Optimize _gcry_burn_stack for 32-bit and 64-bit architectures

* src/misc.c (_gcry_burn_stack): Add optimization for 32-bit and 64-bit
architectures.
--

Busy looping 'tests/benchmark --cipher-repetitions 10 cipher blowfish' on ARM
Cortex-A8 shows that _gcry_burn_stack takes 21% of CPU time. With this patch,
that number drops to 3.4%.

On AMD64 (Intel i5-4570) CPU usage for _gcry_burn_stack in the same test drops
from 3.5% to 1.1%.

Signed-off-by: Jussi Kivilinna <jussi.kivilinna@iki.fi>

1 files changed, 26 insertions, 0 deletions

diff --git a/src/misc.c b/src/misc.c
index ed72ed63..67c2e80e 100644
--- a/src/misc.c
+++ b/src/misc.c
@@ -290,9 +290,35 @@ _gcry_log_printhex (const char *text, const void *buffer, size_t length)
 void
 _gcry_burn_stack (int bytes)
 {
+#if SIZEOF_UNSIGNED_LONG == 4 || SIZEOF_UNSIGNED_LONG == 8
+    /* Optimized burn_stack for 32-bit and 64-bit architectures.  In addition
+       to loop unrolling, compiler sees that writes are within 'buf' and
+       generation of stack-protection code is avoided.  */
+    volatile unsigned long buf[64 / SIZEOF_UNSIGNED_LONG];
+
+    buf[0] = 0;
+    buf[1] = 0;
+    buf[2] = 0;
+    buf[3] = 0;
+    buf[4] = 0;
+    buf[5] = 0;
+    buf[6] = 0;
+    buf[7] = 0;
+# if SIZEOF_UNSIGNED_LONG == 4
+    buf[8] = 0;
+    buf[9] = 0;
+    buf[10] = 0;
+    buf[11] = 0;
+    buf[12] = 0;
+    buf[13] = 0;
+    buf[14] = 0;
+    buf[15] = 0;
+# endif
+#else
     char buf[64];
 
     wipememory (buf, sizeof buf);
+#endif
     bytes -= sizeof buf;
     if (bytes > 0)
         _gcry_burn_stack (bytes);