Fix emulation of misaligned access on big endian target (#224)

machine/misaligned_ldst.c

@@ -10,6 +10,7 @@
 union byte_array {
   uint8_t bytes[8];
   uintptr_t intx;
+  uint32_t int32;
   uint64_t int64;
 };
 
@@ -75,11 +76,15 @@ void misaligned_load_trap(uintptr_t* regs, uintptr_t mcause, uintptr_t mepc)
     val.bytes[i] = load_uint8_t((void *)(addr + i), mepc);
 
   if (!fp)
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
     SET_RD(insn, regs, (intptr_t)val.intx << shift >> shift);
+#else
+    SET_RD(insn, regs, (intptr_t)val.intx >> shift);
+#endif
   else if (len == 8)
     SET_F64_RD(insn, regs, val.int64);
   else
-    SET_F32_RD(insn, regs, val.intx);
+    SET_F32_RD(insn, regs, val.int32);
 
   write_csr(mepc, npc);
 }
@@ -138,8 +143,13 @@ void misaligned_store_trap(uintptr_t* regs, uintptr_t mcause, uintptr_t mepc)
   }
 
   uintptr_t addr = read_csr(mbadaddr);
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  intptr_t offs = (len == 8? 0 : sizeof(val.intx) - len);
+#else
+  intptr_t offs = 0;
+#endif
   for (int i = 0; i < len; i++)
-    store_uint8_t((void *)(addr + i), val.bytes[i], mepc);
+    store_uint8_t((void *)(addr + i), val.bytes[offs + i], mepc);
 
   write_csr(mepc, npc);
 }

machine/unprivileged_memory.h

@@ -79,22 +79,37 @@ static uintptr_t __attribute__((always_inline)) get_insn(uintptr_t mepc, uintptr
        : [mstatus] "+&r" (__mstatus), [insn] "=&r" (val)
        : [mprv] "r" (__mstatus_adjust), [addr] "r" (__mepc));
 #else
-  uintptr_t rvc_mask = 3, tmp;
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  uintptr_t rvc_mask = 3 << 24;
+#else
+  uintptr_t rvc_mask = 3;
+#endif
+  uintptr_t tmp;
   asm ("csrrs %[mstatus], mstatus, %[mprv]\n"
        "and %[tmp], %[addr], 2\n"
        "bnez %[tmp], 1f\n"
        STR(LWU) " %[insn], (%[addr])\n"
        "and %[tmp], %[insn], %[rvc_mask]\n"
        "beq %[tmp], %[rvc_mask], 2f\n"
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
        "sll %[insn], %[insn], %[xlen_minus_16]\n"
        "srl %[insn], %[insn], %[xlen_minus_16]\n"
+#else
+       "srl %[insn], %[insn], 16\n"
+       "sll %[insn], %[insn], 16\n"
+#endif
        "j 2f\n"
        "1:\n"
        "lhu %[insn], (%[addr])\n"
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+       "sll %[insn], %[insn], 16\n"
+#endif
        "and %[tmp], %[insn], %[rvc_mask]\n"
        "bne %[tmp], %[rvc_mask], 2f\n"
        "lhu %[tmp], 2(%[addr])\n"
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
        "sll %[tmp], %[tmp], 16\n"
+#endif
        "add %[insn], %[insn], %[tmp]\n"
        "2: csrw mstatus, %[mstatus]"
        : [mstatus] "+&r" (__mstatus), [insn] "=&r" (val), [tmp] "=&r" (tmp)
@@ -102,7 +117,11 @@ static uintptr_t __attribute__((always_inline)) get_insn(uintptr_t mepc, uintptr
          [rvc_mask] "r" (rvc_mask), [xlen_minus_16] "i" (__riscv_xlen - 16));
 #endif
   *mstatus = __mstatus;
+#if __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+  return __builtin_bswap32(val);
+#else
   return val;
+#endif
 }
 
 #endif