Re: [PATCH bpf-next 2/6] bpf: Re-enable unit_size checking for global per-cpu allocator

[Alexei Starovoitov <alexei.starovoitov@gmail.com>]

On Sat, Oct 07, 2023 at 09:51:02PM +0800, Hou Tao wrote:
> From: Hou Tao <houtao1@huawei.com>
> 
> With alloc_size_percpu() in place, check whether or not the size of
> the dynamic per-cpu area is matched with unit_size.
> 
> Signed-off-by: Hou Tao <houtao1@huawei.com>
> ---
>  kernel/bpf/memalloc.c | 25 ++++++++++++++-----------
>  1 file changed, 14 insertions(+), 11 deletions(-)
> 
> diff --git a/kernel/bpf/memalloc.c b/kernel/bpf/memalloc.c
> index 6cf61ea55c27..af9ff0755959 100644
> --- a/kernel/bpf/memalloc.c
> +++ b/kernel/bpf/memalloc.c
> @@ -497,21 +497,17 @@ static int check_obj_size(struct bpf_mem_cache *c, unsigned int idx)
>  	struct llist_node *first;
>  	unsigned int obj_size;
>  
> -	/* For per-cpu allocator, the size of free objects in free list doesn't
> -	 * match with unit_size and now there is no way to get the size of
> -	 * per-cpu pointer saved in free object, so just skip the checking.
> -	 */
> -	if (c->percpu_size)
> -		return 0;
> -
>  	first = c->free_llist.first;
>  	if (!first)
>  		return 0;
>  
> -	obj_size = ksize(first);
> +	if (c->percpu_size)
> +		obj_size = alloc_size_percpu(((void **)first)[1]);
> +	else
> +		obj_size = ksize(first);
>  	if (obj_size != c->unit_size) {
> -		WARN_ONCE(1, "bpf_mem_cache[%u]: unexpected object size %u, expect %u\n",
> -			  idx, obj_size, c->unit_size);
> +		WARN_ONCE(1, "bpf_mem_cache[%u]: percpu %d, unexpected object size %u, expect %u\n",
> +			  idx, c->percpu_size, obj_size, c->unit_size);
>  		return -EINVAL;
>  	}
>  	return 0;
> @@ -979,7 +975,14 @@ void notrace *bpf_mem_cache_alloc_flags(struct bpf_mem_alloc *ma, gfp_t flags)
>  	return !ret ? NULL : ret + LLIST_NODE_SZ;
>  }
>  
> -/* Most of the logic is taken from setup_kmalloc_cache_index_table() */
> +/* The alignment of dynamic per-cpu area is 8 and c->unit_size and the
> + * actual size of dynamic per-cpu area will always be matched, so there is
> + * no need to adjust size_index for per-cpu allocation. However for the
> + * simplicity of the implementation, use an unified size_index for both
> + * kmalloc and per-cpu allocation.
> + *
> + * Most of the logic is taken from setup_kmalloc_cache_index_table().

Since this logic is removed in bpf tree you probably need to wait for
bpf tree to get merged into bpf-next before respinning to avoid conflicts.