Revert "Revert "Use implicit null checks inside try blocks.""

Fix implicit checks in try blocks to emit stack maps.
Fix arm64 null expection from signal entrypoint to call
the runtime handler instead or simply jumping there.

On Nexus 9, AOSP ToT, the boot.oat size reduction is
  prebuilt multi-part boot image:
    - 32-bit boot.oat: -448KiB (-1.3%)
    - 64-bit boot.oat: -528KiB (-1.2%)
  on-device built single boot image:
    - 32-bit boot.oat: -448KiB (-1.4%)
    - 64-bit boot.oat: -528KiB (-1.3%)
Note that the oat files no longer contain dex files which
have been moved to vdex, so the percentages are not directly
comparable with the those reported in the original commit.

Test: Run ART test suite including gc-stress on host and Nexus 9.
Bug: 30212852
Bug: 31468464

This reverts commit 0719b5b9b458cb3eb9f0823f0dacdfe1a71214dd.

Change-Id: If8a9da8c11adf2aad203e93b6684ce16ed776285
diff --git a/runtime/arch/arm/fault_handler_arm.cc b/runtime/arch/arm/fault_handler_arm.cc
index befdd48..daa2dff 100644
--- a/runtime/arch/arm/fault_handler_arm.cc
+++ b/runtime/arch/arm/fault_handler_arm.cc
@@ -122,13 +122,16 @@
   struct ucontext *uc = reinterpret_cast<struct ucontext*>(context);
   struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
   uint8_t* ptr = reinterpret_cast<uint8_t*>(sc->arm_pc);
-
   uint32_t instr_size = GetInstructionSize(ptr);
-  sc->arm_lr = (sc->arm_pc + instr_size) | 1;      // LR needs to point to gc map location
+  uintptr_t gc_map_location = (sc->arm_pc + instr_size) | 1;
+
+  // Push the gc map location to the stack and pass the fault address in LR.
+  sc->arm_sp -= sizeof(uintptr_t);
+  *reinterpret_cast<uintptr_t*>(sc->arm_sp) = gc_map_location;
+  sc->arm_lr = reinterpret_cast<uintptr_t>(info->si_addr);
   sc->arm_pc = reinterpret_cast<uintptr_t>(art_quick_throw_null_pointer_exception_from_signal);
   // Pass the faulting address as the first argument of
   // art_quick_throw_null_pointer_exception_from_signal.
-  sc->arm_r0 = reinterpret_cast<uintptr_t>(info->si_addr);
   VLOG(signals) << "Generating null pointer exception";
   return true;
 }
diff --git a/runtime/arch/arm/quick_entrypoints_arm.S b/runtime/arch/arm/quick_entrypoints_arm.S
index a3f053b..3fc83ba 100644
--- a/runtime/arch/arm/quick_entrypoints_arm.S
+++ b/runtime/arch/arm/quick_entrypoints_arm.S
@@ -173,6 +173,29 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     * when core registers are already saved.
+     */
+.macro SETUP_SAVE_EVERYTHING_FRAME_CORE_REGS_SAVED rTemp
+                                        @ 14 words of callee saves and args already saved.
+    vpush {d0-d15}                      @ 32 words, 2 for each of the 16 saved doubles.
+    .cfi_adjust_cfa_offset 128
+    sub sp, #8                          @ 2 words of space, alignment padding and Method*
+    .cfi_adjust_cfa_offset 8
+    RUNTIME_CURRENT1 \rTemp             @ Load Runtime::Current into rTemp.
+    @ Load kSaveEverything Method* into rTemp.
+    ldr \rTemp, [\rTemp, #RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET]
+    str \rTemp, [sp, #0]                @ Place Method* at bottom of stack.
+    str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET]  @ Place sp in Thread::Current()->top_quick_frame.
+
+    // Ugly compile-time check, but we only have the preprocessor.
+#if (FRAME_SIZE_SAVE_EVERYTHING != 56 + 128 + 8)
+#error "FRAME_SIZE_SAVE_EVERYTHING(ARM) size not as expected."
+#endif
+.endm
+
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything)
      */
 .macro SETUP_SAVE_EVERYTHING_FRAME rTemp
     push {r0-r12, lr}                   @ 14 words of callee saves and args.
@@ -191,20 +214,7 @@
     .cfi_rel_offset r11, 44
     .cfi_rel_offset ip, 48
     .cfi_rel_offset lr, 52
-    vpush {d0-d15}                      @ 32 words, 2 for each of the 16 saved doubles.
-    .cfi_adjust_cfa_offset 128
-    sub sp, #8                          @ 2 words of space, alignment padding and Method*
-    .cfi_adjust_cfa_offset 8
-    RUNTIME_CURRENT1 \rTemp             @ Load Runtime::Current into rTemp.
-    @ Load kSaveEverything Method* into rTemp.
-    ldr \rTemp, [\rTemp, #RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET]
-    str \rTemp, [sp, #0]                @ Place Method* at bottom of stack.
-    str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET]  @ Place sp in Thread::Current()->top_quick_frame.
-
-    // Ugly compile-time check, but we only have the preprocessor.
-#if (FRAME_SIZE_SAVE_EVERYTHING != 56 + 128 + 8)
-#error "FRAME_SIZE_SAVE_EVERYTHING(ARM) size not as expected."
-#endif
+    SETUP_SAVE_EVERYTHING_FRAME_CORE_REGS_SAVED \rTemp
 .endm
 
