Re: [commited] small changes to fix hpux-cc compile

Re: [commited] small changes to fix hpux-cc compile

[Michael Elizabeth Chastain]

> Yes, this was indeed my thinking and why i hardcoded it.

So what happens if your hardcoded auto array of size 4 ever overflows?
We get a nasty bit of stack corruption.  I hate seeing code like this
creep into gdb.  It works until something else changes and then it
quietly crashes or prints random bad results.

The reason this code stopped compiling is that TARGET_INT_BIT became
more dynamic; it changed from a manifest constant to a function call.

  char dld_flags_buffer[4];
  ...
  status = target_read_memory (addr, dld_flags_buffer, TARGET_INT_BIT / TARGET_CHAR_BIT);
  
That's just wrong!

Michael C

Re: [commited] small changes to fix hpux-cc compile

[Randolph Chung]

> So what happens if your hardcoded auto array of size 4 ever overflows?
> We get a nasty bit of stack corruption.  I hate seeing code like this
> creep into gdb.  It works until something else changes and then it
> quietly crashes or prints random bad results.

ok, how about like this then? this is consistent with how __dld_flags
isused in other parts of somsolib.c

2004-05-16  Randolph Chung  <tausq@debian.org>

	* somsolib.c (som_solib_remove_inferior_hook): Make all the size
	references to dld_flags_buffer consistent.

Index: somsolib.c
===================================================================
RCS file: /cvs/src/src/gdb/somsolib.c,v
retrieving revision 1.33
diff -u -p -r1.33 somsolib.c
--- somsolib.c	16 May 2004 04:33:41 -0000	1.33
+++ somsolib.c	16 May 2004 17:19:04 -0000
@@ -1081,16 +1081,13 @@ som_solib_remove_inferior_hook (int pid)
   msymbol = lookup_minimal_symbol ("__dld_flags", NULL, NULL);
 
   addr = SYMBOL_VALUE_ADDRESS (msymbol);
-  status = target_read_memory (addr, dld_flags_buffer, TARGET_INT_BIT / TARGET_CHAR_BIT);
+  status = target_read_memory (addr, dld_flags_buffer, 4);
 
-  dld_flags_value = extract_unsigned_integer (dld_flags_buffer,
-					      sizeof (dld_flags_value));
+  dld_flags_value = extract_unsigned_integer (dld_flags_buffer, 4);
 
   dld_flags_value &= ~DLD_FLAGS_HOOKVALID;
-  store_unsigned_integer (dld_flags_buffer,
-			  sizeof (dld_flags_value),
-			  dld_flags_value);
-  status = target_write_memory (addr, dld_flags_buffer, TARGET_INT_BIT / TARGET_CHAR_BIT);
+  store_unsigned_integer (dld_flags_buffer, 4, dld_flags_value);
+  status = target_write_memory (addr, dld_flags_buffer, 4);
 
   do_cleanups (old_cleanups);
 }

-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

Re: [commited] small changes to fix hpux-cc compile

[Michael Elizabeth Chastain]

Yes, this would make me happy, because it's consistent and there's
no chance of memory corruption.  Thank you for this.

Michael C

===

2004-05-16  Randolph Chung  <tausq@debian.org>

	* somsolib.c (som_solib_remove_inferior_hook): Make all the size
	references to dld_flags_buffer consistent.

Re: [commited] small changes to fix hpux-cc compile

[Michael Elizabeth Chastain]

Hi Randolph,

In this change:

  -  char dld_flags_buffer[TARGET_INT_BIT / TARGET_CHAR_BIT];
  +  char dld_flags_buffer[4];

I just ran into the same problem with TARGET_INT_BIT so I understand
why this needs to change.  But why 4?  What if the target configuration
has 64-bit ints?

How about this instead:

  char * dld_flags_buffer = alloca(TARGET_INT_BIT/TARGET_CHAR_BIT);

Michael C

Re: [commited] small changes to fix hpux-cc compile

[Mark Kettenis]

   Date: Sun, 16 May 2004 06:12:23 -0400 (EDT)
   From: mec.gnu@mindspring.com (Michael Elizabeth Chastain)

   Hi Randolph,

   In this change:

     -  char dld_flags_buffer[TARGET_INT_BIT / TARGET_CHAR_BIT];
     +  char dld_flags_buffer[4];

   I just ran into the same problem with TARGET_INT_BIT so I understand
   why this needs to change.  But why 4?  What if the target configuration
   has 64-bit ints?

