Messages in this thread | | | Subject | Re: [PATCH 1/3] slqb: Do not use DEFINE_PER_CPU for per-node data | From | Pekka Enberg <> | Date | Mon, 21 Sep 2009 09:24:00 +0300 |
| |
Hi Tejun,
On Mon, 2009-09-21 at 00:55 +0900, Tejun Heo wrote: > Pekka Enberg wrote: > > Tejun Heo wrote: > >> Pekka Enberg wrote: > >>> On Fri, Sep 18, 2009 at 10:34 PM, Mel Gorman <mel@csn.ul.ie> wrote: > >>>> SLQB used a seemingly nice hack to allocate per-node data for the > >>>> statically > >>>> initialised caches. Unfortunately, due to some unknown per-cpu > >>>> optimisation, these regions are being reused by something else as the > >>>> per-node data is getting randomly scrambled. This patch fixes the > >>>> problem but it's not fully understood *why* it fixes the problem at the > >>>> moment. > >>> Ouch, that sounds bad. I guess it's architecture specific bug as x86 > >>> works ok? Lets CC Tejun. > >> > >> Is the corruption being seen on ppc or s390? > > > > On ppc. > > Can you please post full dmesg showing the corruption? Also, if you > apply the attached patch, does the added BUG_ON() trigger?
I don't have the affected machines, Sachin and Mel do.
Pekka
> diff --git a/include/linux/percpu.h b/include/linux/percpu.h > index 878836c..fb690d2 100644 > --- a/include/linux/percpu.h > +++ b/include/linux/percpu.h > @@ -127,7 +127,7 @@ extern int __init pcpu_page_first_chunk(size_t reserved_size, > * dynamically allocated. Non-atomic access to the current CPU's > * version should probably be combined with get_cpu()/put_cpu(). > */ > -#define per_cpu_ptr(ptr, cpu) SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))) > +#define per_cpu_ptr(ptr, cpu) ({ BUG_ON(!(ptr)); SHIFT_PERCPU_PTR((ptr), per_cpu_offset((cpu))); }) > > extern void *__alloc_reserved_percpu(size_t size, size_t align);
| |