 .macro RESTORE_SAVE_EVERYTHING_FRAME
@@ -356,7 +366,34 @@
     /*
      * Call installed by a signal handler to create and deliver a NullPointerException.
      */
-ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal
+    .extern art_quick_throw_null_pointer_exception_from_signal
+ENTRY art_quick_throw_null_pointer_exception_from_signal
+    // The fault handler pushes the gc map address, i.e. "return address", to stack
+    // and passes the fault address in LR. So we need to set up the CFI info accordingly.
+    .cfi_def_cfa_offset __SIZEOF_POINTER__
+    .cfi_rel_offset lr, 0
+    push {r0-r12}                   @ 13 words of callee saves and args; LR already saved.
+    .cfi_adjust_cfa_offset 52
+    .cfi_rel_offset r0, 0
+    .cfi_rel_offset r1, 4
+    .cfi_rel_offset r2, 8
+    .cfi_rel_offset r3, 12
+    .cfi_rel_offset r4, 16
+    .cfi_rel_offset r5, 20
+    .cfi_rel_offset r6, 24
+    .cfi_rel_offset r7, 28
+    .cfi_rel_offset r8, 32
+    .cfi_rel_offset r9, 36
+    .cfi_rel_offset r10, 40
+    .cfi_rel_offset r11, 44
+    .cfi_rel_offset ip, 48
+
+    @ save all registers as basis for long jump context
+    SETUP_SAVE_EVERYTHING_FRAME_CORE_REGS_SAVED r1
+    mov r0, lr                      @ pass the fault address stored in LR by the fault handler.
+    mov r1, r9                      @ pass Thread::Current
+    b   artThrowNullPointerExceptionFromSignal  @ (Thread*)
+END art_quick_throw_null_pointer_exception_from_signal
 
     /*
      * Called by managed code to create and deliver an ArithmeticException.
diff --git a/runtime/arch/arm64/fault_handler_arm64.cc b/runtime/arch/arm64/fault_handler_arm64.cc
index 6724d6d..c02be87 100644
--- a/runtime/arch/arm64/fault_handler_arm64.cc
+++ b/runtime/arch/arm64/fault_handler_arm64.cc
@@ -96,12 +96,12 @@
   struct ucontext *uc = reinterpret_cast<struct ucontext*>(context);
   struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
 
-  sc->regs[30] = sc->pc + 4;      // LR needs to point to gc map location
+  // Push the gc map location to the stack and pass the fault address in LR.
+  sc->sp -= sizeof(uintptr_t);
+  *reinterpret_cast<uintptr_t*>(sc->sp) = sc->pc + 4;
+  sc->regs[30] = reinterpret_cast<uintptr_t>(info->si_addr);
 
