Re: BUG messages with real-time kernel 2.6.25.8-rt7

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 1 Jul 2008, Jürgen Mell wrote:

>
> I am getting this bug message from a real-time kernel 2.6.25.8-rt7:
>
> Jul  1 07:58:30 cwc-003 kernel: BUG: sleeping function called from invalid
> context IRQ-15(416) at arch/x86/mm/highmem_32.c:8
> Jul  1 07:58:30 cwc-003 kernel: in_atomic():0 [00000000], irqs_disabled():1
> Jul  1 07:58:30 cwc-003 kernel: Pid: 416, comm: IRQ-15 Not tainted
> 2.6.25.8-rt7-hws #2
> Jul  1 07:58:30 cwc-003 kernel:  [<c01125f5>] kmap+0x44/0x54
> Jul  1 07:58:30 cwc-003 kernel:  [<f88507fd>] ata_hsm_move+0x3d5/0x666

[...]

>
> Any ideas how to fix this? If you need any more information or if I can
> test something, please let me know.

I found the problem. In PREEMPT_RT we define kmap_atomic to kmap since we
don't disable interrutps with spin_lock_irqsave nor do we disable
preemption.

In drivers/ata/libata-core.c we have:


		/* FIXME: use bounce buffer */
		local_irq_save(flags);
		buf = kmap_atomic(page, KM_IRQ0);


But PREEMPT_RT even kmap_atomic will call "might_sleep". So the fix here
is for us to get rid of the local_irq_save. Perhaps simply add a
spin_lock_irqsave(local_lock) here.

Thanks for the report.

-- Steve