Need to apply patch: "ARM: Do not call flush_cache_user_range with mmap_sem held"

[Olof Johansson]

There's a well-known bug on ARM that if you have an SMP system, you can hang a process by doing a cache flush syscall at the same time as a page fault. It's a known issue, with a patch posted that unfortunately breaks pre-ARMv6 systems so it has not yet been applied.

Many other projects, such as Android and Chrome OS, have applied this locally in their tree. The patch is good and valid for ARMv6 and v7 systems since they have the appropriate exception handling tables in the page flush code. Until that's added to v5 and before though, the patch can't be upstreamed.

Please pick this up for Ubuntu on the v7 platforms, Panda in particular. I have a user here that can very reliably hit this when running a specific workload, and I would prefer if he could keep running a distro kernel instead of a locally built one.

Patch is at:

http://marc.info/?l=linux-arm-kernel&m=132068730012063&w=2

Note that the discussion went in a tangent -- the patch is valid for v6 and v7, and some of the misunderstanding in the discussion was because of this.

Thanks!

-Olof

[Adam Conrad (adconrad)]

I'm told via IRC that this is theoretically triggerable (though less frequently) on UP systems, so adding tasks for omap and mx5 kernels as well.

[Brad Figg (brad-figg)]

This bug is missing log files that will aid in diagnosing the problem. From a terminal window please run:

apport-collect 965840

and then change the status of the bug to 'Confirmed'.

If, due to the nature of the issue you have encountered, you are unable to run this command, please add a comment stating that fact and change the bug status to 'Confirmed'.

This change has been made by an automated script, maintained by the Ubuntu Kernel Team.

[Olof Johansson (olof)]

I'm not local on the machine in question, so I can't collect the logs. And I see no way to mark this as Confirmed on my own, I'm guessing I lack permission to do so?

[Olof Johansson (olof)]

Ah, there it was. Horrible UI.

[Andy Whitcroft (apw)]

Ok. I've added the suggested patch to an 'omap' build from precise master:
    http://people.canonical.com/~apw/lp965840-precise/

And to an omap4 build from ti-omap4:
    http://people.canonical.com/~apw/lp965840+ti-omap4-precise/

If those of you with the h/w could at least confirm they boot ok. If you know how to reproduce this then if you could confirm it also fixes things then great. Please report any testing back here. Thanks.

[Derek Schuff (dschuff)]

I tested the omap4 kernel deb on a pandaboard running Oneiric; it boots, and the small reproducer we have seems to work. I've started up our full bot suite on it, will let you know.

[Mark Seaborn (mrs)]

For context, we hit the bug when testing Native Client's dynamic code loading support; the issue tracking this for Native Client is http://code.google.com/p/nativeclient/issues/detail?id=2678.

Attached is a small test case that reproduces the hang.

[Derek Schuff (dschuff)]

Update: As far as I can tell the patched kernel works with respect to this particular bug; I just can't use it because the board seems to lock up (for presumably unrelated reasons) when I leave it running overnight. no output on the serial console even, just stops responding and has to be reset.

[Adam Conrad (adconrad)]

Derek, can you confirm that the lockups are unrelated to this patch by installing the precise 1411 kernel from the archive and make sure it also locks up similarly for you?

[Paolo Pisati (p-pisati)]

commit 4340608237cd7dc7437f7edaeceaccf1672104dc "ARM: 7409/1: Do not call flush_cache_user_range with mmap_sem held"

and

commit 99a98086bb950128e611aefa8d042caf81954f00 "ARM: 7365/1: drop unused parameter from flush_cache_user_range"

are both present in origin/master (and [OP]/omap4 are rebased trees), Natty is EOL and Q got it from upstream (the two patches were upstreamed).

Tester didn't report any problem in 6 months, closing here.

[Rolf Leggewie (r0lf)]

oneiric has seen the end of its life and is no longer receiving any updates. Marking the oneiric task for this ticket as "Won't Fix".

[Steve Langasek (vorlon)]

The Precise Pangolin has reached end of life, so this bug will not be fixed for that release