Skip to contentSkip to conversation

ehci calling put_device from irq handler

DISCUSSION Greg KH 3 participants 4 messages Jan 18, 2006

lore
Share & tools
lore

Greg KH

author
#0 lore raw

We can not call put_device() from irq context :(

I added a "might_sleep()" to the driver core and get the following from the ehci driver. Any thoughts?

thanks,

greg k-h

Debug: sleeping function called from invalid context at drivers/base/core.c:343 in_atomic():1, irqs_disabled():0

 [<c012006d>] __might_sleep+0x9e/0xa6
 [<c0270400>] put_device+0x1c/0x35
 [<f883ebde>] usb_hcd_giveback_urb+0x23/0x84 [usbcore]
 [<f8820f40>] ehci_urb_done+0x8d/0xcf [ehci_hcd]
 [<f8821138>] qh_completions+0x1b6/0x359 [ehci_hcd]
 [<f882208d>] scan_async+0x8a/0x13f [ehci_hcd]
 [<f8824e77>] ehci_work+0x59/0xd2 [ehci_hcd]
 [<f88255d7>] ehci_irq+0x185/0x336 [ehci_hcd]
 [<c0106a10>] do_gettimeofday+0x1e/0xbf
 [<c0127ae5>] getnstimeofday+0x14/0x2a
 [<c0139f6c>] ktime_get_ts+0x5e/0x66
 [<c0139ea3>] ktime_get+0x1b/0x43
 [<c013a521>] hrtimer_run_queues+0x50/0xf9
 [<f883ec6f>] usb_hcd_irq+0x30/0x68 [usbcore]
 [<c0144b18>] handle_IRQ_event+0x39/0x6d
 [<c0144bd4>] __do_IRQ+0x88/0xfc
 [<c0105789>] do_IRQ+0x19/0x24
 [<c0103b82>] common_interrupt+0x1a/0x20
 [<c01012a5>] mwait_idle+0x2a/0x34
 [<c0101103>] cpu_idle+0x7f/0xb4
 [<c041e898>] start_kernel+0x166/0x17f
 [<c041e2b6>] unknown_bootoption+0x0/0x1e0
#1 lore raw
Quoted 4 lines · Greg KH

We can not call put_device() from irq context :(

I added a "might_sleep()" to the driver core and get the following from the ehci driver. Any thoughts?

In principle the put_device and corresponding get_device calls aren't needed. We don't release a usb_device structure until after disabling all its endpoints, and disabling an endpoint will wait until all the URBs for that endpoint have completed. So there's no reason to keep a reference to the device structure for each URB.

I see that uhci-hcd is guilty of the same thing (reference acquired for each QH, released while holding a spinlock). Probably each of the host controller drivers is.

Alan Stern

author
#2 lore raw
Quoted 13 lines · Alan Stern
Quoted 4 lines · Greg KH

We can not call put_device() from irq context :(

I added a "might_sleep()" to the driver core and get the following from the ehci driver. Any thoughts?

In principle the put_device and corresponding get_device calls aren't needed. We don't release a usb_device structure until after disabling all its endpoints, and disabling an endpoint will wait until all the URBs for that endpoint have completed. So there's no reason to keep a reference to the device structure for each URB.

Quoted 3 lines

I see that uhci-hcd is guilty of the same thing (reference acquired for each QH, released while holding a spinlock). Probably each of the host controller drivers is.

Great, care to make up a patch to fix this?

:)

thanks,

greg k-h

#3 lore raw
In reply to Alan Stern
Quoted 13 lines · Alan Stern
Quoted 4 lines · Greg KH

We can not call put_device() from irq context :(

I added a "might_sleep()" to the driver core and get the following from the ehci driver. Any thoughts?

In principle the put_device and corresponding get_device calls aren't needed. We don't release a usb_device structure until after disabling all its endpoints, and disabling an endpoint will wait until all the URBs for that endpoint have completed. So there's no reason to keep a reference to the device structure for each URB.

Quoted 3 lines

I see that uhci-hcd is guilty of the same thing (reference acquired for each QH, released while holding a spinlock). Probably each of the host controller drivers is.

I think it dated from the days before we had any mechanism for disabling endpoints properly, and was (important!) insurance against the inevitable "driver disconnect() didn't work right" class of bugs.

So it should be safe to remove those calls now.

- Dave