Re: PATCH -RCU locking on last_VFP_context[cpu] in vfp_notifier [2.6.32]

[Russell King - ARM Linux]

On Mon, Aug 11, 2014 at 03:24:18PM -0700, Sadasivan Shaiju wrote:
> Hi ,
> 
> I  work for Montavista (Cavium Inc) as  a  Technical  Lead .  I want to
> push some  of the kernel  patches to  rt community (2.6.32 kernel 2.6.33
> rt patch)  , so  that  It  will  go  to  the  main line These patches  are
> reviewed  and approved  by  our system Architect.  I request  you to
> include  in the main line .  These  issues  were  reported  during econa
> board bringup at montavista.
> 
> Problem Description:
> Using cge60-econa-cns3420-2.6.32, the kernel failed to boot with the
> following
> error:
> 
> Internal error: Oops: 817 [#1] from cpu 1 PREEMPT SMP
> last sysfs file: /sys/devices/virtual/bdi/0:19/uevent
> Modules linked in: hmac ctr deflate
> CPU: 1    Tainted: G        W   (2.6.32.46.cge #1)
> PC is at vfp_notifier+0x48/0xbc
> LR is at vfp_notifier+0x44/0xbc
> pc : []    lr : []    psr: 60000013
> sp : aeee1d30  ip : aeee1d50  fp : aeee1d4c
> r10: af8d6460  r9 : ffffffff  r8 : af88c000
> r7 : a05ba584  r6 : af88c000  r5 : 00000001  r4 : 40000000
> r3 : 00000000  r2 : 00000000  r1 : 40000000  r0 : aeee0230
> Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment user
> Control: 00c5787d  Table: 2eeec00a  DAC: 00000017
> Process grep (pid: 1710, stack limit = 0xaeee0270)
> Stack:  from cpu 1 (0xaeee1d30 to 0xaeee2000)
> 
> Root Cause:
> On the SMP architecture, last_VFP_context[cpu] becomes NULL because it
> gets
> released on a different CPU.
> 
> How Solved:
> Fixed by exiting the thread instead of releasing the thread in the
> vfp_notifier.
> 
> I  request you  to include  the above patch to  the main line kernel .  If
> any questions  please contact me  at  sshaiju@xxxxxxxxxx
> (shaiju_sada@xxxxxxxxx)

This is totally insufficient for fixing a bug in a complex piece of code.

You fail to explain exactly _how_ the bug arises.  You say
"last_VFP_context[cpu] becomes NULL because it gets released on a
different CPU" - how does that happen?

The only places that last_VFP_context[cpu] is set to NULL is within a
cpu = get_cpu()..put_cpu() region, which by definition *must* be running
on the CPU specified by 'cpu'.

Without a proper diagnosis showing exactly what the race is which causes
the above oops, there's nothing I can do.  Sorry.

-- 
FTTC broadband for 0.8mile line: currently at 9.5Mbps down 400kbps up
according to speedtest.net.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/