Re: [PATCH] mm/slab: simplify SLAB_* flag handling

[Kevin Brodsky]

On 17/01/2025 23:13, Christoph Lameter (Ampere) wrote:
> On Fri, 17 Jan 2025, Kevin Brodsky wrote:
>
>> index a29457bef626..3b07cdaac3ae 100644
>> --- a/mm/slab_common.c
>> +++ b/mm/slab_common.c
>> @@ -305,18 +305,6 @@ struct kmem_cache *__kmem_cache_create_args(const char *name,
>>  		goto out_unlock;
>>  	}
>>
>> -	/* Refuse requests with allocator specific flags */
>> -	if (flags & ~SLAB_FLAGS_PERMITTED) {
>> -		err = -EINVAL;
>> -		goto out_unlock;
>> -	}
> I think we should keep checking for invalid flags.

The commit that introduced this check [1] aimed to ensure that no
allocator-specific flag is passed to kmem_cache_create(), so it seemed
to me it was no longer needed now that allocator-specific flags are gone.

Having said that, we could keep it in order to reject flags that are not
supposed to be passed to kmem_cache_create() (e.g. SLAB_SKIP_KFENCE).
With that approach we'd just need to clear SLAB_DEBUG_FLAGS below if
!CONFIG_SLUB_DEBUG (and get rid of CACHE_CREATE_MASK).

- Kevin

[1]
https://lore.kernel.org/all/1478553075-120242-2-git-send-email-thgarnie@xxxxxxxxxx/

>> -	/*
>> -	 * Some allocators will constraint the set of valid flags to a subset
>> -	 * of all flags. We expect them to define CACHE_CREATE_MASK in this
>> -	 * case, and we'll just provide them with a sanitized version of the
>> -	 * passed flags.
>> -	 */
>>  	flags &= CACHE_CREATE_MASK;
> This would silently clear some flags instead of creating an error.