Committed, sim/cris: correct interrupt-preparation sequence for v32.

Committed, sim/cris: correct interrupt-preparation sequence for v32.

[Hans-Peter Nilsson]

Stupid copy-paste-bug; when copying the interrupt-preparation code
from the (local-only) SID port, I missed the part to actually set CCS
to the calculated new value.

sim/testsuite:
	* sim/cris/hw/rv-n-cris/irq6.ms: New test.

sim:
	* cris/crisv32f.c (MY (deliver_interrupt)): Set CCS to new_ccs.

Index: sim/cris/hw/rv-n-cris/irq6.ms
===================================================================
RCS file: sim/cris/hw/rv-n-cris/irq6.ms
diff -N sim/cris/hw/rv-n-cris/irq6.ms
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ sim/cris/hw/rv-n-cris/irq6.ms	8 Apr 2006 17:31:10 -0000
@@ -0,0 +1,99 @@
+#mach: crisv10 crisv32
+#sim(crisv10): --hw-device "/rv/trace? true"
+#sim(crisv32): --hw-device "/rv/trace? true"
+#output: /rv: WD\n
+#output: /rv: REG R 0xd0000032\n
+#output: /rv: := 0xabcdef01\n
+#output: /rv: IRQ 0x4\n
+#output: /rv: REG R 0xd0000036\n
+#output: /rv: := 0x76543210\n
+#output: /rv: REG R 0xd0000030\n
+#output: /rv: IRQ 0x0\n
+#output: /rv: IRQ 0x8\n
+#output: /rv: := 0xeeff4455\n
+#output: /rv: REG R 0xd0000034\n
+#output: /rv: := 0xdd001122\n
+#output: /rv: REG R 0xd0000038\n
+#output: /rv: := 0xaaeeff44\n
+#output: /rv: REG R 0xd000003c\n
+#output: /rv: := 0xff445511\n
+#output: pass\n
+
+# Test two successive ints; that flags are disabled when an interrupt
+# is taken, and then automatically (or by register restore) enabled at
+# return.
+
+#r W,
+#r r,a8832,abcdef01
+#r I,4
+#r r,a8836,76543210
+#r I,0
+#r I,8
+#r r,a8830,eeff4455
+#r r,a8834,dd001122
+#r r,a8838,aaeeff44
+#r r,a883c,ff445511
+
+ .lcomm dummy,4
+
+ .include "testutils.inc"
+ start
+ test_h_mem 0xabcdef01 0xd0000032
+ moveq -1,$r4
+
+ .if ..asm.arch.cris.v32
+ move irqvec1,$ebp
+ .else
+ move irqvec1,$ibr
+ .endif
+
+ ei
+ test_h_mem 0,dummy
+
+ ; Here after the first interrupt, or perhaps the second interrupt is
+ ; taken directly; leave it optional.  Anyway, the second interrupt
+ ; should be taken no later than this branch.
+ test_h_mem 0,dummy
+
+killme:
+ fail
+
+irq0x33:
+ .if ..asm.arch.cris.v32
+ ; Nothing needed to save flags - "shift" should happen, and back at rfe.
+ .else
+ ; The missing sim support for interrupt-excluding instructions is matched
+ ; by the flaw that sim doesn't service interrupts in straight code.
+ ; So, we can use a sequence that would work on actual hardware.
+ move $dccr,$r5
+ di
+ .endif
+
+ test_h_mem 0x76543210 0xd0000036
+ test_h_mem 0xeeff4455 0xd0000030
+ test_h_mem 0xdd001122 0xd0000034
+ moveq -22,$r4
+
+ .if ..asm.arch.cris.v32
+ move irqvec2,$ebp
+ rete
+ rfe
+ .else
+ move irqvec2,$ibr
+ reti
+ move $r5,$dccr
+ .endif
+
+ pass
+
+irq0x34:
+ test_h_mem 0xaaeeff44 0xd0000038
+ test_h_mem 0xff445511 0xd000003c
+ cmpq -22,$r4
+ bne killme
+ nop
+ pass
+
+ singlevec irqvec1,0x33,irq0x33
+
+ singlevec irqvec2,0x34,irq0x34


Index: cris/crisv32f.c
===================================================================
RCS file: /cvs/src/src/sim/cris/crisv32f.c,v
retrieving revision 1.2
diff -p -u -r1.2 crisv32f.c
--- cris/crisv32f.c	3 Apr 2006 03:01:45 -0000	1.2
+++ cris/crisv32f.c	8 Apr 2006 17:27:49 -0000
@@ -590,6 +590,8 @@ MY (deliver_interrupt) (SIM_CPU *current
      recognize that we're in kernel mode.  */
   CPU (h_ubit_v32) = 0;
 
+  SET_H_SR (H_SR_CCS, new_ccs);
+
   if (was_user)
     {
       /* These methods require that user mode is unset.  */

brgds, H-P

Re: Committed, sim/cris: correct interrupt-preparation sequence for v32.

[Frank Ch. Eigler]

Hans-Peter Nilsson <hans-peter.nilsson@axis.com> writes:

> Stupid copy-paste-bug; when copying the interrupt-preparation code
> from the (local-only) SID port, [...]

By the way, can you release the sid port instead of / in addition to
this effort?

- FChE

Re: Committed, sim/cris: correct interrupt-preparation sequence for v32.

[Hans-Peter Nilsson]

> From: fche@redhat.com (Frank Ch. Eigler)
> Date: 09 Apr 2006 10:19:40 -0400

> Hans-Peter Nilsson <hans-peter.nilsson@axis.com> writes:
> 
> > Stupid copy-paste-bug; when copying the interrupt-preparation code
> > from the (local-only) SID port, [...]
> 
> By the way, can you release the sid port instead of / in addition to
> this effort?

IIRC I mentioned in passing that there are issues why I didn't
do that: first of all, last time I looked and asked, the
copyright of SID wasn't FSF, but Red Hat, and contributors were
asked to assign copyright of major contributions to Red Hat.

Second, CGEN-SID has incompatibly (with src/sim) changed syntax
of the "delay" CGEN construct to the effect that it doesn't
support non-parallel machines with delay-slots anymore.  We can
discuss the latter on the CGEN list; the former I think I've
heard from someone (IIRC you in person, years ago) was subject
to change.

When/if those issues are fixed, it may seem more justifiable to
put time into updating the SID port and submitting it.  Maybe
I'll make a try at the CGEN issue, but Not Right Now.

brgds, H-P

Re: Committed, sim/cris: correct interrupt-preparation sequence for v32.

[Frank Ch. Eigler]

Hi -

On Sun, Apr 09, 2006 at 07:00:25PM +0200, Hans-Peter Nilsson wrote:
> [...]
> IIRC I mentioned in passing that there are issues why I didn't
> do that: first of all, last time I looked and asked, the
> copyright of SID wasn't FSF, but Red Hat, and contributors were
> asked to assign copyright of major contributions to Red Hat.

SID contributions are not copyright-assigned to Red Hat, though CGEN
makes erroneous guesses about this in its generated files.

> Second, CGEN-SID has incompatibly (with src/sim) changed syntax of
> the "delay" CGEN construct to the effect that it doesn't support
> non-parallel machines with delay-slots anymore.  [...]

OK.  I didn't realize that you considered this a show-stopper.

- FChE