Re: start_kernel(): bug: interrupts were enabled early

[Linus Torvalds <torvalds@linux-foundation.org>]

On Fri, 2 Apr 2010, David Howells wrote:

> Linus Torvalds <torvalds@linux-foundation.org> wrote:
> 
> > Ahh, yes. In this case, that doesn't likely change anything. The 
> > save/restore versions of the irq-safe locks shouldn't be appreciably more 
> > expensive than the non-saving ones. And architectures that really care 
> > should have done their own per-arch optimized version anyway.
> 
> That depends on the CPU.  Some CPUs have quite expensive interrupt disablement
> instructions.  FRV does for instance; but fortunately, on the FRV, I can use
> some of the excessive quantities of conditional registers to pretend that I
> disable interrupts, and only actually do so if an interrupt actually happens.

I think you're missing the part where we're not _adding_ any irq disables: 
we're just changing the unconditional irq disable to a save-and-disable 
(and the unconditional irq enable to a restore).

So even if irq's are expensive to disable, the change from

	spin_lock_irq()

to

	spin_lock_irqsave()

won't make that code any more expensive.

> > Maybe we should even document that - so that nobody else makes the mistake 
> > x86-64 did of thinking that the "generic spinlock" version of the rwsem's 
> > is anything but a hacky and bad fallback case.
> 
> In some cases, it's actually the best way.  On a UP machine, for instance,
> where they reduce to nothing or where your only atomic instruction is an XCHG
> equivalent.

Again, you seem to think that we used to have just a plain spin_lock. Not 
so. We currently have a spin_lock_irq(), and it is NOT a no-op even on UP. 
It does that irq disable.

Anyway, I suspect that even with just an atomic xchg, you can do a better 
job at doing down_read() than using the generic spin-lock version (likely 
by busy-looping on a special "we're busy" value). But if you do want to 
use the generic spin-lock version, I doubt any architecture makes that 
irqsave version noticeable slower than the unconditional irq version.

		Linus