Re: BUG: using __this_cpu_read() in preemptible [00000000] code: mm_percpu_wq/7

[Tejun Heo]

Hello,

On Wed, Aug 16, 2017 at 11:13:07AM +0200, Heiko Carstens wrote:
> [ 5968.010352] WARNING: CPU: 54 PID: 7 at kernel/workqueue.c:2041 process_one_work+0x6d4/0x718
> 
> (I don't remember we have seen the warning above in the first report) and then
> 
> [ 5968.010913] Kernel panic - not syncing: preempt check
> [ 5968.010919] CPU: 54 PID: 7 Comm: mm_percpu_wq Tainted: G        W       4.13.0-rc4-dirty #3
> [ 5968.010923] Hardware name: IBM 3906 M03 703 (z/VM 6.4.0)
> [ 5968.010927] Workqueue: mm_percpu_wq vmstat_update
> [ 5968.010933] Call Trace:
> [ 5968.010937] ([<0000000000113fbe>] show_stack+0x8e/0xe0)
> [ 5968.010942]  [<0000000000a514be>] dump_stack+0x96/0xd8
> [ 5968.010947]  [<000000000014302a>] panic+0x102/0x248
> [ 5968.010952]  [<00000000007836d8>] check_preemption_disabled+0xf8/0x110
> [ 5968.010956]  [<00000000002ee8e2>] refresh_cpu_vm_stats+0x1b2/0x400
> [ 5968.010961]  [<00000000002ef8be>] vmstat_update+0x2e/0x98
> [ 5968.010965]  [<0000000000166374>] process_one_work+0x3d4/0x718
> [ 5968.010970]  [<000000000016708c>] rescuer_thread+0x214/0x390
> [ 5968.010974]  [<000000000016edbc>] kthread+0x16c/0x180
> [ 5968.010978]  [<0000000000a7273a>] kernel_thread_starter+0x6/0xc
> [ 5968.010983]  [<0000000000a72734>] kernel_thread_starter+0x0/0xc
> 
> On cpu 54 we have mm_percpu_wq with:
> 
>     nr_cpus_allowed = 0x1,
>     cpus_allowed = {
> 	      bits = {0x4, 0x0, 0x0, 0x0}
>     },
> 
> We also have CONFIG_NR_CPUS=256, so the above translates to cpu 3, which
> obviously is not cpu 54 and explains the preempt check warning.

Looks like the same issue Paul was hitting.

 http://lkml.kernel.org/r/1501541603-4456-3-git-send-email-paulmck@xxxxxxxxxxxxxxxxxx

Can you see whether the above patch helps?

Thank.s

-- 
tejun