* Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> wrote:

quoted text
> Although I have not considered how to implement proper fix deeply, I 
> don't think it's so simple. For example, current pciehp is doing like 
> this:
>
> (1) some initialization
> (2) request_irq()
> (3) issue command
> (4) initialize slot data structure
>
> Maybe we want to do (2) after (4) to fix the problem. But if we simply 
> move (2) after (4), we cannot detect the command completion event at 
> (3) and it will cause command timeout.
>
> It's just an example, and there might be other things like this. This 
> example might be fixed simply, but all my worry is that fixing this 
> quickly might cause another regressions. This is why I think Ingo's 
> approach is better in a short term.
>
> And another reason is I'm very nervous because I already caused many 
> problems in pciehp since 2.6.26-rcX... :(

no need to feel bad about it - i dont think what i reported is a serious 
problem in any fashion - it's a very narrow race as long as DEBUG_SHIRQ 
is off that very likely wont trigger in practice. We can do a minimal 
fix in v2.6.26 and something more structural in v2.6.27.

	Ingo
--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/