[patch,rfc,6] Revert to legacy when REGISTER_BYTE

[patch,rfc,6] Revert to legacy when REGISTER_BYTE

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 468 bytes --]

Sigh,

This patch changes the regcache so that, when REGISTER_BYTE or 
REGISTER_RAW_SIZE are present, the old style legacy regcache layout is used.

This will stop those regcache assertion failures.  Fixing the problem 
(especially for MIPS) will just get too messy for the branch :-(

My thought is to commit this to 6.0 branch, while on the mainline delete 
REGISTER_BYTE and REGISTER_RAW_SIZE and just wear the consequences ...

comments and/or suggestions?
Andrew

[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 3198 bytes --]

2003-06-29  Andrew Cagney  <cagney@redhat.com>

	* gdbarch.sh (REGISTER_RAW_SIZE, REGISTER_VIRTUAL_SIZE): Add
	predicate.
	* gdbarch.h, gdbarch.c: Re-generate.
	* regcache.c (init_regcache_descr): Use legacy code when either
	REGISTER_BYTE or REGISTER_RAW_SIZE is set.

Index: gdbarch.sh
===================================================================
RCS file: /cvs/src/src/gdb/gdbarch.sh,v
retrieving revision 1.253
diff -u -r1.253 gdbarch.sh
--- gdbarch.sh	26 Jun 2003 17:18:42 -0000	1.253
+++ gdbarch.sh	29 Jun 2003 20:27:51 -0000
@@ -486,12 +486,12 @@
 # sizes agree with the value computed from REGISTER_TYPE,
 # DEPRECATED_REGISTER_RAW_SIZE can be deleted.  See: maint print
 # registers.
-f:2:REGISTER_RAW_SIZE:int:deprecated_register_raw_size:int reg_nr:reg_nr::generic_register_size:generic_register_size::0
+F:2:REGISTER_RAW_SIZE:int:deprecated_register_raw_size:int reg_nr:reg_nr::generic_register_size:generic_register_size
 # If all registers have identical raw and virtual sizes and those
 # sizes agree with the value computed from REGISTER_TYPE,
 # DEPRECATED_REGISTER_VIRTUAL_SIZE can be deleted.  See: maint print
 # registers.
-f:2:REGISTER_VIRTUAL_SIZE:int:deprecated_register_virtual_size:int reg_nr:reg_nr::generic_register_size:generic_register_size::0
+F:2:REGISTER_VIRTUAL_SIZE:int:deprecated_register_virtual_size:int reg_nr:reg_nr::generic_register_size:generic_register_size
 # DEPRECATED_MAX_REGISTER_RAW_SIZE can be deleted.  It has been
 # replaced by the constant MAX_REGISTER_SIZE.
 V:2:DEPRECATED_MAX_REGISTER_RAW_SIZE:int:deprecated_max_register_raw_size
Index: regcache.c
===================================================================
RCS file: /cvs/src/src/gdb/regcache.c,v
retrieving revision 1.88
diff -u -r1.88 regcache.c
--- regcache.c	16 Jun 2003 13:19:36 -0000	1.88
+++ regcache.c	29 Jun 2003 20:28:07 -0000
@@ -178,16 +178,17 @@
 
   /* If an old style architecture, fill in the remainder of the
      register cache descriptor using the register macros.  */
-  if (!gdbarch_pseudo_register_read_p (gdbarch)
-      && !gdbarch_pseudo_register_write_p (gdbarch)
-      && !gdbarch_register_type_p (gdbarch))
+  /* NOTE: cagney/2003-06-29: If either of REGISTER_BYTE or
+     REGISTER_RAW_SIZE are still present, things are most likely
+     totally screwed.  Ex: an architecture with raw register sizes
+     smaller than what REGISTER_BYTE indicates; non monotonic
+     REGISTER_BYTE values.  For GDB 6 check for these nasty methods
+     and fall back to legacy code when present.  Sigh!  */
+  if ((!gdbarch_pseudo_register_read_p (gdbarch)
+       && !gdbarch_pseudo_register_write_p (gdbarch)
+       && !gdbarch_register_type_p (gdbarch))
+      || REGISTER_BYTE_P () || REGISTER_RAW_SIZE_P ())
     {
-      /* NOTE: cagney/2003-05-02: Don't add a test for REGISTER_BYTE_P
-	 to the above.  Doing that would cause all the existing
-	 architectures to revert back to the legacy regcache
-	 mechanisms, and that is not a good thing.  Instead just,
-	 later, check that the register cache's layout is consistent
-	 with REGISTER_BYTE.  */
       descr->legacy_p = 1;
       init_legacy_regcache_descr (gdbarch, descr);
       return descr;

