialized.

Quoted 10 lines

Once the HC hardware detects the disconnect, future URBs will end with -84 error. However, the HID driver does not do anything about it. It continues to attempt to resubmit until the khubd does its processing and enters its disconnect method. In extreme cases, it is possible to have this submit-and-error-and-repeat loop to monopolize the CPU and prevent khubd from working ever, thus effectively locking up the box. Fortunately, in 2.6 kernel we standardized error codes, and thus drivers like hid can rely on -84 meaning a disconnect and not something else. In such case, hid has to stop resubmitting before its disconnect method is executed.

Can you test this patch? It seems to work for me.

Signature · 2 lines
Vojtech Pavlik
SuSE Labs, SuSE CR

[email protected], 2005-01-11 17:33:17+01:00, [email protected]
input: Handle -EILSEQ return code in the HID driver completion
handlers as unplug.
Flush request queue on unplug, too.

 hid-core.c |   23 ++++++++++++++++++-----
 1 files changed, 18 insertions(+), 5 deletions(-)
drivers/usb/input/hid-core.c +18 -5
diff -Nru a/drivers/usb/input/hid-core.c b/drivers/usb/input/hid-core.c
@@ -925,8 +925,9 @@
 			break;
 		case -ECONNRESET:	/* unlink */
 		case -ENOENT:
-		case -ESHUTDOWN:
 		case -EPERM:
+		case -ESHUTDOWN:	/* unplug */
+		case -EILSEQ:		/* unplug timeout on uhci */
 			return;
 		case -ETIMEDOUT:	/* NAK */
 			break;
@@ -1136,12 +1137,15 @@
 {
 	struct hid_device *hid = urb->context;
 	unsigned long flags;
+	int unplug = 0;
 
 	switch (urb->status) {
 		case 0:			/* success */
+		case -ESHUTDOWN:	/* unplug */
+		case -EILSEQ:		/* unplug timeout on uhci */
+			unplug = 1;
 		case -ECONNRESET:	/* unlink */
 		case -ENOENT:
-		case -ESHUTDOWN:
 			break;
 		default:		/* error */
 			warn("output irq status %d received", urb->status);
@@ -1149,7 +1153,10 @@
 
 	spin_lock_irqsave(&hid->outlock, flags);
 
-	hid->outtail = (hid->outtail + 1) & (HID_OUTPUT_FIFO_SIZE - 1);
+	if (unplug)
+		hid->outtail = hid->outhead;
+	else
+		hid->outtail = (hid->outtail + 1) & (HID_OUTPUT_FIFO_SIZE - 1);
 
 	if (hid->outhead != hid->outtail) {
 		if (hid_submit_out(hid)) {
@@ -1173,6 +1180,7 @@
 {
 	struct hid_device *hid = urb->context;
 	unsigned long flags;
+	int unplug = 0;
 
 	spin_lock_irqsave(&hid->ctrllock, flags);
 
@@ -1180,16 +1188,21 @@
 		case 0:			/* success */
 			if (hid->ctrl[hid->ctrltail].dir == USB_DIR_IN) 
 				hid_input_report(hid->ctrl[hid->ctrltail].report->type, urb, regs);
+		case -ESHUTDOWN:	/* unplug */
+		case -EILSEQ:		/* unplug timectrl on uhci */
+			unplug = 1;
 		case -ECONNRESET:	/* unlink */
 		case -ENOENT:
-		case -ESHUTDOWN:
 		case -EPIPE:		/* report not available */
 			break;
 		default:		/* error */
 			warn("ctrl urb status %d received", urb->status);
 	}
 
-	hid->ctrltail = (hid->ctrltail + 1) & (HID_CONTROL_FIFO_SIZE - 1);
+	if (unplug)
+		hid->ctrltail = hid->ctrlhead;
+	else
+		hid->ctrltail = (hid->ctrltail + 1) & (HID_CONTROL_FIFO_SIZE - 1);
 
 	if (hid->ctrlhead != hid->ctrltail) {
 		if (hid_submit_ctrl(hid)) {