Reland^3 "Don't use AOT code for native methods for java debuggable runtime"

This reverts commit fb1b08cbb9c6ac149d75de16c14fdaa8b68baaa4.

Reason for revert: Reland after a fix. We had to update untagging in jni_dlsym_lookup_stub as well.

Change-Id: Id936e9e60f9e87e96f1a9a79cd2118631ad1616b
diff --git a/compiler/jni/jni_compiler_test.cc b/compiler/jni/jni_compiler_test.cc
index 0a1f017..d05324b 100644
--- a/compiler/jni/jni_compiler_test.cc
+++ b/compiler/jni/jni_compiler_test.cc
@@ -414,7 +414,7 @@
   CHECK(!caller->IsCriticalNative());
   CHECK(caller->IsSynchronized());
   ObjPtr<mirror::Object> lock;
-  if (self->GetManagedStack()->GetTopQuickFrameTag()) {
+  if (self->GetManagedStack()->GetTopQuickFrameGenericJniTag()) {
     // Generic JNI.
     lock = GetGenericJniSynchronizationObject(self, caller);
   } else if (caller->IsStatic()) {
diff --git a/compiler/jni/quick/jni_compiler.cc b/compiler/jni/quick/jni_compiler.cc
index 6cb5021..b88ebaf 100644
--- a/compiler/jni/quick/jni_compiler.cc
+++ b/compiler/jni/quick/jni_compiler.cc
@@ -95,6 +95,13 @@
   const InstructionSetFeatures* instruction_set_features =
       compiler_options.GetInstructionSetFeatures();
 
+  // When  walking the stack the top frame doesn't have a pc associated with it. We then depend on
+  // the invariant that we don't have JITed code when AOT code is available. In debuggable runtimes
+  // this invariant doesn't hold. So we tag the SP for JITed code to indentify if we are executing
+  // JITed code or AOT code. Since tagging involves additional instructions we tag only in
+  // debuggable runtimes.
+  bool should_tag_sp = compiler_options.GetDebuggable() && compiler_options.IsJitCompiler();
+
   // i.e. if the method was annotated with @FastNative
   const bool is_fast_native = (access_flags & kAccFastNative) != 0u;
 
@@ -219,7 +226,7 @@
   //       because garbage collections are disabled within the execution of a
   //       @CriticalNative method.
   if (LIKELY(!is_critical_native)) {
-    __ StoreStackPointerToThread(Thread::TopOfManagedStackOffset<kPointerSize>());
+    __ StoreStackPointerToThread(Thread::TopOfManagedStackOffset<kPointerSize>(), should_tag_sp);
   }
 
   // 2. Lock the object (if synchronized) and transition out of Runnable (if normal native).
@@ -605,7 +612,7 @@
     if (reference_return) {
       // Suspend check entry point overwrites top of managed stack and leaves it clobbered.
       // We need to restore the top for subsequent runtime call to `JniDecodeReferenceResult()`.
-      __ StoreStackPointerToThread(Thread::TopOfManagedStackOffset<kPointerSize>());
+      __ StoreStackPointerToThread(Thread::TopOfManagedStackOffset<kPointerSize>(), should_tag_sp);
     }
     if (reference_return && main_out_arg_size != 0) {
       __ IncreaseFrameSize(main_out_arg_size);
diff --git a/compiler/utils/arm/jni_macro_assembler_arm_vixl.cc b/compiler/utils/arm/jni_macro_assembler_arm_vixl.cc
index 6e6d40d..61151fe 100644
--- a/compiler/utils/arm/jni_macro_assembler_arm_vixl.cc
+++ b/compiler/utils/arm/jni_macro_assembler_arm_vixl.cc
@@ -428,8 +428,15 @@
   asm_.StoreToOffset(kStoreWord, scratch, tr, thr_offs.Int32Value());
 }
 
