[hwasan] Don't require untagged pointer in __hwasan_tag_memory

[vitalybuka]

Summary

Use likely have tagged pointer. I see no value making user to untag it.

[vitalybuka]

ping

[fmayer]

Seems that this is only API for testing, so fine either way. I was pondering whether this restriction was on purpose to prevent people from retagging pointers that already have shadow memory associated.

compiler-rt/lib/hwasan/hwasan.cpp

 void __hwasan_store8_noabort(uptr p) {
   CheckAddress<ErrorAction::Recover, AccessType::Store, 3>(p);
 }
 void __hwasan_store16_noabort(uptr p) {
   CheckAddress<ErrorAction::Recover, AccessType::Store, 4>(p);
 }
 
 void __hwasan_tag_memory(uptr p, u8 tag, uptr sz) {
-  TagMemoryAligned(p, sz, tag);
+  TagMemoryAligned(UntagAddr(p), sz, tag);
 }
 
 uptr __hwasan_tag_pointer(uptr p, u8 tag) {
   return AddTagToPointer(p, tag);
 }
 
 void __hwasan_handle_longjmp(const void *sp_dst) {
   uptr dst = (uptr)sp_dst;

compiler-rt/test/hwasan/TestCases/hwasan-print-shadow.cpp

int main() {
   // Simulate short granule tags.
   alloc[15] = 0x00;
   alloc[31] = 0xbb;
   alloc[47] = 0xcc;
   alloc[63] = 0xdd;
   alloc[79] = 0xee;
   alloc[95] = 0xff;
 
-  // __hwasan_tag_memory expects untagged pointers.
-  char *p = (char *)__hwasan_tag_pointer(alloc, 0);
-  assert(p);
+  char *p = alloc;
 
   // Write tags to shadow.
   __hwasan_tag_memory(p, 1, 32);
   __hwasan_tag_memory(p + 32, 16, 16);
   __hwasan_tag_memory(p + 48, 0, 32);
   __hwasan_tag_memory(p + 80, 4, 16);
 
   char *q = (char *)__hwasan_tag_pointer(p, 7);