Gitiles
Code Review Sign In
LeafOS / LeafOS-Project / android_art / 97c46466aea25ab63a99b3d1afc558f0d9f55abb / . / runtime / arch / arm / quick_entrypoints_arm.S
blob: b909bda3f791dc77d4bcadb8e31adaadc9f2a0ac [file] [log] [blame]
/*
* Copyright (C) 2012 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#include "asm_support_arm.S"
#include "arch/quick_alloc_entrypoints.S"
/* Deliver the given exception */
.extern artDeliverExceptionFromCode
/* Deliver an exception pending on a thread */
.extern artDeliverPendingException
/*
* Macro to spill the GPRs.
*/
.macro SPILL_ALL_CALLEE_SAVE_GPRS
push {r4-r11, lr} @ 9 words (36 bytes) of callee saves.
.cfi_adjust_cfa_offset 36
.cfi_rel_offset r4, 0
.cfi_rel_offset r5, 4
.cfi_rel_offset r6, 8
.cfi_rel_offset r7, 12
.cfi_rel_offset r8, 16
.cfi_rel_offset r9, 20
.cfi_rel_offset r10, 24
.cfi_rel_offset r11, 28
.cfi_rel_offset lr, 32
.endm
/*
* Macro that sets up the callee save frame to conform with
* Runtime::CreateCalleeSaveMethod(kSaveAllCalleeSaves)
*/
.macro SETUP_SAVE_ALL_CALLEE_SAVES_FRAME rTemp
SPILL_ALL_CALLEE_SAVE_GPRS @ 9 words (36 bytes) of callee saves.
vpush {s16-s31} @ 16 words (64 bytes) of floats.
.cfi_adjust_cfa_offset 64
sub sp, #12 @ 3 words of space, bottom word will hold Method*
.cfi_adjust_cfa_offset 12
RUNTIME_CURRENT1 \rTemp @ Load Runtime::Current into rTemp.
@ Load kSaveAllCalleeSaves Method* into rTemp.
ldr \rTemp, [\rTemp, #RUNTIME_SAVE_ALL_CALLEE_SAVES_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_ALL_CALLEE_SAVES != 36 + 64 + 12)
#error "FRAME_SIZE_SAVE_ALL_CALLEE_SAVES(ARM) size not as expected."
#endif
.endm
/*
* Macro that sets up the callee save frame to conform with
* Runtime::CreateCalleeSaveMethod(kSaveRefsOnly).
*/
.macro SETUP_SAVE_REFS_ONLY_FRAME rTemp
push {r5-r8, r10-r11, lr} @ 7 words of callee saves
.cfi_adjust_cfa_offset 28
.cfi_rel_offset r5, 0
.cfi_rel_offset r6, 4
.cfi_rel_offset r7, 8
.cfi_rel_offset r8, 12
.cfi_rel_offset r10, 16
.cfi_rel_offset r11, 20
.cfi_rel_offset lr, 24
sub sp, #4 @ bottom word will hold Method*
.cfi_adjust_cfa_offset 4
RUNTIME_CURRENT2 \rTemp @ Load Runtime::Current into rTemp.
@ Load kSaveRefsOnly Method* into rTemp.
ldr \rTemp, [\rTemp, #RUNTIME_SAVE_REFS_ONLY_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_REFS_ONLY != 28 + 4)
#error "FRAME_SIZE_SAVE_REFS_ONLY(ARM) size not as expected."
#endif
.endm
.macro RESTORE_SAVE_REFS_ONLY_FRAME
add sp, #4 @ bottom word holds Method*
.cfi_adjust_cfa_offset -4
pop {r5-r8, r10-r11, lr} @ 7 words of callee saves
.cfi_restore r5
.cfi_restore r6
.cfi_restore r7
.cfi_restore r8
.cfi_restore r10
.cfi_restore r11
.cfi_restore lr
.cfi_adjust_cfa_offset -28
.endm
.macro RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN
RESTORE_SAVE_REFS_ONLY_FRAME
bx lr @ return
.endm
/*
* Macro that sets up the callee save frame to conform with
* Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs).
