S390x support (#1103)

* s390x support

* - Fix for FrodoKEM-SHAKE for big endian support
- Fix unused variable in Keccak code on big endian

.CMake/compiler_opts.cmake

@@ -22,8 +22,15 @@ if(CMAKE_C_COMPILER_ID MATCHES "Clang|GNU")
     endif()
 
     if(OQS_OPT_TARGET STREQUAL "generic")
-        # Assume sensible default like -march=x86-64, -march=armv8-a, etc.
+        if(ARCH_S390X)
-        set(OQS_OPT_FLAG "")
+            # At least z9-109 is needed for 'stckf' in benchmarking code.
+            # gcc's default is z900 (older than z9-109), clang's default and minimum is z10.
+            # setting to z10 as sensible default.
+            set(OQS_OPT_FLAG "-march=z10")
+        else()
+            # Assume sensible default like -march=x86-64, -march=armv8-a, etc.
+            set(OQS_OPT_FLAG "")
+        endif()
     elseif(OQS_OPT_TARGET STREQUAL "auto")
       if(ARCH_X86_64)
           set(OQS_OPT_FLAG "-march=native")
@@ -31,6 +38,8 @@ if(CMAKE_C_COMPILER_ID MATCHES "Clang|GNU")
           set(OQS_OPT_FLAG "-mcpu=native")
       elseif(ARCH_ARM64v8 AND CMAKE_SYSTEM_NAME STREQUAL "Darwin")
           set(OQS_OPT_FLAG "-mcpu=native")
+      elseif(ARCH_S390X)
+          set(OQS_OPT_FLAG "-march=native")
       else()
           message(WARNING "Setting OQS_OPT_TARGET=AUTO may not produce optimized code on this system.")
       endif()
@@ -39,6 +48,8 @@ if(CMAKE_C_COMPILER_ID MATCHES "Clang|GNU")
         set(OQS_OPT_FLAG "-march=${OQS_OPT_TARGET}")
       elseif(ARCH_ARM64v8 OR ARCH_ARM32v7)
         set(OQS_OPT_FLAG "-mcpu=${OQS_OPT_TARGET}")
+      elseif(ARCH_S390X)
+        set(OQS_OPT_FLAG "-march=${OQS_OPT_TARGET}")
       endif()
     endif()
 

.travis.yml

@@ -11,3 +11,11 @@ jobs:
       script:
         - mkdir build && cd build && cmake -GNinja .. && cmake -LA .. && ninja
         - cd build & ninja run_tests
+    - arch: s390x
+      os: linux
+      dist: focal
+      compiler: gcc
+      if: NOT branch =~ /^ghactionsonly-/
+      script:
+        - mkdir build && cd build && cmake -DOQS_ENABLE_KEM_BIKE=OFF -GNinja .. && cmake -LA .. && ninja
+        - cd build & ninja run_tests

CMakeLists.txt

@@ -59,6 +59,12 @@ elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "ppc64le")
     if(${OQS_DIST_BUILD})
         set(OQS_DIST_PPC64LE_BUILD ON)
     endif()
+elseif(CMAKE_SYSTEM_PROCESSOR MATCHES "s390x")
+    set(ARCH "s390x")
+    set(ARCH_S390X ON)
+    if(${OQS_DIST_BUILD})
+        set(OQS_DIST_S390X_BUILD ON)
+    endif()
 elseif(OQS_PERMIT_UNSUPPORTED_ARCHITECTURE)
     message(WARNING "Unknown or unsupported processor: " ${CMAKE_SYSTEM_PROCESSOR})
     message(WARNING "Compilation on an unsupported processor should only be used for testing, as it may result an insecure configuration, for example due to variable-time instructions leaking secret information.")

src/common/common.c

@@ -135,6 +135,11 @@ static void set_available_cpu_extensions(void) {
 	/* mark that this function has been called */
 	cpu_ext_data[OQS_CPU_EXT_INIT] = 1;
 }
+#elif defined(OQS_DIST_S390X_BUILD)
+static void set_available_cpu_extensions(void) {
+	/* mark that this function has been called */
+	cpu_ext_data[OQS_CPU_EXT_INIT] = 1;
+}
 #elif defined(OQS_DIST_BUILD)
 static void set_available_cpu_extensions(void) {
 }

src/common/sha3/xkcp_low/KeccakP-1600/plain-64bits/KeccakP-1600-opt64.c

@@ -435,10 +435,11 @@ void KeccakP1600_ExtractAndAddBytesInLane(const void *state, unsigned int lanePo
 
 void KeccakP1600_ExtractAndAddLanes(const void *state, const unsigned char *input, unsigned char *output, unsigned int laneCount) {
 	unsigned int i;
-	uint64_t lane;
 #if (PLATFORM_BYTE_ORDER != IS_LITTLE_ENDIAN)
 	unsigned char temp[8];
 	unsigned int j;
+#else
+	uint64_t lane;
 #endif
 
 	for (i = 0; i < laneCount; i++) {

src/kem/frodokem/external/frodo_macrify_shake_portable.c

@@ -39,7 +39,7 @@ int frodo_mul_add_sa_plus_e_shake_portable(uint16_t *out, const uint16_t *s, con
             for (j = 0; j < 4; j++) {
                 uint16_t sp = s[i*PARAMS_N + kk + j];
                 for (int k = 0; k < PARAMS_N; k++) {                // Matrix-vector multiplication
-                    sum[k] += (uint16_t) ((uint32_t) sp * (uint32_t) a_cols[(t+j)*PARAMS_N + k]);
+                    sum[k] += (uint16_t) ((uint32_t) sp * (uint32_t) UINT16_TO_LE(a_cols[(t+j)*PARAMS_N + k]));
                 }
              }
             for (int k = 0; k < PARAMS_N; k++){

src/kem/sike/CMakeLists.txt

@@ -128,6 +128,8 @@ if(OQS_ENABLE_KEM_sike_p434 OR
         endif()
     elseif(ARCH_PPC64LE)
         target_compile_definitions(sike PRIVATE _GENERIC_ _PPC64LE_)
+    elseif(ARCH_S390X)
+        target_compile_definitions(sike PRIVATE _GENERIC_ _S390X_)
     endif()
 
     if(CMAKE_C_COMPILER_ID STREQUAL "GNU")

tests/ds_benchmark.h

@@ -135,6 +135,11 @@ static uint64_t _bench_rdtsc(void) {
 	__asm__ volatile("mrc p15, 0, %0, c9, c13, 0\t\n"
 	                 : "=r"(value));
 	return value;
+#elif defined(__s390x__)
+#define USING_TIME_RATHER_THAN_CYCLES
+	uint64_t tod;
+	__asm__ volatile("stckf %0\n" : "=Q" (tod) : : "cc");
+	return (tod * 1000 / 4096);
 #else
 #define USING_TIME_RATHER_THAN_CYCLES
 	struct timespec time;