Index: lib/Target/X86/X86ISelLowering.h
===================================================================
--- lib/Target/X86/X86ISelLowering.h
+++ lib/Target/X86/X86ISelLowering.h
@@ -1138,6 +1138,9 @@
     MachineBasicBlock *EmitLoweredSegAlloca(MachineInstr *MI,
                                             MachineBasicBlock *BB) const;
 
+    MachineBasicBlock *EmitLoweredTLSAddr(MachineInstr *MI,
+                                          MachineBasicBlock *BB) const;
+
     MachineBasicBlock *EmitLoweredTLSCall(MachineInstr *MI,
                                           MachineBasicBlock *BB) const;
 
Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -22660,6 +22660,35 @@
 }
 
 MachineBasicBlock *
+X86TargetLowering::EmitLoweredTLSAddr(MachineInstr *MI,
+                                      MachineBasicBlock *BB) const {
+  // So, here we replace TLSADDR with the sequence:
+  // adjust_stackdown -> TLSADDR -> adjust_stackup.
+  // We need this because TLSADDR is lowered into calls
+  // inside MC, therefore without the two markers shrink-wrapping
+  // may push the prologue/epilogue pass them.
+  const TargetInstrInfo &TII = *Subtarget.getInstrInfo();
+  DebugLoc DL = MI->getDebugLoc();
+  MachineFunction &MF = *BB->getParent();
+
+  // Emit CALLSEQ_START right before the instruction.
+  unsigned AdjStackDown = TII.getCallFrameSetupOpcode();
+  MachineInstrBuilder CallseqStart =
+    BuildMI(MF, DL, TII.get(AdjStackDown)).addImm(0);
+  BB->insert(MachineBasicBlock::iterator(MI), CallseqStart);
+
+  // Emit CALLSEQ_END  right after the instruction.
+  // We don't call erase from parent because we want to keep the
+  // original instruction around.
+  unsigned AdjStackUp = TII.getCallFrameDestroyOpcode();
+  MachineInstrBuilder CallseqEnd =
+    BuildMI(MF, DL, TII.get(AdjStackUp)).addImm(0).addImm(0);
+  BB->insertAfter(MachineBasicBlock::iterator(MI), CallseqEnd);
+
+  return BB;
+}
+
+MachineBasicBlock *
 X86TargetLowering::EmitLoweredTLSCall(MachineInstr *MI,
                                       MachineBasicBlock *BB) const {
   // This is pretty easy.  We're taking the value that we received from
@@ -23039,6 +23068,11 @@
   case X86::TCRETURNri64:
   case X86::TCRETURNmi64:
     return BB;
+  case X86::TLS_addr32:
+  case X86::TLS_addr64:
+  case X86::TLS_base_addr32:
+  case X86::TLS_base_addr64:
+    return EmitLoweredTLSAddr(MI, BB);
   case X86::WIN_ALLOCA:
     return EmitLoweredWinAlloca(MI, BB);
   case X86::CATCHRET:
Index: lib/Target/X86/X86InstrCompiler.td
===================================================================
--- lib/Target/X86/X86InstrCompiler.td
+++ lib/Target/X86/X86InstrCompiler.td
@@ -436,7 +436,7 @@
             MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
             XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
             XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
-    Uses = [ESP] in {
+    usesCustomInserter = 1, Uses = [ESP] in {
 def TLS_addr32 : I<0, Pseudo, (outs), (ins i32mem:$sym),
                   "# TLS_addr32",
                   [(X86tlsaddr tls32addr:$sym)]>,
@@ -456,7 +456,7 @@
             MM0, MM1, MM2, MM3, MM4, MM5, MM6, MM7,
             XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7,
             XMM8, XMM9, XMM10, XMM11, XMM12, XMM13, XMM14, XMM15, EFLAGS],
-    Uses = [RSP] in {
+    usesCustomInserter = 1, Uses = [RSP] in {
 def TLS_addr64 : I<0, Pseudo, (outs), (ins i64mem:$sym),
                    "# TLS_addr64",
                   [(X86tlsaddr tls64addr:$sym)]>,
Index: test/CodeGen/X86/tls-shrink-wrapping.ll
===================================================================
--- test/CodeGen/X86/tls-shrink-wrapping.ll
+++ test/CodeGen/X86/tls-shrink-wrapping.ll
@@ -0,0 +1,47 @@
+; RUN: llc < %s -march=x86-64 -mtriple=x86_64-linux-gnu -relocation-model=pic | FileCheck %s
+
+@i = external thread_local global i32, align 4
+
+define i32 @g() {
+entry:
+  %0 = load i32, i32* @i, align 4, !tbaa !2
+  %tobool = icmp eq i32 %0, 0
+  br i1 %tobool, label %if.end, label %if.then
+
+if.then:                                          ; preds = %entry
+  store i32 0, i32* @i, align 4, !tbaa !2
+  tail call void (...) @f()
+  %.pre = load i32, i32* @i, align 4, !tbaa !2
+  br label %if.end
+
+if.end:                                           ; preds = %entry, %if.then
+  %1 = phi i32 [ 0, %entry ], [ %.pre, %if.then ]
+  ret i32 %1
+}
+
+; CHECK: g:                                      # @g
+; CHECK-NEXT:         .cfi_startproc
+; CHECK-NEXT: # BB#0:                                 # %entry
+; CHECK-NEXT:         pushq   %rbx
+; CHECK-NEXT: .Ltmp0:
+; CHECK-NEXT:         .cfi_def_cfa_offset 16
+; CHECK-NEXT: .Ltmp1:
+; CHECK-NEXT:         .cfi_offset %rbx, -16
+; CHECK-NEXT:         data16
+; CHECK-NEXT:         leaq    i@TLSGD(%rip), %rdi
+; CHECK-NEXT:         data16
+; CHECK-NEXT:         data16
+; CHECK-NEXT:         rex64
+; CHECK-NEXT:         callq   __tls_get_addr@PLT
+
+declare void @f(...)
+
+!llvm.module.flags = !{!0}
+!llvm.ident = !{!1}
+
+!0 = !{i32 1, !"PIC Level", i32 2}
+!1 = !{!"clang version 3.9.0 (trunk 261236)"}
+!2 = !{!3, !3, i64 0}
+!3 = !{!"int", !4, i64 0}
+!4 = !{!"omnipotent char", !5, i64 0}
+!5 = !{!"Simple C/C++ TBAA"}