8172049: [s390] Implement "JEP 270: Reserved Stack Areas for Critical Sections".
authorgoetz
Tue, 27 Dec 2016 16:10:59 +0100
changeset 43420 a056d6465ef9
parent 43419 a62a0cc7f41a
child 43421 077af5bb5fa4
8172049: [s390] Implement "JEP 270: Reserved Stack Areas for Critical Sections". Reviewed-by: mdoerr
hotspot/src/cpu/s390/vm/c1_LIRAssembler_s390.cpp
hotspot/src/cpu/s390/vm/globalDefinitions_s390.hpp
hotspot/src/cpu/s390/vm/globals_s390.hpp
hotspot/src/cpu/s390/vm/interp_masm_s390.cpp
hotspot/src/cpu/s390/vm/macroAssembler_s390.cpp
hotspot/src/cpu/s390/vm/macroAssembler_s390.hpp
hotspot/src/cpu/s390/vm/s390.ad
hotspot/src/cpu/s390/vm/stubGenerator_s390.cpp
hotspot/src/cpu/s390/vm/templateInterpreterGenerator_s390.cpp
hotspot/src/os_cpu/linux_s390/vm/os_linux_s390.cpp
hotspot/test/runtime/ReservedStack/ReservedStackTest.java
--- a/hotspot/src/cpu/s390/vm/c1_LIRAssembler_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/c1_LIRAssembler_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -1105,16 +1105,16 @@
       }
     case T_FLOAT :
       if (short_disp) {
-                    __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
+        __ z_ste(from->as_float_reg(),  disp_value, disp_reg, dest);
       } else {
-                    __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
+        __ z_stey(from->as_float_reg(), disp_value, disp_reg, dest);
       }
       break;
     case T_DOUBLE:
       if (short_disp) {
-                    __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
+        __ z_std(from->as_double_reg(),  disp_value, disp_reg, dest);
       } else {
-                    __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
+        __ z_stdy(from->as_double_reg(), disp_value, disp_reg, dest);
       }
       break;
     default: ShouldNotReachHere();
@@ -1148,6 +1148,10 @@
     __ restore_return_pc();
   }
 
+  if (StackReservedPages > 0 && compilation()->has_reserved_stack_access()) {
+    __ reserved_stack_check(Z_R14);
+  }
+
   // We need to mark the code position where the load from the safepoint
   // polling page was emitted as relocInfo::poll_return_type here.
   __ relocate(relocInfo::poll_return_type);
--- a/hotspot/src/cpu/s390/vm/globalDefinitions_s390.hpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/globalDefinitions_s390.hpp	Tue Dec 27 16:10:59 2016 +0100
@@ -52,4 +52,6 @@
 // The expected size in bytes of a cache line, used to pad data structures.
 #define DEFAULT_CACHE_LINE_SIZE 256
 
+#define SUPPORT_RESERVED_STACK_AREA
+
 #endif // CPU_S390_VM_GLOBALDEFINITIONS_S390_HPP
--- a/hotspot/src/cpu/s390/vm/globals_s390.hpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/globals_s390.hpp	Tue Dec 27 16:10:59 2016 +0100
@@ -56,7 +56,7 @@
 // Java_java_net_SocketOutputStream_socketWrite0() uses a 64k buffer on the
 // stack. To pass stack overflow tests we need 20 shadow pages.
 #define DEFAULT_STACK_SHADOW_PAGES   (20 DEBUG_ONLY(+2))
-#define DEFAULT_STACK_RESERVED_PAGES (0)
+#define DEFAULT_STACK_RESERVED_PAGES (1)
 
 #define MIN_STACK_YELLOW_PAGES     DEFAULT_STACK_YELLOW_PAGES
 #define MIN_STACK_RED_PAGES        DEFAULT_STACK_RED_PAGES
--- a/hotspot/src/cpu/s390/vm/interp_masm_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/interp_masm_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -860,16 +860,39 @@
                                                   bool throw_monitor_exception,
                                                   bool install_monitor_exception,
                                                   bool notify_jvmti) {
-
+  BLOCK_COMMENT("remove_activation {");
   unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception);
 