*/
.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY
push {r1-r3, r5-r8, r10-r11, lr} @ 10 words of callee saves and args.
.cfi_adjust_cfa_offset 40
.cfi_rel_offset r1, 0
.cfi_rel_offset r2, 4
.cfi_rel_offset r3, 8
.cfi_rel_offset r5, 12
.cfi_rel_offset r6, 16
.cfi_rel_offset r7, 20
.cfi_rel_offset r8, 24
.cfi_rel_offset r10, 28
.cfi_rel_offset r11, 32
.cfi_rel_offset lr, 36
vpush {s0-s15} @ 16 words of float args.
.cfi_adjust_cfa_offset 64
sub sp, #8 @ 2 words of space, alignment padding and Method*
.cfi_adjust_cfa_offset 8
// Ugly compile-time check, but we only have the preprocessor.
#if (FRAME_SIZE_SAVE_REFS_AND_ARGS != 40 + 64 + 8)
#error "FRAME_SIZE_SAVE_REFS_AND_ARGS(ARM) size not as expected."
#endif
.endm
.macro SETUP_SAVE_REFS_AND_ARGS_FRAME rTemp
SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY
RUNTIME_CURRENT3 \rTemp @ Load Runtime::Current into rTemp.
@ Load kSaveRefsAndArgs Method* into rTemp.
ldr \rTemp, [\rTemp, #RUNTIME_SAVE_REFS_AND_ARGS_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.
.endm
.macro SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_R0
SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY
str r0, [sp, #0] @ Store ArtMethod* to bottom of stack.
str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame.
.endm
.macro RESTORE_SAVE_REFS_AND_ARGS_FRAME
add sp, #8 @ rewind sp
.cfi_adjust_cfa_offset -8
vpop {s0-s15}
.cfi_adjust_cfa_offset -64
pop {r1-r3, r5-r8, r10-r11, lr} @ 10 words of callee saves
.cfi_restore r1
.cfi_restore r2
.cfi_restore r3
.cfi_restore r5
.cfi_restore r6
.cfi_restore r7
.cfi_restore r8
.cfi_restore r10
.cfi_restore r11
.cfi_restore lr
.cfi_adjust_cfa_offset -40
.endm
/*
* 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.
.cfi_adjust_cfa_offset 56
.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
.cfi_rel_offset lr, 52
SETUP_SAVE_EVERYTHING_FRAME_CORE_REGS_SAVED \rTemp
.endm
.macro RESTORE_SAVE_EVERYTHING_FRAME
add sp, #8 @ rewind sp
.cfi_adjust_cfa_offset -8
vpop {d0-d15}
.cfi_adjust_cfa_offset -128
pop {r0-r12, lr} @ 14 words of callee saves
.cfi_restore r0
.cfi_restore r1
.cfi_restore r2
.cfi_restore r3
.cfi_restore r5
.cfi_restore r6
.cfi_restore r7
.cfi_restore r8
.cfi_restore r9
.cfi_restore r10
.cfi_restore r11
.cfi_restore r12
.cfi_restore lr
.cfi_adjust_cfa_offset -56
.endm
.macro RESTORE_SAVE_EVERYTHING_FRAME_KEEP_R0
add sp, #8 @ rewind sp
.cfi_adjust_cfa_offset -8
vpop {d0-d15}
.cfi_adjust_cfa_offset -128
add sp, #4 @ skip r0
.cfi_adjust_cfa_offset -4
.cfi_restore r0 @ debugger can no longer restore caller's r0
pop {r1-r12, lr} @ 13 words of callee saves
.cfi_restore r1
.cfi_restore r2
.cfi_restore r3
.cfi_restore r5
.cfi_restore r6
.cfi_restore r7
.cfi_restore r8
.cfi_restore r9
.cfi_restore r10
.cfi_restore r11
.cfi_restore r12
.cfi_restore lr
.cfi_adjust_cfa_offset -52
.endm
.macro RETURN_IF_RESULT_IS_ZERO
cbnz r0, 1f @ result non-zero branch over
bx lr @ return
1:
.endm
.macro RETURN_IF_RESULT_IS_NON_ZERO
cbz r0, 1f @ result zero branch over
bx lr @ return
1:
.endm
/*
* Macro that calls through to artDeliverPendingExceptionFromCode, where the pending
* exception is Thread::Current()->exception_ when the runtime method frame is ready.
