Re: [PATCH 2/4] frontswap: using vzalloc instead of vmalloc

[Bob Liu <lliubbo@gmail.com>]

On Fri, Aug 5, 2011 at 12:47 AM, Dan Magenheimer
<dan.magenheimer@oracle.com> wrote:
>> From: Michal Hocko [mailto:mhocko@suse.cz]
>> Sent: Thursday, August 04, 2011 3:00 AM
>> Subject: Re: [PATCH 2/4] frontswap: using vzalloc instead of vmalloc
>>
>> On Thu 04-08-11 10:14:07, Johannes Weiner wrote:
>> > On Thu, Aug 04, 2011 at 09:57:30AM +0200, Michal Hocko wrote:
>> > > On Thu 04-08-11 11:09:48, Bob Liu wrote:
>> > > > This patch also add checking whether alloc frontswap_map memory
>> > > > failed.
>> > > >
>> > > > Signed-off-by: Bob Liu <lliubbo@gmail.com>
>> > > > ---
>> > > >  mm/swapfile.c |    6 +++---
>> > > >  1 files changed, 3 insertions(+), 3 deletions(-)
>> > > >
>> > > > diff --git a/mm/swapfile.c b/mm/swapfile.c
>> > > > index ffdd06a..8fe9e88 100644
>> > > > --- a/mm/swapfile.c
>> > > > +++ b/mm/swapfile.c
>> > > > @@ -2124,9 +2124,9 @@ SYSCALL_DEFINE2(swapon, const char __user *, specialfile, int, swap_flags)
>> > > >         }
>> > > >         /* frontswap enabled? set up bit-per-page map for frontswap */
>> > > >         if (frontswap_enabled) {
>> > > > -               frontswap_map = vmalloc(maxpages / sizeof(long));
>> > > > -               if (frontswap_map)
>> > > > -                       memset(frontswap_map, 0, maxpages / sizeof(long));
>> > > > +               frontswap_map = vzalloc(maxpages / sizeof(long));
>> > > > +               if (!frontswap_map)
>> > > > +                       goto bad_swap;
>> > >
>> > > vzalloc part looks good but shouldn't we disable frontswap rather than
>> > > fail?
>> >
>> > Silently dropping explicitely enabled features is not a good idea,
>> > IMO.
>>
>> Sure, I didn't mean silently. It should be a big fat warning that there
>> is not enough memory to enable the feature.
>>
>> > But from a quick look, this seems to be actually happening as
>> > frontswap's bitmap tests check for whether there is even a bitmap
>> > allocated and it should essentially never do anything for real if
>> > there isn't.
>>
>> Yes, that was my impression as well. I wasn't 100% sure about that
>> though, because there are many places which check frontswap_enabled and
>> do not check the map. I though that disabling the feature should be
>> safer.
>>
>> > How about printing a warning as to why the swapon fails and give the
>> > admin a choice to disable it on her own?
>>
>> I am not that familiar with the code but drivers/staging/zcache/zcache.c
>> says:
>> /*
>>  * zcache initialization
>>  * NOTE FOR NOW zcache MUST BE PROVIDED AS A KERNEL BOOT PARAMETER OR
>>  * NOTHING HAPPENS!
>>  */
>>
>> Is there something admin can do about it?
>>
>> >
>> > It's outside this patch's scope, though, just as changing the
>> > behaviour to fail swapon is.
>>
>> Agreed. The patch should just use vzalloc and the allocation failure
>> should be handled separately - if needed at all.
>
> Agreed here too.  The frontswap_enabled flag is global (enabling frontswap
> across all frontswap devices) whereas failure to allocate the frontswap_map
> will disable frontswap for only one swap device.  And since frontswap is
> strictly a performance enhancement, there's no reason to fail the swapon
> for the entire swap device.
>

Agreed.

> I am fairly sure that the failed allocation is handled gracefully
> through the remainder of the frontswap code, but will re-audit to
> confirm.  A warning might be nice though.
>

There is a place i think maybe have problem.
function __frontswap_flush_area() in file frontswap.c called
memset(sis->frontswap_map, .., ..);
But if frontswap_map allocation fail there is a null pointer access ?

> In any case:
>
>> -             frontswap_map = vmalloc(maxpages / sizeof(long));
>> -             if (frontswap_map)
>> -                     memset(frontswap_map, 0, maxpages / sizeof(long));
>> +             frontswap_map = vzalloc(maxpages / sizeof(long));
>
> Acked-by: Dan Magenheimer <dan.magenheimer@oracle.com>
>

Thanks,

>> +             if (!frontswap_map)
>> +                     goto bad_swap;
>
> NAK
>
> Dan
>
> Thanks... for the memory!
> I really could use more / my throughput's on the floor
> The balloon is flat / my swap disk's fat / I've OOM's in store
> Overcommitted so much
> (with apologies to Bob Hope)
>

:)

-- 
Regards,
--Bob

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>