Re: CVS commit: src/sys/arch/cesfic

[Garrett D'Amore <garrett_damore%tadpole.com@localhost>]

Daniel Carosone wrote:
> On Thu, Sep 14, 2006 at 10:33:59AM -0700, Garrett D'Amore wrote:
>   
>> I still think even NFS time isn't necessarily good.  You don't know the
>> last time the filesystem being exported was touched...
>>     
>
> From down in the depths of memory, possibly from sun3's or apollos
> or.. something of that vintage, ISTR the idea was that the kernel was
> stored in the NFS root filesystem, and that the lookup and load of
> that made the server set the atime appropriately.
>
> --
> Dan.
>   

But clearly this isn't always the case.  Also, with modern filesystems,
it isn't uncommon to disable setting of the atime bit to avoid thrashing
the filesystem, and it also isn't uncommon to export read-only.

Blindly assuming that the root filesystem time is good is not a good
practice in my opinion.  Sure, we should use it when we have no other
source, but we should tell people (admins) that we're doing that. 
Otherwise they might be lulled into believing that they have a battery
backed system clock, which is patently not true.  (There are some
systems where a battery backed clock might be present, but we might not
have a driver for it.)

In any case the thing is that we are only issuing a warning to the
system admin saying, "hey we're using the filesystem time because we
have no better source".  I think that is reasonable.

But, I am willing to imagine a tunable to turn it off.  I'd be opposed
to having it set by default in GENERIC, though.  (A sysctl would work,
if we can make sysctl's persistent.  I'm not sure about that -- I don't
know anything about sysctls.)

-- 
Garrett D'Amore, Principal Software Engineer
Tadpole Computer / Computing Technologies Division,
General Dynamics C4 Systems
http://www.tadpolecomputer.com/
Phone: 951 325-2134  Fax: 951 325-2191