> >
> >Thanks; I still see a potential problem with MMU context allocation here:
> >Paul's new code uses a bitmap to keep track of contexts in use
> >(context_map). MOL ignores this bit and just keeps allocating contexts by
> >relying on next_mmu_context to be unused (which it should be, most
> >of the time), plus the next one also being unused. Is that always
> >guaranteed?
>
> The MOL developer tree contains the necessary fixes:
That's the one I lookes at ...
>
> Basically, MOL uses a different range of context than the
> kernel (due to the context bitmap, only 32.000 contexts are
> used by the kernel so a lot of the contexts are free for MOL).
OK, I wasn't sure that was what the mol vs. Linux contexts meant. Cleaner
and simpler than atomically checking for two unused contexts at once. I'll
give it a shot as soon as mozilla-cvs has finished building.
> >And what's the penalty on using a context ID that's already in use?
> That would be very bad. From experience, the symptom is a lot
> of segfaults (but the kernel could go down too).
Mine just locked hard (both with and without checking for two consecutive
free context numbers), bad enough.
Michael
