Re: Kernel 2.5.6 Interactive performance

[Andrea Arcangeli]

On Sun, Mar 10, 2002 at 02:11:49AM +0100, Dieter Nützel wrote:
> On Sonntag, 10. März 2002 02:00:02, Charles Heselton wrote:
> > How would you implement these thing?  I'm not on the same technical level
> > that you guys are, and when/if things are out of context, I don't follow.
> > Can you help?
> 
> If you are somewhat open for "new" (experimental) stuff I can prepare a patch 
> on top of 2.4.18 or 2.4.19-pre2 for you.
> 
> But Andrea Arcangeli informed me that vm-29 had a deadlock bug in the recent 
> fixes for the bh headers. vm-28 is fine or soon to be available vm-30.

Confirm. I only wanted to add that only 2.4.19pre2aa1 and vm-29 can
deadlock, all previous -aa kernels and vm-?? patches are rock solid
AFIK. The bug is just fixed, it was a missing UnlockPage. The bug was
introduced while fixing an highmem balancing thing (that nobody ever
reported on production machines so I don't consider such problem a
showstopper, but nevertheless vm-30 will fix both the new pratical and
the old mostly theorical problem).

Andrea