Re: [patch,rfc,6] Revert to legacy when REGISTER_BYTE

[Daniel Jacobowitz]

On Sun, Jun 29, 2003 at 04:43:40PM -0400, Andrew Cagney wrote:
> Sigh,
> 
> This patch changes the regcache so that, when REGISTER_BYTE or 
> REGISTER_RAW_SIZE are present, the old style legacy regcache layout is used.
> 
> This will stop those regcache assertion failures.  Fixing the problem 
> (especially for MIPS) will just get too messy for the branch :-(
> 
> My thought is to commit this to 6.0 branch, while on the mainline delete 
> REGISTER_BYTE and REGISTER_RAW_SIZE and just wear the consequences ...
> 
> comments and/or suggestions?
> Andrew

Makes sense to me.  You've been making steady progress on this problem,
and it looks like it's just finally time to kill REGISTER_BYTE.


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: [patch,rfc,6] Revert to legacy when REGISTER_BYTE

[Andrew Cagney]

> On Sun, Jun 29, 2003 at 04:43:40PM -0400, Andrew Cagney wrote:
> 
>> Sigh,
>> 
>> This patch changes the regcache so that, when REGISTER_BYTE or 
>> REGISTER_RAW_SIZE are present, the old style legacy regcache layout is used.
>> 
>> This will stop those regcache assertion failures.  Fixing the problem 
>> (especially for MIPS) will just get too messy for the branch :-(
>> 
>> My thought is to commit this to 6.0 branch, while on the mainline delete 
>> REGISTER_BYTE and REGISTER_RAW_SIZE and just wear the consequences ...
>> 
>> comments and/or suggestions?
>> Andrew
> 
> 
> Makes sense to me.  You've been making steady progress on this problem,
> and it looks like it's just finally time to kill REGISTER_BYTE.

If I delete them, I think the MIPS nat code will break.  There is an 
assumption that, at least on IRIX, when a 32 bit ABI on a 64 bit ISA, 
all the registers are 32 bit left aligned within the 64 bit field (I 
think they should be sign extended).

:-(

Andrew

Re: [patch,rfc,6] Revert to legacy when REGISTER_BYTE

[Andrew Cagney]

> Sigh,
> 
> This patch changes the regcache so that, when REGISTER_BYTE or REGISTER_RAW_SIZE are present, the old style legacy regcache layout is used.
> 
> This will stop those regcache assertion failures.  Fixing the problem (especially for MIPS) will just get too messy for the branch :-(
> 
> My thought is to commit this to 6.0 branch, while on the mainline delete REGISTER_BYTE and REGISTER_RAW_SIZE and just wear the consequences ...
> 
> comments and/or suggestions?

(none against)

> 2003-06-29  Andrew Cagney  <cagney@redhat.com>
> 
> 	* gdbarch.sh (REGISTER_RAW_SIZE, REGISTER_VIRTUAL_SIZE): Add
> 	predicate.
> 	* gdbarch.h, gdbarch.c: Re-generate.
> 	* regcache.c (init_regcache_descr): Use legacy code when either
> 	REGISTER_BYTE or REGISTER_RAW_SIZE is set.

I've checked this in to the 6.0 branch (and for the sake of consistency, 
the trunk).

Andrew