Re: [PATCH] slub: Fix sysfs duplicate filename creation when slub_debug=O

[Miles Chen <miles.chen@mediatek.com>]

On Wed, 2017-11-08 at 09:05 -0600, Christopher Lameter wrote:
> On Wed, 8 Nov 2017, Miles Chen wrote:
> 
> > > Ok then the aliasing failed for some reason. The creation of the unique id
> > > and the alias detection needs to be in sync otherwise duplicate filenames
> > > are created. What is the difference there?
> >
> > The aliasing failed because find_mergeable() returns if (flags &
> > SLAB_NEVER_MERGE) is true. So we do not go to search for alias caches.
> >
> > __kmem_cache_alias()
> >   find_mergeable()
> >     kmem_cache_flags()  --> setup flag by the slub_debug
> >     if (flags & SLAB_NEVER_MERGE) return NULL;
> >     ...
> >     search alias logic...
> >
> >
> > The flags maybe changed if disable_higher_order_debug=1. So the
> > unmergeable cache becomes mergeable later.
> 
> Ok so make sure taht the aliasing logic also clears those flags before
> checking for SLAB_NEVER_MERGE.
> 
> > > The clearing of the DEBUG_METADATA_FLAGS looks ok to me. kmem_cache_alias
> > > should do the same right?
> > >
> > Yes, I think clearing DEBUG_METADATA flags in kmem_cache_alias is
> > another solution for this issue.
> >
> > We will need to do calculate_sizes() by using original flags and compare
> > the order of s->size and s->object_size when
> > disable_higher_order_debug=1.
> 
> Hmmm... Or move the aliasing check to a point where we know the size of
> the slab objects?

The biggest concern is that we may have some merged caches even if we
enable CONFIG_SLUB_DEBUG_ON and slub_debug=O. So a developer cannot say
"I set CONFIG_SLUB_DEBUG_ON=y to stop all slab merging". 
(https://www.spinics.net/lists/linux-mm/msg77919.html)

In this fix patch, it disables slab merging if SLUB_DEBUG=O and
CONFIG_SLUB_DEBUG_ON=y but the debug features are disabled by the
disable_higher_order_debug logic and it holds the "slab merging is off
if any debug features are enabled" behavior.