Re: next BUG: using smp_processor_id() in preemptible

On Fri, 23 Dec 2011, Benjamin Herrenschmidt wrote:

On Thu, 2011-12-22 at 04:07 -0800, Hugh Dickins wrote:

quoted

On Mon, 5 Dec 2011, Hugh Dickins wrote:

quoted

3.2.0-rc3-next-20111202 with CONFIG_DEBUG_PREEMPT=y gives me lots of

Dec  4 20:03:19 thorn kernel: BUG: using smp_processor_id() in preemptible [00000000] code: startpar/1365
Dec  4 20:03:19 thorn kernel: caller is .arch_local_irq_restore+0x44/0x90
Dec  4 20:03:19 thorn kernel: Call Trace:
Dec  4 20:03:19 thorn kernel: [c0000001b45a7c60] [c000000000011fe8] .show_stack+0x6c/0x16c (unreliable)
Dec  4 20:03:19 thorn kernel: [c0000001b45a7d10] [c00000000024318c] .debug_smp_processor_id+0xe4/0x11c
Dec  4 20:03:19 thorn kernel: [c0000001b45a7da0] [c00000000000e2e8] .arch_local_irq_restore+0x44/0x90
Dec  4 20:03:19 thorn kernel: [c0000001b45a7e30] [c000000000005870] .do_hash_page+0x70/0x74
Dec  4 20:03:21 thorn kernel: debug_smp_processor_id: 21950 callbacks suppressed

from the u64 *next_tb = &__get_cpu_var(decrementers_next_tb)
in decrementer_check_overflow(): I've no idea whether it's safe
just to use get_cpu_var then put_cpu_var there instead,
but no hurry, I can survive with DEBUG_PREEMPT off.

Still a problem in 3.2.0-rc6-next-20111222

Ah forgot about that, I'll have a look. Thanks for the reminder.

I'm afraid it's now blighting Linus's tree for 3.3.

Hugh