On Wed, Sep 19, 2001 at 14:28:18 -0500, Jeff Dike wrote:
> jorgen.cederlof@... said:
> > Sure. I really need to get the slowdown down to an acceptable level,
> > like 2-3x.=20
> You can see the major problems by ^C-ing UML during this time and looki=
> at the stack. I bet that 90% of the time, you see fix_range running, a=
> it's busy remapping the kernel vm when there is no remapping needed.
Actually, it's far above 90%. I tried pressing ^C 41 times. 15 times
fix_range() was running, and the other 26 times UML was running
__munmap(), called from fix_range().
> So, that's the first thing I'd fix. What I have planned is a sequence =
> on kernel vm changes. It'll be stored in a global, and also in each pr=
> thread structure. The processes store the last sequence number that th=
> If a process' sequence number is the same as the global, then fix_range=
> need to look at kernel vm for that process.
OK, I'll do my best. VM hacking is new to me, so expect some stupid