7 files changed, 158 insertions, 28 deletions

diff --git a/doc/extensions.html b/doc/extensions.html
index 7f712a62..368b527c 100644
--- a/doc/extensions.html
+++ b/doc/extensions.html
@@ -365,25 +365,30 @@ the toolchain used to compile LuaJIT:
 </tr>
 <tr class="odd separate">
 <td class="excplatform">POSIX/x64, DWARF2 unwinding</td>
-<td class="exccompiler">GCC 4.3+</td>
+<td class="exccompiler">GCC 4.3+, Clang</td>
 <td class="excinterop"><b style="color: #00a000;">Full</b></td>
 </tr>
 <tr class="even">
+<td class="excplatform">ARM <tt>-DLUAJIT_UNWIND_EXTERNAL</tt></td>
+<td class="exccompiler">GCC, Clang</td>
+<td class="excinterop"><b style="color: #00a000;">Full</b></td>
+</tr>
+<tr class="odd">
 <td class="excplatform">Other platforms, DWARF2 unwinding</td>
-<td class="exccompiler">GCC</td>
+<td class="exccompiler">GCC, Clang</td>
 <td class="excinterop"><b style="color: #c06000;">Limited</b></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td class="excplatform">Windows/x64</td>
 <td class="exccompiler">MSVC or WinSDK</td>
 <td class="excinterop"><b style="color: #00a000;">Full</b></td>
 </tr>
-<tr class="even">
+<tr class="odd">
 <td class="excplatform">Windows/x86</td>
 <td class="exccompiler">Any</td>
-<td class="excinterop"><b style="color: #a00000;">No</b></td>
+<td class="excinterop"><b style="color: #00a000;">Full</b></td>
 </tr>
-<tr class="odd">
+<tr class="even">
 <td class="excplatform">Other platforms</td>
 <td class="exccompiler">Other compilers</td>
 <td class="excinterop"><b style="color: #a00000;">No</b></td>
@@ -432,14 +437,6 @@ C++ destructors.</li>
 <li>Lua errors <b>cannot</b> be caught on the C++ side.</li>
 <li>Throwing Lua errors across C++ frames will <b>not</b> call
 C++ destructors.</li>
-<li>Additionally, on Windows/x86 with SEH-based C++&nbsp;exceptions:
-it's <b>not</b> safe to throw a Lua error across any frames containing
-a C++ function with any try/catch construct or using variables with
-(implicit) destructors. This also applies to any functions which may be
-inlined in such a function. It doesn't matter whether <tt>lua_error()</tt>
-is called inside or outside of a try/catch or whether any object actually
-needs to be destroyed: the SEH chain is corrupted and this will eventually
-lead to the termination of the process.</li>
 </ul>
 <br class="flush">
 </div>
diff --git a/src/host/buildvm.c b/src/host/buildvm.c
index 6d9e09e1..57b4dc97 100644
--- a/src/host/buildvm.c
+++ b/src/host/buildvm.c
@@ -110,7 +110,7 @@ static const char *sym_decorate(BuildCtx *ctx,
   if (p) {
 #if LJ_TARGET_X86ORX64
     if (!LJ_64 && (ctx->mode == BUILD_coffasm || ctx->mode == BUILD_peobj))
-      name[0] = '@';
+      name[0] = name[1] == 'R' ? '_' : '@';  /* Just for _RtlUnwind@16. */
     else
       *p = '\0';
 #elif LJ_TARGET_PPC && !LJ_TARGET_CONSOLE
diff --git a/src/host/buildvm_peobj.c b/src/host/buildvm_peobj.c
index e8c927d8..42f6ac84 100644
--- a/src/host/buildvm_peobj.c
+++ b/src/host/buildvm_peobj.c
@@ -109,6 +109,8 @@ enum {
 #if LJ_TARGET_X64
   PEOBJ_SECT_PDATA,
   PEOBJ_SECT_XDATA,
+#elif LJ_TARGET_X86
+  PEOBJ_SECT_SXDATA,
 #endif
   PEOBJ_SECT_RDATA_Z,
   PEOBJ_NSECTIONS
@@ -208,6 +210,13 @@ void emit_peobj(BuildCtx *ctx)
   sofs += (pesect[PEOBJ_SECT_XDATA].nreloc = 1) * PEOBJ_RELOC_SIZE;
   /* Flags: 40 = read, 30 = align4, 40 = initialized data. */
   pesect[PEOBJ_SECT_XDATA].flags = 0x40300040;
+#elif LJ_TARGET_X86
+  memcpy(pesect[PEOBJ_SECT_SXDATA].name, ".sxdata", sizeof(".sxdata")-1);
+  pesect[PEOBJ_SECT_SXDATA].ofs = sofs;
+  sofs += (pesect[PEOBJ_SECT_SXDATA].size = 4);
+  pesect[PEOBJ_SECT_SXDATA].relocofs = sofs;
+  /* Flags: 40 = read, 30 = align4, 02 = lnk_info, 40 = initialized data. */
+  pesect[PEOBJ_SECT_SXDATA].flags = 0x40300240;
 #endif
 
