This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Fix remaining Java failures on SPARC64/Linux


This fixes the 3 remaining Java failures on SPARC64/Linux, which pertain to 
unwinding through a signal raised from a leaf function.  A small adjustment 
is necessary on SPARC64 because of the stack bias; it is already present in 
the Solaris unwinder.

Tested on SPARC64/Linux, applied on the mainline.


2009-12-11  Eric Botcazou  <ebotcazou@adacore.com>

	* config/sparc/linux-unwind.h (MD_FROB_UPDATE_CONTEXT, 64-bit): Define.
	(sparc64_frob_update_context): New function.


-- 
Eric Botcazou
Index: config/sparc/linux-unwind.h
===================================================================
--- config/sparc/linux-unwind.h	(revision 155123)
+++ config/sparc/linux-unwind.h	(working copy)
@@ -98,6 +98,23 @@ sparc64_fallback_frame_state (struct _Un
   return _URC_NO_REASON;
 }
 
+#define MD_FROB_UPDATE_CONTEXT sparc64_frob_update_context
+
+static void
+sparc64_frob_update_context (struct _Unwind_Context *context,
+			     _Unwind_FrameState *fs)
+{
+  /* The column of %sp contains the old CFA, not the old value of %sp.
+     The CFA offset already comprises the stack bias so, when %sp is the
+     CFA register, we must avoid counting the stack bias twice.  Do not
+     do that for signal frames as the offset is artificial for them.  */
+  if (fs->regs.cfa_reg == __builtin_dwarf_sp_column ()
+      && fs->regs.cfa_how == CFA_REG_OFFSET
+      && fs->regs.cfa_offset != 0
+      && !fs->signal_frame)
+    context->cfa -= 2047;
+}
+
 #else
 
 /* 32-bit SPARC version */

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]