On 01/08/2024 12:15, Andrew Jones wrote:

On Thu, Aug 01, 2024 at 10:43:03AM GMT, Alexandre Ghiti wrote:
...

quoted

quoted

quoted

diff --git a/include/asm-generic/qspinlock.h b/include/asm-generic/qspinlock.h
index 0655aa5b57b2..bf47cca2c375 100644
--- a/include/asm-generic/qspinlock.h
+++ b/include/asm-generic/qspinlock.h

@@ -136,6 +136,7 @@ static __always_inline bool virt_spin_lock(struct qspinlock *lock)
  }
  #endif

+#ifndef __no_arch_spinlock_redefine

I'm not sure what's better/worse, but instead of inventing this
__no_arch_spinlock_redefine thing we could just name all the functions
something like __arch_spin* and then add defines for both to asm/spinlock.h,
i.e.

#define queued_spin_lock(l) __arch_spin_lock(l)
...

#define ticket_spin_lock(l) __arch_spin_lock(l)
...

__arch_spin_lock() would use queued_spin_lock() so that would make an
"infinite recursive definition" right? And that would override the
"real" queued_spin_lock() implementation too.

But maybe I missed something!

It depends on where the definition is done. It should work if the
preprocessor expands the implementation of __arch_spin_* before
evaluating the #define of queued_spin_*. IOW, we just need to put
the defines after the static inline constructions.

So I have just given it a try, both qspinlock.h and ticket_spinlock.h 
define arch_spin_XXXX(). That triggers a bunch of warnings.

I'll drop this suggestion as I find it harder to understand and because 
of the warnings that would need the introduction of a new preprocessor 
variable (or something else?).  And the solution with 
__no_arch_spinlock_redefine is really straightforward and lightweight.

Thanks,

Alex

Thanks,
drew

_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv