From: Andreas Dilger <adilger@whamcloud.com>
Date: Tue, 11 Aug 2020 00:32:18 +0000 (-0600)
Subject: LU-13127 ptlrpc: prefer crc32_le() over CryptoAPI
X-Git-Tag: 2.13.56~79
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=1dda0ef6a70b2c1f6b01054d851ace22e99bb048;ds=sidebyside

LU-13127 ptlrpc: prefer crc32_le() over CryptoAPI

Prefer to call the crc32_le() library function directly if available,
instead of cfs_crypto_hash(CFS_HASH_ALG_CRC32). It is about 10x faster
for the 156-byte struct ptlrpc_body being checked in this function.
A test of small buffers in that compares the two implementations, run
on a 2.9GHz Core i7-7820 shows the difference is significant here:

  buffer size   156 bytes   1536 bytes   4096 bytes     1 MiB
  -----------+------------+------------+-----------+-----------
  cfs_crypto |  182 MiB/s | 1794 MiB/s | 4163 MB/s | 9631 MiB/s
  crc32_le   | 1947 MiB/s | 1871 MiB/s | 1867 MB/s | 1823 MiB/s

This corresponds to 10x faster or 1/10 as many cycles for ptlrpc_body.
The CryptoAPI speed crosses over around 1536 bytes, which is still 10x
larger than the ptlrpc_body size, so it is unlikely to be faster here.

Signed-off-by: Andreas Dilger <adilger@whamcloud.com>
Change-Id: I116fd6c148f15660dd7b7faefb86f9dd603ebbe5
Reviewed-on: https://review.whamcloud.com/39614
Reviewed-by: Sebastien Buisson <sbuisson@ddn.com>
Tested-by: jenkins <devops@whamcloud.com>
Reviewed-by: Alexey Lyashkov <alexey.lyashkov@hpe.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
---

diff --git a/lustre/ptlrpc/pack_generic.c b/lustre/ptlrpc/pack_generic.c
index 0e382c4..33b491c 100644
--- a/lustre/ptlrpc/pack_generic.c
+++ b/lustre/ptlrpc/pack_generic.c
@@ -40,9 +40,7 @@
 
 #define DEBUG_SUBSYSTEM S_RPC
 
-#ifndef CONFIG_CRYPTO_CRC32
 #include <linux/crc32.h>
-#endif
 
 #include <libcfs/libcfs.h>
 
@@ -1387,13 +1385,16 @@ __u32 lustre_msg_calc_cksum(struct lustre_msg *msg, __u32 buf)
 		__u32 crc;
 
 		LASSERTF(pb != NULL, "invalid msg %p: no ptlrpc body!\n", msg);
-#ifdef CONFIG_CRYPTO_CRC32
+#if IS_ENABLED(CONFIG_CRC32)
+		/* about 10x faster than crypto_hash for small buffers */
+		crc = crc32_le(~(__u32)0, (unsigned char *)pb, len);
+#elif IS_ENABLED(CONFIG_CRYPTO_CRC32)
 		unsigned int hsize = 4;
 		cfs_crypto_hash_digest(CFS_HASH_ALG_CRC32, (unsigned char *)pb,
 				       len, NULL, 0, (unsigned char *)&crc,
 				       &hsize);
 #else
-		crc = crc32_le(~(__u32)0, (unsigned char *)pb, len);
+#error "need either CONFIG_CRC32 or CONFIG_CRYPTO_CRC32 enabled in the kernel"
 #endif
 		return crc;
 	}