Skip to contentSkip to conversation

slab: Adds two missing kmalloc() checks.

PATCH In review Luiz Fernando Capitulino 3 participants 6 messages Jan 23, 2006 Memory management

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 slab: Adds two missing kmalloc() checks.

Adds two missing kmalloc() checks in kmem_cache_init(). The check is worth
because if kmalloc() fails we'll have a nice message instead of a OOPS (which will make us look for a bug).

We're using BUG_ON() so that embedded people can disable it.

 mm/slab.c |    2 ++
 1 file changed, 2 insertions(+)
mm/slab.c +2 -0
index 6f8495e..2fcfd08 100644
@@ -1130,6 +1130,7 @@ void __init kmem_cache_init(void)
 		void *ptr;
 
 		ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL);
+		BUG_ON(!ptr);
 
 		local_irq_disable();
 		BUG_ON(ac_data(&cache_cache) != &initarray_cache.cache);
@@ -1139,6 +1140,7 @@ void __init kmem_cache_init(void)
 		local_irq_enable();
 
 		ptr = kmalloc(sizeof(struct arraycache_init), GFP_KERNEL);
+		BUG_ON(!ptr);
 
 		local_irq_disable();
 		BUG_ON(ac_data(malloc_sizes[INDEX_AC].cs_cachep)
Signature · 1 line
Luiz Fernando N. Capitulino
Acked-by on v1 1/1
#1 lore raw

Hi,

Quoted 7 lines · Luiz Fernando Capitulino

Adds two missing kmalloc() checks in kmem_cache_init(). The check is worth
because if kmalloc() fails we'll have a nice message instead of a OOPS (which will make us look for a bug).

We're using BUG_ON() so that embedded people can disable it.

Looks good to me. Arjan, you had some objections last time around. Are you okay with the change?

Pekka

#2 lore raw
Quoted 7 lines · Pekka Enberg

Hi,

Quoted 4 lines · Luiz Fernando Capitulino

Adds two missing kmalloc() checks in kmem_cache_init(). The check is worth
because if kmalloc() fails we'll have a nice message instead of a OOPS (which will make us look for a bug).

Quoted 6 lines
Quoted 3 lines

We're using BUG_ON() so that embedded people can disable it.

Looks good to me. Arjan, you had some objections last time around. Are you okay with the change?

I still fail to see the point. Has anyone EVER seen these trigger????

#3 lore raw
Quoted 2 lines · Pekka Enberg
Quoted 2 lines

Looks good to me. Arjan, you had some objections last time around. Are you okay with the change?

Quoted 1 line · Arjan van de Ven

I still fail to see the point. Has anyone EVER seen these trigger????

Yeah, we probably won't. They seem useful for people who hunt unchecked kmalloc() calls, though.

Pekka

#4 lore raw

Hi Pekka, Arjan,

On Mon, 23 Jan 2006 18:52:02 +0200
Pekka Enberg <[email protected]> wrote:

| On Mon, 2006-01-23 at 18:38 +0200, Pekka Enberg wrote:
| > > Looks good to me. Arjan, you had some objections last time around. Are
| > > you okay with the change?
| 
| On Mon, 2006-01-23 at 17:44 +0100, Arjan van de Ven wrote:
| > I still fail to see the point. Has anyone EVER seen these trigger????
| 
| Yeah, we probably won't. They seem useful for people who hunt unchecked
| kmalloc() calls, though.

It really looks useful to me. You don't check for fail because someone has
seen the fail happen. You check for fail in order to have a robust program.

There are zilions of checks in the kernel and in programs out there which I
think they will never fail. But they are there.

On the other hand, I'm not going to make too much noise for a such trivial
patch. If you think it's not useful, let's drop it. No problem.

Signature · 1 line
Luiz Fernando N. Capitulino
#5 lore raw
Quoted 15 lines · Luiz Fernando Capitulino

Hi Pekka, Arjan,

On Mon, 23 Jan 2006 18:52:02 +0200
Pekka Enberg <[email protected]> wrote:

| On Mon, 2006-01-23 at 18:38 +0200, Pekka Enberg wrote:
| > > Looks good to me. Arjan, you had some objections last time around. Are
| > > you okay with the change?
| 
| On Mon, 2006-01-23 at 17:44 +0100, Arjan van de Ven wrote:
| > I still fail to see the point. Has anyone EVER seen these trigger????
| 
| Yeah, we probably won't. They seem useful for people who hunt unchecked
| kmalloc() calls, though.
Quoted 2 lines

It really looks useful to me. You don't check for fail because someone has
seen the fail happen. You check for fail in order to have a robust program.

Generally you are right. You check for fail because you can recover the failure. In this case the code happens during real early boot, and if it fails you CANNOT BOOT. And you have so little memory, that it's highly unlikely that you even got this far. -> You only make the kernel bigger without any win at all.

Don't get me wrong. Most of the null pointer checks are useful. Just the ones where you can't recover ANYWAY are not. A BUG_ON() is not better than just hitting a GPF due to a null pointer, in either case you don't boot, and probably so early you don't get to see the message either, and when you do see it.. you get no additional information.