On Wednesday 10 October 2012 14:56:04 Jiri Kosina wrote: > On Wed, 10 Oct 2012, Oliver Neukum wrote: > > > For mice with card readers the HID driver can deadlock as its > > post_reset() method allocates memory. GFP_NOIO must be used in a block > > layer error handler, which usbhid can be indirectly part of. > > Hi Oliver, > > umm ... could you please elaborate how this can actually happen? > > I guess the USB mouse has a hub inside, and the card reader is connected > to that HUB, right? (or perhaps different USB interface?). > > In any case, how does HID get bound to the card reader interface? What am > I missing? Hi, HID isn't bound to the to the card reader interface. However if the card reader requests a reset, all the interfaces are reset. Before the reset is done pre_reset() is called for all interfaces and before usb_reset_device() returns post_reset() must run for all interfaces. The following sequence happens: SCSI error -> bus_reset() -> usb storage: usb_reset_device() -> reset going on -> usbhid: post_reset() -> kmalloc(..., GFP_KERNEL) -> page is paged out -> sd waits for reset to finish --> DEADLOCK So this can happen only if there is no hub. But Alan Stern reported having such a device. I audited a patch from Ming Lei which introduces the same problem for usbnet. So I checked usbhid. You need to be unlucky, but it may happen, so here's the patch. Regards Oliver -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
