[NFC][compiler-rt][hwasan] Move __hwasan_tag_mismatch4 to hwasan.cpp

[leonardchan]

Summary

This can be moved out of hwasan_linux.cpp because it can be re-used in the fuchsia port.

[DavidSpickett]

@leonardchan Sorry for the late report, took us some time to get to this but - this caused a test failure on Linaro's AArch64 bots.

I've fixed this with https://reviews.llvm.org/rGbb4f7b916661d2517fb8dfadfc8629e6bed56ce4

What happened is that we now see frame 0 as a tag mismatch function and frame 1 is main. With my limited knowledge of sanitizers I think this is fine and the test is just ensuring that there is some main frame shown. So allowing it to be frame 0 or 1 is what I did.

If that's not the case let me know. If, for instance, the __hwasan_tag_mismatch functions are supposed to be hidden from the backtrace somehow.

[leonardchan]

In D105669#2892913, @DavidSpickett wrote:

@leonardchan Sorry for the late report, took us some time to get to this but - this caused a test failure on Linaro's AArch64 bots.

I've fixed this with https://reviews.llvm.org/rGbb4f7b916661d2517fb8dfadfc8629e6bed56ce4

What happened is that we now see frame 0 as a tag mismatch function and frame 1 is main. With my limited knowledge of sanitizers I think this is fine and the test is just ensuring that there is some main frame shown. So allowing it to be frame 0 or 1 is what I did.

If that's not the case let me know. If, for instance, the __hwasan_tag_mismatch functions are supposed to be hidden from the backtrace somehow.

Thanks for catching this! And sorry I didn't notice it earlier. I *think* __hwasan_tag_mismatch appearing in the backtrace might not be too important. It looks like the purpose of that test is more to check that hwasan hits those errors. @eugenis or @vitalybuka may be able to correct me on this.

[eugenis]

In D105669#2894003, @leonardchan wrote:

In D105669#2892913, @DavidSpickett wrote:

If that's not the case let me know. If, for instance, the __hwasan_tag_mismatch functions are supposed to be hidden from the backtrace somehow.

Thanks for catching this! And sorry I didn't notice it earlier. I *think* __hwasan_tag_mismatch appearing in the backtrace might not be too important. It looks like the purpose of that test is more to check that hwasan hits those errors. @eugenis or @vitalybuka may be able to correct me on this.

Yes this should be fine.

compiler-rt/lib/hwasan/hwasan.cpp

 #endif
 
 extern "C" {
 SANITIZER_INTERFACE_ATTRIBUTE
 void __sanitizer_print_stack_trace() {
   GET_FATAL_STACK_TRACE_PC_BP(StackTrace::GetCurrentPc(), GET_CURRENT_FRAME());
   stack.Print();
 }
+
+// Entry point for interoperability between __hwasan_tag_mismatch (ASM) and the
+// rest of the mismatch handling code (C++).
+void __hwasan_tag_mismatch4(uptr addr, uptr access_info, uptr *registers_frame,
+                            size_t outsize) {
+  __hwasan::HwasanTagMismatch(addr, access_info, registers_frame, outsize);
+}
+
 } // extern "C"

compiler-rt/lib/hwasan/hwasan_linux.cpp

void Thread::InitStackAndTls(const InitState *) {
   GetThreadStackAndTls(IsMainThread(), &stack_bottom_, &stack_size, &tls_begin_,
                        &tls_size);
   stack_top_ = stack_bottom_ + stack_size;
   tls_end_ = tls_begin_ + tls_size;
 }
 
 } // namespace __hwasan
 
-// Entry point for interoperability between __hwasan_tag_mismatch (ASM) and the
-// rest of the mismatch handling code (C++).
-void __hwasan_tag_mismatch4(uptr addr, uptr access_info, uptr *registers_frame,
-                            size_t outsize) {
-  __hwasan::HwasanTagMismatch(addr, access_info, registers_frame, outsize);
-}
-
 #endif // SANITIZER_FREEBSD || SANITIZER_LINUX || SANITIZER_NETBSD