From: Matthew Dillon <dillon@apollo.backplane.com>
Date: Fri, 14 Feb 2020 06:13:57 +0000 (-0800)
Subject: rdrand - Document massive improvement in performance
X-Git-Tag: v5.9.0~39
X-Git-Url: https://gitweb.dragonflybsd.org/dragonfly.git/commitdiff_plain/f1af78531d94cf7d1abe2b0db2ad4f248e3cf008

rdrand - Document massive improvement in performance

* Document the huge difference going from 512 to 16 bytes.  General
  system performance is improved by 9.3% on a TR3990X.

  This is not entirely the fault of rdrand.  It is also in a large
  part due to the overhead of add_buffer_randomness().
---

diff --git a/sys/dev/crypto/rdrand/rdrand.c b/sys/dev/crypto/rdrand/rdrand.c
index ccfb92e0cc..e778dfb546 100644
--- a/sys/dev/crypto/rdrand/rdrand.c
+++ b/sys/dev/crypto/rdrand/rdrand.c
@@ -45,11 +45,28 @@
  * The RDRAND instruction is a very slow instruction, burning approximately
  * 0.79uS per 64-bit word on a modern ryzen cpu.  Intel cpu's run this
  * instruction far more quickly.  The quality of the results are unknown
- * either way.  The add_buffer_randomness() call is also not cheap.
+ * either way.
+ *
+ * However, possibly an even bigger problem is the cost of calling
+ * add_buffer_randomness(), which takes an enormous amount of time
+ * when handed a large buffer.
  *
  * Our code harvests at a 10hz rate on every single core, and also chains
  * some entropy from core to core so honestly it doesn't take much to really
  * mix things up.  Use a decent size (16 or 32 bytes should be good).
+ *
+ * On a TR3990 going from 512 to 16 gave userland almost 10% additional
+ * performance... a very stark difference.  A simple test loop:
+ *
+ * BEFORE: (RDRAND_SIZE 512)
+ *	7.258u 0.000s 0:07.69 94.2%     2+70k 2+0io 0pf+0w
+ *	7.222u 0.000s 0:07.67 94.1%     2+70k 0+0io 0pf+0w
+ *	7.239u 0.000s 0:07.69 94.0%     2+70k 0+0io 0pf+0w
+ *
+ * AFTER: (RDRAND_SIZE 16)		(9.3% faster)
+ *	7.019u 0.000s 0:07.02 99.8%     2+66k 0+0io 0pf+0w
+ *	7.019u 0.000s 0:07.02 99.8%     2+66k 0+0io 0pf+0w
+ *	7.028u 0.000s 0:07.02 100.0%    2+66k 0+0io 0pf+0w
  */
 #define	RDRAND_ALIGN(p)	(void *)(roundup2((uintptr_t)(p), 16))
 #define RDRAND_SIZE	16