Re: [parisc-linux] atomic_t

[Grant Grundler <grundler@dsl2.external.hp.com>]

Matthew Wilcox wrote:
> Just thinking about atomic_set taking the spinlock to assure
> exclusivity...  I thought we could eliminate it.

You are only looking at interactions between atomic_set
and atomic_add_return. Aren't there potentialially other
types of interactions with other atomic operations?
But I don't know. Alan Cox?

...
> Before step b leads to case 2.  after step d leads to case 1.  between
> steps b and d, it's as if the atomic_set _never_happened_.  It results
> in v=4, a_a_r returns 4.

I agree with the analysis.

But someone thought it was time to reset the counter. And if
it "never happened", then whoever is looking for 'v == 1' will
never see it.

> The question is whether any code relies on this behaviour.  It probably
> does in some of the messier bits of networking :-(

Even if that code doesn't exist today, it might exist in the future.
I don't want to bet on corner cases with the semantics.
Those types of bugs are hard to reproduce and hard to debug.


thanks,
grant