RFA: fix simulator's handling of exceptions

RFA: fix simulator's handling of exceptions

[Jim Blandy]

This patch could be totally inept, but with luck, you can see what I
have in mind.  Please clue me in.

2001-05-05  Jim Blandy  <jimb@redhat.com>

	* interp.c (program_interrupt): Don't try to emulate the
	processor's behavior when it hits a breakpoint, unless we're
	in the OPERATING_ENVIRONMENT. 

Index: sim/mn10300/interp.c
===================================================================
RCS file: /cvs/cvsfiles/devo/sim/mn10300/interp.c,v
retrieving revision 1.48
diff -c -r1.48 interp.c
*** sim/mn10300/interp.c	2000/05/29 18:53:02	1.48
--- sim/mn10300/interp.c	2001/05/05 20:58:50
***************
*** 1348,1376 ****
  
    /* avoid infinite recursion */
    if (in_interrupt)
      {
-       (*mn10300_callback->printf_filtered) (mn10300_callback, 
- 					    "ERROR: recursion in program_interrupt during software exception dispatch.");
-     }
-   else
-     {
-       in_interrupt = 1;
        /* copy NMI handler code from dv-mn103cpu.c */
        store_word (SP - 4, CIA_GET (cpu));
        store_half (SP - 8, PSW);
  
        /* Set the SYSEF flag in NMICR by backdoor method.  See
! 	 dv-mn103int.c:write_icr().  This is necessary because
           software exceptions are not modelled by actually talking to
           the interrupt controller, so it cannot set its own SYSEF
           flag. */
!      if ((NULL != board) && (strcmp(board, BOARD_AM32) == 0))
!        store_byte (0x34000103, 0x04);
!     }
  
!   PSW &= ~PSW_IE;
!   SP = SP - 8;
!   CIA_SET (cpu, 0x40000008);
  
    in_interrupt = 0;
    sim_engine_halt(sd, cpu, NULL, cia, sim_stopped, sig);
--- 1348,1382 ----
  
    /* avoid infinite recursion */
    if (in_interrupt)
+     (*mn10300_callback->printf_filtered) (mn10300_callback, 
+                                           "ERROR: recursion in program_interrupt during software exception dispatch.");
+ 
+   in_interrupt = 1;
+ 
+   /* Don't try to simulate the processor's response to the interrupt,
+      unless we're in OPERATING_ENVIRONMENT, where the sim is
+      explicitly responsible for such things.  When running under GDB,
+      we don't actually need this behavior, and if our SP is garbage,
+      the stores can raise exceptions of their own that mask the
+      underlying interrupt.  */
+   if (STATE_ENVIRONMENT (sd) == OPERATING_ENVIRONMENT)
      {
        /* copy NMI handler code from dv-mn103cpu.c */
        store_word (SP - 4, CIA_GET (cpu));
        store_half (SP - 8, PSW);
  
        /* Set the SYSEF flag in NMICR by backdoor method.  See
!          dv-mn103int.c:write_icr().  This is necessary because
           software exceptions are not modelled by actually talking to
           the interrupt controller, so it cannot set its own SYSEF
           flag. */
!       if ((NULL != board) && (strcmp(board, BOARD_AM32) == 0))
!         store_byte (0x34000103, 0x04);
  
!       PSW &= ~PSW_IE;
!       SP = SP - 8;
!       CIA_SET (cpu, 0x40000008);
!     }
  
    in_interrupt = 0;
    sim_engine_halt(sd, cpu, NULL, cia, sim_stopped, sig);

Re: RFA: fix simulator's handling of exceptions

[Andrew Cagney]

This patch could be totally inept, but with luck, you can see what I
have in mind.  Please clue me in.

2001-05-05  Jim Blandy  <jimb@redhat.com>

* interp.c (program_interrupt): Don't try to emulate the
	processor's behavior when it hits a breakpoint, unless we're
	in the OPERATING_ENVIRONMENT.

I'm not sure this is correct.  The mn10300 doesn't have an operating 
environment (user/supervisor modes).  Have a look at either the d10v or 
the tx39.

In general what happens is:

	o	simulator encounters a breakpoint

	o	simulator saves all state relevant
		to the breakpoint in case it has
		to be delivered

	o	breakpoint encountered (SIGBREAK)
		returned to GDB

	o	GDB checks to see if it is a
		breakpoint on its breakpoint list.

	o	if it is.
		Process the breakpoint and then,
		later when the target is resumed,
		pass in 0 as the signal.

	o	if it isn't
		Depending on the current state of
		SIGBREAK (according to (gdb) handle?)
		possibly resume the target with
		SIGBREAK as the signal.  The target
		simulator should then use that saved
		breakpoint state to deliver a breakpoint
		interrupt to the target.

Andrew