   sc->pc = reinterpret_cast<uintptr_t>(art_quick_throw_null_pointer_exception_from_signal);
-  // Pass the faulting address as the first argument of
-  // art_quick_throw_null_pointer_exception_from_signal.
-  sc->regs[0] = reinterpret_cast<uintptr_t>(info->si_addr);
   VLOG(signals) << "Generating null pointer exception";
   return true;
 }
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index 25aa8ce..5a92659 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -265,10 +265,10 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     * when the SP has already been decremented by FRAME_SIZE_SAVE_EVERYTHING
+     * and saving registers x29 and LR is handled elsewhere.
      */
-.macro SETUP_SAVE_EVERYTHING_FRAME
-    INCREASE_FRAME 512
-
+.macro SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR
     // Ugly compile-time check, but we only have the preprocessor.
 #if (FRAME_SIZE_SAVE_EVERYTHING != 512)
 #error "FRAME_SIZE_SAVE_EVERYTHING(ARM64) size not as expected."
@@ -310,7 +310,6 @@
     SAVE_TWO_REGS x23, x24, 448
     SAVE_TWO_REGS x25, x26, 464
     SAVE_TWO_REGS x27, x28, 480
-    SAVE_TWO_REGS x29, xLR, 496
 
     // art::Runtime** xIP0 = &art::Runtime::instance_
     adrp xIP0, :got:_ZN3art7Runtime9instance_E
@@ -328,6 +327,16 @@
     str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET]
 .endm
 
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     */
+.macro SETUP_SAVE_EVERYTHING_FRAME
+    INCREASE_FRAME 512
+    SAVE_TWO_REGS x29, xLR, 496
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR
+.endm
+
 .macro RESTORE_SAVE_EVERYTHING_FRAME
     // Restore FP registers.
     // For better performance, load d0 and d31 separately, so that all LDPs are 16-byte aligned.
@@ -462,7 +471,22 @@
     /*
      * Call installed by a signal handler to create and deliver a NullPointerException.
      */
-ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal
+    .extern art_quick_throw_null_pointer_exception_from_signal
+ENTRY art_quick_throw_null_pointer_exception_from_signal
+    // The fault handler pushes the gc map address, i.e. "return address", to stack
+    // and passes the fault address in LR. So we need to set up the CFI info accordingly.
+    .cfi_def_cfa_offset __SIZEOF_POINTER__
+    .cfi_rel_offset lr, 0
+    // Save all registers as basis for long jump context.
+    INCREASE_FRAME (FRAME_SIZE_SAVE_EVERYTHING - __SIZEOF_POINTER__)
+    SAVE_REG x29, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__)  // LR already saved.
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP_SKIP_X29_LR
+    mov x0, lr                        // pass the fault address stored in LR by the fault handler.
+    mov x1, xSELF                     // pass Thread::Current.
+    // TODO: Change other throwing entrypoints to use BL instead of B. http://b/31468464
+    bl   artThrowNullPointerExceptionFromSignal  // (arg, Thread*).
+    brk 0
+END art_quick_throw_null_pointer_exception_from_signal
 
     /*
      * Called by managed code to create and deliver an ArithmeticException.
diff --git a/runtime/arch/mips/asm_support_mips.S b/runtime/arch/mips/asm_support_mips.S
index 801f708..7955b1d 100644
--- a/runtime/arch/mips/asm_support_mips.S
+++ b/runtime/arch/mips/asm_support_mips.S
@@ -43,7 +43,7 @@
 .endm
 
      // Declare a function called name, doesn't set up $gp.
-.macro ENTRY_NO_GP name
+.macro ENTRY_NO_GP_CUSTOM_CFA name, cfa_offset
     .type \name, %function
     .global \name
     // Cache alignment for function entry.
@@ -51,7 +51,12 @@
 \name:
     .cfi_startproc
      // Ensure we get a sane starting CFA.
-    .cfi_def_cfa $sp,0
+    .cfi_def_cfa $sp, \cfa_offset
+.endm
+
+     // Declare a function called name, doesn't set up $gp.
+.macro ENTRY_NO_GP name
+    ENTRY_NO_GP_CUSTOM_CFA \name, 0
 .endm
 
 .macro END name
diff --git a/runtime/arch/mips/fault_handler_mips.cc b/runtime/arch/mips/fault_handler_mips.cc
index 7969a8f..b6a63ca 100644
--- a/runtime/arch/mips/fault_handler_mips.cc
+++ b/runtime/arch/mips/fault_handler_mips.cc
@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-
+#include "arch/mips/quick_method_frame_info_mips.h"
 #include "fault_handler.h"
 #include <sys/ucontext.h>
 #include "art_method-inl.h"
@@ -82,12 +82,15 @@
   struct ucontext *uc = reinterpret_cast<struct ucontext*>(context);
   struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
 
