Prev: [PATCH resend] x86, amd: Check X86_FEATURE_OSVW bit before accessing OSVW MSRs
Next: rwsem: wake queued readers when other readers are active
From: David Howells on 12 May 2010 08:20
Michel Lespinasse <walken(a)google.com> wrote:

> static inline void __up_write(struct rw_semaphore *sem)
> {
> ...
> LOCK_PREFIX " xadd %1,(%2)\n\t"
> - /* tries to transition
> - 0xffff0001 -> 0x00000000 */
> - " jz 1f\n"
> + /* substracts 0xffff0001, returns the old value */
> + " cmp %4,%1\n\t"
> + /* are there waiting threads and no active threads ? */
> + " jne 1f\n"
> " call call_rwsem_wake\n"

It looks like you're betting on there being fast-path contention. If the
common case is no fast-path contention, then you'd be better off putting the
comparison out of line in the medium-path and retaining the JZ instruction.

The same goes for __up_read(): you could retain the JNS there and put the
comparison out of line into the medium-path.

Doing this also saves you some code space.

David
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo(a)vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
 | 
Pages: 1
Prev: [PATCH resend] x86, amd: Check X86_FEATURE_OSVW bit before accessing OSVW MSRs
Next: rwsem: wake queued readers when other readers are active