8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.

Summary: Fixed Implicit NULL check optimization for AIX, where the page at address '0' is only write-protected.
Reviewed-by: kvn

11 files changed, 65 insertions, 8 deletions

diff --git a/src/cpu/ppc/vm/globals_ppc.hpp b/src/cpu/ppc/vm/globals_ppc.hpp
index 2dd4b1df1..b39151004 100644
--- a/src/cpu/ppc/vm/globals_ppc.hpp
+++ b/src/cpu/ppc/vm/globals_ppc.hpp
@@ -92,7 +92,7 @@ define_pd_global(uintx, TypeProfileLevel, 0);
   /* Trap based checks use the ppc trap instructions to check certain */    \
   /* conditions. This instruction raises a SIGTRAP caught by the      */    \
   /* exception handler of the VM.                                     */    \
-  product(bool, UseSIGTRAP, false,                                          \
+  product(bool, UseSIGTRAP, true,                                           \
           "Allow trap instructions that make use of SIGTRAP. Use this to "  \
           "switch off all optimizations requiring SIGTRAP.")                \
   product(bool, TrapBasedICMissChecks, true,                                \
diff --git a/src/cpu/ppc/vm/macroAssembler_ppc.cpp b/src/cpu/ppc/vm/macroAssembler_ppc.cpp
index eb716b5b8..f4b7c5e76 100644
--- a/src/cpu/ppc/vm/macroAssembler_ppc.cpp
+++ b/src/cpu/ppc/vm/macroAssembler_ppc.cpp
@@ -2433,7 +2433,7 @@ void MacroAssembler::load_klass(Register dst, Register src) {
 }
 
 void MacroAssembler::load_klass_with_trap_null_check(Register dst, Register src) {
-  if (false  NOT_LINUX(|| true) /*!os::zero_page_read_protected()*/) {
+  if (!os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       trap_null_check(src);
     }
diff --git a/src/cpu/ppc/vm/macroAssembler_ppc.inline.hpp b/src/cpu/ppc/vm/macroAssembler_ppc.inline.hpp
index d80e4ec44..529f82468 100644
--- a/src/cpu/ppc/vm/macroAssembler_ppc.inline.hpp
+++ b/src/cpu/ppc/vm/macroAssembler_ppc.inline.hpp
@@ -269,7 +269,7 @@ inline void MacroAssembler::trap_ic_miss_check(Register a, Register b) {
 // No support for compressed oops (base page of heap).  Does not distinguish
 // loads and stores.
 inline void MacroAssembler::null_check_throw(Register a, int offset, Register temp_reg, address exception_entry) {
-  if (!ImplicitNullChecks || needs_explicit_null_check(offset) NOT_LINUX(|| true) /*!os::zero_page_read_protected()*/) {
+  if (!ImplicitNullChecks || needs_explicit_null_check(offset) || !os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       assert(UseSIGTRAP, "sanity");
       trap_null_check(a);
@@ -286,7 +286,7 @@ inline void MacroAssembler::null_check_throw(Register a, int offset, Register te
 }
 