*/
.macro DELIVER_PENDING_EXCEPTION_FRAME_READY
mov r0, r9 @ pass Thread::Current
bl artDeliverPendingExceptionFromCode @ artDeliverPendingExceptionFromCode(Thread*)
.endm
/*
* Macro that calls through to artDeliverPendingExceptionFromCode, where the pending
* exception is Thread::Current()->exception_.
*/
.macro DELIVER_PENDING_EXCEPTION
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r0 @ save callee saves for throw
DELIVER_PENDING_EXCEPTION_FRAME_READY
.endm
.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name
.extern \cxx_name
ENTRY \c_name
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r0 @ save all registers as basis for long jump context
mov r0, r9 @ pass Thread::Current
bl \cxx_name @ \cxx_name(Thread*)
END \c_name
.endm
.macro NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING c_name, cxx_name
.extern \cxx_name
ENTRY \c_name
SETUP_SAVE_EVERYTHING_FRAME r0 @ save all registers as basis for long jump context
mov r0, r9 @ pass Thread::Current
bl \cxx_name @ \cxx_name(Thread*)
END \c_name
.endm
.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name
.extern \cxx_name
ENTRY \c_name
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r1 @ save all registers as basis for long jump context
mov r1, r9 @ pass Thread::Current
bl \cxx_name @ \cxx_name(Thread*)
END \c_name
.endm
.macro TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING c_name, cxx_name
.extern \cxx_name
ENTRY \c_name
SETUP_SAVE_EVERYTHING_FRAME r2 @ save all registers as basis for long jump context
mov r2, r9 @ pass Thread::Current
bl \cxx_name @ \cxx_name(Thread*)
END \c_name
.endm
.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg
ldr \reg, [r9, #THREAD_EXCEPTION_OFFSET] // Get exception field.
cbnz \reg, 1f
bx lr
1:
DELIVER_PENDING_EXCEPTION
.endm
.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
RETURN_OR_DELIVER_PENDING_EXCEPTION_REG r1
.endm
.macro RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
.endm
.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
RETURN_IF_RESULT_IS_NON_ZERO
DELIVER_PENDING_EXCEPTION
.endm
// Macros taking opportunity of code similarities for downcalls.
.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r1 @ save callee saves in case of GC
mov r1, r9 @ pass Thread::Current
bl \entrypoint @ (uint32_t field_idx, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r2 @ save callee saves in case of GC
mov r2, r9 @ pass Thread::Current
bl \entrypoint @ (field_idx, Object*, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r3 @ save callee saves in case of GC
mov r3, r9 @ pass Thread::Current
bl \entrypoint @ (field_idx, Object*, new_val, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME @ TODO: we can clearly save an add here
\return
END \name
.endm
/*
* Called by managed code, saves callee saves and then calls artThrowException
* that will place a mock Method* at the bottom of the stack. Arg1 holds the exception.
*/
ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode
/*
* Called by managed code to create and deliver a NullPointerException.
*/
NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode
/*
* Call installed by a signal handler to create and deliver a NullPointerException.
*/
.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
bl artThrowNullPointerExceptionFromSignal @ (Thread*)
END art_quick_throw_null_pointer_exception_from_signal
/*
* Called by managed code to create and deliver an ArithmeticException.