+  // Decrement $sp by the frame size of the kSaveEverything method and store
+  // the fault address in the padding right after the ArtMethod*.
+  sc->sc_regs[mips::SP] -= mips::MipsCalleeSaveFrameSize(Runtime::kSaveEverything);
+  uintptr_t* padding = reinterpret_cast<uintptr_t*>(sc->sc_regs[mips::SP]) + /* ArtMethod* */ 1;
+  *padding = reinterpret_cast<uintptr_t>(info->si_addr);
+
   sc->sc_regs[mips::RA] = sc->sc_pc + 4;      // RA needs to point to gc map location
   sc->sc_pc = reinterpret_cast<uintptr_t>(art_quick_throw_null_pointer_exception_from_signal);
   sc->sc_regs[mips::T9] = sc->sc_pc;          // make sure T9 points to the function
-  // Pass the faulting address as the first argument of
-  // art_quick_throw_null_pointer_exception_from_signal.
-  sc->sc_regs[mips::A0] = reinterpret_cast<uintptr_t>(info->si_addr);
   VLOG(signals) << "Generating null pointer exception";
   return true;
 }
diff --git a/runtime/arch/mips/quick_entrypoints_mips.S b/runtime/arch/mips/quick_entrypoints_mips.S
index 4bd1314..71b8ae2 100644
--- a/runtime/arch/mips/quick_entrypoints_mips.S
+++ b/runtime/arch/mips/quick_entrypoints_mips.S
@@ -279,6 +279,7 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything).
+     * when the $sp has already been decremented by FRAME_SIZE_SAVE_EVERYTHING.
      * Callee-save: $at, $v0-$v1, $a0-$a3, $t0-$t7, $s0-$s7, $t8-$t9, $gp, $fp $ra, $f0-$f31;
      *              28(GPR)+ 32(FPR) + 3 words for padding and 1 word for Method*
      * Clobbers $t0 and $t1.
@@ -286,10 +287,7 @@
      * Reserves FRAME_SIZE_SAVE_EVERYTHING + ARG_SLOT_SIZE bytes on the stack.
      * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP.
      */
-.macro SETUP_SAVE_EVERYTHING_FRAME
-    addiu  $sp, $sp, -256
-    .cfi_adjust_cfa_offset 256
-
+.macro SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
      // Ugly compile-time check, but we only have the preprocessor.
 #if (FRAME_SIZE_SAVE_EVERYTHING != 256)
 #error "FRAME_SIZE_SAVE_EVERYTHING(MIPS) size not as expected."
@@ -388,6 +386,22 @@
     .cfi_adjust_cfa_offset ARG_SLOT_SIZE
 .endm
 
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything).
+     * Callee-save: $at, $v0-$v1, $a0-$a3, $t0-$t7, $s0-$s7, $t8-$t9, $gp, $fp $ra, $f0-$f31;
+     *              28(GPR)+ 32(FPR) + 3 words for padding and 1 word for Method*
+     * Clobbers $t0 and $t1.
+     * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots.
+     * Reserves FRAME_SIZE_SAVE_EVERYTHING + ARG_SLOT_SIZE bytes on the stack.
+     * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP.
+     */
+.macro SETUP_SAVE_EVERYTHING_FRAME
+    addiu  $sp, $sp, -(FRAME_SIZE_SAVE_EVERYTHING)
+    .cfi_adjust_cfa_offset (FRAME_SIZE_SAVE_EVERYTHING)
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
+.endm
+
 .macro RESTORE_SAVE_EVERYTHING_FRAME
     addiu  $sp, $sp, ARG_SLOT_SIZE                # remove argument slots on the stack
     .cfi_adjust_cfa_offset -ARG_SLOT_SIZE
@@ -708,8 +722,10 @@
      * Call installed by a signal handler to create and deliver a NullPointerException.
      */
     .extern artThrowNullPointerExceptionFromSignal
-ENTRY art_quick_throw_null_pointer_exception_from_signal
-    SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
+ENTRY_NO_GP_CUSTOM_CFA art_quick_throw_null_pointer_exception_from_signal, FRAME_SIZE_SAVE_EVERYTHING
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
+    # Retrieve the fault address from the padding where the signal handler stores it.
+    lw   $a0, (ARG_SLOT_SIZE + __SIZEOF_POINTER__)($sp)
     la   $t9, artThrowNullPointerExceptionFromSignal
     jalr $zero, $t9                 # artThrowNullPointerExceptionFromSignal(uintptr_t, Thread*)
     move $a1, rSELF                 # pass Thread::Current
diff --git a/runtime/arch/mips64/asm_support_mips64.S b/runtime/arch/mips64/asm_support_mips64.S
index 786e860..6c58fcf 100644
--- a/runtime/arch/mips64/asm_support_mips64.S
+++ b/runtime/arch/mips64/asm_support_mips64.S
@@ -45,8 +45,8 @@
 .L\name\()_gp_set:
 .endm
 
