Re: TAKE 950027 - xfs_icsb_lock_all_counters fails with CONFIG_PREEMPT and >=256p

[Andi Kleen <ak@xxxxxxx>]

On Thursday 02 March 2006 17:53, Linus Torvalds wrote:
>nything worse.
> 
> > I don't see anything that would rely on the count being positive
> > so using the sign bit is probably ok. Hardirq 11 might be a bit
> > tight though, so maybe it would be better to move 64bit machines
> > to 64bit here.

Yes agreed, it just would be a bigger patch probably affecting many
architectures.
 
> And yes, I think it may make sense to just use the full 64 bits on a 
> 64-bit machine. Eventually. Somebody should check what the larger 
> constants result in, though.
> 
> And not for now, but obviously the 11 bits will run out for even bigger 
> machines. But for a "let's fix it quickly", adding three bits should be 
> plenty good enough, no?

That would be spinlock nesting of 2 per CPU on a 1024 CPU machine. 
Not exactly plenty. Better 12-14 at least.

-Andi