decrease RAM usage using PROGMEM

ADiea · igrr · commit feed1ca219c6 · 2017-04-29T18:42:22.000+08:00
diff --git a/Makefile b/Makefile
@@ -4,8 +4,6 @@ AR := $(TOOLCHAIN_PREFIX)ar
 LD := $(TOOLCHAIN_PREFIX)gcc
 OBJCOPY := $(TOOLCHAIN_PREFIX)objcopy
 
-MFORCE32 := $(shell $(CC) --help=target | grep mforce-l32)
-
 XTENSA_LIBS ?= $(shell $(CC) -print-sysroot)
 
 
@@ -43,13 +41,17 @@ LDFLAGS  += 	-L$(XTENSA_LIBS)/lib \
 
 CFLAGS+=-std=c99 -DESP8266
 
-CFLAGS += -Wall -Os -g -O2 -Wpointer-arith -Wno-implicit-function-declaration -Wl,-EL -fno-inline-functions -nostdlib -mlongcalls -mno-text-section-literals  -D__ets__ -DICACHE_FLASH
+CFLAGS += -Wall -Os -g -O2 -Wpointer-arith -Wl,-EL -nostdlib -mlongcalls -mno-text-section-literals  -D__ets__ -DICACHE_FLASH
 
+MFORCE32 := $(shell $(CC) --help=target | grep mforce-l32)
 ifneq ($(MFORCE32),)
-    # Your compiler supports the -mforce-l32 flag which means that 
-    # constants can be stored in flash (program) memory instead of SRAM.
-    # See: https://www.arduino.cc/en/Reference/PROGMEM
-    CFLAGS += -DPROGMEM="__attribute__((aligned(4))) __attribute__((section(\".irom.text\")))" -mforce-l32
+    # If the compiler supports the -mforce-l32 flag, the compiler will generate correct code for loading
+    # 16- and 8-bit constants from program memory. So in the code we can directly access the arrays
+    # placed into program memory.
+    CFLAGS +=  -mforce-l32
+else
+	# Otherwise we need to use a helper function to load 16- and 8-bit constants from program memory.
+    CFLAGS += -DWITH_PGM_READ_HELPER
 endif
 
 BIN_DIR := bin
diff --git a/crypto/aes.c b/crypto/aes.c
@@ -220,20 +220,20 @@ void AES_set_key(AES_CTX *ctx, const uint8_t *key,
 
         if ((i % words) == 0)
         {
-            tmp2 =(uint32_t)aes_sbox[(tmp    )&0xff]<< 8;
-            tmp2|=(uint32_t)aes_sbox[(tmp>> 8)&0xff]<<16;
-            tmp2|=(uint32_t)aes_sbox[(tmp>>16)&0xff]<<24;
-            tmp2|=(uint32_t)aes_sbox[(tmp>>24)     ];
+            tmp2 =(uint32_t)(ax_array_read_u8(aes_sbox, (tmp    )&0xff))<< 8;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>> 8)&0xff))<<16;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>>16)&0xff))<<24;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>>24)));
             tmp=tmp2^(((unsigned int)*ip)<<24);
             ip++;
         }
 
         if ((words == 8) && ((i % words) == 4))
         {
-            tmp2 =(uint32_t)aes_sbox[(tmp    )&0xff]    ;
-            tmp2|=(uint32_t)aes_sbox[(tmp>> 8)&0xff]<< 8;
-            tmp2|=(uint32_t)aes_sbox[(tmp>>16)&0xff]<<16;
-            tmp2|=(uint32_t)aes_sbox[(tmp>>24)     ]<<24;
+            tmp2 =(uint32_t)(ax_array_read_u8(aes_sbox, (tmp    )&0xff))    ;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>> 8)&0xff))<< 8;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>>16)&0xff))<<16;
+            tmp2|=(uint32_t)(ax_array_read_u8(aes_sbox, (tmp>>24)     ))<<24;
             tmp=tmp2;
         }
 
@@ -369,10 +369,10 @@ static void AES_encrypt(const AES_CTX *ctx, uint32_t *data)
         /* Perform ByteSub and ShiftRow operations together */
         for (row = 0; row < 4; row++)
         {
-            a0 = (uint32_t)aes_sbox[(data[row%4]>>24)&0xFF];
-            a1 = (uint32_t)aes_sbox[(data[(row+1)%4]>>16)&0xFF];
-            a2 = (uint32_t)aes_sbox[(data[(row+2)%4]>>8)&0xFF]; 
-            a3 = (uint32_t)aes_sbox[(data[(row+3)%4])&0xFF];
+            a0 = (uint32_t)(ax_array_read_u8(aes_sbox, (data[row%4]>>24)&0xFF));
+            a1 = (uint32_t)(ax_array_read_u8(aes_sbox, (data[(row+1)%4]>>16)&0xFF));
+            a2 = (uint32_t)(ax_array_read_u8(aes_sbox, (data[(row+2)%4]>>8)&0xFF));
+            a3 = (uint32_t)(ax_array_read_u8(aes_sbox, (data[(row+3)%4])&0xFF));
 
