From: Angus Gratton <angus@espressif.com>
Date: Tue, 15 Aug 2017 23:06:52 +0000 (+1000)
Subject: esp32 hwcrypto: Use spinlock instead of lock to protect AES
X-Git-Tag: v3.1-dev~347^2~3
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=50e0a54630747f76c0d94455bf1a0d7a4908856a;p=esp-idf

esp32 hwcrypto: Use spinlock instead of lock to protect AES

More than doubles performance of mbedTLS AES self-tests.
---

diff --git a/components/esp32/hwcrypto/aes.c b/components/esp32/hwcrypto/aes.c
index 7a9c93f32f..716bc636fe 100644
--- a/components/esp32/hwcrypto/aes.c
+++ b/components/esp32/hwcrypto/aes.c
@@ -32,12 +32,26 @@
 #include "soc/hwcrypto_reg.h"
 #include <sys/lock.h>
 
-static _lock_t aes_lock;
+#include <freertos/FreeRTOS.h>
+
+#include "soc/cpu.h"
+#include <stdio.h>
+
+
+/* AES uses a spinlock mux not a lock as the underlying block operation
+   only takes 208 cycles (to write key & compute block), +600 cycles
+   for DPORT protection but +3400 cycles again if you use a full sized lock.
+
+   For CBC, CFB, etc. this may mean that interrupts are disabled for a longer
+   period of time for bigger lengths. However at the moment this has to happen
+   anyway due to DPORT protection...
+*/
+static portMUX_TYPE aes_spinlock = portMUX_INITIALIZER_UNLOCKED;
 
 void esp_aes_acquire_hardware( void )
 {
     /* newlib locks lazy initialize on ESP-IDF */
-    _lock_acquire(&aes_lock);
+    portENTER_CRITICAL(&aes_spinlock);
 
     DPORT_STALL_OTHER_CPU_START();
     {
@@ -65,7 +79,7 @@ void esp_aes_release_hardware( void )
     }
     DPORT_STALL_OTHER_CPU_END();
 
-    _lock_release(&aes_lock);
+    portEXIT_CRITICAL(&aes_spinlock);
 }
 
 void esp_aes_init( esp_aes_context *ctx )
@@ -178,6 +192,7 @@ int esp_aes_crypt_ecb( esp_aes_context *ctx,
     esp_aes_setkey_hardware(ctx, mode);
     esp_aes_block(input, output);
     esp_aes_release_hardware();
+
     return 0;
 }