Re: [PATCH] mm/swapfile.c: use spin_lock_bh with swap_lock to avoid deadlocks

[Michal Hocko]

On Wed 04-02-15 01:48:23, Pasi Sjöholm wrote:
> On 03.02.2015 15:14, Michal Hocko wrote:
> >> It is possible to get kernel in deadlock-state if swap_lock is not locked
> >> with spin_lock_bh by calling si_swapinfo() simultaneously through
> >> timer_function and registered vm shinker callback-function.
> >>
> >> BUG: spinlock recursion on CPU#0, main/2447
> >> lock: swap_lock+0x0/0x10, .magic: dead4ead, .owner: main/2447, .owner_cpu: 0
> >> [<c010b938>] (unwind_backtrace+0x0/0x11c) from [<c03e9be0>] (do_raw_spin_lock+0x48/0x154)
> >> [<c03e9be0>] (do_raw_spin_lock+0x48/0x154) from [<c0226e10>] (si_swapinfo+0x10/0x90)
> >> [<c0226e10>] (si_swapinfo+0x10/0x90) from [<c04d7e18>] (timer_function+0x24/0x258)
> > Who is calling si_swapinfo from timer_function? AFAICS the vanilla
> > kernel doesn't do that. Or am I missing something?
> 
> Nothing in vanilla kernel, but "memnotify"
> (https://lkml.org/lkml/2012/1/17/182) together with modified
> lowmemorykiller (drivers/staging/android/lowmemorykiller.c) which takes
> in account also the available swap (calling si_swapinfo as well) will
> cause the deadlock.
> 
> Memnotify uses timer (with backoff) for checking the memory pressure
> which can be then used to let the processes itself adjust their memory
> pressure before getting killed by the modified lowmemorykiller.

We are not usually changing the core kernel for an out of tree
functionality. So NAK to this patch.
-- 
Michal Hocko
SUSE Labs
--
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/