Skip to contentSkip to conversation

USB deadlock in v2.5.67

Linux-ia64 Applied Greg KH 3 participants 4 messages Apr 18, 2003 USB

lore
Share & tools
lore

Greg KH

author
#0 lore raw
Quoted 2 lines · David Mosberger

So my 2.5.67 kernel was humming along nicely when I got the idea to unplug a USB keyboard. Instant deadlock. The backtrace shows this:

Ah, I just got this too, but you have the better debug trace.

Quoted 19 lines
Call Trace:
 [<e0000000048c4260>] hcd_free_dev+0x140/0x240
 [<e0000000048b96e0>] usb_release_dev+0x100/0x140
 [<e000000004743420>] device_release+0x80/0xa0
 [<e000000004697600>] kobject_cleanup+0x100/0x120
 [<e0000000048c34d0>] urb_unlink+0x110/0x1a0
 [<e0000000048c4390>] usb_hcd_giveback_urb+0x30/0x1a0
 [<e0000000048da730>] dl_done_list+0x230/0x2a0
 [<e0000000048dbfd0>] ohci_irq+0x290/0x340
 [<e0000000048c4580>] usb_hcd_irq+0x80/0x100
 [<e000000004414e00>] handle_IRQ_event+0xa0/0x120
 [<e0000000044155e0>] do_IRQ+0x360/0x460
 [<e0000000044174b0>] ia64_handle_irq+0x70/0x140
 [<e000000004411e40>] ia64_leave_kernel+0x0/0x240

<hcd_free_dev+0x140> translates into line 1249 in hcd.c, where it does:

Quoted 6 lines
	spin_lock_irqsave (&hcd_data_lock, flags);

The deadlock is pretty obvious: the same lock has already been acquired urb_unlink(), 4 levels up in the call-chain.

Anybody have a fix for this?

David (Brownell, that is), does this help with the trace I sent you a few hours ago?

thanks,

greg k-h

#1 lore raw
Quoted 7 lines · Greg KH
Quoted 5 lines · David Mosberger

<hcd_free_dev+0x140> translates into line 1249 in hcd.c, where it does:

	spin_lock_irqsave (&hcd_data_lock, flags);
Quoted 8 lines
Quoted 4 lines

The deadlock is pretty obvious: the same lock has already been acquired urb_unlink(), 4 levels up in the call-chain.

Anybody have a fix for this?

David (Brownell, that is), does this help with the trace I sent you a few hours ago?

Seems to be a different problem. The patch below should resolve the keyboard problem -- just reorders two lines so the lock isn't held after the device's records get deleted, so the order is what it should always have been.

- Dave

drivers/usb/core/hcd.c +1 -1
@@ -961,8 +961,8 @@
 	spin_lock_irqsave (&hcd_data_lock, flags);
 	list_del_init (&urb->urb_list);
 	dev = urb->dev;
-	usb_put_dev (dev);
 	spin_unlock_irqrestore (&hcd_data_lock, flags);
+	usb_put_dev (dev);
 }
 
 
#2 lore raw
In reply to Greg KH
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · David Brownell

David> Seems to be a different problem. The patch below should David> resolve the keyboard problem -- just reorders two lines so David> the lock isn't held after the device's records get deleted, David> so the order is what it should always have been.

With the patch applied, I can now add/remove USB keyboards without crashing the kernel.

Thanks!

	--david
author Applied
#3 lore raw
In reply to Greg KH
Quoted 14 lines · David Brownell
Quoted 13 lines · Greg KH
Quoted 9 lines · David Mosberger

<hcd_free_dev+0x140> translates into line 1249 in hcd.c, where it does:

	spin_lock_irqsave (&hcd_data_lock, flags);

The deadlock is pretty obvious: the same lock has already been acquired urb_unlink(), 4 levels up in the call-chain.

Anybody have a fix for this?

Quoted 7 lines
Quoted 2 lines

David (Brownell, that is), does this help with the trace I sent you a few hours ago?

Seems to be a different problem. The patch below should resolve the keyboard problem -- just reorders two lines so the lock isn't held after the device's records get deleted, so the order is what it should always have been.

Applied to my trees, thanks.

greg k-h