   memcpy(pesect[PEOBJ_SECT_RDATA_Z].name, ".rdata$Z", sizeof(".rdata$Z")-1);
@@ -232,7 +241,7 @@ void emit_peobj(BuildCtx *ctx)
   nrsym = ctx->nrelocsym;
   pehdr.nsyms = 1+PEOBJ_NSECTIONS*2 + 1+ctx->nsym + nrsym;
 #if LJ_TARGET_X64
-  pehdr.nsyms += 1;  /* Symbol for lj_err_unwind_win64. */
+  pehdr.nsyms += 1;  /* Symbol for lj_err_unwind_win. */
 #endif
 
   /* Write PE object header and all sections. */
@@ -312,6 +321,19 @@ void emit_peobj(BuildCtx *ctx)
     reloc.type = PEOBJ_RELOC_ADDR32NB;
     owrite(ctx, &reloc, PEOBJ_RELOC_SIZE);
   }
+#elif LJ_TARGET_X86
+  /* Write .sxdata section. */
+  for (i = 0; i < nrsym; i++) {
+    if (!strcmp(ctx->relocsym[i], "_lj_err_unwind_win")) {
+      uint32_t symidx = 1+2+i;
+      owrite(ctx, &symidx, 4);
+      break;
+    }
+  }
+  if (i == nrsym) {
+    fprintf(stderr, "Error: extern lj_err_unwind_win not used\n");
+    exit(1);
+  }
 #endif
 
   /* Write .rdata$Z section. */
@@ -333,8 +355,10 @@ void emit_peobj(BuildCtx *ctx)
 #if LJ_TARGET_X64
     emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_PDATA);
     emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_XDATA);
-    emit_peobj_sym(ctx, "lj_err_unwind_win64", 0,
+    emit_peobj_sym(ctx, "lj_err_unwind_win", 0,
                    PEOBJ_SECT_UNDEF, PEOBJ_TYPE_FUNC, PEOBJ_SCL_EXTERN);
+#elif LJ_TARGET_X86
+    emit_peobj_sym_sect(ctx, pesect, PEOBJ_SECT_SXDATA);
 #endif
 
     emit_peobj_sym(ctx, ctx->beginsym, 0,
diff --git a/src/lj_err.c b/src/lj_err.c
index a847ca07..1314c8db 100644
--- a/src/lj_err.c
+++ b/src/lj_err.c
@@ -46,7 +46,8 @@
 **   the wrapper function feature. Lua errors thrown through C++ frames
 **   cannot be caught by C++ code and C++ destructors are not run.
 **
-** EXT is the default on x64 systems, INT is the default on all other systems.
+** EXT is the default on x64 systems and on Windows, INT is the default on all
+** other systems.
 **
 ** EXT can be manually enabled on POSIX systems using GCC and DWARF2 stack
 ** unwinding with -DLUAJIT_UNWIND_EXTERNAL. *All* C code must be compiled
@@ -55,7 +56,6 @@
 ** and all C libraries that have callbacks which may be used to call back
 ** into Lua. C++ code must *not* be compiled with -fno-exceptions.
 **
-** EXT cannot be enabled on WIN32 since system exceptions use code-driven SEH.
 ** EXT is mandatory on WIN64 since the calling convention has an abundance
 ** of callee-saved registers (rbx, rbp, rsi, rdi, r12-r15, xmm6-xmm15).
 ** The POSIX/x64 interpreter only saves r12/r13 for INT (e.g. PS4).
@@ -63,7 +63,7 @@
 
