1 files changed, 15 insertions, 4 deletions

diff --git a/src/buildvm_x86.dasc b/src/buildvm_x86.dasc
index b64617f9..67282163 100644
--- a/src/buildvm_x86.dasc
+++ b/src/buildvm_x86.dasc
@@ -691,12 +691,12 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |->vm_call_dispatch:
   |  mov LFUNC:RB, [RA-8]
   |  cmp dword [RA-4], LJ_TFUNC
-  |  jne ->vmeta_call           // Ensure KBASE defined and != BASE.
+  |  jne ->vmeta_call                   // Ensure KBASE defined and != BASE.
   |
   |->vm_call_dispatch_f:
   |  mov BASE, RA
   |  ins_call
-  |  // BASE = new base, RD = nargs+1
+  |  // BASE = new base, RB = func, RD = nargs+1, PC = caller PC
   |
   |->vm_cpcall:                         // Setup protected C frame, call C.
   |  // (lua_State *L, lua_CFunction func, void *ud, lua_CPFunction cp)
@@ -760,10 +760,14 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  mov PC, [RB-12]                    // Restore PC from [cont|PC].
   |.if X64
   |  movsxd RAa, dword [RB-16]          // May be negative on WIN64 with debug.
+  |  test RA, RA
+  |  jz >1
   |  lea KBASEa, qword [=>0]
   |  add RAa, KBASEa
   |.else
   |  mov RA, dword [RB-16]
+  |  test RA, RA
+  |  jz >1
   |.endif
   |  mov LFUNC:KBASE, [BASE-8]
   |  mov KBASE, LFUNC:KBASE->pc
@@ -771,6 +775,12 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  // BASE = base, RC = result, RB = meta base
   |  jmp RAa                            // Jump to continuation.
   |
+  |1:  // Tail call from C function.
+  |  sub RB, BASE
+  |  shr RB, 3
+  |  lea RD, [RB-1]
+  |  jmp ->vm_call_tail
+  |
   |->cont_cat:                          // BASE = base, RC = result, RB = mbase
   |  movzx RA, PC_RB
   |  sub RB, 16
@@ -2735,10 +2745,11 @@ static void build_subroutines(BuildCtx *ctx, int cmov, int sse)
   |  test RD, RD
   |  lea NARGS:RD, [RA+1]
   |  mov LFUNC:RB, [BASE-8]
-  |  jne >2                             // Returned -1?
+  |  jne ->vm_call_tail                 // Returned -1?
   |  ins_callt                          // Returned 0: retry fast path.
   |
-  |2:  // Reconstruct previous base for vmeta_call during tailcall.
+  |// Reconstruct previous base for vmeta_call during tailcall.
+  |->vm_call_tail:
   |  mov RA, BASE
   |  test PC, FRAME_TYPE
   |  jnz >3