Re: [rfc patch] i386: don't save eflags on task switch

[Andi Kleen]

On Sunday 05 November 2006 17:12, Linus Torvalds wrote:

> And changing restore-flags to a "conditional branch around sti" 

Yes of course.

> is likely  
> not much better 

We'll see.

It used to be a bad idea because everything was inline, but these
days with out of line code one can be much more flexible.

> - mispredicted branches on a P4 are potentially worse than  
> the popf cost.

They are far less than 48 cycles. The P4 is not _that_ bad in this
area.

> Side note: for the netburst microarchitecture - aka P4 - in general, 
> something like 48 cycles is a _good_ thing. I measured a internal 
> micro-fault for marking a page table entry dirty at over 1500 cycles! 
> There's a reason Intel dropped Netburst in favour of Core 2 - which is 
> largely just an improved Pentium Pro uarch. Admittedly, the "just" is a 
> bit unfair, because there's a _lot_ of improvement, but still..
> 
> So you should never actually make any real code design decisions based on 
> a P4 result. The P4 is goign away, and it was odd. 

There are millions and millions of P4s out there running 
Linux and I don't think that will change any time soon (in fact Intel will
be still shipping many new ones for a long time) If there are cheap 
valuable optimizations for P4 that don't impact others much I'm all for them.
 
-Andi
-
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/