The question should be: will there ever be a target that uses SOM that
doesn't use 32-bit ints?  I think we can be fairly certain that there
won't be such target in the future.  HP already has abandoned SOM in
favour of ELF of 64-bit HP-UX, and even that still has 32-bit ints.
As a matter of fact, I'm not aware of any ABI that has 64-bit ints.
But in general:

   How about this instead:

     char * dld_flags_buffer = alloca(TARGET_INT_BIT/TARGET_CHAR_BIT);

This is indeed the right approach.

Mark

Re: [commited] small changes to fix hpux-cc compile

[Randolph Chung]

> The question should be: will there ever be a target that uses SOM that
> doesn't use 32-bit ints?  I think we can be fairly certain that there
> won't be such target in the future.  HP already has abandoned SOM in
> favour of ELF of 64-bit HP-UX, and even that still has 32-bit ints.
> As a matter of fact, I'm not aware of any ABI that has 64-bit ints.

Yes, this was indeed my thinking and why i hardcoded it.

> But in general:
> 
>    How about this instead:
> 
>      char * dld_flags_buffer = alloca(TARGET_INT_BIT/TARGET_CHAR_BIT);
> 
> This is indeed the right approach.

actually in the code:

  unsigned int dld_flags_value;
[...]
  dld_flags_value = extract_unsigned_integer (dld_flags_buffer,
                                              sizeof (dld_flags_value));

so the code does not anyway allow dld_flags_value to be 64-bit. (i don't
think anyone ever does 64-bit int's, right?)

randolph
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/

[commited] small changes to fix hpux-cc compile

[Randolph Chung]

Some small changes were needed to get gdb to compile with the HPUX ANSI
C compiler.

Committed as obvious.

thanks,
randolph

2004-05-15  Randolph Chung  <tausq@debian.org>

	* hppa-tdep.h (hppa_frame_prev_register_helper): Pass save_regs
	as pointer instead of array reference since HPUX compiler does
	not accept unsized array arguments.
	* somsolib.c (dld_flags_buffer): Use constant array size.

Index: hppa-tdep.h
===================================================================
RCS file: /cvs/src/src/gdb/hppa-tdep.h,v
retrieving revision 1.8
diff -u -p -r1.8 hppa-tdep.h
--- hppa-tdep.h	9 May 2004 20:56:41 -0000	1.8
+++ hppa-tdep.h	16 May 2004 04:26:38 -0000
@@ -194,7 +194,7 @@ int hppa_sign_extend (unsigned int, unsi
 
 void
 hppa_frame_prev_register_helper (struct frame_info *next_frame,
-			         struct trad_frame_saved_reg saved_regs[],
+			         struct trad_frame_saved_reg *saved_regs,
 				 int regnum, int *optimizedp,
 				 enum lval_type *lvalp, CORE_ADDR *addrp,
 				 int *realnump, void *valuep);
Index: somsolib.c
===================================================================
RCS file: /cvs/src/src/gdb/somsolib.c,v
retrieving revision 1.32
diff -u -p -r1.32 somsolib.c
--- somsolib.c	7 May 2004 05:48:49 -0000	1.32
+++ somsolib.c	16 May 2004 04:26:38 -0000
@@ -1064,7 +1064,7 @@ som_solib_remove_inferior_hook (int pid)
   CORE_ADDR addr;
   struct minimal_symbol *msymbol;
   int status;
-  char dld_flags_buffer[TARGET_INT_BIT / TARGET_CHAR_BIT];
+  char dld_flags_buffer[4];
   unsigned int dld_flags_value;
   struct cleanup *old_cleanups = save_inferior_ptid ();
 
-- 
Randolph Chung
Debian GNU/Linux Developer, hppa/ia64 ports
http://www.tausq.org/