Fix PR target/100402

Message ID 36824219.10thIPus4b@fomalhaut
State New
Headers show
Series
  • Fix PR target/100402
Related show

Commit Message

Eric Botcazou May 5, 2021, 9:02 p.m.
This is a regression for 64-bit Windows present from mainline down to the 9
branch and introduced by my fix for PR target/99234.  Again SEH, but with a
twist related to the way MinGW implements setjmp/longjmp that I discovered
while debugging it: setjmp/longjmp is piggybacked on SEH with recent versions
of MinGW, i.e. the longjmp performs a bona-fide unwinding of the stack, as it
calls RtlUnwindEx with the second argument passed to setjmp, which is the
result of __builtin_frame_address (0) in the MinGW header file:

  define setjmp(BUF) _setjmp((BUF), __builtin_frame_address (0))

This means that we directly expose the frame pointer to the SEH machinery here
(unlike with regular exception handling where we use an intermediate CFA) and
thus that we cannot do whatever we want with it.  The old code would leave it
unaligned, i.e. not multiple of 16, whereas the new code aligns it, but this
breaks for some reason; at least it appears that a .seh_setframe directive
with 0 as second argument always works, so the fix aligns it this way.

Tested on x86-64/Windows, applied on the affected branches as obvious.


2021-05-05  Eric Botcazou  <ebotcazou@adacore.com>

	PR target/100402
	* config/i386/i386.c (ix86_compute_frame_layout): For a SEH target,
	always return the establisher frame for __builtin_frame_address (0).


2021-05-05  Eric Botcazou  <ebotcazou@adacore.com>

	* gcc.c-torture/execute/20210505-1.c: New test.

-- 
Eric Botcazou

Patch

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index 06b0f5814ea..ecc15358efe 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -6672,12 +6672,29 @@  ix86_compute_frame_layout (void)
 	 area, see the SEH code in config/i386/winnt.c for the rationale.  */
       frame->hard_frame_pointer_offset = frame->sse_reg_save_offset;
 
-      /* If we can leave the frame pointer where it is, do so.  Also, return
+      /* If we can leave the frame pointer where it is, do so; however return
 	 the establisher frame for __builtin_frame_address (0) or else if the
-	 frame overflows the SEH maximum frame size.  */
+	 frame overflows the SEH maximum frame size.
+
+	 Note that the value returned by __builtin_frame_address (0) is quite
+	 constrained, because setjmp is piggybacked on the SEH machinery with
+	 recent versions of MinGW:
+
+	  #    elif defined(__SEH__)
+	  #     if defined(__aarch64__) || defined(_ARM64_)
+	  #      define setjmp(BUF) _setjmp((BUF), __builtin_sponentry())
+	  #     elif (__MINGW_GCC_VERSION < 40702)
+	  #      define setjmp(BUF) _setjmp((BUF), mingw_getsp())
+	  #     else
+	  #      define setjmp(BUF) _setjmp((BUF), __builtin_frame_address (0))
+	  #     endif
+
+	 and the second argument passed to _setjmp, if not null, is forwarded
+	 to the TargetFrame parameter of RtlUnwindEx by longjmp (after it has
+	 built an ExceptionRecord on the fly describing the setjmp buffer).  */
       const HOST_WIDE_INT diff
 	= frame->stack_pointer_offset - frame->hard_frame_pointer_offset;
-      if (diff <= 255)
+      if (diff <= 255 && !crtl->accesses_prior_frames)
 	{
 	  /* The resulting diff will be a multiple of 16 lower than 255,
 	     i.e. at most 240 as required by the unwind data structure.  */