-void ArmVIXLJNIMacroAssembler::StoreStackPointerToThread(ThreadOffset32 thr_offs) {
-  asm_.StoreToOffset(kStoreWord, sp, tr, thr_offs.Int32Value());
+void ArmVIXLJNIMacroAssembler::StoreStackPointerToThread(ThreadOffset32 thr_offs, bool tag_sp) {
+  if (tag_sp) {
+    UseScratchRegisterScope temps(asm_.GetVIXLAssembler());
+    vixl32::Register reg = temps.Acquire();
+    ___ Orr(reg, sp, 0x2);
+    asm_.StoreToOffset(kStoreWord, reg, tr, thr_offs.Int32Value());
+  } else {
+    asm_.StoreToOffset(kStoreWord, sp, tr, thr_offs.Int32Value());
+  }
 }
 
 void ArmVIXLJNIMacroAssembler::SignExtend(ManagedRegister mreg ATTRIBUTE_UNUSED,
diff --git a/compiler/utils/arm/jni_macro_assembler_arm_vixl.h b/compiler/utils/arm/jni_macro_assembler_arm_vixl.h
index ed453ae..980de41 100644
--- a/compiler/utils/arm/jni_macro_assembler_arm_vixl.h
+++ b/compiler/utils/arm/jni_macro_assembler_arm_vixl.h
@@ -70,7 +70,7 @@
 
   void StoreStackOffsetToThread(ThreadOffset32 thr_offs, FrameOffset fr_offs) override;
 
-  void StoreStackPointerToThread(ThreadOffset32 thr_offs) override;
+  void StoreStackPointerToThread(ThreadOffset32 thr_offs, bool tag_sp) override;
 
   void StoreSpanning(FrameOffset dest, ManagedRegister src, FrameOffset in_off) override;
 
diff --git a/compiler/utils/arm64/jni_macro_assembler_arm64.cc b/compiler/utils/arm64/jni_macro_assembler_arm64.cc
index 50ca468..323a01e 100644
--- a/compiler/utils/arm64/jni_macro_assembler_arm64.cc
+++ b/compiler/utils/arm64/jni_macro_assembler_arm64.cc
@@ -218,10 +218,13 @@
   ___ Str(scratch, MEM_OP(reg_x(TR), tr_offs.Int32Value()));
 }
 
-void Arm64JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset64 tr_offs) {
+void Arm64JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset64 tr_offs, bool tag_sp) {
   UseScratchRegisterScope temps(asm_.GetVIXLAssembler());
   Register scratch = temps.AcquireX();
   ___ Mov(scratch, reg_x(SP));
+  if (tag_sp) {
+    ___ Orr(scratch, scratch, 0x2);
+  }
   ___ Str(scratch, MEM_OP(reg_x(TR), tr_offs.Int32Value()));
 }
 
diff --git a/compiler/utils/arm64/jni_macro_assembler_arm64.h b/compiler/utils/arm64/jni_macro_assembler_arm64.h
index 2c04184..daea95d 100644
--- a/compiler/utils/arm64/jni_macro_assembler_arm64.h
+++ b/compiler/utils/arm64/jni_macro_assembler_arm64.h
@@ -72,7 +72,7 @@
   void StoreRawPtr(FrameOffset dest, ManagedRegister src) override;
   void StoreImmediateToFrame(FrameOffset dest, uint32_t imm) override;
   void StoreStackOffsetToThread(ThreadOffset64 thr_offs, FrameOffset fr_offs) override;
-  void StoreStackPointerToThread(ThreadOffset64 thr_offs) override;
+  void StoreStackPointerToThread(ThreadOffset64 thr_offs, bool tag_sp) override;
   void StoreSpanning(FrameOffset dest, ManagedRegister src, FrameOffset in_off) override;
 
