Re: [PATCH] remove LOCK_SECTION from x86_64 spin_lock asm

[Ingo Molnar]

* Zwane Mwaikambo <zwane@xxxxxxxxxxxxx> wrote:

> On Thu, 16 Sep 2004, Ingo Molnar wrote:
> 
> > the ebp trick is nice, but forcing a formal stack frame for every
> > function has global performance implications. Couldnt we define some
> > sort of current-> field [or current_thread_info() field] that the
> > spinlock code could set and clear, which field would be listened to by
> > profile_pc(), so that the time spent spinning would be attributed to the
> > callee? Something like:
> 
> I think the generic route is nice but wouldn't this break with the 
> following.
> 
> taskA:
> spin_lock(lockA); // contended
> <interrupt>
> int1:
> spin_lock(lockB)
> 
> I was thinking along the likes of a per_cpu real_pc, but realised it
> falls prey to the same problem as above... Unless we have irq threads,
> then of course your solution works.

you mean the nesting? spin_lock() should save/restore the value instead
of setting/clearing it - and fork() should initialize it to zero.

	Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/