> From: Jason Thorpe <thorpej@nas.nasa.gov>
> Date: Mon, 24 Mar 1997 15:16:07 -0800
> 
> On Mon, 24 Mar 97 16:55:15 EST 
>  gwr@mc.com (Gordon W. Ross) wrote:
> 
>  > While tsleep() does not really do as the comment says it will,
>  > I'm not sure we want it to.  Enabling interrupts while cold is
>  > not necessarily a good idea (safer to wait until autoconfig is
>  > finished).  My hope is that with the code movement I suggested,
>  > we could change that bit of code in tsleep() to this:
>  > 
>  > 	if (cold)
>  > 		panic("tsleep cold");
> 
> Yah, I'm with Gordon on this one... IMO, the whole motion of calling
> tsleep() while cold is, at best, counter-intuitive.
> 
> Jason R. Thorpe                                       thorpej@nas.nasa.gov
> NASA Ames Research Center                               Home: 408.866.1912
> NAS: M/S 258-6                                          Work: 415.604.0935
> Moffett Field, CA 94035                                Pager: 415.428.6939


I've since tried out the code movement idea, and it works very nicely!
If I move the code that does:  findroot/setroot, swapconf, dumpconf
from where it is now into a new, separate function that is called just
before vfs_mountroot, then the kernel does not call tsleep while cold.
The code that I moved does things like read disk labels, and is more
robust against missing interrupt problems if it is done later, so that
timeouts will work.  Therefore, I propose the following changes.

Comments?
Gordon


diff -c sys/systm.h.orig sys/systm.h
*** sys/systm.h.orig	Fri Jan 31 11:05:18 1997
--- sys/systm.h	Tue Mar 25 15:33:03 1997
***************
*** 260,265 ****
--- 260,266 ----
  
  void	consinit __P((void));
  
+ void	cpu_rootconf __P((void));
  void	cpu_startup __P((void));
  void	cpu_set_kpc __P((struct proc *, void (*)(struct proc *)));
  

diff -c kern/init_main.c.orig kern/init_main.c
*** kern/init_main.c.orig	Mon Feb  3 11:11:20 1997
--- kern/init_main.c	Tue Mar 25 15:33:24 1997
***************
*** 320,325 ****
--- 320,326 ----
  	schedcpu(NULL);
  
  	/* Mount the root file system. */
+ 	cpu_rootconf();
  	do {
  		domountroothook();
  		if ((error = vfs_mountroot())) {

I just included one representative sample change based on the sun3
port here.  All ports would need a little code moved into the new
cpu_rootconf() function, something like the change shown here.

diff -c arch/sun3/sun3/autoconf.c.orig arch/sun3/sun3/autoconf.c
*** arch/sun3/sun3/autoconf.c.orig	Tue Mar 25 15:44:15 1997
--- arch/sun3/sun3/autoconf.c	Tue Mar 25 15:46:09 1997
***************
*** 87,96 ****
--- 87,106 ----
  	printf("enabling interrupts\n");
  	(void)spl0();
  
+ #if 0
  	rootconf(); 	/* local findroot/setroot code */
  	swapconf();
  	dumpconf();
+ #endif
  	cold = 0;
+ }
+ 
+ void
+ cpu_rootconf()
+ {
+ 	rootconf(); 	/* local findroot/setroot code */
+ 	swapconf();
+ 	dumpconf();
  }
  
  /****************************************************************/