On Wednesday, January 08, 2014 03:41:52 PM Yinghai Lu wrote: > On Mon, Jan 6, 2014 at 12:41 PM, Rafael J. Wysocki <rjw@xxxxxxxxxxxxx> wrote: > > >> Not sure how that could happen. > >> > >> If it would really happen, we could set dev->match_driver to 0 in pci_stop_dev. > > > > Simply, run "modprobe -r driver && modprobe driver" in a loop and > > remove the PCI host bridge the given device is on in parallel to that. Chances > > are, you'll see some nice breakage. > > I would suggest using match_driver prevent driver from attaching again. Yes, we can do that. Some locking is needed for it to be non-racy, however. Anyway, we still have the problem with race conditions between different PCI removal/rescan code paths. And I'm still going to prepare a patch to use the remove-rescan mutex to address those race conditions and that patch should help here too. > > --- > drivers/pci/remove.c | 1 + > 1 file changed, 1 insertion(+) > > Index: linux-2.6/drivers/pci/remove.c > =================================================================== > --- linux-2.6.orig/drivers/pci/remove.c > +++ linux-2.6/drivers/pci/remove.c > @@ -11,6 +11,7 @@ static void pci_stop_dev(struct pci_dev > pci_proc_detach_device(dev); > pci_remove_sysfs_dev_files(dev); > device_release_driver(&dev->dev); > + dev->match_driver = false; > dev->is_added = 0; > } > > > > > > Also what happens if somebody uses the "remove" sysfs attribute on a device > > needed by ioapic/dmar? > > Good question, we will have problem in that case. > To make it simple, we may hide the "remove" in sysfs for ioapic pci device ? Yeah, we need to do that if using that attribute may lead to problems. Thanks, Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