             /* Perform MixColumn iff not last round */
             if (curr_rnd < (rounds - 1))
@@ -417,10 +417,10 @@ static void AES_decrypt(const AES_CTX *ctx, uint32_t *data)
         /* Perform ByteSub and ShiftRow operations together */
         for (row = 4; row > 0; row--)
         {
-            a0 = aes_isbox[(data[(row+3)%4]>>24)&0xFF];
-            a1 = aes_isbox[(data[(row+2)%4]>>16)&0xFF];
-            a2 = aes_isbox[(data[(row+1)%4]>>8)&0xFF];
-            a3 = aes_isbox[(data[row%4])&0xFF];
+            a0 = ax_array_read_u8(aes_isbox, (data[(row+3)%4]>>24)&0xFF);
+            a1 = ax_array_read_u8(aes_isbox, (data[(row+2)%4]>>16)&0xFF);
+            a2 = ax_array_read_u8(aes_isbox, (data[(row+1)%4]>>8)&0xFF);
+            a3 = ax_array_read_u8(aes_isbox, (data[row%4])&0xFF);
 
             /* Perform MixColumn iff not last round */
             if (curr_rnd<(rounds-1))
diff --git a/crypto/crypto_misc.c b/crypto/crypto_misc.c
@@ -334,7 +334,7 @@ EXP_FUNC int STDCALL base64_decode(const char *in, int len,
     g = 3;
     for (x = y = z = t = 0; x < len; x++)
     {
-        if ((c = map[in[x]&0x7F]) == 0xff)
+        if ((c = ax_array_read_u8(map, in[x]&0x7F)) == 0xff)
             continue;
 
         if (c == 254)   /* this is the end... */
diff --git a/crypto/md5.c b/crypto/md5.c
@@ -60,7 +60,7 @@ static void MD5Transform(uint32_t state[4], const uint8_t block[64]);
 static void Encode(uint8_t *output, uint32_t *input, uint32_t len);
 static void Decode(uint32_t *output, const uint8_t *input, uint32_t len);
 
-static const uint8_t PADDING[64] = 
+static const uint8_t PADDING[64] PROGMEM = 
 {
     0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
@@ -158,15 +158,22 @@ EXP_FUNC void STDCALL MD5_Final(uint8_t *digest, MD5_CTX *ctx)
 {
     uint8_t bits[8];
     uint32_t x, padLen;
-
+#ifdef WITH_PGM_READ_HELPER
+    uint8_t PADDING_stack[64];
+    memcpy(PADDING_stack, PADDING, 64);
+#endif
     /* Save number of bits */
     Encode(bits, ctx->count, 8);
 
     /* Pad out to 56 mod 64.
      */
     x = (uint32_t)((ctx->count[0] >> 3) & 0x3f);
     padLen = (x < 56) ? (56 - x) : (120 - x);
+#ifdef WITH_PGM_READ_HELPER
+    MD5_Update(ctx, PADDING_stack, padLen);
+#else
     MD5_Update(ctx, PADDING, padLen);
+#endif
 
     /* Append length (before padding) */
     MD5_Update(ctx, bits, 8);
diff --git a/crypto/sha256.c b/crypto/sha256.c
@@ -48,7 +48,7 @@
     (b)[(i) + 3] = (uint8_t) ((n)      );       \
 }
 
-static const uint8_t sha256_padding[64] =
+static const uint8_t sha256_padding[64] PROGMEM =
 {
  0x80, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
@@ -249,6 +249,10 @@ void SHA256_Final(uint8_t *digest, SHA256_CTX *ctx)
     uint32_t last, padn;
     uint32_t high, low;
     uint8_t msglen[8];
+#ifdef WITH_PGM_READ_HELPER
+    uint8_t sha256_padding_ram[64];
+    memcpy(sha256_padding_ram, sha256_padding, 64);
+#endif
 
     high = (ctx->total[0] >> 29)
          | (ctx->total[1] <<  3);
@@ -259,8 +263,11 @@ void SHA256_Final(uint8_t *digest, SHA256_CTX *ctx)
 
     last = ctx->total[0] & 0x3F;
     padn = (last < 56) ? (56 - last) : (120 - last);
-
+#ifdef WITH_PGM_READ_HELPER
+    SHA256_Update(ctx, sha256_padding_ram, padn);
+#else
     SHA256_Update(ctx, sha256_padding, padn);
+#endif
     SHA256_Update(ctx, msglen, 8);
 
     PUT_UINT32(ctx->state[0], digest,  0);
diff --git a/ssl/config.h b/ssl/config.h
@@ -45,7 +45,7 @@
 #define CONFIG_SSL_HAS_PEM 1
 #undef CONFIG_SSL_USE_PKCS12
 #define CONFIG_SSL_EXPIRY_TIME 24
-#define CONFIG_X509_MAX_CA_CERTS 150
+#define CONFIG_X509_MAX_CA_CERTS 10
 #define CONFIG_SSL_MAX_CERTS 1
 #undef CONFIG_SSL_CTX_MUTEXING
 #undef CONFIG_USE_DEV_URANDOM
diff --git a/ssl/os_port.h b/ssl/os_port.h
@@ -98,6 +98,31 @@ extern "C" {
 
 void ax_wdt_feed();
 
