Subject: Re: CVS commit: syssrc/sys/arch/arm/arm32
To: <>
From: Richard Earnshaw <rearnsha@buzzard.freeserve.co.uk>
List: tech-kern
Date: 10/18/2001 21:42:43
> In article <200110181818.TAA07617@cam-mail2.cambridge.arm.com> you write:
> >> pmap_activate() should be able to notice a pmap_collect()ed pmap and undo
> >> the action as necessary, shouldn't it?
> >
> >Not necessarily, the pmap may even be the active pmap; I've seen this
> >happen (since a test I put into pmap_activate for a missing page 0 never
> >triggered).
>
> On arm26 I have a flag in each pmap to indicate whether it's active (set by
> pmap_activate(), cleared by pmap_deactivate(). Could you not do this?
Yes, I'm sure that could be done.
Though there is still the question of whether you really want to
pmap_collect a pmap that is part of a uvmspace share. I guess really, uvm
should be handling this case -- if it swaps one process out in this form,
it really ought to swap the other out as well, since they are practically
two threads in the same program.
R.