gdb 7.6: Fix info mem command for 32 bits host/64 bits target

gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Udma Catalin-Dan-B32721]

Hi,
I found an issue with the "info mem" command when running gdb on 32 bits host for 64 bits target architecture.
"info mem" command truncates the target address to 32 bits, like in the example below:

    (gdb) set architecture powerpc:common64
    The target architecture is assumed to be powerpc:common64
    (gdb) mem 0x100000000 0x200000000 rw
    (gdb) info mem
    Using user-defined memory regions.
    Num Enb Low Addr           High Addr          Attrs
    1   y   0x0000000000000000 0x0000000000000000 rw nocache

Please find below the proposed patch. Please let me know your comments.

Regards,
Catalin


Subject: [PATCHH] Fix info mem command for 32 bits host/64 bits target

When running gdb on 32 bits host for 64 bits target, info mem command
truncates the target address to 32 bits, like in the example below
	(gdb) set architecture powerpc:common64 
	(gdb) mem 0x100000000 0x200000000 rw
	(gdb) info mem
	1   y  	0x0000000000000000 0x0000000000000000 rw nocache 

Signed-off-by: Catalin Udma <catalin.udma@freescale.com>
---
--- gdb/memattr.c.orig	2013-06-27 17:28:20.000000000 +0300
+++ gdb/memattr.c	2013-06-27 17:29:16.000000000 +0300
@@ -447,9 +447,9 @@
 		       m->number,
 		       m->enabled_p ? 'y' : 'n');
       if (gdbarch_addr_bit (target_gdbarch ()) <= 32)
-	tmp = hex_string_custom ((unsigned long) m->lo, 8);
+	tmp = hex_string_custom ((LONGEST) m->lo, 8);
       else
-	tmp = hex_string_custom ((unsigned long) m->lo, 16);
+	tmp = hex_string_custom ( (LONGEST) m->lo, 16);
       
       printf_filtered ("%s ", tmp);
 
@@ -458,14 +458,14 @@
 	  if (m->hi == 0)
 	    tmp = "0x100000000";
 	  else
-	    tmp = hex_string_custom ((unsigned long) m->hi, 8);
+	    tmp = hex_string_custom ((LONGEST) m->hi, 8);
 	}
       else
 	{
 	  if (m->hi == 0)
 	    tmp = "0x10000000000000000";
 	  else
-	    tmp = hex_string_custom ((unsigned long) m->hi, 16);
+	    tmp = hex_string_custom ((LONGEST) m->hi, 16);
 	}
 
       printf_filtered ("%s ", tmp);

Re: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Pedro Alves]

On 06/26/2013 04:07 PM, Udma Catalin-Dan-B32721 wrote:
> Hi,
> I found an issue with the "info mem" command when running gdb on 32 bits host for 64 bits target architecture.
> "info mem" command truncates the target address to 32 bits, like in the example below:
> 
>     (gdb) set architecture powerpc:common64
>     The target architecture is assumed to be powerpc:common64
>     (gdb) mem 0x100000000 0x200000000 rw
>     (gdb) info mem
>     Using user-defined memory regions.
>     Num Enb Low Addr           High Addr          Attrs
>     1   y   0x0000000000000000 0x0000000000000000 rw nocache
> 
> Please find below the proposed patch. Please let me know your comments.

Thanks!  The (LONGEST) casts look unnecessary, as
hex_string_custom already takes LONGEST.

This needs a ChangeLog entry.  See
<http://sourceware.org/gdb/wiki/ContributionChecklist>.  Could
you send one?

In order to backport to the 7.6 branch, we'll need a bugzilla entry
filed, reporting the bug.  Could you do that?  That will be used
to record the bug fix in <http://sourceware.org/gdb/wiki/GDB_7.6_Release>.

-- 
Pedro Alves

Re: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Pedro Alves]

BTW,

$ grep hex_string_custom * | grep long

finds:

breakpoint.c:  /* The longest string possibly returned by hex_string_custom
breakpoint.c:  strcpy (astr1, hex_string_custom ((unsigned long) from_addr, 8));
breakpoint.c:  strcpy (astr2, hex_string_custom ((unsigned long) to_addr, 8));

Those casts are likewise bogus:

static void
breakpoint_adjustment_warning (CORE_ADDR from_addr, CORE_ADDR to_addr,
                               int bnum, int have_bnum)
{
  /* The longest string possibly returned by hex_string_custom
     is 50 chars.  These must be at least that big for safety.  */
  char astr1[64];
  char astr2[64];

  strcpy (astr1, hex_string_custom ((unsigned long) from_addr, 8));
  strcpy (astr2, hex_string_custom ((unsigned long) to_addr, 8));
  if (have_bnum)
    warning (_("Breakpoint %d address previously adjusted from %s to %s."),
             bnum, astr1, astr2);
  else
    warning (_("Breakpoint address adjusted from %s to %s."), astr1, astr2);
}

-- 
Pedro Alves

Re: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Pedro Alves]

On 06/26/2013 05:07 PM, Pedro Alves wrote:
> BTW,
> 
> $ grep hex_string_custom * | grep long
> 
> finds:
> 
> breakpoint.c:  /* The longest string possibly returned by hex_string_custom
> breakpoint.c:  strcpy (astr1, hex_string_custom ((unsigned long) from_addr, 8));
> breakpoint.c:  strcpy (astr2, hex_string_custom ((unsigned long) to_addr, 8));
> 
> Those casts are likewise bogus:

Filed <http://sourceware.org/bugzilla/show_bug.cgi?id=15683>.