 inline void MacroAssembler::ld_with_trap_null_check(Register d, int si16, Register s1) {
-  if ( NOT_LINUX(true) LINUX_ONLY(false)/*!os::zero_page_read_protected()*/) {
+  if (!os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       trap_null_check(s1);
     }
@@ -297,7 +297,7 @@ inline void MacroAssembler::ld_with_trap_null_check(Register d, int si16, Regist
 // Attention: No null check for loaded uncompressed OOP. Can be used for loading klass field.
 inline void MacroAssembler::load_heap_oop_with_trap_null_check(Register d, RegisterOrConstant si16,
                                                                    Register s1) {
-  if ( NOT_LINUX(true)LINUX_ONLY(false) /*!os::zero_page_read_protected()*/) {
+  if ( !os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       trap_null_check(s1);
     }
diff --git a/src/cpu/ppc/vm/sharedRuntime_ppc.cpp b/src/cpu/ppc/vm/sharedRuntime_ppc.cpp
index e9d217601..c2a971118 100644
--- a/src/cpu/ppc/vm/sharedRuntime_ppc.cpp
+++ b/src/cpu/ppc/vm/sharedRuntime_ppc.cpp
@@ -1128,7 +1128,7 @@ AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm
   assert(!MacroAssembler::needs_explicit_null_check(oopDesc::klass_offset_in_bytes()),
          "klass offset should reach into any page");
   // Check for NULL argument if we don't have implicit null checks.
-  if (!ImplicitNullChecks NOT_LINUX(|| true) /*!os::zero_page_read_protected()*/) {
+  if (!ImplicitNullChecks || !os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       __ trap_null_check(R3_ARG1);
     } else {
diff --git a/src/cpu/ppc/vm/vtableStubs_ppc_64.cpp b/src/cpu/ppc/vm/vtableStubs_ppc_64.cpp
index 5323e4ae5..6a608e932 100644
--- a/src/cpu/ppc/vm/vtableStubs_ppc_64.cpp
+++ b/src/cpu/ppc/vm/vtableStubs_ppc_64.cpp
@@ -221,7 +221,7 @@ VtableStub* VtableStubs::create_itable_stub(int vtable_index) {
 
   // Must do an explicit check if implicit checks are disabled.
   assert(!MacroAssembler::needs_explicit_null_check(in_bytes(Method::from_compiled_offset())), "sanity");
-  if (!ImplicitNullChecks NOT_LINUX(|| true) /*!os::zero_page_read_protected()*/) {
+  if (!ImplicitNullChecks || !os::zero_page_read_protected()) {
     if (TrapBasedNullChecks) {
       __ trap_null_check(R19_method);
     } else {
diff --git a/src/os/bsd/vm/os_bsd.hpp b/src/os/bsd/vm/os_bsd.hpp
index a906a30e9..c33968a0a 100644
--- a/src/os/bsd/vm/os_bsd.hpp
+++ b/src/os/bsd/vm/os_bsd.hpp
@@ -27,6 +27,9 @@
 
 // Bsd_OS defines the interface to Bsd operating systems
 
+// Information about the protection of the page at address '0' on this os.
+static bool zero_page_read_protected() { return true; }
+
 /* pthread_getattr_np comes with BsdThreads-0.9-7 on RedHat 7.1 */
 typedef int (*pthread_getattr_func_type) (pthread_t, pthread_attr_t *);
 
diff --git a/src/os/linux/vm/os_linux.hpp b/src/os/linux/vm/os_linux.hpp
index 5eed54f8c..829852f91 100644
--- a/src/os/linux/vm/os_linux.hpp
+++ b/src/os/linux/vm/os_linux.hpp
@@ -30,6 +30,9 @@
 /* pthread_getattr_np comes with LinuxThreads-0.9-7 on RedHat 7.1 */
 typedef int (*pthread_getattr_func_type) (pthread_t, pthread_attr_t *);
 
+// Information about the protection of the page at address '0' on this os.
+static bool zero_page_read_protected() { return true; }
+
 class Linux {
   friend class os;
   friend class TestReserveMemorySpecial;
diff --git a/src/os/solaris/vm/os_solaris.hpp b/src/os/solaris/vm/os_solaris.hpp
index 330d4fbc5..048534691 100644
--- a/src/os/solaris/vm/os_solaris.hpp
+++ b/src/os/solaris/vm/os_solaris.hpp
@@ -27,6 +27,9 @@
 
 // Solaris_OS defines the interface to Solaris operating systems
 
+// Information about the protection of the page at address '0' on this os.
+static bool zero_page_read_protected() { return true; }
+
 class Solaris {
   friend class os;
 
diff --git a/src/os/windows/vm/os_windows.hpp b/src/os/windows/vm/os_windows.hpp
index 67ed8a4af..c9c4840bb 100644
--- a/src/os/windows/vm/os_windows.hpp
+++ b/src/os/windows/vm/os_windows.hpp
@@ -26,6 +26,9 @@
 #define OS_WINDOWS_VM_OS_WINDOWS_HPP
 // Win32_OS defines the interface to windows operating systems
 