 #if defined(__GNUC__) && (LJ_TARGET_X64 || defined(LUAJIT_UNWIND_EXTERNAL)) && !LJ_NO_UNWIND
 #define LJ_UNWIND_EXT   1
-#elif LJ_TARGET_X64 && LJ_TARGET_WINDOWS
+#elif LJ_TARGET_WINDOWS
 #define LJ_UNWIND_EXT   1
 #endif
 
@@ -384,7 +384,7 @@ static void err_raise_ext(int errcode)
 
 #endif /* LJ_TARGET_ARM */
 
-#elif LJ_TARGET_X64 && LJ_ABI_WIN
+#elif LJ_ABI_WIN
 
 /*
 ** Someone in Redmond owes me several days of my life. A lot of this is
@@ -402,6 +402,7 @@ static void err_raise_ext(int errcode)
 #define WIN32_LEAN_AND_MEAN
 #include <windows.h>
 
+#if LJ_TARGET_X64
 /* Taken from: http://www.nynaeve.net/?p=99 */
 typedef struct UndocumentedDispatcherContext {
   ULONG64 ControlPc;
@@ -416,11 +417,14 @@ typedef struct UndocumentedDispatcherContext {
   ULONG ScopeIndex;
   ULONG Fill0;
 } UndocumentedDispatcherContext;
+#else
+typedef void *UndocumentedDispatcherContext;
+#endif
 
 /* Another wild guess. */
 extern void __DestructExceptionObject(EXCEPTION_RECORD *rec, int nothrow);
 
-#ifdef MINGW_SDK_INIT
+#if LJ_TARGET_X64 && defined(MINGW_SDK_INIT)
 /* Workaround for broken MinGW64 declaration. */
 VOID RtlUnwindEx_FIXED(PVOID,PVOID,PVOID,PVOID,PVOID,PVOID) asm("RtlUnwindEx");
 #define RtlUnwindEx RtlUnwindEx_FIXED
@@ -434,10 +438,15 @@ VOID RtlUnwindEx_FIXED(PVOID,PVOID,PVOID,PVOID,PVOID,PVOID) asm("RtlUnwindEx");
 #define LJ_EXCODE_CHECK(cl)     (((cl) ^ LJ_EXCODE) <= 0xff)
 #define LJ_EXCODE_ERRCODE(cl)   ((int)((cl) & 0xff))
 
-/* Win64 exception handler for interpreter frame. */
-LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
-  void *cf, CONTEXT *ctx, UndocumentedDispatcherContext *dispatch)
+/* Windows exception handler for interpreter frame. */
+LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win(EXCEPTION_RECORD *rec,
+  void *f, CONTEXT *ctx, UndocumentedDispatcherContext *dispatch)
 {
+#if LJ_TARGET_X64
+  void *cf = f;
+#else
+  void *cf = (char *)f - CFRAME_OFS_SEH;
+#endif
   lua_State *L = cframe_L(cf);
   int errcode = LJ_EXCODE_CHECK(rec->ExceptionCode) ?
                 LJ_EXCODE_ERRCODE(rec->ExceptionCode) : LUA_ERRRUN;
@@ -457,6 +466,7 @@ LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
         /* Don't catch access violations etc. */
         return ExceptionContinueSearch;
       }
+#if LJ_TARGET_X64
       /* Unwind the stack and call all handlers for all lower C frames
       ** (including ourselves) again with EH_UNWINDING set. Then set
       ** rsp = cf, rax = errcode and jump to the specified target.
@@ -466,6 +476,18 @@ LJ_FUNCA EXCEPTION_DISPOSITION lj_err_unwind_win64(EXCEPTION_RECORD *rec,
                                lj_vm_unwind_c_eh),
                   rec, (void *)(uintptr_t)errcode, ctx, dispatch->HistoryTable);
       /* RtlUnwindEx should never return. */
+#else
+      UNUSED(ctx);
+      UNUSED(dispatch);
+      /* Call all handlers for all lower C frames (including ourselves) again
+      ** with EH_UNWINDING set. Then call the specified function, passing cf
+      ** and errcode.
+      */
+      lj_vm_rtlunwind(cf, (void *)rec,
+        (cframe_unwind_ff(cf2) && errcode != LUA_YIELD) ?
+        (void *)lj_vm_unwind_ff : (void *)lj_vm_unwind_c, errcode);
+      /* lj_vm_rtlunwind does not return. */
+#endif
     }
   }
   return ExceptionContinueSearch;
diff --git a/src/lj_frame.h b/src/lj_frame.h
index fc0e281c..db2e4da1 100644
--- a/src/lj_frame.h
+++ b/src/lj_frame.h
@@ -116,6 +116,17 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
 
