Re: [patch, 2.6.10-rc2] fix __flush_tlb*() preemption bug onCONFIG_PREEMPT

[Linus Torvalds]

On Thu, 18 Nov 2004, Ingo Molnar wrote:
> 
> note that reproducing this bug was only possible under PREEMPT_RT (there
> it can be triggered in 30 seconds, with the right reproducer) - it needs
> a really unlikely scenario which PREEMPT_RT's high concurrency does
> offer but which is apparently much harder to reproduce in the vanilla
> kernel. The patch fixes x86 and x64. Other architectures are most likely
> safe, but they need review as well.

Ok, that's a pretty race.

However, I'm wondering whether this is the proper approach. After all, a
lazy-tlb process should never have any reason to flush its TLB, since "its
TLB" just aint there, and it ends up flushing somebody elses.

So I assume that this happens only with kswapd or similar? It really might 
be interesting to make the "we were a lazy tlb, and we're flushing 
somebody else" case do a stack dump, because I _suspect_ that this really 
is a special thing, and maybe the right thing to do is to make it special 
in _that_ path.

Very impressive debugging, btw. That must have been painful.

		Linus
-
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/