-     // Declare a function called name, doesn't set up $gp.
-.macro ENTRY_NO_GP name
+    // Declare a function called name, doesn't set up $gp.
+.macro ENTRY_NO_GP_CUSTOM_CFA name, cfa_offset
     .type \name, %function
     .global \name
     // Cache alignment for function entry.
@@ -54,7 +54,12 @@
 \name:
     .cfi_startproc
      // Ensure we get a sane starting CFA.
-    .cfi_def_cfa $sp,0
+    .cfi_def_cfa $sp, \cfa_offset
+.endm
+
+    // Declare a function called name, doesn't set up $gp.
+.macro ENTRY_NO_GP name
+    ENTRY_NO_GP_CUSTOM_CFA \name, 0
 .endm
 
 .macro END name
diff --git a/runtime/arch/mips64/fault_handler_mips64.cc b/runtime/arch/mips64/fault_handler_mips64.cc
index 0bbb6e1..e52dc73 100644
--- a/runtime/arch/mips64/fault_handler_mips64.cc
+++ b/runtime/arch/mips64/fault_handler_mips64.cc
@@ -14,7 +14,7 @@
  * limitations under the License.
  */
 
-
+#include "arch/mips64/quick_method_frame_info_mips64.h"
 #include "fault_handler.h"
 #include <sys/ucontext.h>
 #include "art_method-inl.h"
@@ -83,12 +83,15 @@
   struct ucontext *uc = reinterpret_cast<struct ucontext*>(context);
   struct sigcontext *sc = reinterpret_cast<struct sigcontext*>(&uc->uc_mcontext);
 
+  // Decrement $sp by the frame size of the kSaveEverything method and store
+  // the fault address in the padding right after the ArtMethod*.
+  sc->sc_regs[mips64::SP] -= mips64::Mips64CalleeSaveFrameSize(Runtime::kSaveEverything);
+  uintptr_t* padding = reinterpret_cast<uintptr_t*>(sc->sc_regs[mips64::SP]) + /* ArtMethod* */ 1;
+  *padding = reinterpret_cast<uintptr_t>(info->si_addr);
+
   sc->sc_regs[mips64::RA] = sc->sc_pc + 4;      // RA needs to point to gc map location
   sc->sc_pc = reinterpret_cast<uintptr_t>(art_quick_throw_null_pointer_exception_from_signal);
   sc->sc_regs[mips64::T9] = sc->sc_pc;          // make sure T9 points to the function
-  // Pass the faulting address as the first argument of
-  // art_quick_throw_null_pointer_exception_from_signal.
-  sc->sc_regs[mips64::A0] = reinterpret_cast<uintptr_t>(info->si_addr);
   VLOG(signals) << "Generating null pointer exception";
   return true;
 }
diff --git a/runtime/arch/mips64/quick_entrypoints_mips64.S b/runtime/arch/mips64/quick_entrypoints_mips64.S
index 26717ad..61c9019 100644
--- a/runtime/arch/mips64/quick_entrypoints_mips64.S
+++ b/runtime/arch/mips64/quick_entrypoints_mips64.S
@@ -316,14 +316,12 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything).
+     * when the $sp has already been decremented by FRAME_SIZE_SAVE_EVERYTHING.
      * callee-save: $at + $v0-$v1 + $a0-$a7 + $t0-$t3 + $s0-$s7 + $t8-$t9 + $gp + $s8 + $ra + $s8,
      *              $f0-$f31; 28(GPR)+ 32(FPR) + 1x8 bytes padding + method*
      * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP.
      */
-.macro SETUP_SAVE_EVERYTHING_FRAME
-    daddiu $sp, $sp, -496
-    .cfi_adjust_cfa_offset 496
-
+.macro SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
      // Ugly compile-time check, but we only have the preprocessor.
 #if (FRAME_SIZE_SAVE_EVERYTHING != 496)
 #error "FRAME_SIZE_SAVE_EVERYTHING(MIPS64) size not as expected."
@@ -436,6 +434,19 @@
     sd      $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF)
 .endm
 
