RE: smp_call_function/flush_tlb_all hang on large memory system

[Zwane Mwaikambo]

On Tue, 23 Nov 2004, Deepak Kumar Gupta, Noida wrote:

> Hi Robin 
> 
> The output of CPU is 
> 
> CPU  A: 0x02:   Kernel: CPU busy
>         0x03:   Kernel: CPU busy
> CPU  C: 0x03:   Kernel: CPU busy
> 
> well regarding filing the issue.. i haven't yet contactated support
> persons.. send the mail to just know whether there is already a solution
> available or not.. 
> 
> If you are interested in stack trace.. then it is as follows:-
> 
> [0]kdb> bt
> Stack traceback for pid 7
> 0xe00000307b818000        7        1  1    0   R  0xe00000307b8185a0 *kswapd
> 0xe00000000444b120 smp_call_function+0x5e0
>         args (0xe000000005033698, 0xe000000005033698, 0x1,
> 0xa000000000008000, 0x1)
>         kernel .text 0xe000000004400000 0xe00000000444ab40
> 0xe00000000444b160
> 0xe00000000444a330 smp_flush_tlb_all+0x30
>         args (0xe0000000044545a0, 0x288)
>         kernel .text 0xe000000004400000 0xe00000000444a300
> 0xe00000000444a360
> 0xe0000000044545a0 flush_tlb_range+0x40
>         args (0xe00000307a5b64c8, 0x2000000002128000, 0x200000000212c000,
> 0xe000000004559880, 0x58e)
>         kernel .text 0xe000000004400000 0xe000000004454560
> 0xe000000004454700
> 0xe000000004559880 try_to_swap_out+0x320
>         args (0xe00000307a5b64c8, 0xe00000303b910468, 0x27be00,
> 0xe000003045638250, 0xa0007fffffe20300)
>         kernel .text 0xe000000004400000 0xe000000004559560

This function holds mm->page_table_lock which is acquired with interrupts 
disabled. As a result there is a window for deadlock when you descend into 
smp_call_function. I suggest you run fast from crusty kernels ;)

-
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/