   // Load routines.
diff --git a/compiler/utils/assembler_thumb_test.cc b/compiler/utils/assembler_thumb_test.cc
index b2d4dcd..f867a05 100644
--- a/compiler/utils/assembler_thumb_test.cc
+++ b/compiler/utils/assembler_thumb_test.cc
@@ -159,7 +159,8 @@
   __ StoreRef(FrameOffset(48), scratch_register);
   __ StoreSpanning(FrameOffset(48), method_register, FrameOffset(48));
   __ StoreStackOffsetToThread(ThreadOffset32(512), FrameOffset(4096));
-  __ StoreStackPointerToThread(ThreadOffset32(512));
+  __ StoreStackPointerToThread(ThreadOffset32(512), false);
+  __ StoreStackPointerToThread(ThreadOffset32(512), true);
 
   // Other
   __ Call(method_register, FrameOffset(48));
diff --git a/compiler/utils/assembler_thumb_test_expected.cc.inc b/compiler/utils/assembler_thumb_test_expected.cc.inc
index 1775014..dac21ae 100644
--- a/compiler/utils/assembler_thumb_test_expected.cc.inc
+++ b/compiler/utils/assembler_thumb_test_expected.cc.inc
@@ -41,46 +41,46 @@
   "      7e: 0d f5 80 5c   add.w r12, sp, #4096\n"
   "      82: c9 f8 00 c2   str.w r12, [r9, #512]\n"
   "      86: c9 f8 00 d2   str.w sp, [r9, #512]\n"
-  "      8a: d0 f8 30 e0   ldr.w lr, [r0, #48]\n"
-  "      8e: f0 47         blx lr\n"
-  "      90: dd f8 2c c0   ldr.w r12, [sp, #44]\n"
-  "      94: cd f8 30 c0   str.w r12, [sp, #48]\n"
-  "      98: d9 f8 00 c2   ldr.w r12, [r9, #512]\n"
-  "      9c: cd f8 2c c0   str.w r12, [sp, #44]\n"
-  "      a0: dd f8 2c c0   ldr.w r12, [sp, #44]\n"
-  "      a4: cd f8 30 c0   str.w r12, [sp, #48]\n"
-  "      a8: 48 46         mov r0, r9\n"
-  "      aa: cd f8 30 90   str.w r9, [sp, #48]\n"
-  "      ae: 04 46         mov r4, r0\n"
-  "      b0: 0d f1 30 0c   add.w r12, sp, #48\n"
-  "      b4: bb f1 00 0f   cmp.w r11, #0\n"
-  "      b8: 18 bf         it ne\n"
-  "      ba: e3 46         movne r11, r12\n"
-  "      bc: 0d f1 30 0b   add.w r11, sp, #48\n"
-  "      c0: 5f ea 0b 00   movs.w r0, r11\n"
-  "      c4: 18 bf         it ne\n"
-  "      c6: 0c a8         addne r0, sp, #48\n"
-  "      c8: dd f8 40 c0   ldr.w r12, [sp, #64]\n"
-  "      cc: bc f1 00 0f   cmp.w r12, #0\n"
-  "      d0: 18 bf         it ne\n"
-  "      d2: 0d f1 40 0c   addne.w r12, sp, #64\n"
-  "      d6: cd f8 30 c0   str.w r12, [sp, #48]\n"
-  "      da: 5f ea 0b 00   movs.w r0, r11\n"
-  "      de: 18 bf         it ne\n"
-  "      e0: 00 a8         addne r0, sp, #0\n"
-  "      e2: 0d f2 04 40   addw r0, sp, #1028\n"
-  "      e6: bb f1 00 0f   cmp.w r11, #0\n"
-  "      ea: 08 bf         it eq\n"
-  "      ec: 58 46         moveq r0, r11\n"
-  "      ee: 0d f2 04 4c   addw r12, sp, #1028\n"
-  "      f2: bb f1 00 0f   cmp.w r11, #0\n"
-  "      f6: 18 bf         it ne\n"
-  "      f8: e3 46         movne r11, r12\n"
-  "      fa: d9 f8 9c c0   ldr.w r12, [r9, #156]\n"
-  "      fe: bc f1 00 0f   cmp.w r12, #0\n"
-  "     102: 71 d1         bne 0x1e8     @ imm = #226\n"
-  "     104: cd f8 ff c7   str.w r12, [sp, #2047]\n"
-  "     108: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "      8a: 4d f0 02 0c   orr r12, sp, #2\n"
+  "      8e: c9 f8 00 c2   str.w r12, [r9, #512]\n"
+  "      92: d0 f8 30 e0   ldr.w lr, [r0, #48]\n"
+  "      96: f0 47         blx lr\n"
+  "      98: dd f8 2c c0   ldr.w r12, [sp, #44]\n"
+  "      9c: cd f8 30 c0   str.w r12, [sp, #48]\n"
+  "      a0: d9 f8 00 c2   ldr.w r12, [r9, #512]\n"
+  "      a4: cd f8 2c c0   str.w r12, [sp, #44]\n"
+  "      a8: dd f8 2c c0   ldr.w r12, [sp, #44]\n"
+  "      ac: cd f8 30 c0   str.w r12, [sp, #48]\n"
+  "      b0: 48 46         mov r0, r9\n"
+  "      b2: cd f8 30 90   str.w r9, [sp, #48]\n"
+  "      b6: 04 46         mov r4, r0\n"