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything).
+     * callee-save: $at + $v0-$v1 + $a0-$a7 + $t0-$t3 + $s0-$s7 + $t8-$t9 + $gp + $s8 + $ra + $s8,
+     *              $f0-$f31; 28(GPR)+ 32(FPR) + 1x8 bytes padding + method*
+     * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP.
+     */
+.macro SETUP_SAVE_EVERYTHING_FRAME
+    daddiu $sp, $sp, -(FRAME_SIZE_SAVE_EVERYTHING)
+    .cfi_adjust_cfa_offset (FRAME_SIZE_SAVE_EVERYTHING)
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
+.endm
+
 .macro RESTORE_SAVE_EVERYTHING_FRAME
     // Restore FP registers.
     l.d    $f31, 264($sp)
@@ -818,8 +829,10 @@
      * Call installed by a signal handler to create and deliver a NullPointerException
      */
     .extern artThrowNullPointerExceptionFromSignal
-ENTRY art_quick_throw_null_pointer_exception_from_signal
-    SETUP_SAVE_ALL_CALLEE_SAVES_FRAME
+ENTRY_NO_GP_CUSTOM_CFA art_quick_throw_null_pointer_exception_from_signal, FRAME_SIZE_SAVE_EVERYTHING
+    SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP
+    # Retrieve the fault address from the padding where the signal handler stores it.
+    ld   $a0, (__SIZEOF_POINTER__)($sp)
     dla  $t9, artThrowNullPointerExceptionFromSignal
     jalr $zero, $t9                 # artThrowNullPointerExceptionFromSignal(uinptr_t, Thread*)
     move $a1, rSELF                 # pass Thread::Current
diff --git a/runtime/arch/x86/asm_support_x86.S b/runtime/arch/x86/asm_support_x86.S
index 3e47209..14b01c5 100644
--- a/runtime/arch/x86/asm_support_x86.S
+++ b/runtime/arch/x86/asm_support_x86.S
@@ -114,7 +114,7 @@
     .balign 16
 END_MACRO
 
-MACRO1(DEFINE_FUNCTION, c_name)
+MACRO2(DEFINE_FUNCTION_CUSTOM_CFA, c_name, cfa_offset)
     FUNCTION_TYPE(SYMBOL(\c_name))
     ASM_HIDDEN CALLVAR(c_name)
     .globl CALLVAR(c_name)
@@ -122,7 +122,11 @@
 CALLVAR(c_name):
     CFI_STARTPROC
     // Ensure we get a sane starting CFA.
-    CFI_DEF_CFA(esp, 4)
+    CFI_DEF_CFA(esp, RAW_VAR(cfa_offset))
+END_MACRO
+
+MACRO1(DEFINE_FUNCTION, c_name)
+    DEFINE_FUNCTION_CUSTOM_CFA RAW_VAR(c_name), __SIZEOF_POINTER__
 END_MACRO
 
 MACRO1(END_FUNCTION, c_name)
diff --git a/runtime/arch/x86/fault_handler_x86.cc b/runtime/arch/x86/fault_handler_x86.cc
index c7af249..a4d6bb4 100644
--- a/runtime/arch/x86/fault_handler_x86.cc
+++ b/runtime/arch/x86/fault_handler_x86.cc
@@ -325,21 +325,15 @@
   // next instruction (this instruction + instruction size).  The return address
   // is on the stack at the top address of the current frame.
 