 /* These definitions must match with the arch-specific *.dasc files. */
 #if LJ_TARGET_X86
+#if LJ_ABI_WIN
+#define CFRAME_OFS_ERRF         (19*4)
+#define CFRAME_OFS_NRES         (18*4)
+#define CFRAME_OFS_PREV         (17*4)
+#define CFRAME_OFS_L            (16*4)
+#define CFRAME_OFS_SEH          (9*4)
+#define CFRAME_OFS_PC           (6*4)
+#define CFRAME_OFS_MULTRES      (5*4)
+#define CFRAME_SIZE             (16*4)
+#define CFRAME_SHIFT_MULTRES    0
+#else
 #define CFRAME_OFS_ERRF         (15*4)
 #define CFRAME_OFS_NRES         (14*4)
 #define CFRAME_OFS_PREV         (13*4)
@@ -124,6 +135,7 @@ enum { LJ_CONT_TAILCALL, LJ_CONT_FFI_CALLBACK };  /* Special continuations. */
 #define CFRAME_OFS_MULTRES      (5*4)
 #define CFRAME_SIZE             (12*4)
 #define CFRAME_SHIFT_MULTRES    0
+#endif
 #elif LJ_TARGET_X64
 #if LJ_ABI_WIN
 #define CFRAME_OFS_PREV         (13*8)
diff --git a/src/lj_vm.h b/src/lj_vm.h
index be35295d..d605b143 100644
--- a/src/lj_vm.h
+++ b/src/lj_vm.h
@@ -17,6 +17,10 @@ LJ_ASMF int lj_vm_cpcall(lua_State *L, lua_CFunction func, void *ud,
 LJ_ASMF int lj_vm_resume(lua_State *L, TValue *base, int nres1, ptrdiff_t ef);
 LJ_ASMF_NORET void LJ_FASTCALL lj_vm_unwind_c(void *cframe, int errcode);
 LJ_ASMF_NORET void LJ_FASTCALL lj_vm_unwind_ff(void *cframe);
+#if LJ_ABI_WIN && LJ_TARGET_X86
+LJ_ASMF_NORET void LJ_FASTCALL lj_vm_rtlunwind(void *cframe, void *excptrec,
+                                               void *unwinder, int errcode);
+#endif
 LJ_ASMF void lj_vm_unwind_c_eh(void);
 LJ_ASMF void lj_vm_unwind_ff_eh(void);
 #if LJ_TARGET_X86ORX64
diff --git a/src/vm_x86.dasc b/src/vm_x86.dasc
index f108c0b5..39ccaa2e 100644
--- a/src/vm_x86.dasc
+++ b/src/vm_x86.dasc
@@ -121,19 +121,68 @@
 |//-----------------------------------------------------------------------
 |.if not X64            // x86 stack layout.
 |
-|.define CFRAME_SPACE,  aword*7                 // Delta for esp (see <--).
+|.if WIN
+|
+|.define CFRAME_SPACE,  aword*9                 // Delta for esp (see <--).
 |.macro saveregs_
 |  push edi; push esi; push ebx
+|  push extern lj_err_unwind_win
+|  fs; push dword [0]
+|  fs; mov [0], esp
 |  sub esp, CFRAME_SPACE
 |.endmacro
-|.macro saveregs
-|  push ebp; saveregs_
+|.macro restoreregs
+|  add esp, CFRAME_SPACE
+|  fs; pop dword [0]
+|  pop edi      // Short for esp += 4.
+|  pop ebx; pop esi; pop edi; pop ebp
+|.endmacro
+|
+|.else
+|
+|.define CFRAME_SPACE,  aword*7                 // Delta for esp (see <--).
+|.macro saveregs_
+|  push edi; push esi; push ebx
+|  sub esp, CFRAME_SPACE
 |.endmacro
 |.macro restoreregs
 |  add esp, CFRAME_SPACE
 |  pop ebx; pop esi; pop edi; pop ebp
 |.endmacro
 |
+|.endif
+|
+|.macro saveregs
+|  push ebp; saveregs_
+|.endmacro
+|
+|.if WIN
+|.define SAVE_ERRF,     aword [esp+aword*19]    // vm_pcall/vm_cpcall only.
+|.define SAVE_NRES,     aword [esp+aword*18]
+|.define SAVE_CFRAME,   aword [esp+aword*17]
+|.define SAVE_L,        aword [esp+aword*16]
+|//----- 16 byte aligned, ^^^ arguments from C caller
+|.define SAVE_RET,      aword [esp+aword*15]    //<-- esp entering interpreter.
