On Wed, 2 Jul 2008, Alan Cox wrote:

quoted text
> On Wed, 2 Jul 2008 00:09:22 -0400 (EDT)
> Mikulas Patocka <mpatocka@redhat.com> wrote:
>
>>> Right, that wont work of course. Completions are typically done through
>>> a softirq, so it is not currently done with hard interrupts disabled.
>>
>> I thought, from hardirq - that's what IDE is doing. And they are called
>
> Even IDE will sometimes complete from a timer on an error.
>
>> And does local_irq_restore() need to execute even more costy "popf" when
>> it makes a transition from disabled to disabled state? What's
>> local_irq_restore semantics --- is it allowed to use local_irq_restore for
>> transition from interrupt-enabled state into interrupt-disabled state?
>
> If you are worried about performance the network layer has _irq variants
> of various functions that are faster and can only be called from the
> right context (eg kfree_skb_irq), so you could do two versions of that
> code.
>
> Alan

I was thinking about a general optimization --- something like this. On 
Core2 it doesn't make much sense --- cli is fast there (just 11 ticks). On 
Pentium 4 it would make sense, because cli is much more costy there.

Mikulas

---
  include/asm-x86/irqflags.h |    3 ++-
  1 file changed, 2 insertions(+), 1 deletion(-)

Index: linux-2.6.26-rc8/include/asm-x86/irqflags.h
===================================================================
--- linux-2.6.26-rc8.orig/include/asm-x86/irqflags.h	2008-06-02 23:13:32.000000000 +0200
+++ linux-2.6.26-rc8/include/asm-x86/irqflags.h	2008-07-03 22:53:32.000000000 +0200
@@ -101,7 +101,8 @@ static inline unsigned long __raw_local_
  {
  	unsigned long flags = __raw_local_save_flags();

-	raw_local_irq_disable();
+	if (flags & X86_EFLAGS_IF)
+		raw_local_irq_disable();

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