Re: Regression for gdb.stabs/gdb11479.exp [Re: [patch 1/2] Use custom hash function with bcache]

[sami wagiaalla <swagiaal@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 431 bytes --]

On 09/01/2010 02:37 PM, Tom Tromey wrote:
>>>>>> "Doug" == Doug Evans<dje@google.com>  writes:
>
> Doug>  One would expect the original code to have done a memset too, instead
> Doug>  of using "static".  Presumably it didn't for performance reasons.  Do
> Doug>  we know if the performance concerns were real?
>
> No, we don't know.
> It is safest to just revert to what it was before.
>

Sounds good to me. Patch attached.

Sami

[-- Attachment #2: hash_regression.patch --]
[-- Type: text/x-patch, Size: 1042 bytes --]

Fix custom hash regression.

2010-09-01  Sami Wagiaalla  <swagiaal@redhat.com>

	* psymtab.c (add_psymbol_to_bcache): Declare psymbol
	as static.
	memset psymbol.ginfo.value to 0.

diff --git a/gdb/psymtab.c b/gdb/psymtab.c
index a5d2f98..24dd301 100644
--- a/gdb/psymtab.c
+++ b/gdb/psymtab.c
@@ -1376,7 +1376,15 @@ add_psymbol_to_bcache (char *name, int namelength, int copy_name,
 		       enum language language, struct objfile *objfile,
 		       int *added)
 {
-  struct partial_symbol psymbol;
+  /* psymbol is static so that there will be no uninitialized gaps in the
+     structure which might contain random data, causing cache misses in
+     bcache. */
+  static struct partial_symbol psymbol;
+
+  /* However, we must ensure that the entire 'value' field has been
+     zeroed before assigning to it, because an assignment may not
+     write the entire field.  */
+  memset (&psymbol.ginfo.value, 0, sizeof (psymbol.ginfo.value));
 
   /* val and coreaddr are mutually exclusive, one of them *will* be zero */
   if (val != 0)