Messages in this thread | ![/](/images/icornerl.gif) | | Date | Sun, 12 May 2013 17:37:22 +0200 | From | Borislav Petkov <> | Subject | Re: lockdep: BUG: key ffff880436f00330 not in .data! |
| |
On Thu, May 09, 2013 at 10:01:08PM +0200, Borislav Petkov wrote: > On Thu, May 09, 2013 at 11:28:08AM -0300, Mauro Carvalho Chehab wrote: > > Could you please check if ghes_edac also got registered there?
Ok, I got it:
[ 5.430408] BUG: key ffff88043c320330 not in .data! [ 5.431043] ------------[ cut here ]------------ [ 5.431776] WARNING: at kernel/lockdep.c:2987 lockdep_init_map+0x555/0x590() [ 5.432512] DEBUG_LOCKS_WARN_ON(1)
basically means we're using dynamically allocated memory for the lock_key. And the last can thus disappear, which is Not Good(tm).
This happens because the EDAC instance probing functions do edac_mc_alloc() to allocate the mci which also contains embedded in it:
struct bus_type bus;
and the bus_register() call in edac_create_sysfs_mci_device() hands down this struct bus_type to lockdep.
The patch which introduced the bus_register() call in edac_create_sysfs_mci_device() is
-- commit de3910eb79ac8c0f29a11224661c0ebaaf813039 Author: Mauro Carvalho Chehab <mchehab@redhat.com> Date: Tue Apr 24 15:05:43 2012 -0300
edac: change the mem allocation scheme to make Documentation/kobject.txt happy --
which means we need a fix since this affects all edac drivers, not only sb_edac, as initially surmised.
Thanks.
-- Regards/Gruss, Boris.
Sent from a fat crate under my desk. Formatting is fine. --
| ![\](/images/icornerr.gif) |