svn commit: r208988 - in head/sys: kern sys

John Baldwin jhb at freebsd.org
Fri Jun 11 13:44:53 UTC 2010


On Friday 11 June 2010 8:41:49 am Kostik Belousov wrote:
> On Fri, Jun 11, 2010 at 08:10:40AM -0400, John Baldwin wrote:
> > On Friday 11 June 2010 4:31:37 am Kostik Belousov wrote:
> > > On Thu, Jun 10, 2010 at 04:14:05PM +0000, Alexander Motin wrote:
> > > > Author: mav
> > > > Date: Thu Jun 10 16:14:05 2010
> > > > New Revision: 208988
> > > > URL: http://svn.freebsd.org/changeset/base/208988
> > > > 
> > > > Log:
> > > >   Store interrupt trap frame into struct thread. It allows interrupt 
> > handler
> > > >   to obtain both trap frame and opaque argument submitted on registrction.
> > > >   After kernel and all drivers get used to it, legacy hack can be removed.
> > > >   
> > > >   Reviewed by:	jhb@
> > > Just curious, why td_frame is not enough for your usage ? I believe td_frame
> > > is currently set only by traps and syscalls, and copied on forks.
> > 
> > td_frame is not set on a nested interrupt.  Thus, if you use td_frame and get 
> > an interrupt while in a syscall, statclock() would account the time as a user 
> > tick instead of a system tick.
> > 
> > Put another way, td_frame always references the user --> system
> > trapframe, but td_intr_frame will reference the most recent trapframe
> > on the stack.
> 
> I mean, save old td_frame on the interrupt, and assign interrupt trap
> frame to td_frame. I.e., do what r208988, but replace td_intr_frame
> with td_frame.

Hmmm, I had thought there were reasons that was not safe, but after looking
it may be safe to do this.

-- 
John Baldwin


More information about the svn-src-all mailing list