On Tue, Dec 16, 2025 at 01:32PM +0100, Peter Zijlstra wrote:
> On Mon, Dec 15, 2025 at 02:38:52PM +0100, Marco Elver wrote:
>
> > Working on rebasing this to v6.19-rc1 and saw this new scoped seqlock
> > abstraction. For that one I was able to make it work like I thought we
> > could (below). Some awkwardness is required to make it work in
> > for-loops, which only let you define variables with the same type.
>
> >
> > diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
> > index b5563dc83aba..5162962b4b26 100644
> > --- a/include/linux/seqlock.h
> > +++ b/include/linux/seqlock.h
> > @@ -1249,6 +1249,7 @@ struct ss_tmp {
> > };
> >
> > static __always_inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
> > + __no_context_analysis
> > {
> > if (sst->lock)
> > spin_unlock(sst->lock);
> > @@ -1278,6 +1279,7 @@ extern void __scoped_seqlock_bug(void);
> >
> > static __always_inline void
> > __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t *lock, enum ss_state
> > target)
> > + __no_context_analysis
> > {
> > switch (sst->state) {
> > case ss_done:
> > @@ -1320,9 +1322,18 @@ __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t
> > *lock, enum ss_state target)
> > }
> > }
> >
> > +/*
> > + * Context analysis helper to release seqlock at the end of the for-scope;
> > the
> > + * alias analysis of the compiler will recognize that the pointer @s is is
> > an
> > + * alias to @_seqlock passed to read_seqbegin(_seqlock) below.
> > + */
> > +static __always_inline void __scoped_seqlock_cleanup_ctx(struct ss_tmp **s)
> > + __releases_shared(*((seqlock_t **)s)) __no_context_analysis {}
> > +
> > #define __scoped_seqlock_read(_seqlock, _target, _s)
> > \
> > for (struct ss_tmp _s __cleanup(__scoped_seqlock_cleanup) = \
> > - { .state = ss_lockless, .data = read_seqbegin(_seqlock) }; \
> > + { .state = ss_lockless, .data = read_seqbegin(_seqlock) }, \
> > + *__UNIQUE_ID(ctx) __cleanup(__scoped_seqlock_cleanup_ctx) =
> > (struct ss_tmp *)_seqlock; \
> > _s.state != ss_done; \
> > __scoped_seqlock_next(&_s, _seqlock, _target))
> >
>
> I am ever so confused.. where is the __acquire_shared(), in read_seqbegin() ?
Ah this is just a diff on top of this v4 series. The read_seqbegin()
already had it:
static inline unsigned read_seqbegin(const seqlock_t *sl)
__acquires_shared(sl) __no_context_analysis
{
> Also, why do we need this second variable with cleanup; can't the
> existing __scoped_seqlock_cleanup() get the __releases_shared()
> attribute?
The existing __scoped_seqlock_cleanup() receives &_s (struct ss_tmp *),
and we can't refer to the _seqlock from __scoped_seqlock_cleanup(). Even
if I create a member seqlock_t* ss_tmp::seqlock and initialize it with
_seqlock, the compiler can't track that the member would be an alias of
_seqlock. The function __scoped_seqlock_next() does receive _seqlock to
effectively release it executes for every loop, so there'd be a "lock
imbalance" in the compiler's eyes.
So having the direct alias (even if we cast it to make it work in the
single-statement multi-definition, the compiler doesn't care) is
required for it to work.
