On Fri, Jul 21, 2006 at 12:33:15 +0900, Masao Uebayashi wrote:

> A co-worker found a problem that system calls in kernel mode are
> running in the highest interrupt level (SR:I[3-0] == 1111).

Oops.


> Can't they safely run in the lowest interrupt level?

They should.


> Index: arch/sh3/sh3/exception_vector.S
> ===================================================================
> RCS file: /cvsroot/src/sys/arch/sh3/sh3/exception_vector.S,v
> retrieving revision 1.18
> diff -u -r1.18 exception_vector.S
> --- arch/sh3/sh3/exception_vector.S	8 Feb 2006 02:19:57 -0000	1.18
> +++ arch/sh3/sh3/exception_vector.S	21 Jul 2006 03:23:31 -0000
> @@ -104,6 +104,7 @@
>  	mov.l	r2,	@(TF_UBC, r14)	/* clear trapframe->tf_ubc */
>  #endif /* DDB */
>  	__EXCEPTION_UNBLOCK(r0, r1)
> +	__INTR_UNMASK(r0, r1)
>  	mov.l	_L.general, r0
>  	jsr	@r0
>  	 mov	r14,	r5	/* 2nd arg */

I think that's not a correct fix.  We can hit general_excepion() from
kernel code, so we cannot set sr.imask to zero in that place.

general_excepion() should call spl0() if the trap is coming from the
user mode.

I'm trying to wrap my head around this code.

SY, Uwe
-- 
uwe@ptc.spbu.ru                         |       Zu Grunde kommen
http://snark.ptc.spbu.ru/~uwe/          |       Ist zu Grunde gehen