*/
NO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_div_zero, artThrowDivZeroFromCode
/*
* Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds
* index, arg2 holds limit.
*/
TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_array_bounds, artThrowArrayBoundsFromCode
/*
* Called by managed code to create and deliver a StringIndexOutOfBoundsException
* as if thrown from a call to String.charAt(). Arg1 holds index, arg2 holds limit.
*/
TWO_ARG_RUNTIME_EXCEPTION_SAVE_EVERYTHING art_quick_throw_string_bounds, artThrowStringBoundsFromCode
/*
* Called by managed code to create and deliver a StackOverflowError.
*/
NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode
/*
* All generated callsites for interface invokes and invocation slow paths will load arguments
* as usual - except instead of loading arg0/r0 with the target Method*, arg0/r0 will contain
* the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper.
* NOTE: "this" is first visible argument of the target, and so can be found in arg1/r1.
*
* The helper will attempt to locate the target and return a 64-bit result in r0/r1 consisting
* of the target Method* in r0 and method->code_ in r1.
*
* If unsuccessful, the helper will return null/null. There will bea pending exception in the
* thread and we branch to another stub to deliver it.
*
* On success this wrapper will restore arguments and *jump* to the target, leaving the lr
* pointing back to the original caller.
*/
.macro INVOKE_TRAMPOLINE_BODY cxx_name
.extern \cxx_name
SETUP_SAVE_REFS_AND_ARGS_FRAME r2 @ save callee saves in case allocation triggers GC
mov r2, r9 @ pass Thread::Current
mov r3, sp
bl \cxx_name @ (method_idx, this, Thread*, SP)
mov r12, r1 @ save Method*->code_
RESTORE_SAVE_REFS_AND_ARGS_FRAME
cbz r0, 1f @ did we find the target? if not go to exception delivery
bx r12 @ tail call to target
1:
DELIVER_PENDING_EXCEPTION
.endm
.macro INVOKE_TRAMPOLINE c_name, cxx_name
ENTRY \c_name
INVOKE_TRAMPOLINE_BODY \cxx_name
END \c_name
.endm
INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck
INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck
INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck
INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck
INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck
/*
* Quick invocation stub internal.
* On entry:
* r0 = method pointer
* r1 = argument array or null for no argument methods
* r2 = size of argument array in bytes
* r3 = (managed) thread pointer
* [sp] = JValue* result
* [sp + 4] = result_in_float
* [sp + 8] = core register argument array
* [sp + 12] = fp register argument array
* +-------------------------+
* | uint32_t* fp_reg_args |
* | uint32_t* core_reg_args |
* | result_in_float | <- Caller frame
* | Jvalue* result |
* +-------------------------+
* | lr |
* | r11 |
* | r9 |
* | r4 | <- r11
* +-------------------------+
* | uint32_t out[n-1] |
* | : : | Outs
* | uint32_t out[0] |
* | StackRef<ArtMethod> | <- SP value=null
* +-------------------------+
*/
ENTRY art_quick_invoke_stub_internal
SPILL_ALL_CALLEE_SAVE_GPRS @ spill regs (9)
mov r11, sp @ save the stack pointer
.cfi_def_cfa_register r11
mov r9, r3 @ move managed thread pointer into r9
add r4, r2, #4 @ create space for method pointer in frame
sub r4, sp, r4 @ reserve & align *stack* to 16 bytes: native calling
and r4, #0xFFFFFFF0 @ convention only aligns to 8B, so we have to ensure ART
mov sp, r4 @ 16B alignment ourselves.
