2009-06-12 09:22:39

by Pekka Enberg

Subject: [PATCH] powerpc: don't pass GFP_WAIT to kmalloc() from *_maybe_bootmem()

From: Pekka Enberg <[email protected]>

As slab is initialized much earlier now, we must not pass GFP_WAIT to kmalloc()
from *_maybe_bootmem() functions; otherwise slab allocators will enable
interrupts.

Signed-off-by: Pekka Enberg <[email protected]>
---
Ben, while we're discussing the slab side of things, can we go ahead and
merge this one? It should fix plenty of cases for powerpc.

arch/powerpc/lib/alloc.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/lib/alloc.c b/arch/powerpc/lib/alloc.c
index f53e09c..56b35e3 100644
--- a/arch/powerpc/lib/alloc.c
+++ b/arch/powerpc/lib/alloc.c
@@ -9,7 +9,7 @@
void * __init_refok alloc_maybe_bootmem(size_t size, gfp_t mask)
{
if (mem_init_done)
- return kmalloc(size, mask);
+ return kmalloc(size, mask & ~__GFP_WAIT);
else
return alloc_bootmem(size);
}
@@ -19,7 +19,7 @@ void * __init_refok zalloc_maybe_bootmem(size_t size, gfp_t mask)
void *p;

if (mem_init_done)
- p = kzalloc(size, mask);
+ p = kzalloc(size, mask & ~__GFP_WAIT);
else {
p = alloc_bootmem(size);
if (p)
--
1.6.0.4

2009-06-12 09:35:20

by Benjamin Herrenschmidt

Subject: Re: [PATCH] powerpc: don't pass GFP_WAIT to kmalloc() from *_maybe_bootmem()

On Fri, 2009-06-12 at 12:22 +0300, Pekka J Enberg wrote:
> From: Pekka Enberg <[email protected]>
>
> As slab is initialized much earlier now, we must not pass GFP_WAIT to kmalloc()
> from *_maybe_bootmem() functions; otherwise slab allocators will enable
> interrupts.
>
> Signed-off-by: Pekka Enberg <[email protected]>
> ---
> Ben, while we're discussing the slab side of things, can we go ahead and
> merge this one? It should fix plenty of cases for powerpc.

We can, but it's a plaster... I won't fix it all, and again, what tells
you that things calling alloc_maybe_bootmem() don't do that specifically
-because- they can be called also much later ? (In this case, it's
__init so I'm not going to object, but I still think it's unnecessary if
we fix it properly).

Cheers,
Ben.

> arch/powerpc/lib/alloc.c | 4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/arch/powerpc/lib/alloc.c b/arch/powerpc/lib/alloc.c
> index f53e09c..56b35e3 100644
> --- a/arch/powerpc/lib/alloc.c
> +++ b/arch/powerpc/lib/alloc.c
> @@ -9,7 +9,7 @@
> void * __init_refok alloc_maybe_bootmem(size_t size, gfp_t mask)
> {
> if (mem_init_done)
> - return kmalloc(size, mask);
> + return kmalloc(size, mask & ~__GFP_WAIT);
> else
> return alloc_bootmem(size);
> }
> @@ -19,7 +19,7 @@ void * __init_refok zalloc_maybe_bootmem(size_t size, gfp_t mask)
> void *p;
>
> if (mem_init_done)
> - p = kzalloc(size, mask);
> + p = kzalloc(size, mask & ~__GFP_WAIT);
> else {
> p = alloc_bootmem(size);
> if (p)