svn commit: r191733 - in head/sys: amd64/isa isa

[M. Warner Losh]

In message: <49FCC404.6090105 at FreeBSD.org>
            Alexander Motin <mav at FreeBSD.org> writes:
: M. Warner Losh wrote:
: > : Author: mav
: > : Date: Fri May  1 21:43:04 2009
: > : New Revision: 191733
: > : URL: http://svn.freebsd.org/changeset/base/191733
: > : 
: > : Log:
: > :   Add resume methods to i8254 and atrtc devices.
: > 
: > This likely obviates the need for pmtimer.c now.  You might want to
: > investigate...
: 
: Yes, I have seen it.
: 
: > : Modified:
: > :   head/sys/amd64/isa/clock.c
: > :   head/sys/isa/atrtc.c
: > 
: > Shouldn't there be one for i386 too?
: 
: For i386 it is done by pmtimer now (that's why I haven't done it there), 
: but there is no pmtimer driver for amd64. Actually both ways are not so 
: perfect, as both restore timer interrupts quite late on resume process. 

Yes.  That's a problem...

: In my case it is not fatal as i8254 is anyway ticking by default, just 
: slower. But it seems to increase my system resume time to 10 seconds 
: instead of usual 4-5. May be we should somehow enforce order of device 
: resuming, or build some special event timers control infrastructure 
: alike to PIC one.

There's some work to do multi-pass probe/attach.  Perhaps it should be
generalized to do suspend/resume as well...

: Also, except restoring clocks interrupts, pmtimer restores system time 
: on wakeup. For amd64 it is implemented in MD resume code now. We should 
: decide which way to go. I don't very like pmtimer approach, as there is 
: no any newbus relations between it and i8254/atrtc drivers.

pmtimer should likely die, or at least be only used by APM.  It was a
hack that was tuned to APM and never really updated for ACPI...

Warner