On Mon, Jul 22, 2002 at 02:21:13PM +0200, Maciej W. Rozycki wrote:
> I hope so. I'm begging Ralf for about half a year, sigh... The subject
> was beaten to death at the list and the feedback looked positive to me
> (after a few doubts were resolved), but Ralf seems to be unhappy with the
> changes due to some ia64 interactions (I'm still not sure which ones,
> though). Ralf, could you please elaborate?
> Otherwise, your case convinces me I should not care about purity or
> cross-platform consistency of code in this area, anymore. I've been
> observing problems with interrupts due to the lack of iomem access
> synchronization already, but unlike for your PMAZ-A problem, their result
> was more of annoyance than instability. Since stability is a priority,
> although reluctantly, I will rework the changes to apply to the DECstation
> code only, to keep others happy. The interface won't change, apart from
> resolving namespace clashes.
We had some discussion with the IA64 guys at SGI on how to handle this
kind of I/O ordering issues. We never came to a final conclusion but
the proposal was the introduction of separate memory barriers macros for
I/O stuff. Anyway, I think for the moment we should go with your proposal.