   // Save result (push state before jvmti call and pop it afterwards) and notify jvmti.
   notify_method_exit(false, state, notify_jvmti ? NotifyJVMTI : SkipNotifyJVMTI);
 
+  if (StackReservedPages > 0) {
+    BLOCK_COMMENT("reserved_stack_check:");
+    // Test if reserved zone needs to be enabled.
+    Label no_reserved_zone_enabling;
+
+    // Compare frame pointers. There is no good stack pointer, as with stack
+    // frame compression we can get different SPs when we do calls. A subsequent
+    // call could have a smaller SP, so that this compare succeeds for an
+    // inner call of the method annotated with ReservedStack.
+    z_lg(Z_R0, Address(Z_SP, (intptr_t)_z_abi(callers_sp)));
+    z_clg(Z_R0, Address(Z_thread, JavaThread::reserved_stack_activation_offset())); // Compare with frame pointer in memory.
+    z_brl(no_reserved_zone_enabling);
+
+    // Enable reserved zone again, throw stack overflow exception.
+    call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
+    call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_delayed_StackOverflowError));
+
+    should_not_reach_here();
+
+    bind(no_reserved_zone_enabling);
+  }
+
   verify_oop(Z_tos, state);
   verify_thread();
 
   pop_interpreter_frame(return_pc, Z_ARG2, Z_ARG3);
+  BLOCK_COMMENT("} remove_activation");
 }
 
 // lock object
--- a/hotspot/src/cpu/s390/vm/macroAssembler_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/macroAssembler_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -2666,6 +2666,32 @@
   }
 }
 
+void MacroAssembler::reserved_stack_check(Register return_pc) {
+  // Test if reserved zone needs to be enabled.
+  Label no_reserved_zone_enabling;
+  assert(return_pc == Z_R14, "Return pc must be in R14 before z_br() to StackOverflow stub.");
+  BLOCK_COMMENT("reserved_stack_check {");
+
+  z_clg(Z_SP, Address(Z_thread, JavaThread::reserved_stack_activation_offset()));
+  z_brl(no_reserved_zone_enabling);
+
+  // Enable reserved zone again, throw stack overflow exception.
+  save_return_pc();
+  push_frame_abi160(0);
+  call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), Z_thread);
+  pop_frame();
+  restore_return_pc();
+
+  load_const_optimized(Z_R1, StubRoutines::throw_delayed_StackOverflowError_entry());
+  // Don't use call() or z_basr(), they will invalidate Z_R14 which contains the return pc.
+  z_br(Z_R1);
+
+  should_not_reach_here();
+
+  bind(no_reserved_zone_enabling);
+  BLOCK_COMMENT("} reserved_stack_check");
+}
+
 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes.
 void MacroAssembler::tlab_allocate(Register obj,
                                    Register var_size_in_bytes,
--- a/hotspot/src/cpu/s390/vm/macroAssembler_s390.hpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/macroAssembler_s390.hpp	Tue Dec 27 16:10:59 2016 +0100
@@ -627,6 +627,11 @@
   // Stack overflow checking
   void bang_stack_with_offset(int offset);
 
+  // Check for reserved stack access in method being exited. If the reserved
+  // stack area was accessed, protect it again and throw StackOverflowError.
+  // Uses Z_R1.
+  void reserved_stack_check(Register return_pc);
+
   // Atomics
   // -- none?
 
--- a/hotspot/src/cpu/s390/vm/s390.ad	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/s390.ad	Tue Dec 27 16:10:59 2016 +0100
@@ -909,15 +909,8 @@
   // If this does safepoint polling, then do it here.
   bool need_polling = do_polling() && C->is_method_compilation();
 
-  // Touch the polling page.
-  // Part 1: get the page's address.
-  if (need_polling) {
-    AddressLiteral pp(os::get_polling_page());
-    __ load_const_optimized(Z_R1_scratch, pp);
-  }
-
   // Pop frame, restore return_pc, and all stuff needed by interpreter.
-  // Pop frame by add insted of load (a penny saved is a penny got :-).
+  // Pop frame by add instead of load (a penny saved is a penny got :-).
   int frame_size_in_bytes = Assembler::align((C->frame_slots() << LogBytesPerInt), frame::alignment_in_bytes);
   int retPC_offset        = frame_size_in_bytes + _z_abi16(return_pc);
   if (Displacement::is_validDisp(retPC_offset)) {
@@ -928,9 +921,14 @@
     __ restore_return_pc();
   }
 
