sp: support ECC non-blocking for the large implementation with tables and windowing

.github/configs/multi-arch.json

@@ -234,5 +234,29 @@
   "--enable-sp=yes,nonblock", "--enable-riscv-asm", "--enable-sp-asm", "--disable-examples",
   "CPPFLAGS=-Wdeclaration-after-statement -DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"],
  "check": false,
+ "run": [["env", "QEMU_LD_PREFIX=/usr/riscv64-linux-gnu", "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "arm64-sp-nonblock-fast", "minutes": 2.5,
+ "cc": "ccache aarch64-linux-gnu-gcc",
+ "comment": "ECC non-blocking on the SP assembly implementation without the small math: the table and window non-blocking multiplies.  Key gen, sign and the base point half of verify use sp_<curve>_ecc_mulmod_base_<n>_nb; shared secret and the other half of verify use sp_<curve>_ecc_mulmod_<n>_nb.  No WOLFSSL_VALIDATE_ECC_KEYGEN here, so the non-blocking key gen and shared secret test cases run.",
+ "configure": ["--host=aarch64-linux-gnu", "--enable-ecc=nonblock",
+  "--enable-sp=yes,nonblocklarge", "--enable-sp-asm", "--disable-examples",
+  "CPPFLAGS=-Wdeclaration-after-statement -DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"],
+ "check": false,
+ "run": [["env", "QEMU_LD_PREFIX=/usr/aarch64-linux-gnu", "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "armhf-sp-nonblock-fast", "minutes": 2.5,
+ "cc": "ccache arm-linux-gnueabihf-gcc",
+ "comment": "ECC non-blocking on the SP assembly implementation without the small math: the table and window non-blocking multiplies.  Key gen, sign and the base point half of verify use sp_<curve>_ecc_mulmod_base_<n>_nb; shared secret and the other half of verify use sp_<curve>_ecc_mulmod_<n>_nb.  No WOLFSSL_VALIDATE_ECC_KEYGEN here, so the non-blocking key gen and shared secret test cases run.",
+ "configure": ["--host=arm-linux-gnueabihf", "--enable-ecc=nonblock",
+  "--enable-sp=yes,nonblocklarge", "--enable-sp-asm", "--disable-examples",
+  "CPPFLAGS=-Wdeclaration-after-statement -DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"],
+ "check": false,
+ "run": [["env", "QEMU_LD_PREFIX=/usr/arm-linux-gnueabihf", "./wolfcrypt/test/testwolfcrypt"]]},
+{"name": "riscv64-sp-nonblock-fast", "minutes": 2.5,
+ "cc": "ccache riscv64-linux-gnu-gcc",
+ "comment": "ECC non-blocking on the SP assembly implementation without the small math: the table and window non-blocking multiplies.  Key gen, sign and the base point half of verify use sp_<curve>_ecc_mulmod_base_<n>_nb; shared secret and the other half of verify use sp_<curve>_ecc_mulmod_<n>_nb.  No WOLFSSL_VALIDATE_ECC_KEYGEN here, so the non-blocking key gen and shared secret test cases run.",
+ "configure": ["--host=riscv64-linux-gnu", "--enable-ecc=nonblock",
+  "--enable-sp=yes,nonblocklarge", "--enable-riscv-asm", "--enable-sp-asm", "--disable-examples",
+  "CPPFLAGS=-Wdeclaration-after-statement -DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK -DWC_ECC_NONBLOCK_ONLY"],
+ "check": false,
  "run": [["env", "QEMU_LD_PREFIX=/usr/riscv64-linux-gnu", "./wolfcrypt/test/testwolfcrypt"]]}
 ]

