[libunwind] Fix Unwind-EHABI.cpp:getByte on big-endian targets

[miyuki]

Summary

The function getByte is dependent on endianness and the current
behavior is incorrect on big-endian targets.

This patch fixes the issue.

[ostannard]

This won't work for cross-unwinding between a big and little endian system, I think it would be better to load the whole word, and mask out the correct byte.

[chill]

LGTM, that's what EHABI spec says:

9.2 Exception-handling table entries
...
Unwinding instruction bytes are executed in order of significance within their containing word (most significant byte
first) and in increasing word address order. An implicit Finish instruction is assumed to be positioned after all the
explicitly present instructions.

[chill]

Sorry, messages crossed.

[ostannard]

Actually, I think cross-unwinding with EHABI is already completely broken, because we are doing raw memory accesses (not using the AddressSpace), so this is fine.

libunwind/src/Unwind-EHABI.cpp

 #include "unwind.h"
 
 namespace {
 
 // Strange order: take words in order, but inside word, take from most to least
 // signinficant byte.
 uint8_t getByte(const uint32_t* data, size_t offset) {
   const uint8_t* byteData = reinterpret_cast<const uint8_t*>(data);
+#ifdef __LITTLE_ENDIAN__
   return byteData[(offset & ~(size_t)0x03) + (3 - (offset & (size_t)0x03))];
+#else
+  return byteData[offset];
+#endif
 }
 
 const char* getNextWord(const char* data, uint32_t* out) {
   *out = *reinterpret_cast<const uint32_t*>(data);
   return data + 4;
 }
 
 const char* getNextNibble(const char* data, uint32_t* out) {