Skip to contentSkip to conversation

Add __GFP_ZERO to GFP_LEVEL_MASK

RFC PATCH In review Jonathan Corbet 2 participants 4 messages Oct 13, 2006

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 Add __GFP_ZERO to GFP_LEVEL_MASK

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.

include/linux/gfp.h renamed from k/t/2.6.19-rc2/include/linux/gfp.h +2 -1
@@ -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)
#1 lore raw
Quoted 8 lines · On Fri, 13 Oct 2006 14:25:16 -0600 Jonathan Corbet <[email protected]> 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.

#2 lore raw
Quoted 4 lines · Andrew Morton

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

#3 lore raw
Quoted 11 lines · On Fri, 13 Oct 2006 15:00:22 -0600 [email protected] (Jonathan Corbet) wrote:
Quoted 4 lines · Andrew Morton

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?

mm/vmalloc.c renamed from mm/vmalloc.c~vmalloc-dont-pass-__gfp_zero-to-slab +5 -2
diff -puN mm/vmalloc.c~vmalloc-dont-pass-__gfp_zero-to-slab 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.)