+#ifndef PROGMEM
+#define PROGMEM __attribute__((aligned(4))) __attribute__((section(".irom.text")))
+#endif
+
+#ifndef WITH_PGM_READ_HELPER
+#define ax_array_read_u8(x, y) x[y]
+#else
+
+static inline uint8_t pgm_read_byte(const void* addr) {
+  register uint32_t res;
+  __asm__("extui    %0, %1, 0, 2\n"     /* Extract offset within word (in bytes) */
+      "sub      %1, %1, %0\n"       /* Subtract offset from addr, yielding an aligned address */
+      "l32i.n   %1, %1, 0x0\n"      /* Load word from aligned address */
+      "slli     %0, %0, 3\n"        /* Multiply offset by 8, yielding an offset in bits */
+      "ssr      %0\n"               /* Prepare to shift by offset (in bits) */
+      "srl      %0, %1\n"           /* Shift right; now the requested byte is the first one */
+      :"=r"(res), "=r"(addr)
+      :"1"(addr)
+      :);
+  return (uint8_t) res;     /* This masks the lower byte from the returned word */
+}
+
+#define ax_array_read_u8(x, y) pgm_read_byte((x)+(y))
+#endif //WITH_PGM_READ_HELPER
+
 #elif defined(WIN32)
 
 /* Windows CE stuff */
diff --git a/ssl/ssl.h b/ssl/ssl.h
@@ -233,6 +233,20 @@ EXP_FUNC void STDCALL ssl_ctx_free(SSL_CTX *ssl_ctx);
  */
 EXP_FUNC SSL_EXTENSIONS * STDCALL ssl_ext_new();
 
+/**
+ * @brief Set the host name for SNI extension
+ * @param ssl_ext pointer returned by ssl_ext_new
+ * @param host_name pointer to a zero-terminated string containing host name
+ */
+EXP_FUNC void STDCALL ssl_ext_set_host_name(SSL_EXTENSIONS * ext, const char* host_name);
+
+/**
+ * @brief Set the maximum fragment size for the fragment size negotiation extension
+ * @param ssl_ext pointer returned by ssl_ext_new
+ * @param fragment_size fragment size, allowed values: 2^9, 2^10 ... 2^14
+ */
+EXP_FUNC void STDCALL ssl_ext_set_max_fragment_size(SSL_EXTENSIONS * ext, unsigned fragment_size);
+
 /**
  * @brief Frees SSL extensions structure
  *
diff --git a/ssl/tls1.c b/ssl/tls1.c
@@ -140,7 +140,7 @@ void DISPLAY_BYTES(SSL *ssl, const char *format,
 #endif
 
 /**
- * Allocates new SSL extensions structure and returns pointer to it
+ * Allocate new SSL extensions structure and return pointer to it
  *
  */
 EXP_FUNC SSL_EXTENSIONS * STDCALL ssl_ext_new()
@@ -153,7 +153,7 @@ EXP_FUNC SSL_EXTENSIONS * STDCALL ssl_ext_new()
 }
 
 /**
- * Allocates new SSL extensions structure and returns pointer to it
+ * Free SSL extensions structure
  *
  */
 EXP_FUNC void STDCALL ssl_ext_free(SSL_EXTENSIONS *ssl_ext)
@@ -168,6 +168,23 @@ EXP_FUNC void STDCALL ssl_ext_free(SSL_EXTENSIONS *ssl_ext)
     free(ssl_ext);
 }
 
+EXP_FUNC void STDCALL ssl_ext_set_host_name(SSL_EXTENSIONS * ext, const char* host_name)
+{
+    free(ext->host_name);
+    ext->host_name = NULL;
+    if (host_name) {
+        ext->host_name = strdup(host_name);
+    }
+}
+
+/**
+ * Set the maximum fragment size for the fragment size negotiation extension
+ */
+EXP_FUNC void STDCALL ssl_ext_set_max_fragment_size(SSL_EXTENSIONS * ext, unsigned fragment_size)
+{
+    ext->max_fragment_size = fragment_size;
+}
+
 /**
  * Establish a new client/server context.
  */
@@ -2426,10 +2443,6 @@ EXP_FUNC void STDCALL ssl_display_error(int error_code)
     printf("\n");
 }
 
-/**
- * Debugging routine to display alerts.
- */
-
 /**
  * Debugging routine to display alerts.
  */

Original file line number	Diff line number	Diff line change
`@@ -334,7 +334,7 @@ EXP_FUNC int STDCALL base64_decode(const char *in, int len,`
`334`	`334`	`g = 3;`
`335`	`335`	`for (x = y = z = t = 0; x < len; x++)`
`336`	`336`	`{`
`337`		`- if ((c = map[in[x]&0x7F]) == 0xff)`
	`337`	`+ if ((c = ax_array_read_u8(map, in[x]&0x7F)) == 0xff)`
`338`	`338`	`continue;`
`339`	`339`
`340`	`340`	`if (c == 254) /* this is the end... */`