[PATCH 0/1] MR408: ntdll: Fix x64 single step exception detection on macOS.

[Piotr Caban (＠piotr)]

On macOS si_code is always set to TRAP_BRKPT.

[PATCH 1/1] ntdll: Fix x64 single step exception detection on macOS.

[Piotr Caban]

From: Piotr Caban

On macOS si_code is always set to TRAP_BRKPT.
---
 dlls/ntdll/unix/signal_x86_64.c | 20 +++++++++++++++++++-
 1 file changed, 19 insertions(+), 1 deletion(-)

diff --git a/dlls/ntdll/unix/signal_x86_64.c b/dlls/ntdll/unix/signal_x86_64.c
index 6c87e347eac..b54a3ca88f8 100644
--- a/dlls/ntdll/unix/signal_x86_64.c
+++ b/dlls/ntdll/unix/signal_x86_64.c
@@ -2687,13 +2687,31 @@ static void trap_handler( int signal, siginfo_t *siginfo, void *sigcontext )
     EXCEPTION_RECORD rec = { 0 };
     struct xcontext context;
     ucontext_t *ucontext = sigcontext;
+    int si_code;

if (handle_syscall_trap( sigcontext )) return;

rec.ExceptionAddress = (void *)RIP_sig(ucontext);
     save_context( &context, sigcontext );

-    switch (siginfo->si_code)
+#if defined(__APPLE__)
+    switch (TRAP_sig(ucontext))
+    {
+    case TRAP_x86_TRCTRAP:
+        si_code = TRAP_TRACE;
+        break;
+    case TRAP_x86_BPTFLT:
+        si_code = TRAP_BRKPT;
+        break;
+    default:
+        si_code = siginfo->si_code;
+        break;
+    }
+#else
+    si_code = siginfo->si_code;
+#endif
+
+    switch (si_code)
     {
     case TRAP_TRACE:  /* Single-step exception */
     case 4 /* TRAP_HWBKPT */: /* Hardware breakpoint exception */

[Jinoh Kang (＠iamahuman)]

Jinoh Kang (@iamahuman) commented about dlls/ntdll/unix/signal_x86_64.c:
>  EXCEPTION_RECORD rec = { 0 };
>  struct xcontext context;
>  ucontext_t *ucontext = sigcontext;
> -
>
> int si_code;
>
> if (handle_syscall_trap( sigcontext )) return;
>
> rec.ExceptionAddress = (void *)RIP_sig(ucontext);
> save_context( &context, sigcontext );
> - switch (siginfo->si_code)
>
> +#if defined(__APPLE__)
> - switch (TRAP_sig(ucontext))

I noticed that we already rely on `TRAP_SIG(ucontext)` to determine the correct exception in `segv_handler`.

If we can trust `TRAP_SIG(ucontext)` over `siginfo->si_code` for `SIGTRAP` too in all platforms, perhaps it makes sense to make this switch unconditional.

[Piotr Caban (＠piotr)]

On Sat Jul  9 06:52:04 2022 +0000, Jinoh Kang wrote:
> I noticed that we already rely on `TRAP_SIG(ucontext)` to determine the
> correct exception in `segv_handler`.
> If we can trust `TRAP_SIG(ucontext)` over `siginfo->si_code` for
> `SIGTRAP` too in all platforms, perhaps it makes sense to make this
> switch unconditional.

I haven't done it this way because I was unsure of what happens in case of single step and hardware breakpoint on the same instruction. I turns out it doesn't need special handling in x86_64 case. I've also done some testing with icebp instruction, after changing the code it no longer needs special handling. I will push changed version.