+  "      b8: 0d f1 30 0c   add.w r12, sp, #48\n"
+  "      bc: bb f1 00 0f   cmp.w r11, #0\n"
+  "      c0: 18 bf         it ne\n"
+  "      c2: e3 46         movne r11, r12\n"
+  "      c4: 0d f1 30 0b   add.w r11, sp, #48\n"
+  "      c8: 5f ea 0b 00   movs.w r0, r11\n"
+  "      cc: 18 bf         it ne\n"
+  "      ce: 0c a8         addne r0, sp, #48\n"
+  "      d0: dd f8 40 c0   ldr.w r12, [sp, #64]\n"
+  "      d4: bc f1 00 0f   cmp.w r12, #0\n"
+  "      d8: 18 bf         it ne\n"
+  "      da: 0d f1 40 0c   addne.w r12, sp, #64\n"
+  "      de: cd f8 30 c0   str.w r12, [sp, #48]\n"
+  "      e2: 5f ea 0b 00   movs.w r0, r11\n"
+  "      e6: 18 bf         it ne\n"
+  "      e8: 00 a8         addne r0, sp, #0\n"
+  "      ea: 0d f2 04 40   addw r0, sp, #1028\n"
+  "      ee: bb f1 00 0f   cmp.w r11, #0\n"
+  "      f2: 08 bf         it eq\n"
+  "      f4: 58 46         moveq r0, r11\n"
+  "      f6: 0d f2 04 4c   addw r12, sp, #1028\n"
+  "      fa: bb f1 00 0f   cmp.w r11, #0\n"
+  "      fe: 18 bf         it ne\n"
+  "     100: e3 46         movne r11, r12\n"
+  "     102: d9 f8 9c c0   ldr.w r12, [r9, #156]\n"
+  "     106: bc f1 00 0f   cmp.w r12, #0\n"
+  "     10a: 71 d1         bne 0x1f0     @ imm = #226\n"
   "     10c: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     110: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     114: cd f8 ff c7   str.w r12, [sp, #2047]\n"
@@ -135,26 +135,28 @@
   "     1d8: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     1dc: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     1e0: cd f8 ff c7   str.w r12, [sp, #2047]\n"
-  "     1e4: 00 f0 02 b8   b.w 0x1ec     @ imm = #4\n"
-  "     1e8: 00 f0 1b b8   b.w 0x222     @ imm = #54\n"
-  "     1ec: cd f8 ff c7   str.w r12, [sp, #2047]\n"
-  "     1f0: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "     1e4: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "     1e8: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "     1ec: 00 f0 02 b8   b.w 0x1f4     @ imm = #4\n"
+  "     1f0: 00 f0 1b b8   b.w 0x22a     @ imm = #54\n"
   "     1f4: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     1f8: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     1fc: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     200: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     204: cd f8 ff c7   str.w r12, [sp, #2047]\n"
   "     208: cd f8 ff c7   str.w r12, [sp, #2047]\n"
-  "     20c: 0d f5 80 5d   add.w sp, sp, #4096\n"
-  "     210: 08 b0         add sp, #32\n"
-  "     212: 01 b0         add sp, #4\n"
-  "     214: bd ec 10 8a   vpop {s16, s17, s18, s19, s20, s21, s22, s23, s24, s25, s26, s27, s28, s29, s30, s31}\n"
-  "     218: bd e8 e0 4d   pop.w {r5, r6, r7, r8, r10, r11, lr}\n"
-  "     21c: d9 f8 24 80   ldr.w r8, [r9, #36]\n"
-  "     220: 70 47         bx lr\n"
-  "     222: d9 f8 9c 00   ldr.w r0, [r9, #156]\n"
-  "     226: d9 f8 d0 e2   ldr.w lr, [r9, #720]\n"
-  "     22a: f0 47         blx lr\n"
+  "     20c: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "     210: cd f8 ff c7   str.w r12, [sp, #2047]\n"
+  "     214: 0d f5 80 5d   add.w sp, sp, #4096\n"
+  "     218: 08 b0         add sp, #32\n"
+  "     21a: 01 b0         add sp, #4\n"
+  "     21c: bd ec 10 8a   vpop {s16, s17, s18, s19, s20, s21, s22, s23, s24, s25, s26, s27, s28, s29, s30, s31}\n"
+  "     220: bd e8 e0 4d   pop.w {r5, r6, r7, r8, r10, r11, lr}\n"
+  "     224: d9 f8 24 80   ldr.w r8, [r9, #36]\n"
+  "     228: 70 47         bx lr\n"
+  "     22a: d9 f8 9c 00   ldr.w r0, [r9, #156]\n"
+  "     22e: d9 f8 d0 e2   ldr.w lr, [r9, #720]\n"
+  "     232: f0 47         blx lr\n"
 };
 