-  // Push the return address onto the stack.
+  // Push the return address and fault address onto the stack.
   uintptr_t retaddr = reinterpret_cast<uintptr_t>(pc + instr_size);
-  uintptr_t* next_sp = reinterpret_cast<uintptr_t*>(sp - sizeof(uintptr_t));
-  *next_sp = retaddr;
+  uintptr_t* next_sp = reinterpret_cast<uintptr_t*>(sp - 2 * sizeof(uintptr_t));
+  next_sp[1] = retaddr;
+  next_sp[0] = reinterpret_cast<uintptr_t>(sig->si_addr);
   uc->CTX_ESP = reinterpret_cast<uintptr_t>(next_sp);
 
   uc->CTX_EIP = reinterpret_cast<uintptr_t>(
       art_quick_throw_null_pointer_exception_from_signal);
-  // Pass the faulting address as the first argument of
-  // art_quick_throw_null_pointer_exception_from_signal.
-#if defined(__x86_64__)
-  uc->CTX_RDI = reinterpret_cast<uintptr_t>(sig->si_addr);
-#else
-  uc->CTX_EAX = reinterpret_cast<uintptr_t>(sig->si_addr);
-#endif
   VLOG(signals) << "Generating null pointer exception";
   return true;
 }
diff --git a/runtime/arch/x86/quick_entrypoints_x86.S b/runtime/arch/x86/quick_entrypoints_x86.S
index 646a80c..0beb2a4 100644
--- a/runtime/arch/x86/quick_entrypoints_x86.S
+++ b/runtime/arch/x86/quick_entrypoints_x86.S
@@ -224,10 +224,11 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     * when EDI is already saved.
      */
-MACRO2(SETUP_SAVE_EVERYTHING_FRAME, got_reg, temp_reg)
-    // Save core registers.
-    PUSH edi
+MACRO2(SETUP_SAVE_EVERYTHING_FRAME_EDI_SAVED, got_reg, temp_reg)
+    // Save core registers from highest to lowest to agree with core spills bitmap.
+    // EDI, or at least a placeholder for it, is already on the stack.
     PUSH esi
     PUSH ebp
     PUSH ebx
@@ -264,6 +265,15 @@
 #endif
 END_MACRO
 
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     */
+MACRO2(SETUP_SAVE_EVERYTHING_FRAME, got_reg, temp_reg)
+    PUSH edi
+    SETUP_SAVE_EVERYTHING_FRAME_EDI_SAVED RAW_VAR(got_reg), RAW_VAR(temp_reg)
+END_MACRO
+
 MACRO0(RESTORE_SAVE_EVERYTHING_FRAME)
     // Restore FPRs. Method and padding is still on the stack.
     movsd 16(%esp), %xmm0
@@ -320,7 +330,6 @@
 MACRO2(ONE_ARG_RUNTIME_EXCEPTION, c_name, cxx_name)
     DEFINE_FUNCTION VAR(c_name)
     SETUP_SAVE_ALL_CALLEE_SAVES_FRAME ebx, ebx // save all registers as basis for long jump context
-    mov %esp, %ecx
     // Outgoing argument set up
     subl MACRO_LITERAL(8), %esp                // alignment padding
     CFI_ADJUST_CFA_OFFSET(8)
@@ -354,7 +363,23 @@
     /*
      * Call installed by a signal handler to create and deliver a NullPointerException.
      */
-ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal
+DEFINE_FUNCTION_CUSTOM_CFA art_quick_throw_null_pointer_exception_from_signal, 2 * __SIZEOF_POINTER__
+    // Fault address and return address were saved by the fault handler.
+    // Save all registers as basis for long jump context; EDI will replace fault address later.
+    SETUP_SAVE_EVERYTHING_FRAME_EDI_SAVED ebx, ebx
+    // Retrieve fault address and save EDI.
+    movl (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__)(%esp), %eax
+    movl %edi, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__)(%esp)
+    CFI_REL_OFFSET(%edi, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__))
+    // Outgoing argument set up
+    subl MACRO_LITERAL(8), %esp                           // alignment padding
+    CFI_ADJUST_CFA_OFFSET(8)
+    pushl %fs:THREAD_SELF_OFFSET                          // pass Thread::Current()
+    CFI_ADJUST_CFA_OFFSET(4)
+    PUSH eax                                              // pass arg1
+    call SYMBOL(artThrowNullPointerExceptionFromSignal)   // (addr, self)
+    UNREACHABLE
+END_FUNCTION art_quick_throw_null_pointer_exception
 