-- 
Pedro Alves

RE: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Udma Catalin-Dan-B32721]

> Thanks!  The (LONGEST) casts look unnecessary, as
> hex_string_custom already takes LONGEST.
> 
> This needs a ChangeLog entry.  See
> <http://sourceware.org/gdb/wiki/ContributionChecklist>.  Could
> you send one?
> 
> In order to backport to the 7.6 branch, we'll need a bugzilla entry
> filed, reporting the bug.  Could you do that?  That will be used
> to record the bug fix in
> <http://sourceware.org/gdb/wiki/GDB_7.6_Release>.
> 
> --
> Pedro Alves
[CU] Sure, I'll update the patch and the ChangeLog. And I'll open the bug for this.
[CU] Catalin

Re: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Tom Tromey]

Pedro> static void
Pedro> breakpoint_adjustment_warning (CORE_ADDR from_addr, CORE_ADDR to_addr,
Pedro>                                int bnum, int have_bnum)
Pedro> {
Pedro>   /* The longest string possibly returned by hex_string_custom
Pedro>      is 50 chars.  These must be at least that big for safety.  */
Pedro>   char astr1[64];
Pedro>   char astr2[64];

Pedro>   strcpy (astr1, hex_string_custom ((unsigned long) from_addr, 8));
Pedro>   strcpy (astr2, hex_string_custom ((unsigned long) to_addr, 8));

These strcpy calls are also bogus.
hex_string_custom uses the "cell" mechanism, so this code can assume
that these calls don't interfere.

Tom

RE: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Udma Catalin-Dan-B32721]

Hi,
Re-sending the patch including: bug entry, changeLog and the patch.

Regards,
Catalin


Bug entry:
http://sourceware.org/bugzilla/show_bug.cgi?id=15684 


gdb/ChangeLog:

2013-06-26  Catalin Udma  <catalin.udma@freescale.com>

	* memattr.c (mem_info_command): Fix info mem command
	for 32 bits host/64 bits target by removing incorrect cast


Patch:

Subject: [PATCH] Fix info mem command for 32 bits host/64 bits target

When running gdb on 32 bits host for 64 bits target, info mem command
truncates the target address to 32 bits, like in the example below
	(gdb) set architecture powerpc:common64 
	(gdb) mem 0x100000000 0x200000000 rw
	(gdb) info mem
	1   y  	0x0000000000000000 0x0000000000000000 rw nocache 

Signed-off-by: Catalin Udma <catalin.udma@freescale.com>
---
--- gdb/memattr.c.orig	2013-06-27 17:28:20.000000000 +0300
+++ gdb/memattr.c	2013-06-27 17:29:16.000000000 +0300
@@ -447,9 +447,9 @@
 		       m->number,
 		       m->enabled_p ? 'y' : 'n');
       if (gdbarch_addr_bit (target_gdbarch ()) <= 32)
-	tmp = hex_string_custom ((unsigned long) m->lo, 8);
+	tmp = hex_string_custom (m->lo, 8);
       else
-	tmp = hex_string_custom ((unsigned long) m->lo, 16);
+	tmp = hex_string_custom (m->lo, 16);
       
       printf_filtered ("%s ", tmp);
 
@@ -458,14 +458,14 @@
 	  if (m->hi == 0)
 	    tmp = "0x100000000";
 	  else
-	    tmp = hex_string_custom ((unsigned long) m->hi, 8);
+	    tmp = hex_string_custom (m->hi, 8);
 	}
       else
 	{
 	  if (m->hi == 0)
 	    tmp = "0x10000000000000000";
 	  else
-	    tmp = hex_string_custom ((unsigned long) m->hi, 16);
+	    tmp = hex_string_custom (m->hi, 16);
 	}
 
       printf_filtered ("%s ", tmp);

Re: gdb 7.6: Fix info mem command for 32 bits host/64 bits target

[Pedro Alves]

On 06/27/2013 09:19 AM, Udma Catalin-Dan-B32721 wrote:
> Hi,
> Re-sending the patch including: bug entry, changeLog and the patch.
> 
> Regards,
> Catalin
> 
> 
> Bug entry:
> http://sourceware.org/bugzilla/show_bug.cgi?id=15684 
> 
> 
> gdb/ChangeLog:
> 
> 2013-06-26  Catalin Udma  <catalin.udma@freescale.com>
> 
> 	* memattr.c (mem_info_command): Fix info mem command
> 	for 32 bits host/64 bits target by removing incorrect cast

ChangeLog entries say "what" changed.  It also includes the PR
associated with the patch.  I.e., write:

2013-06-26  Catalin Udma  <catalin.udma@freescale.com>

	PR gdb/15684
 	* memattr.c (mem_info_command): Remove "unsigned long" casts.

OK with that change.

Thanks,
-- 
Pedro Alves