On 05/22/2012 07:22 AM, David Rientjes wrote:
- if (setup_cpu_cache(cachep, gfp)) {
> + /* Can't do strdup while kmalloc is not up */
> + if (g_cpucache_up> EARLY)
> + cachep->name = kstrdup(name, GFP_KERNEL);
> + else
> + cachep->name = name;
> +
> + if (!cachep->name || setup_cpu_cache(cachep, gfp)) {
> __kmem_cache_destroy(cachep);
> cachep = NULL;
> goto oops;
This doesn't work if you kmem_cache_destroy() a cache that was created
when g_cpucache_cpu<= EARLY, the kfree() will explode. That never
happens for any existing cache created in kmem_cache_init(), but this
would introduce the first roadblock in doing so. So you'll need some
magic to determine whether the cache was allocated statically and suppress
the kfree() in such a case.
David,
I tried to do something like I was doing for the memcg caches: after
creation of the kmalloc + cache_cache, I loop through them and duplicate
the name. So instead of conditionally freeing the late caches - that
could cause consistency headaches in the future - kfree'ing the name
string will just work for all of them. I will send it shortly.
Cristoph, I am dropping your ack since this change is quite significant.
If you agree with it, would you ack it again?
Thanks.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>