Re: [PATCH] slob: add size header to all allocations

[Hyeonggon Yoo]

On Thu, Oct 21, 2021 at 07:36:26PM +0200, Vlastimil Babka wrote:
> On 10/20/21 13:46, Hyeonggon Yoo wrote:
> > On Sun, Oct 17, 2021 at 08:38:41PM -0700, Rustam Kovhaev wrote:
> >> Let's prepend all  allocations of (PAGE_SIZE - align_offset) and less
> >> with the size header. This way kmem_cache_alloc() memory can be freed
> >> with kfree() and the other way around, as long as they are less than
> >> (PAGE_SIZE - align_offset).
> > 
> > Hello Rustam, I measured its impact on memory usage on
> > tiny kernel configuration as SLOB is used in very small machine.
> > 
> > on x86 32 bit + tinyconfig:
> >     Before:
> >     Slab:                668 kB
> > 
> >     After:
> >     Slab:                688~692 kB
> > 
> > it adds 20~24kB.
> 
> Thanks for the measurement. That's 3.5% increase.
> 

You're welcome.

> > 
> >> 
> >> The main reason for this change is to simplify SLOB a little bit, make
> >> it a bit easier to debug whenever something goes wrong.
> >>
> > 
> > It seems acceptable But I wonder it is worth to increase memory usage
> > to allow freeing kmem_cache_alloc-ed objects by kfree()?
> 
> Not for the reason above, but for providing a useful API guarantee
> regardless of selected slab allocator IMHO yes.
> 

Mm.. that means some callers free kmem_cache_alloc-ed object using
kfree, and SLAB/SLUB already support that, and SLOB doesn't.

In what situations is freeing using kfree needed?
Wouldn't this make code confusing?

> > Thanks,
> > Hyeonggon
> > 
> >> meminfo right after the system boot, without the patch:
> >> Slab:              35500 kB
> >> 
> >> the same, with the patch:
> >> Slab:              36396 kB
> >> 
> > 
>