1. Patchew
  2. linux
  3. View series

seqlock: Cure some more scoped_seqlock() optimization fails

Peter Zijlstra posted 1 patch 2 weeks, 1 day ago
There is a newer version of this series
include/linux/seqlock.h | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
seqlock: Cure some more scoped_seqlock() optimization fails
Posted by Peter Zijlstra 2 weeks, 1 day ago 

Arnd reported an x86 randconfig using gcc-15 tripped over
__scoped_seqlock_bug(). Turns out GCC chose not to inline the
scoped_seqlock helper functions and as such was not able to optimize
properly.

Reported-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
---
For tip/locking/urgent

 include/linux/seqlock.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index a8a8661839b6..221123660e71 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -1224,7 +1224,7 @@ struct ss_tmp {
 	spinlock_t	*lock_irqsave;
 };
 
-static inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
+static __always_inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
 {
 	if (sst->lock)
 		spin_unlock(sst->lock);
@@ -1252,7 +1252,7 @@ static inline void __scoped_seqlock_bug(void) { }
 extern void __scoped_seqlock_bug(void);
 #endif
 
-static inline void
+static __always_inline void
 __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t *lock, enum ss_state target)
 {
 	switch (sst->state) {
Re: seqlock: Cure some more scoped_seqlock() optimization fails
Posted by Ingo Molnar 1 week, 6 days ago 
* Peter Zijlstra <peterz@infradead.org> wrote:

> Arnd reported an x86 randconfig using gcc-15 tripped over
> __scoped_seqlock_bug(). Turns out GCC chose not to inline the
> scoped_seqlock helper functions and as such was not able to optimize
> properly.

BTW., I found a Clang randconfig too that fails the build, so it's
not limited to GCC.

Thanks,

	Ingo
Re: seqlock: Cure some more scoped_seqlock() optimization fails
Posted by Arnd Bergmann 2 weeks, 1 day ago 
On Thu, Dec 4, 2025, at 11:43, Peter Zijlstra wrote:
> Arnd reported an x86 randconfig using gcc-15 tripped over
> __scoped_seqlock_bug(). Turns out GCC chose not to inline the
> scoped_seqlock helper functions and as such was not able to optimize
> properly.
>
> Reported-by: Arnd Bergmann <arnd@arndb.de>
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>

Tested-by: Arnd Bergmann <arnd@arndb.de>

> 
> -static inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
> +static __always_inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
>  {
>  	if (sst->lock)
>  		spin_unlock(sst->lock);
> @@ -1252,7 +1252,7 @@ static inline void __scoped_seqlock_bug(void) { }
>  extern void __scoped_seqlock_bug(void);
>  #endif
> 
> -static inline void
> +static __always_inline void
>  __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t *lock, enum 
> ss_state target)
>  {
>  	switch (sst->state) {

It looks I got close: I had tried the __always_inline on
__scoped_seqlock_next but missed the one on __scoped_seqlock_cleanup,
so that was not enough.

Your version addresses the issue for me, thanks a lot for the fix!

      Arnd
Re: seqlock: Cure some more scoped_seqlock() optimization fails
Posted by Ingo Molnar 1 week, 6 days ago 
* Arnd Bergmann <arnd@arndb.de> wrote:

> On Thu, Dec 4, 2025, at 11:43, Peter Zijlstra wrote:
> > Arnd reported an x86 randconfig using gcc-15 tripped over
> > __scoped_seqlock_bug(). Turns out GCC chose not to inline the
> > scoped_seqlock helper functions and as such was not able to optimize
> > properly.
> >
> > Reported-by: Arnd Bergmann <arnd@arndb.de>
> > Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
>
> Tested-by: Arnd Bergmann <arnd@arndb.de>
>
> >
> > -static inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
> > +static __always_inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
> >  {
> >	if (sst->lock)
> >		spin_unlock(sst->lock);
> > @@ -1252,7 +1252,7 @@ static inline void __scoped_seqlock_bug(void) { }
> >  extern void __scoped_seqlock_bug(void);
> >  #endif
> >
> > -static inline void
> > +static __always_inline void
> >  __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t *lock, enum
> > ss_state target)
> >  {
> >	switch (sst->state) {
>
> It looks I got close: I had tried the __always_inline on
> __scoped_seqlock_next but missed the one on __scoped_seqlock_cleanup,
> so that was not enough.

Same here, I ran into that build failure and ended up finding this
as a side-effect:

  24bc5ea5c01a ("seqlock, procfs: Match scoped_seqlock_read() critical section vs. RCU ordering in do_task_stat() to do_io_accounting()")

And like you I was trying to work around the compiler failure
via forced-inlining of __scoped_seqlock_next(), but missed
__scoped_seqlock_cleanup() ... :-)

>
> Your version addresses the issue for me, thanks a lot for the fix!

Works for me too, and I've applied the fix to tip:locking/urgent.

Thanks,

	Ingo
[tip: locking/urgent] seqlock: Cure some more scoped_seqlock() optimization fails
Posted by tip-bot2 for Peter Zijlstra 1 week, 6 days ago 
The following commit has been merged into the locking/urgent branch of tip:

Commit-ID:     90dfeef1cd38dff19f8b3a752d13bfd79f0f7694
Gitweb:        https://git.kernel.org/tip/90dfeef1cd38dff19f8b3a752d13bfd79f0f7694
Author:        Peter Zijlstra <peterz@infradead.org>
AuthorDate:    Thu, 04 Dec 2025 11:43:32 +01:00
Committer:     Ingo Molnar <mingo@kernel.org>
CommitterDate: Sat, 06 Dec 2025 09:53:05 +01:00

seqlock: Cure some more scoped_seqlock() optimization fails

Arnd reported an x86 randconfig using gcc-15 tripped over
__scoped_seqlock_bug(). Turns out GCC chose not to inline the
scoped_seqlock helper functions and as such was not able to optimize
properly.

[ mingo: Clang fails the build too in some circumstances. ]

Reported-by: Arnd Bergmann <arnd@arndb.de>
Tested-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Link: https://patch.msgid.link/20251204104332.GG2528459@noisy.programming.kicks-ass.net
---
 include/linux/seqlock.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/include/linux/seqlock.h b/include/linux/seqlock.h
index a8a8661..2211236 100644
--- a/include/linux/seqlock.h
+++ b/include/linux/seqlock.h
@@ -1224,7 +1224,7 @@ struct ss_tmp {
 	spinlock_t	*lock_irqsave;
 };
 
-static inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
+static __always_inline void __scoped_seqlock_cleanup(struct ss_tmp *sst)
 {
 	if (sst->lock)
 		spin_unlock(sst->lock);
@@ -1252,7 +1252,7 @@ static inline void __scoped_seqlock_bug(void) { }
 extern void __scoped_seqlock_bug(void);
 #endif
 
-static inline void
+static __always_inline void
 __scoped_seqlock_next(struct ss_tmp *sst, seqlock_t *lock, enum ss_state target)
 {
 	switch (sst->state) {

Patchew 2.1-devel-b67e4c2

© 2016 - 2025 Red Hat, Inc.