     /*
      * Called by managed code to create and deliver an ArithmeticException.
diff --git a/runtime/arch/x86_64/asm_support_x86_64.S b/runtime/arch/x86_64/asm_support_x86_64.S
index 0728f99..af4a6c4 100644
--- a/runtime/arch/x86_64/asm_support_x86_64.S
+++ b/runtime/arch/x86_64/asm_support_x86_64.S
@@ -110,7 +110,7 @@
 
 // TODO: we might need to use SYMBOL() here to add the underscore prefix
 // for mac builds.
-MACRO1(DEFINE_FUNCTION, c_name)
+MACRO2(DEFINE_FUNCTION_CUSTOM_CFA, c_name, cfa_offset)
     FUNCTION_TYPE(SYMBOL(\c_name))
     ASM_HIDDEN CALLVAR(c_name)
     .globl CALLVAR(c_name)
@@ -118,7 +118,11 @@
 CALLVAR(c_name):
     CFI_STARTPROC
     // Ensure we get a sane starting CFA.
-    CFI_DEF_CFA(rsp, 8)
+    CFI_DEF_CFA(rsp, RAW_VAR(cfa_offset))
+END_MACRO
+
+MACRO1(DEFINE_FUNCTION, c_name)
+    DEFINE_FUNCTION_CUSTOM_CFA RAW_VAR(c_name), __SIZEOF_POINTER__
 END_MACRO
 
 MACRO1(END_FUNCTION, c_name)
diff --git a/runtime/arch/x86_64/quick_entrypoints_x86_64.S b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
index 5ea58af..089ed75 100644
--- a/runtime/arch/x86_64/quick_entrypoints_x86_64.S
+++ b/runtime/arch/x86_64/quick_entrypoints_x86_64.S
@@ -263,14 +263,15 @@
     /*
      * Macro that sets up the callee save frame to conform with
      * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     * when R15 is already saved.
      */
-MACRO0(SETUP_SAVE_EVERYTHING_FRAME)
+MACRO0(SETUP_SAVE_EVERYTHING_FRAME_R15_SAVED)
 #if defined(__APPLE__)
     int3
     int3
 #else
     // Save core registers from highest to lowest to agree with core spills bitmap.
-    PUSH r15
+    // R15, or at least a placeholder for it, is already on the stack.
     PUSH r14
     PUSH r13
     PUSH r12
@@ -322,6 +323,15 @@
 #endif  // __APPLE__
 END_MACRO
 
+    /*
+     * Macro that sets up the callee save frame to conform with
+     * Runtime::CreateCalleeSaveMethod(kSaveEverything)
+     */
+MACRO0(SETUP_SAVE_EVERYTHING_FRAME)
+    PUSH r15
+    SETUP_SAVE_EVERYTHING_FRAME_R15_SAVED
+END_MACRO
+
 MACRO0(RESTORE_SAVE_EVERYTHING_FRAME)
     // Restore FPRs. Method and padding is still on the stack.
     movq 16(%rsp), %xmm0
@@ -413,7 +423,19 @@
     /*
      * Call installed by a signal handler to create and deliver a NullPointerException.
      */
-ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception_from_signal, artThrowNullPointerExceptionFromSignal
+DEFINE_FUNCTION_CUSTOM_CFA art_quick_throw_null_pointer_exception_from_signal, 2 * __SIZEOF_POINTER__
+    // Fault address and return address were saved by the fault handler.
+    // Save all registers as basis for long jump context; R15 will replace fault address later.
+    SETUP_SAVE_EVERYTHING_FRAME_R15_SAVED
+    // Retrieve fault address and save R15.
+    movq (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__)(%rsp), %rdi
+    movq %r15, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__)(%rsp)
+    CFI_REL_OFFSET(%r15, (FRAME_SIZE_SAVE_EVERYTHING - 2 * __SIZEOF_POINTER__))
+    // Outgoing argument set up; RDI already contains the fault address.
+    movq %gs:THREAD_SELF_OFFSET, %rsi  // pass Thread::Current()
+    call SYMBOL(artThrowNullPointerExceptionFromSignal)  // (addr, self)
+    UNREACHABLE
+END_FUNCTION art_quick_throw_null_pointer_exception_from_signal
 
     /*
      * Called by managed code to create and deliver an ArithmeticException.