 const char* const VixlLoadFromOffsetResults = {
diff --git a/compiler/utils/jni_macro_assembler.h b/compiler/utils/jni_macro_assembler.h
index 7022e3d..c8c713a 100644
--- a/compiler/utils/jni_macro_assembler.h
+++ b/compiler/utils/jni_macro_assembler.h
@@ -126,7 +126,11 @@
   virtual void StoreStackOffsetToThread(ThreadOffset<kPointerSize> thr_offs,
                                         FrameOffset fr_offs) = 0;
 
-  virtual void StoreStackPointerToThread(ThreadOffset<kPointerSize> thr_offs) = 0;
+  // Stores stack pointer by tagging it if required so we can walk the stack. In debuggable runtimes
+  // we use tag to tell if we are using JITed code or AOT code. In non-debuggable runtimes we never
+  // use JITed code when AOT code is present. So checking for AOT code is sufficient to detect which
+  // code is being executed. We avoid tagging in non-debuggable runtimes to reduce instructions.
+  virtual void StoreStackPointerToThread(ThreadOffset<kPointerSize> thr_offs, bool tag_sp) = 0;
 
   virtual void StoreSpanning(FrameOffset dest,
                              ManagedRegister src,
diff --git a/compiler/utils/x86/jni_macro_assembler_x86.cc b/compiler/utils/x86/jni_macro_assembler_x86.cc
index 685f5f1..55d5428 100644
--- a/compiler/utils/x86/jni_macro_assembler_x86.cc
+++ b/compiler/utils/x86/jni_macro_assembler_x86.cc
@@ -187,8 +187,18 @@
   __ fs()->movl(Address::Absolute(thr_offs), scratch);
 }
 
