mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* Improving mutex_init() optimisation for !lockdep
@ 2025-06-23  9:28 Sebastian Andrzej Siewior
  2025-06-23 10:42 ` Peter Zijlstra
  0 siblings, 1 reply; 2+ messages in thread
From: Sebastian Andrzej Siewior @ 2025-06-23  9:28 UTC (permalink / raw)
  To: linux-kernel
  Cc: Boqun Feng, Ingo Molnar, Peter Zijlstra, Waiman Long, Will Deacon

Hi,

while looking at the assembly of something else I stumbled upon
code that originated from mutex_int() on a !LOCKDEP kernel.
We have this macro:

| #define mutex_init(mutex)                                               \
| do {                                                                    \
|         static struct lock_class_key __key;                             \
|                                                                         \
|         __mutex_init((mutex), #mutex, &__key);                          \
| } while (0)

and the compiler computed an offset for __key and an offset and storage
for #mutex. These two arguments aren't used by __mutex_init() but the
compiler can't know that.
If I remove these two arguments on a x86-64 defconfig, I see:

|    text     data     bss      dec     hex filename
| 29753523 8033942 1306232 39093697 25485c1 vmlinux.before
| 29748880 8021654 1306168 39076702 254435e vmlinux.after
|     4643   12288      64    16995  0x4263 diff in bytes

That is 4KiB in text and 12KiB in data. I don't know why we lost 64
bytes in BSS.

Any objections in redoing this to save some bytes?

Sebastian

^ permalink raw reply	[flat|nested] 2+ messages in thread

* Re: Improving mutex_init() optimisation for !lockdep
  2025-06-23  9:28 Improving mutex_init() optimisation for !lockdep Sebastian Andrzej Siewior
@ 2025-06-23 10:42 ` Peter Zijlstra
  0 siblings, 0 replies; 2+ messages in thread
From: Peter Zijlstra @ 2025-06-23 10:42 UTC (permalink / raw)
  To: Sebastian Andrzej Siewior
  Cc: linux-kernel, Boqun Feng, Ingo Molnar, Waiman Long, Will Deacon

On Mon, Jun 23, 2025 at 11:28:32AM +0200, Sebastian Andrzej Siewior wrote:
> Hi,
> 
> while looking at the assembly of something else I stumbled upon
> code that originated from mutex_int() on a !LOCKDEP kernel.
> We have this macro:
> 
> | #define mutex_init(mutex)                                               \
> | do {                                                                    \
> |         static struct lock_class_key __key;                             \
> |                                                                         \
> |         __mutex_init((mutex), #mutex, &__key);                          \
> | } while (0)
> 
> and the compiler computed an offset for __key and an offset and storage
> for #mutex. These two arguments aren't used by __mutex_init() but the
> compiler can't know that.
> If I remove these two arguments on a x86-64 defconfig, I see:
> 
> |    text     data     bss      dec     hex filename
> | 29753523 8033942 1306232 39093697 25485c1 vmlinux.before
> | 29748880 8021654 1306168 39076702 254435e vmlinux.after
> |     4643   12288      64    16995  0x4263 diff in bytes
> 
> That is 4KiB in text and 12KiB in data. I don't know why we lost 64
> bytes in BSS.
> 
> Any objections in redoing this to save some bytes?

Nope, easy win :-)

^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2025-06-23 10:42 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-06-23  9:28 Improving mutex_init() optimisation for !lockdep Sebastian Andrzej Siewior
2025-06-23 10:42 ` Peter Zijlstra

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®