+// Information about the protection of the page at address '0' on this os.
+static bool zero_page_read_protected() { return true; }
+
 class win32 {
   friend class os;
 
diff --git a/src/share/vm/opto/callGenerator.cpp b/src/share/vm/opto/callGenerator.cpp
index adc27ac02..22807e2c7 100644
--- a/src/share/vm/opto/callGenerator.cpp
+++ b/src/share/vm/opto/callGenerator.cpp
@@ -201,7 +201,7 @@ JVMState* VirtualCallGenerator::generate(JVMState* jvms, Parse* parent_parser) {
   // Block::implicit_null_check() only looks for loads and stores, not calls.
   ciMethod *caller = kit.method();
   ciMethodData *caller_md = (caller == NULL) ? NULL : caller->method_data();
-  if (!UseInlineCaches || !ImplicitNullChecks ||
+  if (!UseInlineCaches || !ImplicitNullChecks || !os::zero_page_read_protected() ||
        ((ImplicitNullCheckThreshold > 0) && caller_md &&
        (caller_md->trap_count(Deoptimization::Reason_null_check)
        >= (uint)ImplicitNullCheckThreshold))) {
diff --git a/src/share/vm/opto/lcm.cpp b/src/share/vm/opto/lcm.cpp
index 8c4f339cd..dfdc663a2 100644
--- a/src/share/vm/opto/lcm.cpp
+++ b/src/share/vm/opto/lcm.cpp
@@ -54,6 +54,43 @@
 
 // Optimization - Graph Style
 
+// Check whether val is not-null-decoded compressed oop,
+// i.e. will grab into the base of the heap if it represents NULL.
+static bool accesses_heap_base_zone(Node *val) {
+  if (Universe::narrow_oop_base() > 0) { // Implies UseCompressedOops.
+    if (val && val->is_Mach()) {
+      if (val->as_Mach()->ideal_Opcode() == Op_DecodeN) {
+        // This assumes all Decodes with TypePtr::NotNull are matched to nodes that
+        // decode NULL to point to the heap base (Decode_NN).
+        if (val->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull) {
+          return true;
+        }
+      }
+      // Must recognize load operation with Decode matched in memory operand.
+      // We should not reach here exept for PPC/AIX, as os::zero_page_read_protected()
+      // returns true everywhere else. On PPC, no such memory operands
+      // exist, therefore we did not yet implement a check for such operands.
+      NOT_AIX(Unimplemented());
+    }
+  }
+  return false;
+}
+
+static bool needs_explicit_null_check_for_read(Node *val) {
+  // On some OSes (AIX) the page at address 0 is only write protected.
+  // If so, only Store operations will trap.
+  if (os::zero_page_read_protected()) {
+    return false;  // Implicit null check will work.
+  }
+  // Also a read accessing the base of a heap-based compressed heap will trap.
+  if (accesses_heap_base_zone(val) &&                    // Hits the base zone page.
+      Universe::narrow_oop_use_implicit_null_checks()) { // Base zone page is protected.
+    return false;
+  }
+
+  return true;
+}
+
 //------------------------------implicit_null_check----------------------------
 // Detect implicit-null-check opportunities.  Basically, find NULL checks
 // with suitable memory ops nearby.  Use the memory op to do the NULL check.
@@ -209,6 +246,14 @@ void PhaseCFG::implicit_null_check(Block* block, Node *proj, Node *val, int allo
       }
       break;
     }
+
+    // On some OSes (AIX) the page at address 0 is only write protected.
+    // If so, only Store operations will trap.
+    // But a read accessing the base of a heap-based compressed heap will trap.
+    if (!was_store && needs_explicit_null_check_for_read(val)) {
+      continue;
+    }
+
     // check if the offset is not too high for implicit exception
     {
       intptr_t offset = 0;