mov r4, r0 @ save method*
add r0, sp, #4 @ pass stack pointer + method ptr as dest for memcpy
bl memcpy @ memcpy (dest, src, bytes)
mov ip, #0 @ set ip to 0
str ip, [sp] @ store null for method* at bottom of frame
ldr ip, [r11, #48] @ load fp register argument array pointer
vldm ip, {s0-s15} @ copy s0 - s15
ldr ip, [r11, #44] @ load core register argument array pointer
mov r0, r4 @ restore method*
add ip, ip, #4 @ skip r0
ldm ip, {r1-r3} @ copy r1 - r3
#ifdef ARM_R4_SUSPEND_FLAG
mov r4, #SUSPEND_CHECK_INTERVAL @ reset r4 to suspend check interval
#endif
ldr ip, [r0, #ART_METHOD_QUICK_CODE_OFFSET_32] @ get pointer to the code
blx ip @ call the method
mov sp, r11 @ restore the stack pointer
.cfi_def_cfa_register sp
ldr r4, [sp, #40] @ load result_is_float
ldr r9, [sp, #36] @ load the result pointer
cmp r4, #0
ite eq
strdeq r0, [r9] @ store r0/r1 into result pointer
vstrne d0, [r9] @ store s0-s1/d0 into result pointer
pop {r4, r5, r6, r7, r8, r9, r10, r11, pc} @ restore spill regs
END art_quick_invoke_stub_internal
/*
* On stack replacement stub.
* On entry:
* r0 = stack to copy
* r1 = size of stack
* r2 = pc to call
* r3 = JValue* result
* [sp] = shorty
* [sp + 4] = thread
*/
ENTRY art_quick_osr_stub
SPILL_ALL_CALLEE_SAVE_GPRS @ Spill regs (9)
mov r11, sp @ Save the stack pointer
mov r10, r1 @ Save size of stack
ldr r9, [r11, #40] @ Move managed thread pointer into r9
mov r8, r2 @ Save the pc to call
sub r7, sp, #12 @ Reserve space for stack pointer,
@ JValue* result, and ArtMethod* slot.
and r7, #0xFFFFFFF0 @ Align stack pointer
mov sp, r7 @ Update stack pointer
str r11, [sp, #4] @ Save old stack pointer
str r3, [sp, #8] @ Save JValue* result
mov ip, #0
str ip, [sp] @ Store null for ArtMethod* at bottom of frame
sub sp, sp, r1 @ Reserve space for callee stack
mov r2, r1
mov r1, r0
mov r0, sp
bl memcpy @ memcpy (dest r0, src r1, bytes r2)
bl .Losr_entry @ Call the method
ldr r10, [sp, #8] @ Restore JValue* result
ldr sp, [sp, #4] @ Restore saved stack pointer
ldr r4, [sp, #36] @ load shorty
ldrb r4, [r4, #0] @ load return type
cmp r4, #68 @ Test if result type char == 'D'.
beq .Losr_fp_result
cmp r4, #70 @ Test if result type char == 'F'.
beq .Losr_fp_result
strd r0, [r10] @ Store r0/r1 into result pointer
b .Losr_exit
.Losr_fp_result:
vstr d0, [r10] @ Store s0-s1/d0 into result pointer
.Losr_exit:
pop {r4, r5, r6, r7, r8, r9, r10, r11, pc}
.Losr_entry:
sub r10, r10, #4
str lr, [sp, r10] @ Store link register per the compiler ABI
bx r8
END art_quick_osr_stub
/*
* On entry r0 is uint32_t* gprs_ and r1 is uint32_t* fprs_
*/
ARM_ENTRY art_quick_do_long_jump
vldm r1, {s0-s31} @ load all fprs from argument fprs_
ldr r2, [r0, #60] @ r2 = r15 (PC from gprs_ 60=4*15)
ldr r14, [r0, #56] @ (LR from gprs_ 56=4*14)
add r0, r0, #12 @ increment r0 to skip gprs_[0..2] 12=4*3
ldm r0, {r3-r13} @ load remaining gprs from argument gprs_
ldr r0, [r0, #-12] @ load r0 value
mov r1, #0 @ clear result register r1
bx r2 @ do long jump
END art_quick_do_long_jump
/*
* Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on
* failure.