-void X86JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset32 thr_offs) {
-  __ fs()->movl(Address::Absolute(thr_offs), ESP);
+void X86JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset32 thr_offs, bool tag_sp) {
+  if (tag_sp) {
+    // There is no free register, store contents onto stack and restore back later.
+    Register scratch = ECX;
+    __ movl(Address(ESP, -32), scratch);
+    __ movl(scratch, ESP);
+    __ orl(scratch, Immediate(0x2));
+    __ fs()->movl(Address::Absolute(thr_offs), scratch);
+    __ movl(scratch, Address(ESP, -32));
+  } else {
+    __ fs()->movl(Address::Absolute(thr_offs), ESP);
+  }
 }
 
 void X86JNIMacroAssembler::StoreSpanning(FrameOffset /*dst*/,
diff --git a/compiler/utils/x86/jni_macro_assembler_x86.h b/compiler/utils/x86/jni_macro_assembler_x86.h
index 29fccfd..f8ce38b 100644
--- a/compiler/utils/x86/jni_macro_assembler_x86.h
+++ b/compiler/utils/x86/jni_macro_assembler_x86.h
@@ -66,7 +66,7 @@
 
   void StoreStackOffsetToThread(ThreadOffset32 thr_offs, FrameOffset fr_offs) override;
 
-  void StoreStackPointerToThread(ThreadOffset32 thr_offs) override;
+  void StoreStackPointerToThread(ThreadOffset32 thr_offs, bool tag_sp) override;
 
   void StoreSpanning(FrameOffset dest, ManagedRegister src, FrameOffset in_off) override;
 
diff --git a/compiler/utils/x86_64/jni_macro_assembler_x86_64.cc b/compiler/utils/x86_64/jni_macro_assembler_x86_64.cc
index d5d1bba..adc431f 100644
--- a/compiler/utils/x86_64/jni_macro_assembler_x86_64.cc
+++ b/compiler/utils/x86_64/jni_macro_assembler_x86_64.cc
@@ -217,8 +217,15 @@
   __ gs()->movq(Address::Absolute(thr_offs, true), scratch);
 }
 
-void X86_64JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset64 thr_offs) {
-  __ gs()->movq(Address::Absolute(thr_offs, true), CpuRegister(RSP));
+void X86_64JNIMacroAssembler::StoreStackPointerToThread(ThreadOffset64 thr_offs, bool tag_sp) {
+  if (tag_sp) {
+    CpuRegister reg = GetScratchRegister();
+    __ movq(reg, CpuRegister(RSP));
+    __ orq(reg, Immediate(0x2));
+    __ gs()->movq(Address::Absolute(thr_offs, true), reg);
+  } else {
+    __ gs()->movq(Address::Absolute(thr_offs, true), CpuRegister(RSP));
+  }
 }
 
 void X86_64JNIMacroAssembler::StoreSpanning(FrameOffset /*dst*/,
diff --git a/compiler/utils/x86_64/jni_macro_assembler_x86_64.h b/compiler/utils/x86_64/jni_macro_assembler_x86_64.h
index e080f0b..feaf27e 100644
--- a/compiler/utils/x86_64/jni_macro_assembler_x86_64.h
+++ b/compiler/utils/x86_64/jni_macro_assembler_x86_64.h
@@ -67,7 +67,7 @@
 
   void StoreStackOffsetToThread(ThreadOffset64 thr_offs, FrameOffset fr_offs) override;
 
-  void StoreStackPointerToThread(ThreadOffset64 thr_offs) override;
+  void StoreStackPointerToThread(ThreadOffset64 thr_offs, bool tag_sp) override;
 
   void StoreSpanning(FrameOffset dest, ManagedRegister src, FrameOffset in_off) override;