[PATCH,RFC] Add __GFP_ZERO to GFP_LEVEL_MASK

[PATCH,RFC] Add __GFP_ZERO to GFP_LEVEL_MASK

[Jonathan Corbet]

There is a very helpful comment in <linux/gfp.h>:

  /* if you forget to add the bitmask here kernel will crash, period */

Well, my kernel has been crashing (period) at the BUG() in cache_grow();
the offending flag is __GFP_ZERO.  I think it needs to be in
GFP_LEVEL_MASK.  Anybody know a good reason why it's not there now?

jon


Add __GFP_ZERO to GFP_LEVEL_MASK and cut down on those unsightly oopses.

Signed-off-by: Jonathan Corbet <corbet@lwn.net>

--- /k/t/2.6.19-rc2/include/linux/gfp.h	2006-10-13 13:04:17.000000000 -0600
+++ 19-rc2.jc/include/linux/gfp.h	2006-10-13 14:17:10.000000000 -0600
@@ -54,7 +54,8 @@ struct vm_area_struct;
 #define GFP_LEVEL_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS| \
 			__GFP_COLD|__GFP_NOWARN|__GFP_REPEAT| \
 			__GFP_NOFAIL|__GFP_NORETRY|__GFP_NO_GROW|__GFP_COMP| \
-			__GFP_NOMEMALLOC|__GFP_HARDWALL|__GFP_THISNODE)
+			__GFP_ZERO|__GFP_NOMEMALLOC|__GFP_HARDWALL|\
+			__GFP_THISNODE)
 
 /* This equals 0, but use constants in case they ever change */
 #define GFP_NOWAIT	(GFP_ATOMIC & ~__GFP_HIGH)

Re: [PATCH,RFC] Add __GFP_ZERO to GFP_LEVEL_MASK

[Andrew Morton]

On Fri, 13 Oct 2006 14:25:16 -0600
Jonathan Corbet <corbet@lwn.net> wrote:

> There is a very helpful comment in <linux/gfp.h>:
> 
>   /* if you forget to add the bitmask here kernel will crash, period */
> 
> Well, my kernel has been crashing (period) at the BUG() in cache_grow();
> the offending flag is __GFP_ZERO.  I think it needs to be in
> GFP_LEVEL_MASK.  Anybody know a good reason why it's not there now?
> 

It would be a bit odd to pass __GFP_ZERO into the slab allocator.  Slab
doesn't need that hint: it has its own ways of initialising the memory.

What is the callsite?

Thanks.

Re: [PATCH,RFC] Add __GFP_ZERO to GFP_LEVEL_MASK

[Jonathan Corbet]

Andrew Morton <akpm@osdl.org> wrote:

> It would be a bit odd to pass __GFP_ZERO into the slab allocator.  Slab
> doesn't need that hint: it has its own ways of initialising the memory.
> 
> What is the callsite?

It's vmalloc_user(), which does this:

  ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);

jon

Re: [PATCH,RFC] Add __GFP_ZERO to GFP_LEVEL_MASK

[Andrew Morton]

On Fri, 13 Oct 2006 15:00:22 -0600
corbet@lwn.net (Jonathan Corbet) wrote:

> Andrew Morton <akpm@osdl.org> wrote:
> 
> > It would be a bit odd to pass __GFP_ZERO into the slab allocator.  Slab
> > doesn't need that hint: it has its own ways of initialising the memory.
> > 
> > What is the callsite?
> 
> It's vmalloc_user(), which does this:
> 
>   ret = __vmalloc(size, GFP_KERNEL | __GFP_HIGHMEM | __GFP_ZERO, PAGE_KERNEL);
> 

oic, yes, that was a recent change.

I guess this will do the trick?


diff -puN mm/vmalloc.c~vmalloc-dont-pass-__gfp_zero-to-slab mm/vmalloc.c
--- a/mm/vmalloc.c~vmalloc-dont-pass-__gfp_zero-to-slab
+++ a/mm/vmalloc.c
@@ -428,8 +428,11 @@ void *__vmalloc_area_node(struct vm_stru
 	if (array_size > PAGE_SIZE) {
 		pages = __vmalloc_node(array_size, gfp_mask, PAGE_KERNEL, node);
 		area->flags |= VM_VPAGES;
-	} else
-		pages = kmalloc_node(array_size, (gfp_mask & ~__GFP_HIGHMEM), node);
+	} else {
+		pages = kmalloc_node(array_size,
+				(gfp_mask & ~(__GFP_HIGHMEM | __GFP_ZERO)),
+				node);
+	}
 	area->pages = pages;
 	if (!area->pages) {
 		remove_vm_area(area->addr);
_



(hmm, __vmalloc_area_node and __vmalloc_node are mutually recursive. ick.)