*/
TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
/*
* Entry from managed code that calls artLockObjectFromCode, may block for GC. r0 holds the
* possibly null object to lock.
*/
.extern artLockObjectFromCode
ENTRY art_quick_lock_object
cbz r0, .Lslow_lock
.Lretry_lock:
ldr r2, [r9, #THREAD_ID_OFFSET]
ldrex r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
mov r3, r1
and r3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED @ zero the gc bits
cbnz r3, .Lnot_unlocked @ already thin locked
@ unlocked case - r1: original lock word that's zero except for the read barrier bits.
orr r2, r1, r2 @ r2 holds thread id with count of 0 with preserved read barrier bits
strex r3, r2, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
cbnz r3, .Llock_strex_fail @ store failed, retry
dmb ish @ full (LoadLoad|LoadStore) memory barrier
bx lr
.Lnot_unlocked: @ r1: original lock word, r2: thread_id with count of 0 and zero read barrier bits
lsr r3, r1, LOCK_WORD_STATE_SHIFT
cbnz r3, .Lslow_lock @ if either of the top two bits are set, go slow path
eor r2, r1, r2 @ lock_word.ThreadId() ^ self->ThreadId()
uxth r2, r2 @ zero top 16 bits
cbnz r2, .Lslow_lock @ lock word and self thread id's match -> recursive lock
@ else contention, go to slow path
mov r3, r1 @ copy the lock word to check count overflow.
and r3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED @ zero the gc bits.
add r2, r3, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ increment count in lock word placing in r2 to check overflow
lsr r3, r2, #LOCK_WORD_GC_STATE_SHIFT @ if the first gc state bit is set, we overflowed.
cbnz r3, .Lslow_lock @ if we overflow the count go slow path
add r2, r1, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ increment count for real
strex r3, r2, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits
cbnz r3, .Llock_strex_fail @ strex failed, retry
bx lr
.Llock_strex_fail:
b .Lretry_lock @ retry
.Lslow_lock:
SETUP_SAVE_REFS_ONLY_FRAME r1 @ save callee saves in case we block
mov r1, r9 @ pass Thread::Current
bl artLockObjectFromCode @ (Object* obj, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_lock_object
ENTRY art_quick_lock_object_no_inline
SETUP_SAVE_REFS_ONLY_FRAME r1 @ save callee saves in case we block
mov r1, r9 @ pass Thread::Current
bl artLockObjectFromCode @ (Object* obj, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_lock_object_no_inline
/*
* Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
* r0 holds the possibly null object to lock.
*/
.extern artUnlockObjectFromCode
ENTRY art_quick_unlock_object
cbz r0, .Lslow_unlock
.Lretry_unlock:
#ifndef USE_READ_BARRIER
ldr r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
#else
ldrex r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ Need to use atomic instructions for read barrier
#endif
lsr r2, r1, #LOCK_WORD_STATE_SHIFT
cbnz r2, .Lslow_unlock @ if either of the top two bits are set, go slow path
ldr r2, [r9, #THREAD_ID_OFFSET]
mov r3, r1 @ copy lock word to check thread id equality
and r3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED @ zero the gc bits
eor r3, r3, r2 @ lock_word.ThreadId() ^ self->ThreadId()
uxth r3, r3 @ zero top 16 bits
cbnz r3, .Lslow_unlock @ do lock word and self thread id's match?
