Re: [PATCH 03/22] slab: Remove dead code in check_slab()

[Vlastimil Babka <vbabka@suse.cz>]

On 9/10/25 13:39, Matthew Wilcox wrote:
> On Wed, Sep 10, 2025 at 08:21:18PM +0900, Harry Yoo wrote:
>> Side question: if someone wants to (though I don't see much value in it)
>> iterate over n->partial and check whether a slab is corrupted and no
>> longer a slab, should they convert it from slab to page and then check
>> PageSlab()?

Your question sounds like a hypothetical scenario, but that's actually what
writing to sysfs "validate" file of a cache does, no? One could say that in
validate_slab() we don't "already know the slab passed in is a valid slab".

> I don't think so.  There are probably other objects allocated from the
> corrupt slab, and they might get freed at some point.  If they are,
> we'll want to handle them gracefully (after all, the caller of kfree()
> probably did nothing wrong) and not print a warning suggesting that
> someone's calling kfree() on a non-slab pointer.

In case of validate_slab() it wouldn't be triggered from a kfree().

However once struct slab stop overlaying struct page/folio then we simply
can't have this validation via page/folio flag anymore, so it's fine to
remove it. Maybe there will be a different way to reintroduce something
similar in the memdesc world.