CrashReason: Add MTE tag check faults to the list of crash reasons.

[pcc]

Summary

As of Linux 5.10, the kernel may report either of the two following
crash reasons:

• SEGV_MTEAERR: async MTE tag check fault
• SEGV_MTESERR: sync MTE tag check fault

Teach LLDB about them.

[DavidSpickett]

I assume that the signal displays without tag bits just like any other SEGV?

I was thinking about testing but I don't see any tests for specific signals so it would only be needed if your change to add the tag bits to siginfo has gone in.
(I have been following it from a distance but not sure of the status)

[pcc]

Show fault address for SEGV_MTESERR

[pcc]

In D93495#2462581, @DavidSpickett wrote:

I assume that the signal displays without tag bits just like any other SEGV?

I was thinking about testing but I don't see any tests for specific signals so it would only be needed if your change to add the tag bits to siginfo has gone in.
(I have been following it from a distance but not sure of the status)

With 5.10 the tag bits don't appear but with the latest version of my patch series, which is due to land in 5.11, the tag bits will appear in si_addr in the siginfo retrieved via ptrace(PTRACE_GETSIGINFO) and therefore will appear in the fault address displayed by the debugger. The tag bits are only hidden from signal handlers, and then only if SA_EXPOSE_TAGBITS is clear, so they will be exposed to ptrace no matter whether the debuggee has SA_EXPOSE_TAGBITS clear or set in its signal handler. Note that the tag bits will be exposed for all faults with a fault address, not just SEGV_MTESERR (though with SEGV_MTESERR we only get the low nibble of the tag due to hardware limitations).

I verified all of this behavior in FVP and also noticed that I didn't add a line of code here that makes the fault address visible in the sync tag check fault case so I've now added it.

[DavidSpickett]

Great, thanks for the explanation.

I don't think this needs a test since (for now) it acts just like existing signals.

The change looks good to me but @labath should approve since I don't know all the signal handling details.

Edit: one thing, the commit message should include a tag at least for lldb, e.g. "[lldb][AArch64] ..."

[emaste]

lldb/source/Plugins/Process/POSIX/CrashReason.cpp

62

This should perhaps include an #ifdef LINUX(sp?) somehow, since 8 and 9 are not necessarily MTEAERR / MTESERR on all POSIXy systems (although that said I'm planning to reserve 8 and 9 on FreeBSD to match).

[emaste]

lldb/source/Plugins/Process/POSIX/CrashReason.cpp

62

(https://reviews.freebsd.org/D27785)

[pcc]

lldb/source/Plugins/Process/POSIX/CrashReason.cpp

62

Makes sense, done.

For now I left SEGV_BNDERR as is, although it doesn't appear that any of FreeBSD, OpenBSD, NetBSD, illumos or XNU have defined anything besides SEGV_MAPERR and SEGV_ACCERR so it might be good to put SEGV_BNDERR under the #ifdef as well.

[emaste]

lldb/source/Plugins/Process/POSIX/CrashReason.cpp

62

What do you think about

#ifdef __linux__
#ifndef SEGV_MTEAERR
#define SEGV_MTEAERR 8
#endif

#ifdef SEGV_MTEAERR
  case SEGV_MTEAERR:
    return CrashReason::eAsyncTagCheckFault;
#endif

We're likely to have SEGV_MTEAERR in FreeBSD soon

lldb/source/Plugins/Process/POSIX/CrashReason.h

 
 enum class CrashReason {
   eInvalidCrashReason,
 
   // SIGSEGV crash reasons.
   eInvalidAddress,
   ePrivilegedAddress,
   eBoundViolation,
+  eAsyncTagCheckFault,
+  eSyncTagCheckFault,
 
   // SIGILL crash reasons.
   eIllegalOpcode,
   eIllegalOperand,
   eIllegalAddressingMode,
   eIllegalTrap,
   ePrivilegedOpcode,
   ePrivilegedRegister,

lldb/source/Plugins/Process/POSIX/CrashReason.cpp

case SEGV_MAPERR:
     return CrashReason::eInvalidAddress;
   case SEGV_ACCERR:
     return CrashReason::ePrivilegedAddress;
 #ifndef SEGV_BNDERR
 #define SEGV_BNDERR 3
 #endif
   case SEGV_BNDERR:
     return CrashReason::eBoundViolation;
+#ifdef __linux__
+#ifndef SEGV_MTEAERR
+#define SEGV_MTEAERR 8
+#endif
+  case SEGV_MTEAERR:
+    return CrashReason::eAsyncTagCheckFault;
+#ifndef SEGV_MTESERR
+#define SEGV_MTESERR 9
+#endif
+  case SEGV_MTESERR:
+    return CrashReason::eSyncTagCheckFault;
+#endif // __linux__
   }
 
   return CrashReason::eInvalidCrashReason;
 }
 
 CrashReason GetCrashReasonForSIGILL(const siginfo_t &info) {
   assert(info.si_signo == SIGILL);
 
case CrashReason::eInvalidAddress:
     break;
   case CrashReason::ePrivilegedAddress:
     str = "signal SIGSEGV: address access protected";
     AppendFaultAddr(str, fault_addr);
     break;
   case CrashReason::eBoundViolation:
     str = "signal SIGSEGV: bound violation";
     break;
+  case CrashReason::eAsyncTagCheckFault:
+    str = "signal SIGSEGV: async tag check fault";
+    break;
+  case CrashReason::eSyncTagCheckFault:
+    str = "signal SIGSEGV: sync tag check fault";
+    AppendFaultAddr(str, fault_addr);
+    break;
   case CrashReason::eIllegalOpcode:
     str = "signal SIGILL: illegal instruction";
     break;
   case CrashReason::eIllegalOperand:
     str = "signal SIGILL: illegal instruction operand";
     break;
   case CrashReason::eIllegalAddressingMode:
     str = "signal SIGILL: illegal addressing mode";
case CrashReason::eInvalidAddress:
     str = "eInvalidAddress";
     break;
   case CrashReason::ePrivilegedAddress:
     str = "ePrivilegedAddress";
     break;
   case CrashReason::eBoundViolation:
     str = "eBoundViolation";
     break;
+  case CrashReason::eAsyncTagCheckFault:
+    str = "eAsyncTagCheckFault";
+    break;
+  case CrashReason::eSyncTagCheckFault:
+    str = "eSyncTagCheckFault";
+    break;
 
   // SIGILL crash reasons.
   case CrashReason::eIllegalOpcode:
     str = "eIllegalOpcode";
     break;
   case CrashReason::eIllegalOperand:
     str = "eIllegalOperand";
     break;