Fwd: [PATCH v2 6/6] slub: only preallocate cpus_with_slabs if offstack

[Gilad Ben-Yossef]

[Re-sending to the list due to some foul up with LKML address on my part]


On Mon, Oct 24, 2011 at 7:19 AM, Andi Kleen <andi@xxxxxxxxxxxxxx> wrote:
>
> Gilad Ben-Yossef <gilad@xxxxxxxxxxxxx> writes:
>
> > We need a cpumask to track cpus with per cpu cache pages
> > to know which cpu to whack during flush_all. For
> > CONFIG_CPUMASK_OFFSTACK=n we allocate the mask on stack.
> > For CONFIG_CPUMASK_OFFSTACK=y we don't want to call kmalloc
> > on the flush_all path, so we preallocate per kmem_cache
> > on cache creation and use it in flush_all.
>
> What's the problem with calling kmalloc in flush_all?
> That's a slow path anyways, isn't it?
>
> I believe the IPI functions usually allocate anyways.
>
> So maybe you can do that much simpler.

That was what the first version of the patch did (use
alloc_cpumask_var in flush_all).

Pekka Enberg pointed out that calling kmalloc on the kmem_cache
shrinking code path is not a good idea
and it does sound like a deadlock waiting to happen.

Gilad

--
Gilad Ben-Yossef
Chief Coffee Drinker
gilad@xxxxxxxxxxxxx
Israel Cell: +972-52-8260388
US Cell: +1-973-8260388
http://benyossef.com

"I've seen things you people wouldn't believe. Goto statements used to
implement co-routines. I watched C structures being stored in
registers. All those moments will be lost in time... like tears in
rain... Time to die. "



-- 
Gilad Ben-Yossef
Chief Coffee Drinker
gilad@xxxxxxxxxxxxx
Israel Cell: +972-52-8260388
US Cell: +1-973-8260388
http://benyossef.com

"I've seen things you people wouldn't believe. Goto statements used to
implement co-routines. I watched C structures being stored in
registers. All those moments will be lost in time... like tears in
rain... Time to die. "
--
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/