Messages in this thread | | | Date | Tue, 14 Oct 2008 23:04:04 +0200 | From | Karel Zak <> | Subject | Re: [kerneloops] regression in 2.6.27 wrt "lock_page" and the "hwclock" program |
| |
On Mon, Oct 13, 2008 at 10:55:59AM -0400, Arjan van de Ven wrote: > Karel: Can you describe what it WANTS to do so we can see if we can
It wants to "atomically" (without context switch) read/write time from CMOS. This is unrealistic of course.
> just extend the linux kernel to do that the right way?
The kernel is already extended and hwclock uses the extension :-) The solution is /dev/rtc.
The code that directly works with CMOS is fallback solution for people who don't want or can not use the standard RTC device. I guess people use this functionally for experiments only. Today the RTC framework should work everywhere.
I'll simply remove the "cli" / "sti" code -- eventually I can try to optimize it by mlockall() and SCHED_FIFO (as suggested by Linus).
Karel
-- Karel Zak <kzak@redhat.com>
| |