.github/configs/os-check-linux.json

@@ -89,6 +89,15 @@
  "configure": ["--enable-ecc=nonblock", "--enable-sp=yes,nonblock",
   "--enable-sp-asm", "--enable-intelasm",
   "CPPFLAGS=-DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"]},
+{"name": "nonblock-sp-asm-fast", "minutes": 3.0,
+ "comment": "Same but without the small math, so the non-blocking ECC state machines and the table-based blocking paths are built together.",
+ "configure": ["--enable-ecc=nonblock", "--enable-sp=yes,nonblocklarge",
+  "--enable-sp-asm", "--enable-intelasm",
+  "CPPFLAGS=-DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"]},
+{"name": "nonblock-sp-c64-fast", "minutes": 4.2,
+ "comment": "Non-blocking ECC on the portable C implementation without the small math; RSA/DH non-block are left off because their mod_exp only exists in the small variant.",
+ "configure": ["--enable-ecc=nonblock", "--enable-sp=yes,nonblocklarge", "--disable-sp-asm",
+  "CPPFLAGS=-DWOLFSSL_PUBLIC_MP -DWOLFSSL_DEBUG_NONBLOCK"]},
 {"name": "all-asn-template", "minutes": 3.5,
  "configure": ["--enable-all", "--enable-asn=template"]},
 {"name": "tls13-dtls13-session-misc", "minutes": 3.3,

configure.ac

@@ -1130,7 +1130,7 @@ fi
 
 # Single Precision maths implementation
 AC_ARG_ENABLE([sp],
-    [AS_HELP_STRING([--enable-sp],[Enable Single Precision maths implementation, comma separated options include small, smallstack, nomalloc, nonblock and notransposedtable (default: disabled)])],
+    [AS_HELP_STRING([--enable-sp],[Enable Single Precision maths implementation, comma separated options include small, smallstack, nomalloc, nonblock, nonblocklarge and notransposedtable (default: disabled)])],
     [ ENABLED_SP=$enableval ],
     [ ENABLED_SP=$ENABLED_SP_DEFAULT ],
     )
@@ -11763,6 +11763,14 @@ do
     ENABLED_SP_SMALL=yes
     ;;
 
+  nonblocklarge)
+    # Non-blocking ECC with the large implementation's tables and windowing.
+    # Requires no malloc.  Small is still forced below when RSA/DH non-blocking
+    # is requested on the C implementation.
+    ENABLED_SP_NONBLOCK=yes
+    ENABLED_SP_NO_MALLOC=yes
+    ;;
+
   asm)
     ENABLED_SP_ASM=yes
     ;;
@@ -12005,6 +12013,17 @@ if test "$ENABLED_SP_ASM" = "yes" && test "$ENABLED_SP" = "yes"; then
   esac
 fi
 
+# Non-blocking RSA/DH mod_exp only exists in the small variant of the C
+# implementation.  Checked here, after the host checks that can still
+# turn the assembly off.
+if test "$ENABLED_SP_NONBLOCK" = "yes" && test "$ENABLED_SP_ASM" != "yes" && \
+   test "$ENABLED_SP_SMALL" != "yes" && \
+   { test "$enable_rsa" = "nonblock" || test "$enable_dh" = "nonblock"; }; then
+    ENABLED_SP_SMALL=yes
+    AM_CFLAGS="$AM_CFLAGS -DWOLFSSL_SP_SMALL"
+    AM_CCASFLAGS="$AM_CCASFLAGS -DWOLFSSL_SP_SMALL"
+fi
+
 if test "$ENABLED_SP_ASM" = "yes" && test "$ENABLED_SP" = "yes"; then
   if test "$ENABLED_SP_NONBLOCK" = "yes" && \
      { test "$enable_rsa" = "nonblock" || test "$enable_dh" = "nonblock"; }; then

wolfssl/wolfcrypt/sp_int.h

@@ -401,7 +401,16 @@ extern "C" {
 #if defined(WOLFSSL_HAVE_SP_ECC) && defined(WOLFSSL_SP_NONBLOCK)
 /* Non-blocking ECC operation context. */
 typedef struct sp_ecc_ctx {
+    #ifndef WOLFSSL_SP_SMALL
+    /* Large implementation: verify keeps a table of 33 points. */
     #ifdef WOLFSSL_SP_521
+    XALIGNED(4) byte data[66*340]; /* stack data */
+    #elif defined(WOLFSSL_SP_384)
+    XALIGNED(4) byte data[48*330]; /* stack data */
+    #else
+    XALIGNED(4) byte data[32*340]; /* stack data */
+    #endif
+    #elif defined(WOLFSSL_SP_521)
     XALIGNED(4) byte data[66*80]; /* stack data */
     #elif defined(WOLFSSL_SP_384)
     XALIGNED(4) byte data[48*80]; /* stack data */