Uninitialised Variable in symfile.c

Uninitialised Variable in symfile.c

[Steven Johnson]

[-- Attachment #1: Type: text/plain, Size: 1041 bytes --]

There is a codepath is "load_section_callback" where the variable "err" 
will not end up being assigned a value, and hence will default to 
whatever happened to be on the stack at its location.  And in my case, 
it was defaulting to a vlaue which signaled an error, when none had occured.

This causes the following problem:

When attempting a "load" to a remote target, "load" fails after first 
packet is successfuly transfered to the target with:

Loading section .text, size 0xdc220 lma 0x0
Sending packet: $X0,400:.....
[$][O][K][#][9][a]Packet received: OK
remote:target_xfer_partial (2, (null), 0x0,  0x40ed9008,  0x0, 1024) = 
1024, bytes = 48 00 40 00 41 50 50 20 ...
Memory access error while loading section .text.

The call to "target_write_memory_partial" will not set "err" if it uses 
"target_xfer_partial" to do the memory transfer, as 
"target_xfer_partial" does not take "err" as a parameter.

The attached patch fixes this, by simply defaulting "err" to 0, the OK 
state. 

This is the subject of PR# 1944

Steven Johnson

[-- Attachment #2: gdb-6.3-load-remote-target-bug.patch --]
[-- Type: text/x-patch, Size: 469 bytes --]

diff -Naur gdb-6.3/gdb/symfile.c gdb-6.3-modified/gdb/symfile.c
--- gdb-6.3/gdb/symfile.c	2004-09-30 23:23:09.000000000 -1100
+++ gdb-6.3-modified/gdb/symfile.c	2005-05-14 21:01:18.959321653 -1100
@@ -1405,7 +1405,7 @@
 	  struct cleanup *old_chain;
 	  CORE_ADDR lma = bfd_section_lma (abfd, asec) + args->load_offset;
 	  bfd_size_type block_size;
-	  int err;
+	  int err = 0;
 	  const char *sect_name = bfd_get_section_name (abfd, asec);
 	  bfd_size_type sent;
 

Re: Uninitialised Variable in symfile.c

[Daniel Jacobowitz]

On Mon, May 16, 2005 at 10:24:18AM -1100, Steven Johnson wrote:
> There is a codepath is "load_section_callback" where the variable "err" 
> will not end up being assigned a value, and hence will default to 
> whatever happened to be on the stack at its location.  And in my case, 
> it was defaulting to a vlaue which signaled an error, when none had occured.

This is already fixed in CVS, several months ago.

-- 
Daniel Jacobowitz
CodeSourcery, LLC