Re: BUG: using __this_cpu_write() in preemptible [00000000] code: systemd-udevd/497

[Luis Henriques]

Christoph Lameter <cl@xxxxxxxxx> writes:

> On Tue, 15 Apr 2014, Andrew Morton wrote:
>
>> On Tue, 15 Apr 2014 00:55:50 +0100 Luis Henriques <luis.henriques@xxxxxxxxxxxxx> wrote:
>>
>> > (Cc'ing both lis3lv02d and ACPI maintainers)
>> >
>> > Since commit 188a81409ff7de1c5aae947a96356ddd8ff4aaa3 ("percpu: add
>> > preemption checks to __this_cpu ops") I've been seeing the following:
>> >
>> > [   10.485588] hp_accel: hardware type HPB64xx found
>> > [   10.485772] BUG: using __this_cpu_write() in preemptible [00000000] code: systemd-udevd/497
>> > [   10.485777] caller is __this_cpu_preempt_check+0x13/0x20
>> > [   10.485781] CPU: 3 PID: 497 Comm: systemd-udevd Tainted: G        W     3.15.0-rc1 #9
>> > [   10.485783] Hardware name: Hewlett-Packard HP EliteBook 8470p/179B, BIOS 68ICF Ver. F.02 04/27/2012
>> > [   10.485785]  ffffffff81a14db5 ffff88022c80b8e0 ffffffff81604ba4 0000000000000003
>> > [   10.485789]  ffff88022c80b908 ffffffff81313431 0000000000000000 0000000000000032
>> > [   10.485793]  00000000000003e8 ffff88022c80b918 ffffffff81313473 ffff88022c80b928
>> > [   10.485796] Call Trace:
>> > [   10.485802]  [<ffffffff81604ba4>] dump_stack+0x4e/0x7a
>> > [   10.485805]  [<ffffffff81313431>] check_preemption_disabled+0xe1/0xf0
>> > [   10.485808]  [<ffffffff81313473>] __this_cpu_preempt_check+0x13/0x20
>> > [   10.485813]  [<ffffffff810e4eb8>] touch_nmi_watchdog+0x28/0x40
>>
>> Presumably touch_softlockup_watchdog() being called with preemption
>> enabled.  Which is a legitimate thing to do and there's no point in
>> disabling preemption just to squish a runtime warning.
>>
>> Christoph, this thing has iirc caught a couple of very minor bugs but
>> it is being quite a pain in the rear.   I'm inclined to revert?
>
> Well this preemption check functionality was strongly desired by Peter
> Zilkstra and Ingo Molnar and made a precondition for more extensive use of
> the this_cpu operations. My patches to various subsystems were rejected
> because these operations were deemed unsafe without those checks.
>
> The simple thing to do in these cases is to use switch from __this_cpu_*
> to raw_cpu ops to squish these warnings.

FWIW, I just gave the -mm tree patch
kernel-watchdogc-touch_softlockup_watchdog-use-raw_cpu_write.patch [1]
a try and it looks like it fixes this issue.

[1] http://ozlabs.org/~akpm/mmots/broken-out/kernel-watchdogc-touch_softlockup_watchdog-use-raw_cpu_write.patch

Cheers,
-- 
LuÃs
--
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/