Re: [PATCH] lowmemorykiller: prevent multiple instances of low memory killer

[Dan Carpenter <dan.carpenter@oracle.com>]

On Mon, Apr 15, 2013 at 04:11:18PM -0700, David Rientjes wrote:
> On Mon, 15 Apr 2013, Greg Kroah-Hartman wrote:
> 
> > > The positive numbers are used to return information on the remaining
> > > cache size (again, see the comment I pasted above). We could use
> > > -EBUSY, but we'd have to change vmscan.c, which checks specifically
> > > for -1. I can't see a technical reason why -EBUSY couldn't have been
> > > chosen instead, but there's also no real reason to change it now.
> > 
> > If it's not the correct thing to do, sure we can change it, just send a
> > patch.  It makes way more sense than some random -1 return value to me.
> > 
> > Care to send a series of patches fixing this up properly?
> > 
> 
> The comment in shrinker.h is misleading, not the source code.
> do_shrinker_shrink() will fail for anything negative and 0.

The comment is correct.  The only acceptable negative return is -1.
Look at the second time do_shrinker_shrink() is called from
shrink_slab().

   283                  while (total_scan >= batch_size) {
   284                          int nr_before;
   285  
   286                          nr_before = do_shrinker_shrink(shrinker, shrink, 0);
   287                          shrink_ret = do_shrinker_shrink(shrinker, shrink,
   288                                                          batch_size);
   289                          if (shrink_ret == -1)
   290                                  break;
   291                          if (shrink_ret < nr_before)
   292                                  ret += nr_before - shrink_ret;
   293                          count_vm_events(SLABS_SCANNED, batch_size);

regards,
dan carpenter