+|.define SAVE_R4,       aword [esp+aword*14]
+|.define SAVE_R3,       aword [esp+aword*13]
+|.define SAVE_R2,       aword [esp+aword*12]
+|//----- 16 byte aligned
+|.define SAVE_R1,       aword [esp+aword*11]
+|.define SEH_FUNC,      aword [esp+aword*10]
+|.define SEH_NEXT,      aword [esp+aword*9]     //<-- esp after register saves.
+|.define UNUSED2,       aword [esp+aword*8]
+|//----- 16 byte aligned
+|.define UNUSED1,       aword [esp+aword*7]
+|.define SAVE_PC,       aword [esp+aword*6]
+|.define TMP2,          aword [esp+aword*5]
+|.define TMP1,          aword [esp+aword*4]
+|//----- 16 byte aligned
+|.define ARG4,          aword [esp+aword*3]
+|.define ARG3,          aword [esp+aword*2]
+|.define ARG2,          aword [esp+aword*1]
+|.define ARG1,          aword [esp]             //<-- esp while in interpreter.
+|//----- 16 byte aligned, ^^^ arguments for C callee
+|.else
 |.define SAVE_ERRF,     aword [esp+aword*15]    // vm_pcall/vm_cpcall only.
 |.define SAVE_NRES,     aword [esp+aword*14]
 |.define SAVE_CFRAME,   aword [esp+aword*13]
@@ -154,6 +203,7 @@
 |.define ARG2,          aword [esp+aword*1]
 |.define ARG1,          aword [esp]             //<-- esp while in interpreter.
 |//----- 16 byte aligned, ^^^ arguments for C callee
+|.endif
 |
 |// FPARGx overlaps ARGx and ARG(x+1) on x86.
 |.define FPARG3,        qword [esp+qword*1]
@@ -554,6 +604,10 @@ static void build_subroutines(BuildCtx *ctx)
   |.else
   |  mov eax, FCARG2                    // Error return status for vm_pcall.
   |  mov esp, FCARG1
+  |.if WIN
+  |  lea FCARG1, SEH_NEXT
+  |  fs; mov [0], FCARG1
+  |.endif
   |.endif
   |->vm_unwind_c_eh:                    // Landing pad for external unwinder.
   |  mov L:RB, SAVE_L
@@ -577,6 +631,10 @@ static void build_subroutines(BuildCtx *ctx)
   |.else
   |  and FCARG1, CFRAME_RAWMASK
   |  mov esp, FCARG1
+  |.if WIN
+  |  lea FCARG1, SEH_NEXT
+  |  fs; mov [0], FCARG1
+  |.endif
   |.endif
   |->vm_unwind_ff_eh:                   // Landing pad for external unwinder.
   |  mov L:RB, SAVE_L
@@ -590,6 +648,19 @@ static void build_subroutines(BuildCtx *ctx)
   |  set_vmstate INTERP
   |  jmp ->vm_returnc                   // Increments RD/MULTRES and returns.
   |
+  |.if WIN and not X64
+  |->vm_rtlunwind@16:                   // Thin layer around RtlUnwind.
+  |  // (void *cframe, void *excptrec, void *unwinder, int errcode)
+  |  mov [esp], FCARG1                  // Return value for RtlUnwind.
+  |  push FCARG2                        // Exception record for RtlUnwind.
+  |  push 0                             // Ignored by RtlUnwind.
+  |  push dword [FCARG1+CFRAME_OFS_SEH]
+  |  call extern RtlUnwind@16           // Violates ABI (clobbers too much).
+  |  mov FCARG1, eax
+  |  mov FCARG2, [esp+4]                // errcode (for vm_unwind_c).
+  |  ret                                // Jump to unwinder.
+  |.endif
+  |
   |//-----------------------------------------------------------------------
   |//-- Grow stack for calls -----------------------------------------------
   |//-----------------------------------------------------------------------