Re: [PATCH bpf-next 07/10] bpf: Switch to bpf mem allocator for LPM trie

["Toke Høiland-Jørgensen" <toke@redhat.com>]

Hou Tao <houtao@huaweicloud.com> writes:

> Fix these warnings by replacing kmalloc()/kfree()/kfree_rcu() with
> equivalent bpf memory allocator APIs. Since intermediate node and leaf
> node have fixed sizes, fixed-size allocation APIs are used.
>
> Two aspects of this change require explanation:
>
> 1. A new flag LPM_TREE_NODE_FLAG_ALLOC_LEAF is added to track the
>    original allocator. This is necessary because during deletion, a leaf
>    node may be used as an intermediate node. These nodes must be freed
>    through the leaf allocator.
> 2. The intermediate node allocator and leaf node allocator may be merged
>    because value_size for LPM trie is usually small. The merging reduces
>    the memory overhead of bpf memory allocator.

This seems like an awfully complicated way to fix this. Couldn't we just
move the node allocations in trie_update_elem() out so they happen
before the trie lock is taken instead?

-Toke