[ On Friday, November 29, 2002 at 17:24:49 (+0000), David Laight wrote: ]
> Subject: Re: fork1(), uvm_fork() and cpu_fork()
>
> > While you're in there it would be really nice to move the process start
> > time back out of struct pstats and into the proc structure where it
> > belongs and where it won't get paged out....  p_ru should probably be
> > kept in-core all the time too, though it's the start time that I miss
> > the most....
> 
> Actually I would be more inclined to move more stuff into the
> user area.

That's all fine and good, just so long as it's not critical information
which is needed for systems administration and diagnostics.

> However something so that 'ps' could display more of the
> information might be useful.
> 
> One possibility would to be to page in the user area if
> ps requested info about a single process.

I'd really Really REALLY rather not have to try to trigger paging
activity just to display the start time of a process, something which I
consider to be critical necessary information at all times, and
something which NetBSD and sadly now FreeBSD are almost unique in not
keeping in-core at all times.

> You certainly don't want 'top' to page in all the user areas...

That's the least of my worries actually (though the bigger you make
struct user, the more memory 'top' would require just to try to watch
the pot come to a boil, and thus the less likely it would be to ever
come to a boil if you're watching it with 'top'.... :-)

I'm more interested in knowing some of these values even when the system
is almost completely wedged (even from DDB), and especially from system
core images after a panic).

-- 
								Greg A. Woods

+1 416 218-0098;            <g.a.woods@ieee.org>;           <woods@robohack.ca>
Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>