Re: [PATCH v2 24/33] mm/slob: Convert SLOB to use struct slab

[Vlastimil Babka <vbabka@suse.cz>]

On 12/10/21 16:29, Hyeonggon Yoo wrote:
> On Fri, Dec 10, 2021 at 12:44:32PM +0100, Vlastimil Babka wrote:
>> On 12/10/21 11:44, Hyeonggon Yoo wrote:
>> > On Wed, Dec 01, 2021 at 07:15:01PM +0100, Vlastimil Babka wrote:
>> >> From: "Matthew Wilcox (Oracle)" <willy@infradead.org>
>> >> 
>> >> Use struct slab throughout the slob allocator.
>> >> 
>> >> [ vbabka@suse.cz: don't introduce wrappers for PageSlobFree in mm/slab.h just
>> >>   for the single callers being wrappers in mm/slob.c ]
>> >> 
>> >> Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
>> >> Signed-off-by: Vlastimil Babka <vbabka@suse.cz>
>> >> ---
>> >>  mm/slob.c | 34 +++++++++++++++++-----------------
>> >>  1 file changed, 17 insertions(+), 17 deletions(-)
>> >> 
>> >> diff --git a/mm/slob.c b/mm/slob.c
>> >> index d2d15e7f191c..d3512bcc3141 100644
>> >> --- a/mm/slob.c
>> >> +++ b/mm/slob.c
>> > 
>> > ...
>> > 
>> >>  		/* Enough room on this page? */
>> >> @@ -358,8 +358,8 @@ static void *slob_alloc(size_t size, gfp_t gfp, int align, int node,
>> >>  		b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
>> >>  		if (!b)
>> >>  			return NULL;
>> >> -		sp = virt_to_page(b);
>> >> -		__SetPageSlab(sp);
>> >> +		sp = virt_to_slab(b);
>> >> +		__SetPageSlab(slab_page(sp));
>> > 
>> > Hello Vlastimil.
>> > 
>> > I've tested this patch on my machine and it causes NULL pointer dereference.
>> > that's because virt_to_slab returns NULL if folio_test_slab is false.
>> > and __SetPageSlab is called with sp = NULL.
>> 
>> Oops, thanks for catching this.
>> 
>> > diff below fixed bug.
>> 
>> Changed it to use folio and will push an updated branch
>> slab-struct_slab-v3r2
>> 
>> Thanks!
> 
> Tested again on slab-struct_slab-v3r2 and everyting works fine!
> And the code overall looks good to me.
> 
> Reviewed-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>
> Tested-by: Hyeonggon Yoo <42.hyeyoo@gmail.com>

Thanks!

> And below is not about this patch,
> but what about something like this?
> 
> Because it's not necessary that page->_mapcount and
> slab->units have same offset, so we can remove
> (somewhat confusing) page_mapcount_reset function call.

Yeah this could be done as new patch on top of 21/33. I wouldn't squash it
there so that it remains just a split without functional change.

> I want to remove calling page_mapcount_reset also in
> SLAB, but I have no idea. (maybe we can do that if we remove
> SLAB colouring but that's going too far)
> 
> diff --git a/mm/slab.h b/mm/slab.h
> index 90d7fceba470..dd0480149d38 100644
> --- a/mm/slab.h
> +++ b/mm/slab.h
> @@ -50,8 +50,8 @@ struct slab {
>         struct list_head slab_list;
>         void * __unused_1;
>         void *freelist;         /* first free block */
> -       void * __unused_2;
> -       int units;
> +       long units;
> +       unsigned int __unused_2;
> 
>  #else
>  #error "Unexpected slab allocator configured"
> diff --git a/mm/slob.c b/mm/slob.c
> index 39b651b3e6e7..7b2d2c7d69cc 100644
> --- a/mm/slob.c
> +++ b/mm/slob.c
> @@ -404,7 +404,6 @@ static void slob_free(void *block, int size)
>                         clear_slob_page_free(sp);
>                 spin_unlock_irqrestore(&slob_lock, flags);
>                 __ClearPageSlab(slab_page(sp));
> -               page_mapcount_reset(slab_page(sp));
>                 slob_free_pages(b, 0);
>                 return;
>         }
> 
>> 
>> > diff --git a/mm/slob.c b/mm/slob.c
>> > index d3512bcc3141..cf669f03440f 100644
>> > --- a/mm/slob.c
>> > +++ b/mm/slob.c
>> > @@ -358,8 +358,8 @@ static void *slob_alloc(size_t size, gfp_t gfp, int a
>> > lign, int node,
>> >                 b = slob_new_pages(gfp & ~__GFP_ZERO, 0, node);
>> >                 if (!b)
>> >                         return NULL;
>> > +               __SetPageSlab(virt_to_page(b));
>> >                 sp = virt_to_slab(b);
>> > -               __SetPageSlab(slab_page(sp));
>> > 
>> >                 spin_lock_irqsave(&slob_lock, flags);
>> >                 sp->units = SLOB_UNITS(PAGE_SIZE);
>> >  
>> >  Thanks,
>> >  Hyeonggon.
>> > 
>> >>  
>> >>  		spin_lock_irqsave(&slob_lock, flags);
>> >>  		sp->units = SLOB_UNITS(PAGE_SIZE);
>> >> @@ -381,7 +381,7 @@ static void *slob_alloc(size_t size, gfp_t gfp, int align, int node,
>> >>   */
>> >>  static void slob_free(void *block, int size)
>> >>  {
>> >> -	struct page *sp;
>> >> +	struct slab *sp;
>> >>  	slob_t *prev, *next, *b = (slob_t *)block;
>> >>  	slobidx_t units;
>> >>  	unsigned long flags;
>> >> @@ -391,7 +391,7 @@ static void slob_free(void *block, int size)
>> >>  		return;
>> >>  	BUG_ON(!size);
>> >>  
>> >> -	sp = virt_to_page(block);
>> >> +	sp = virt_to_slab(block);
>> >>  	units = SLOB_UNITS(size);
>> >>  
>> >>  	spin_lock_irqsave(&slob_lock, flags);
>> >> @@ -401,8 +401,8 @@ static void slob_free(void *block, int size)
>> >>  		if (slob_page_free(sp))
>> >>  			clear_slob_page_free(sp);
>> >>  		spin_unlock_irqrestore(&slob_lock, flags);
>> >> -		__ClearPageSlab(sp);
>> >> -		page_mapcount_reset(sp);
>> >> +		__ClearPageSlab(slab_page(sp));
>> >> +		page_mapcount_reset(slab_page(sp));
>> >>  		slob_free_pages(b, 0);
>> >>  		return;
>> >>  	}
>> >> -- 
>> >> 2.33.1
>> >> 
>> >> 
>>