On Tue, Aug 26, 2025 at 2:32 AM Vlastimil Babka <vba...@suse.cz> wrote:
>
> On 8/22/25 18:28, Suren Baghdasaryan wrote:
> >> diff --git a/tools/testing/shared/linux/rcupdate.h
> >> b/tools/testing/shared/linux/rcupdate.h
> >> index
> >> fed468fb0c78db6f33fb1900c7110ab5f3c19c65..c95e2f0bbd93798e544d7d34e0823ed68414f924
> >> 100644
> >> --- a/tools/testing/shared/linux/rcupdate.h
> >> +++ b/tools/testing/shared/linux/rcupdate.h
> >> @@ -9,4 +9,26 @@
> >> #define rcu_dereference_check(p, cond) rcu_dereference(p)
> >> #define RCU_INIT_POINTER(p, v) do { (p) = (v); } while (0)
> >>
> >> +void kmem_cache_free_active(void *objp);
> >> +static unsigned long kfree_cb_offset = 0;
> >> +
> >> +static inline void kfree_rcu_cb(struct rcu_head *head)
> >> +{
> >> + void *objp = (void *) ((unsigned long)head - kfree_cb_offset);
> >> +
> >> + kmem_cache_free_active(objp);
> >> +}
> >> +
> >> +#ifndef offsetof
> >> +#define offsetof(TYPE, MEMBER) __builtin_offsetof(TYPE, MEMBER)
> >> +#endif
> >> +
> >
> > We need a comment here that concurrent kfree_rcu() calls are not
> > supported because they would override each other's kfree_cb_offset.
>
> I think it's a bit more complex and related to the commit log sentence "This
> only works with one kmem_cache, and only the first one used.". The first
> call to kfree_rcu sets kfree_cb_offset (but what if the rhv offset is
> actually 0?) so the others won't update it. So concurrent calls will work as
> far as from the same cache thus same offset. But I'd like Liam's
> confirmation and the comment text, if possible :)
>
> > Kinda obvious but I think unusual limitations should be explicitly
> > called out.
> >
> >> +#define kfree_rcu(ptr, rhv) \
> >> +do { \
> >> + if (!kfree_cb_offset) \
> >> + kfree_cb_offset = offsetof(typeof(*(ptr)), rhv); \
> >> + \
> >> + call_rcu(&ptr->rhv, kfree_rcu_cb); \
> >> +} while (0)
> >
> > Any specific reason kfree_rcu() is a macro and not a static inline function?
>
> Think it's needed for the typeof() to work. The kernel's kfree_rcu() is
> similar in this aspect.
Ah, got it. Thanks!
>
> >> +
> >> #endif
> >>
> >> --
> >> 2.50.1
> >>
>