-  // Touch the polling page,
-  // part 2: touch the page now.
+  if (StackReservedPages > 0 && C->has_reserved_stack_access()) {
+    __ reserved_stack_check(Z_R14);
+  }
+
+  // Touch the polling page.
   if (need_polling) {
+    AddressLiteral pp(os::get_polling_page());
+    __ load_const_optimized(Z_R1_scratch, pp);
     // We need to mark the code position where the load from the safepoint
     // polling page was emitted as relocInfo::poll_return_type here.
     __ relocate(relocInfo::poll_return_type);
@@ -939,7 +937,7 @@
 }
 
 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
-  // variable size. determine dynamically.
+  // Variable size. determine dynamically.
   return MachNode::size(ra_);
 }
 
--- a/hotspot/src/cpu/s390/vm/stubGenerator_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/stubGenerator_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -2433,13 +2433,12 @@
     StubRoutines::_throw_StackOverflowError_entry          =
       generate_throw_exception("StackOverflowError throw_exception",
                                CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
+    StubRoutines::_throw_delayed_StackOverflowError_entry  =
+      generate_throw_exception("delayed StackOverflowError throw_exception",
+                               CAST_FROM_FN_PTR(address, SharedRuntime::throw_delayed_StackOverflowError), false);
 
     //----------------------------------------------------------------------
     // Entry points that are platform specific.
-    // Build this early so it's available for the interpreter.
-    StubRoutines::_throw_StackOverflowError_entry          =
-      generate_throw_exception("StackOverflowError throw_exception",
-                               CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError), false);
 
     if (UseCRC32Intrinsics) {
       // We have no CRC32 table on z/Architecture.
--- a/hotspot/src/cpu/s390/vm/templateInterpreterGenerator_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/cpu/s390/vm/templateInterpreterGenerator_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -1112,16 +1112,21 @@
   // top_frame_size = TOP_IJAVA_FRAME_ABI + max_stack + size of interpreter state
   __ add2reg(top_frame_size,
              frame::z_top_ijava_frame_abi_size +
-               frame::z_ijava_state_size +
-               frame::interpreter_frame_monitor_size() * wordSize,
+             frame::z_ijava_state_size +
+             frame::interpreter_frame_monitor_size() * wordSize,
              max_stack);
 
-  // Check if there's room for the new frame...
-  Register frame_size = max_stack; // Reuse the regiser for max_stack.
-  __ z_lgr(frame_size, Z_SP);
-  __ z_sgr(frame_size, sp_after_resize);
-  __ z_agr(frame_size, top_frame_size);
-  generate_stack_overflow_check(frame_size, fp/*tmp1*/);
+  if (!native_call) {
+    // Stack overflow check.
+    // Native calls don't need the stack size check since they have no
+    // expression stack and the arguments are already on the stack and
+    // we only add a handful of words to the stack.
+    Register frame_size = max_stack; // Reuse the regiser for max_stack.
+    __ z_lgr(frame_size, Z_SP);
+    __ z_sgr(frame_size, sp_after_resize);
+    __ z_agr(frame_size, top_frame_size);
+    generate_stack_overflow_check(frame_size, fp/*tmp1*/);
+  }
 
   DEBUG_ONLY(__ z_cg(Z_R14, _z_abi16(return_pc), Z_SP));
   __ asm_assert_eq("killed Z_R14", 0);
--- a/hotspot/src/os_cpu/linux_s390/vm/os_linux_s390.cpp	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/src/os_cpu/linux_s390/vm/os_linux_s390.cpp	Tue Dec 27 16:10:59 2016 +0100
@@ -144,6 +144,42 @@
   return frame(sp, epc.pc());
 }
 
