HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Andrew Cagney]

Hello,

In checking various things to do with dummy frames, I noticed this 
potential foo-bar.

I recently changed the default for CALL_DUMMY_LOCATION from ON_STACK to 
AT_ENTRY.  I suspect the HP/PA 64 currently assumes ON_STACK.

Anyone able to check this theory?

Andrew

Re: HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Joel Brobecker]

> I recently changed the default for CALL_DUMMY_LOCATION from ON_STACK to 
> AT_ENTRY.  I suspect the HP/PA 64 currently assumes ON_STACK.
> 
> Anyone able to check this theory?

I'll try to have a look today (thanks for this notice!).

-- 
Joel

Re: HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Andrew Cagney]

>> I recently changed the default for CALL_DUMMY_LOCATION from ON_STACK to 
>> AT_ENTRY.  I suspect the HP/PA 64 currently assumes ON_STACK.
>> 
>> Anyone able to check this theory?
> 
> 
> I'll try to have a look today (thanks for this notice!).

I'd better explain why I think I broke it :-/  I believe the old code 
had the sequence:

tm-hppa64.h:
	#include "tm-hppa.h"
		tm-hppa.h:
		#define CALL_DUMMY_LOCATION ON_STACK
	#undef CALL_DUMMY_LOCATION

gdbarch.h:
	#if !multi-arch
	// do nothing with CALL_DUMMY_LOCATION
	#endif

inferior.h:
	#ifndef CALL_DUMMY_LOCATION
	#define CALL_DUMMY_LOCATION ON_STACK
	#endif

My change effectively changed the middle one so that CALL_DUMMY_LOCATION 
defaults to AT_ENTRY_POINT - good for everyone else, just not for 64 bit 
HP/UX.

Andrew

Re: HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Joel Brobecker]

[-- Attachment #1: Type: text/plain, Size: 1122 bytes --]

Andrew,

I finally had a look at this problem.

> >I'll try to have a look today (thanks for this notice!).
> 
> I'd better explain why I think I broke it :-/  I believe the old code 
> had the sequence:
> 
> tm-hppa64.h:
> 	#include "tm-hppa.h"
> 		tm-hppa.h:
> 		#define CALL_DUMMY_LOCATION ON_STACK
> 	#undef CALL_DUMMY_LOCATION

I believe you are right. I did a bit of archeology, but could not
find the purpose of this undef.

Before you added the '#define CALL_DUMMY_LOCATION ON_STACK' when you
changed the default value for CALL_DUMMY_LOCATION, there was no such
defined in tm-hppa.h... So the undef was useless, and now seems to
cause some breakage on hppa64.

Short of understanding completely the issues at hand, I simply ran the
testsuite with and without the #undef, and found that that results are
sensibly better without (ie, to my relief, you guess right about hppa64).

I have the following change ready:

2003-01-28  J. Brobecker  <brobecker@gnat.com>

	* config/pa/tm-hppa64.h (CALL_DUMMY_LOCATION): Remove #undef
	causing some regressions due to a change in the default value
	for this macro.

-- 
Joel

[-- Attachment #2: tm-hppa64.h --]
[-- Type: text/plain, Size: 525 bytes --]

Index: config/pa/tm-hppa64.h
===================================================================
RCS file: /cvs/src/src/gdb/config/pa/tm-hppa64.h,v
retrieving revision 1.6
diff -c -3 -p -r1.6 tm-hppa64.h
*** config/pa/tm-hppa64.h	14 Nov 2002 00:25:05 -0000	1.6
--- config/pa/tm-hppa64.h	28 Jan 2003 04:12:36 -0000
*************** call_dummy
*** 217,223 ****
  #undef FUNC_LDO_OFFSET
  #undef SR4EXPORT_LDIL_OFFSET
  #undef SR4EXPORT_LDO_OFFSET
- #undef CALL_DUMMY_LOCATION
  
  #undef REG_STRUCT_HAS_ADDR
  
--- 217,222 ----

Re: HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Andrew Cagney]

> I believe you are right. I did a bit of archeology, but could not
> find the purpose of this undef.
> 
> Before you added the '#define CALL_DUMMY_LOCATION ON_STACK' when you
> changed the default value for CALL_DUMMY_LOCATION, there was no such
> defined in tm-hppa.h... So the undef was useless, and now seems to
> cause some breakage on hppa64.
> 
> Short of understanding completely the issues at hand, I simply ran the
> testsuite with and without the #undef, and found that that results are
> sensibly better without (ie, to my relief, you guess right about hppa64).
> 
> I have the following change ready:
> 
> 2003-01-28  J. Brobecker  <brobecker@gnat.com>
> 
> 	* config/pa/tm-hppa64.h (CALL_DUMMY_LOCATION): Remove #undef
> 	causing some regressions due to a change in the default value
> 	for this macro.

Thanks!  Definitly check this in.

Andrew

Re: HP/UX 64 need CALL_DUMMY_LOCATION ON_STACK?

[Joel Brobecker]

> >2003-01-28  J. Brobecker  <brobecker@gnat.com>
> >
> >	* config/pa/tm-hppa64.h (CALL_DUMMY_LOCATION): Remove #undef
> >	causing some regressions due to a change in the default value
> >	for this macro.
> 
> Thanks!  Definitly check this in.

This is now checked in.

-- 
Joel