On Thu, Jul 11, 2024 at 4:51 AM Alex Shi <seakeel@xxxxxxxxx> wrote:
>
>
>
> On 7/11/24 4:13 PM, Vlastimil Babka wrote:
> > On 7/10/24 7:43 AM, alexs@xxxxxxxxxx wrote:
> >> From: "Alex Shi (Tencent)" <alexs@xxxxxxxxxx>
> >>
> >> commit 21c690a349ba ("mm: introduce slabobj_ext to support slab object
> >> extensions") changed the folio/page->memcg_data define condition from
> >> MEMCG to SLAB_OBJ_EXT. And selected SLAB_OBJ_EXT for MEMCG, just for
> >> SLAB_MATCH(memcg_data, obj_exts), even no other relationship between them.
> >>
> >> Above action make memcg_data exposed and include SLAB_OBJ_EXT for
> >> !MEMCG. That's incorrect in logcial and pay on code size.
> >>
> >> As Vlastimil Babka suggested, let's add _unused_slab_obj_ext for
> >> SLAB_MATCH for slab.obj_exts while !MEMCG. That could resolve the match
> >> issue, clean up the feature logical. And decouple the SLAB_OBJ_EXT from
> >> MEMCG in next patch.
> >>
> >> Signed-off-by: Alex Shi (Tencent) <alexs@xxxxxxxxxx>
> >> Cc: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> >> Cc: Yoann Congal <yoann.congal@xxxxxxxx>
> >> Cc: Masahiro Yamada <masahiroy@xxxxxxxxxx>
> >> Cc: Petr Mladek <pmladek@xxxxxxxx>
> >> Cc: Suren Baghdasaryan <surenb@xxxxxxxxxx>
> >> Cc: Vlastimil Babka <vbabka@xxxxxxx>
> >> ---
> >> v1->v3: take Vlastimil's suggestion and move SLAB_OBJ_EXT/MEMCG decouple
> >> to 2nd patch.
> >> ---
> >> include/linux/mm_types.h | 8 ++++++--
> >> mm/slab.h | 4 ++++
> >> 2 files changed, 10 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
> >> index ef09c4eef6d3..4ac3abc673d3 100644
> >> --- a/include/linux/mm_types.h
> >> +++ b/include/linux/mm_types.h
> >> @@ -180,8 +180,10 @@ struct page {
> >> /* Usage count. *DO NOT USE DIRECTLY*. See page_ref.h */
> >> atomic_t _refcount;
> >>
> >> -#ifdef CONFIG_SLAB_OBJ_EXT
> >> +#ifdef CONFIG_MEMCG
> >> unsigned long memcg_data;
> >> +#elif defined(CONFIG_SLAB_OBJ_EXT)
> >> + unsigned long _unused_slab_obj_ext;
> >> #endif
> >>
> >> /*
> >> @@ -343,8 +345,10 @@ struct folio {
> >> };
> >> atomic_t _mapcount;
> >> atomic_t _refcount;
> >> -#ifdef CONFIG_SLAB_OBJ_EXT
> >> +#ifdef CONFIG_MEMCG
> >> unsigned long memcg_data;
> >> +#elif defined(CONFIG_SLAB_OBJ_EXT)
> >> + unsigned long _unused_slab_obj_ext;
> >> #endif
> >> #if defined(WANT_PAGE_VIRTUAL)
> >> void *virtual;
> >> diff --git a/mm/slab.h b/mm/slab.h
> >> index 3586e6183224..8ffdd4f315f8 100644
> >> --- a/mm/slab.h
> >> +++ b/mm/slab.h
> >> @@ -98,7 +98,11 @@ SLAB_MATCH(flags, __page_flags);
> >> SLAB_MATCH(compound_head, slab_cache); /* Ensure bit 0 is clear */
> >> SLAB_MATCH(_refcount, __page_refcount);
> >> #ifdef CONFIG_SLAB_OBJ_EXT
> >> +#ifdef CONFIG_MEMCG
> >> SLAB_MATCH(memcg_data, obj_exts);
> >> +#else
> >> +SLAB_MATCH(_unused_slab_obj_ext, obj_exts);
> >> +#endif
> >> #endif
> >
> > Why not also #ifdef / #elif like above, instead of this nesting?
>
> Uh, it works too if MEMCG/SLAB_OBJ_EXT decoupled.
> but right, it could be written with #ifdef/#elif.
Yes, please keep the same condition, otherwise it gets confusing.
>
> Thanks
> Alex
> >
> >> #undef SLAB_MATCH
> >> static_assert(sizeof(struct slab) <= sizeof(struct page));
> >
