[PATCH] CPEI gets warning at kernel/irq/migration.c:27/move_masked_irq()
authorHidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Fri, 8 Dec 2006 10:35:58 +0000 (02:35 -0800)
committerLinus Torvalds <torvalds@woody.osdl.org>
Fri, 8 Dec 2006 16:28:37 +0000 (08:28 -0800)
While running my MCA test (hardware error injection) on 2.6.19,
I got some warning like following:

> BUG: warning at kernel/irq/migration.c:27/move_masked_irq()
>
> Call Trace:
>  [<a000000100013d20>] show_stack+0x40/0xa0
>                                 sp=e00000006b2578d0 bsp=e00000006b2510b0
>  [<a000000100013db0>] dump_stack+0x30/0x60
>                                 sp=e00000006b257aa0 bsp=e00000006b251098
>  [<a0000001000de430>] move_masked_irq+0xb0/0x240
>                                 sp=e00000006b257aa0 bsp=e00000006b251070
>  [<a0000001000de6a0>] move_native_irq+0xe0/0x180
>                                 sp=e00000006b257aa0 bsp=e00000006b251040
>  [<a00000010004ff50>] iosapic_end_level_irq+0x30/0xe0
>                                 sp=e00000006b257aa0 bsp=e00000006b251020
>  [<a0000001000d94d0>] __do_IRQ+0x170/0x400
>                                 sp=e00000006b257aa0 bsp=e00000006b250fd8
>  [<a0000001000116f0>] ia64_handle_irq+0x1b0/0x260
>                                 sp=e00000006b257aa0 bsp=e00000006b250fa8
>  [<a00000010000c3a0>] ia64_leave_kernel+0x0/0x280
>                                 sp=e00000006b257aa0 bsp=e00000006b250fa8
>  [<a000000100690cf0>] _spin_unlock_irqrestore+0x30/0x60
>                                 sp=e00000006b257c70 bsp=e00000006b250f90

It comes from:

[kernel/irq/migration.c]
  26         if (CHECK_IRQ_PER_CPU(desc->status)) {
  27                 WARN_ON(1);
  28                 return;
  29         }

By putting some printk in kernel, I found that irqbalance is trying to
move CPEI which is handled as PER_CPU irq. That's why.

CPEI(Corrected Platform Error Interrupt) is ia64 specific irq, is
allowed to pin to particular processor which selected by the platform, and
even it is PER_CPU but it has set_affinity handler (=iosapic_set_affinity)
as same as other IO-SAPIC-level interrupts. (I don't know why, but
I guess that there would be typical situation where the handler for
migration is needed, such as hotplug - the processor going to be
offline/hot-removed.)

To shut up this warning, there are 2 way at least:
 a) fix CPEI stuff
 b) prohibit setting affinity to PER_CPU irq

I'm not sure what stuff of CPEI need to be fixed, but I think that
returning error to attempting move PER_CPU irq is useful for all
applications since it will never work.

Following small patch takes b) style.
It works, the warning disappeared and irqbalance still runs well.

Signed-off-by: Hidetoshi Seto <seto.hidetoshi@jp.fujitsu.com>
Cc: Arjan van de Ven <arjan@infradead.org>
Acked-by: Ingo Molnar <mingo@elte.hu>
Acked-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
kernel/irq/proc.c

index 9a352667007ce52fe610c7c72cc190772cb126dd..61f5c717a8f579cbe5dd3c9b3991b7ef2fb5cead 100644 (file)
@@ -54,7 +54,8 @@ static int irq_affinity_write_proc(struct file *file, const char __user *buffer,
        unsigned int irq = (int)(long)data, full_count = count, err;
        cpumask_t new_value, tmp;
 
-       if (!irq_desc[irq].chip->set_affinity || no_irq_affinity)
+       if (!irq_desc[irq].chip->set_affinity || no_irq_affinity ||
+                               CHECK_IRQ_PER_CPU(irq_desc[irq].status))
                return -EIO;
 
        err = cpumask_parse_user(buffer, count, new_value);