Messages in this thread | | | Date | Fri, 7 May 2010 17:26:40 +0200 | From | Daniel Mack <> | Subject | Re: UBIFS: Oops while rebooting 2.6.34-rc6 |
| |
On Fri, May 07, 2010 at 06:23:46PM +0300, Artem Bityutskiy wrote: > On Fri, 2010-05-07 at 15:16 +0200, Daniel Mack wrote: > > Hi, > > > > We've had a kernel Oops today when rebooting an ARM PXA based machine > > while file I/O via SSH was outstanding. > > > > Daniel > > > > # reboot > > # [ 671.190085] UBIFS: un-mount UBI device 0, volume 1 > > The system is going down NOW! > > Sent SIGTERM to all processes > > [ 672.083833] Unable to handle kernel NULL pointer dereference at virtual address 000000ac > > [ 672.094587] pgd = c0004000 > > [ 672.097301] [000000ac] *pgd=00000000 > > [ 672.100850] Internal error: Oops: 817 [#1] > > [ 672.104919] last sysfs file: /sys/devices/platform/spi_gpio.0/spi0.2/value > > It's Firday, and I want to go home, so here is another quick idea for > you where to dig. > > When the system reboots it re-mounts the FS to RO mode, usually. And > there is some emergency remount business (see do_emergency_remount()), > which will re-mount the FS even if there are files opened for writing. > > So, if there is a UBIFS or VFS bug, and somehow one process is in > make_reservation() and is about to write something, and another process > managed to re-mount the FS to R/O mode, then we may ooops, because UBIFS > frees these 'wbuf' objects when it is mounted to R/O (see > ubifs_remount_ro()). > > So, inject printks to ubifs_remount_ro() to check this theory. > > Have a nice weekend and bughunting!
Thanks for your feedback - I'll give that a try next week.
Have a good weekend :)
Daniel
| |