On Tue, Aug 27, 2024 at 11:10:10PM GMT, Vlastimil Babka wrote:
> On 8/27/24 17:59, Christian Brauner wrote:
> > When a kmem cache is created with SLAB_TYPESAFE_BY_RCU the free pointer
> > must be located outside of the object because we don't know what part of
> > the memory can safely be overwritten as it may be needed to prevent
> > object recycling.
> >
> > That has the consequence that SLAB_TYPESAFE_BY_RCU may end up adding a
> > new cacheline. This is the case for .e.g, struct file. After having it
> > shrunk down by 40 bytes and having it fit in three cachelines we still
> > have SLAB_TYPESAFE_BY_RCU adding a fourth cacheline because it needs to
> > accomodate the free pointer and is hardware cacheline aligned.
> >
> > I tried to find ways to rectify this as struct file is pretty much
> > everywhere and having it use less memory is a good thing. So here's a
> > proposal.
> >
> > Signed-off-by: Christian Brauner <brauner@xxxxxxxxxx>
>
> So logistically patch 3 needs stuff in the vfs tree and having 1+2 in slab
> tree and 3 in vfs that depends on 1+2 elsewhere is infeasible, so it will be
> easiest for whole series to be in vfs, right?
Yeah, that's fine by me.
>
> > ---
> > include/linux/slab.h | 9 ++++
> > mm/slab.h | 1 +
> > mm/slab_common.c | 133 ++++++++++++++++++++++++++++++++++++---------------
> > mm/slub.c | 17 ++++---
> > 4 files changed, 114 insertions(+), 46 deletions(-)
> >
> > diff --git a/include/linux/slab.h b/include/linux/slab.h
> > index eb2bf4629157..5b2da2cf31a8 100644
> > --- a/include/linux/slab.h
> > +++ b/include/linux/slab.h
> > @@ -212,6 +212,12 @@ enum _slab_flag_bits {
> > #define SLAB_NO_OBJ_EXT __SLAB_FLAG_UNUSED
> > #endif
> >
> > +/*
> > + * freeptr_t represents a SLUB freelist pointer, which might be encoded
> > + * and not dereferenceable if CONFIG_SLAB_FREELIST_HARDENED is enabled.
> > + */
> > +typedef struct { unsigned long v; } freeptr_t;
> > +
> > /*
> > * ZERO_SIZE_PTR will be returned for zero sized kmalloc requests.
> > *
> > @@ -242,6 +248,9 @@ struct kmem_cache *kmem_cache_create_usercopy(const char *name,
> > slab_flags_t flags,
> > unsigned int useroffset, unsigned int usersize,
> > void (*ctor)(void *));
> > +struct kmem_cache *kmem_cache_create_rcu(const char *name, unsigned int size,
> > + unsigned int freeptr_offset,
> > + slab_flags_t flags);
> > void kmem_cache_destroy(struct kmem_cache *s);
> > int kmem_cache_shrink(struct kmem_cache *s);
> >
> > diff --git a/mm/slab.h b/mm/slab.h
> > index dcdb56b8e7f5..b05512a14f07 100644
> > --- a/mm/slab.h
> > +++ b/mm/slab.h
> > @@ -261,6 +261,7 @@ struct kmem_cache {
> > unsigned int object_size; /* Object size without metadata */
> > struct reciprocal_value reciprocal_size;
> > unsigned int offset; /* Free pointer offset */
> > + unsigned int rcu_freeptr_offset; /* Specific free pointer requested */
>
> More precisely something like:
>
> Specific offset requested (if not
> UINT_MAX)
Yep, added that.
>
> ?
>
> > #ifdef CONFIG_SLUB_CPU_PARTIAL
> > /* Number of per cpu partial objects to keep around */
> > unsigned int cpu_partial;
> > diff --git a/mm/slab_common.c b/mm/slab_common.c
> > index c8dd7e08c5f6..c4beff642fff 100644
> > --- a/mm/slab_common.c
> > +++ b/mm/slab_common.c
> > @@ -202,9 +202,10 @@ struct kmem_cache *find_mergeable(unsigned int size, unsigned int align,
> > }
> >
> > static struct kmem_cache *create_cache(const char *name,
> > - unsigned int object_size, unsigned int align,
> > - slab_flags_t flags, unsigned int useroffset,
> > - unsigned int usersize, void (*ctor)(void *))
> > + unsigned int object_size, unsigned int freeptr_offset,
> > + unsigned int align, slab_flags_t flags,
> > + unsigned int useroffset, unsigned int usersize,
> > + void (*ctor)(void *))
> > {
> > struct kmem_cache *s;
> > int err;
> > @@ -212,6 +213,12 @@ static struct kmem_cache *create_cache(const char *name,
> > if (WARN_ON(useroffset + usersize > object_size))
> > useroffset = usersize = 0;
> >
> > + err = -EINVAL;
> > + if (freeptr_offset < UINT_MAX &&
>
> freeptr_offset != UINT_MAX to be more obvious and match has_freeptr_offset() ?
Done.
>
> > + (freeptr_offset >= object_size ||
> > + (freeptr_offset && !(flags & SLAB_TYPESAFE_BY_RCU))))
>
> and here drop the "freeptr_offset &&" as zero is a valid value
Yes, thank you.
>
> instead we could want alignment to sizeof(freeptr_t) if we were paranoid?
Added a check for that.
>
> > + goto out;
>
> The rest seems good to me now.
Thanks for the review! v3 incoming
