On Sat, May 22, 2004 at 11:11:19PM -0600, M. Warner Losh wrote:
> More like two or three power cycles before mounting /.  At least in
> the 1.6 time frame on my hpcmips machine for the microdrive I use as
> my root device.

I see. Well, we could easily add another function_enable() call at first
attach time, schedule a callout in say 15 seconds and do a function_disable()
then. (In the wd driver). This would cause any intermediate
enable()/disable() calls to not change the power state. But the first enable
would be after the first power off after reading the CIS.

Or we could do that at the pcmcia layer level for socket_enable()/
socket_disable(), so every other type of card would use this feature too - no
idea if there are other cards that need significant time for power cycles. We
could avoid the after-CIS-reading power off this way too.

This would not delay power off at (later) unmount time.

> FreeBSD powers things up once, and it turns out to be a lot faster to
> do it this way for real spinning media.  Turns out to be about a wash
> for non-spinning media or network devices.

I'm not sure I understand this completely - I suppose you don't let the
socket powered on unconditionally - even if the drive is not mounted? Is it a
timeout method, like described above, or do you rely on pccard daemon to do
the mount imediately? Or are you talking only root mounts only?

Martin