binutils bug workaround

binutils bug workaround

[Vivien Chappelier]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 489 bytes --]

Hi,

	There is a binutils bug (as) which produces bad addresses when
getting the address of a struct member for initializing the same struct,
and when there is data or static functions declared before:
int test = 0xdeadbeef;

struct {
        int dummy;
        void *ptr;
} bug =
{
  ptr:  &bug.ptr
};

will produce bad value for bug.ptr.

	This patch move the declaration of kswapd_wait as a workaround to
this compiler bug. This probably affects all mips64 kernels.

Vivien Chappelier.

[-- Attachment #2: Type: TEXT/plain, Size: 554 bytes --]

diff -Naur linux/mm/vmscan.c linux.patch/mm/vmscan.c
--- linux/mm/vmscan.c	Sun Dec  9 15:53:15 2001
+++ linux.patch/mm/vmscan.c	Wed Jan  2 23:16:22 2002
@@ -24,6 +24,8 @@
 
 #include <asm/pgalloc.h>
 
+DECLARE_WAIT_QUEUE_HEAD(kswapd_wait);
+
 /*
  * The "priority" of VM scanning is how much of the queues we
  * will scan in one go. A value of 6 for DEF_PRIORITY implies
@@ -601,8 +603,6 @@
 	out_of_memory();
 	return 0;
 }
-
-DECLARE_WAIT_QUEUE_HEAD(kswapd_wait);
 
 static int check_classzone_need_balance(zone_t * classzone)
 {

Re: binutils bug workaround

[H . J . Lu]

On Thu, Jan 03, 2002 at 10:44:30PM +0100, Vivien Chappelier wrote:
> Hi,
> 
> 	There is a binutils bug (as) which produces bad addresses when
> getting the address of a struct member for initializing the same struct,
> and when there is data or static functions declared before:
> int test = 0xdeadbeef;
> 
> struct {
>         int dummy;
>         void *ptr;
> } bug =
> {
>   ptr:  &bug.ptr
> };
> 
> will produce bad value for bug.ptr.
> 
> 	This patch move the declaration of kswapd_wait as a workaround to
> this compiler bug. This probably affects all mips64 kernels.
> 

Shouldn't you fix the assmbler instead?


H.J.

Re: binutils bug workaround

[Ralf Baechle]

On Thu, Jan 03, 2002 at 01:53:34PM -0800, H . J . Lu wrote:

> > 	This patch move the declaration of kswapd_wait as a workaround to
> > this compiler bug. This probably affects all mips64 kernels.
> 
> Shouldn't you fix the assmbler instead?

I absolutely agree.  The only thing I'd like to add is a some code that makes
the kernel panic if built with broken binutils.

  Ralf