bpf, lpm: make longest_prefix_match() faster
authorEric Dumazet <edumazet@google.com>
Thu, 22 Nov 2018 05:39:52 +0000 (21:39 -0800)
committerDaniel Borkmann <daniel@iogearbox.net>
Thu, 22 Nov 2018 09:35:13 +0000 (10:35 +0100)
At LPC 2018 in Vancouver, Vlad Dumitrescu mentioned that longest_prefix_match()
has a high cost [1].

One reason for that cost is a loop handling one byte at a time.

We can handle more bytes at a time, if enough attention is paid
to endianness.

I was able to remove ~55 % of longest_prefix_match() cpu costs.

[1] https://linuxplumbersconf.org/event/2/contributions/88/attachments/76/87/lpc-bpf-2018-shaping.pdf

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Vlad Dumitrescu <vladum@google.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
kernel/bpf/lpm_trie.c

index 9058317..bfd4882 100644 (file)
@@ -168,20 +168,59 @@ static size_t longest_prefix_match(const struct lpm_trie *trie,
                                   const struct lpm_trie_node *node,
                                   const struct bpf_lpm_trie_key *key)
 {
-       size_t prefixlen = 0;
-       size_t i;
+       u32 limit = min(node->prefixlen, key->prefixlen);
+       u32 prefixlen = 0, i = 0;
 
-       for (i = 0; i < trie->data_size; i++) {
-               size_t b;
+       BUILD_BUG_ON(offsetof(struct lpm_trie_node, data) % sizeof(u32));
+       BUILD_BUG_ON(offsetof(struct bpf_lpm_trie_key, data) % sizeof(u32));
 
-               b = 8 - fls(node->data[i] ^ key->data[i]);
-               prefixlen += b;
+#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && defined(CONFIG_64BIT)
 
-               if (prefixlen >= node->prefixlen || prefixlen >= key->prefixlen)
-                       return min(node->prefixlen, key->prefixlen);
+       /* data_size >= 16 has very small probability.
+        * We do not use a loop for optimal code generation.
+        */
+       if (trie->data_size >= 8) {
+               u64 diff = be64_to_cpu(*(__be64 *)node->data ^
+                                      *(__be64 *)key->data);
+
+               prefixlen = 64 - fls64(diff);
+               if (prefixlen >= limit)
+                       return limit;
+               if (diff)
+                       return prefixlen;
+               i = 8;
+       }
+#endif
+
+       while (trie->data_size >= i + 4) {
+               u32 diff = be32_to_cpu(*(__be32 *)&node->data[i] ^
+                                      *(__be32 *)&key->data[i]);
+
+               prefixlen += 32 - fls(diff);
+               if (prefixlen >= limit)
+                       return limit;
+               if (diff)
+                       return prefixlen;
+               i += 4;
+       }
 
-               if (b < 8)
-                       break;
+       if (trie->data_size >= i + 2) {
+               u16 diff = be16_to_cpu(*(__be16 *)&node->data[i] ^
+                                      *(__be16 *)&key->data[i]);
+
+               prefixlen += 16 - fls(diff);
+               if (prefixlen >= limit)
+                       return limit;
+               if (diff)
+                       return prefixlen;
+               i += 2;
+       }
+
+       if (trie->data_size >= i + 1) {
+               prefixlen += 8 - fls(node->data[i] ^ key->data[i]);
+
+               if (prefixlen >= limit)
+                       return limit;
        }
 
        return prefixlen;