+bool os::Linux::get_frame_at_stack_banging_point(JavaThread* thread, ucontext_t* uc, frame* fr) {
+  address pc = (address) os::Linux::ucontext_get_pc(uc);
+  if (Interpreter::contains(pc)) {
+    // Interpreter performs stack banging after the fixed frame header has
+    // been generated while the compilers perform it before. To maintain
+    // semantic consistency between interpreted and compiled frames, the
+    // method returns the Java sender of the current frame.
+    *fr = os::fetch_frame_from_context(uc);
+    if (!fr->is_first_java_frame()) {
+      assert(fr->safe_for_sender(thread), "Safety check");
+      *fr = fr->java_sender();
+    }
+  } else {
+    // More complex code with compiled code.
+    assert(!Interpreter::contains(pc), "Interpreted methods should have been handled above");
+    CodeBlob* cb = CodeCache::find_blob(pc);
+    if (cb == NULL || !cb->is_nmethod() || cb->is_frame_complete_at(pc)) {
+      // Not sure where the pc points to, fallback to default
+      // stack overflow handling. In compiled code, we bang before
+      // the frame is complete.
+      return false;
+    } else {
+      intptr_t* fp = os::Linux::ucontext_get_fp(uc);
+      intptr_t* sp = os::Linux::ucontext_get_sp(uc);
+      *fr = frame(sp, (address)*sp);
+      if (!fr->is_java_frame()) {
+        assert(fr->safe_for_sender(thread), "Safety check");
+        assert(!fr->is_first_frame(), "Safety check");
+        *fr = fr->java_sender();
+      }
+    }
+  }
+  assert(fr->is_java_frame(), "Safety check");
+  return true;
+}
+
 frame os::get_sender_for_C_frame(frame* fr) {
   if (*fr->sp() == 0) {
     // fr is the last C frame.
@@ -279,13 +315,31 @@
       if (thread->on_local_stack(addr)) {
         // stack overflow
         if (thread->in_stack_yellow_reserved_zone(addr)) {
-          thread->disable_stack_yellow_reserved_zone();
           if (thread->thread_state() == _thread_in_Java) {
+            if (thread->in_stack_reserved_zone(addr)) {
+              frame fr;
+              if (os::Linux::get_frame_at_stack_banging_point(thread, uc, &fr)) {
+                assert(fr.is_java_frame(), "Must be a Javac frame");
+                frame activation =
+                  SharedRuntime::look_for_reserved_stack_annotated_method(thread, fr);
+                if (activation.sp() != NULL) {
+                  thread->disable_stack_reserved_zone();
+                  if (activation.is_interpreted_frame()) {
+                    thread->set_reserved_stack_activation((address)activation.fp());
+                  } else {
+                    thread->set_reserved_stack_activation((address)activation.unextended_sp());
+                  }
+                  return 1;
+                }
+              }
+            }
             // Throw a stack overflow exception.
             // Guard pages will be reenabled while unwinding the stack.
+            thread->disable_stack_yellow_reserved_zone();
             stub = SharedRuntime::continuation_for_implicit_exception(thread, pc, SharedRuntime::STACK_OVERFLOW);
           } else {
             // Thread was in the vm or native code. Return and try to finish.
+            thread->disable_stack_yellow_reserved_zone();
             return 1;
           }
         } else if (thread->in_stack_red_zone(addr)) {
--- a/hotspot/test/runtime/ReservedStack/ReservedStackTest.java	Tue Jan 03 11:22:37 2017 +0100
+++ b/hotspot/test/runtime/ReservedStack/ReservedStackTest.java	Tue Dec 27 16:10:59 2016 +0100
@@ -199,7 +199,8 @@
             // corruptions are still possible.
             boolean supportedPlatform =
                 Platform.isAix() ||
-                (Platform.isLinux() && (Platform.isPPC() || Platform.isX64() || Platform.isX86())) ||
+                (Platform.isLinux() &&
+                  (Platform.isPPC() || Platform.isS390x() || Platform.isX64() || Platform.isX86())) ||
                 Platform.isOSX() ||
                 Platform.isSolaris();
             if (supportedPlatform && !result.contains("PASSED")) {