mov r3, r1 @ copy lock word to detect transition to unlocked
and r3, #LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED @ zero the gc bits
cmp r3, #LOCK_WORD_THIN_LOCK_COUNT_ONE
bpl .Lrecursive_thin_unlock
@ transition to unlocked
mov r3, r1
and r3, #LOCK_WORD_GC_STATE_MASK_SHIFTED @ r3: zero except for the preserved gc bits
dmb ish @ full (LoadStore|StoreStore) memory barrier
#ifndef USE_READ_BARRIER
str r3, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
#else
strex r2, r3, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits
cbnz r2, .Lunlock_strex_fail @ store failed, retry
#endif
bx lr
.Lrecursive_thin_unlock: @ r1: original lock word
sub r1, r1, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ decrement count
#ifndef USE_READ_BARRIER
str r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
#else
strex r2, r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits
cbnz r2, .Lunlock_strex_fail @ store failed, retry
#endif
bx lr
.Lunlock_strex_fail:
b .Lretry_unlock @ retry
.Lslow_unlock:
@ save callee saves in case exception allocation triggers GC
SETUP_SAVE_REFS_ONLY_FRAME r1
mov r1, r9 @ pass Thread::Current
bl artUnlockObjectFromCode @ (Object* obj, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_unlock_object
ENTRY art_quick_unlock_object_no_inline
@ save callee saves in case exception allocation triggers GC
SETUP_SAVE_REFS_ONLY_FRAME r1
mov r1, r9 @ pass Thread::Current
bl artUnlockObjectFromCode @ (Object* obj, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_unlock_object_no_inline
/*
* Entry from managed code that calls artInstanceOfFromCode and on failure calls
* artThrowClassCastExceptionForObject.
*/
.extern artInstanceOfFromCode
.extern artThrowClassCastExceptionForObject
ENTRY art_quick_check_instance_of
push {r0-r1, lr} @ save arguments, link register and pad
.cfi_adjust_cfa_offset 12
.cfi_rel_offset r0, 0
.cfi_rel_offset r1, 4
.cfi_rel_offset lr, 8
sub sp, #4
.cfi_adjust_cfa_offset 4
bl artInstanceOfFromCode
cbz r0, .Lthrow_class_cast_exception
add sp, #4
.cfi_adjust_cfa_offset -4
pop {r0-r1, pc}
.cfi_adjust_cfa_offset 4 @ Reset unwind info so following code unwinds.
.Lthrow_class_cast_exception:
add sp, #4
.cfi_adjust_cfa_offset -4
pop {r0-r1, lr}
.cfi_adjust_cfa_offset -12
.cfi_restore r0
.cfi_restore r1
.cfi_restore lr
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r2 @ save all registers as basis for long jump context
mov r2, r9 @ pass Thread::Current
bl artThrowClassCastExceptionForObject @ (Object*, Class*, Thread*)
bkpt
END art_quick_check_instance_of
// Restore rReg's value from [sp, #offset] if rReg is not the same as rExclude.
.macro POP_REG_NE rReg, offset, rExclude
.ifnc \rReg, \rExclude
ldr \rReg, [sp, #\offset] @ restore rReg
.cfi_restore \rReg
.endif
.endm
// Save rReg's value to [sp, #offset].
.macro PUSH_REG rReg, offset
str \rReg, [sp, #\offset] @ save rReg
.cfi_rel_offset \rReg, \offset
.endm
/*
* Macro to insert read barrier, only used in art_quick_aput_obj.
* rObj and rDest are registers, offset is a defined literal such as MIRROR_OBJECT_CLASS_OFFSET.
* TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path.
*/
.macro READ_BARRIER rDest, rObj, offset
#ifdef USE_READ_BARRIER
push {r0-r3, ip, lr} @ 6 words for saved registers (used in art_quick_aput_obj)
.cfi_adjust_cfa_offset 24
.cfi_rel_offset r0, 0
.cfi_rel_offset r1, 4
.cfi_rel_offset r2, 8
.cfi_rel_offset r3, 12
.cfi_rel_offset ip, 16
.cfi_rel_offset lr, 20
sub sp, #8 @ push padding
.cfi_adjust_cfa_offset 8
@ mov r0, \rRef @ pass ref in r0 (no-op for now since parameter ref is unused)
.ifnc \rObj, r1
mov r1, \rObj @ pass rObj
.endif
mov r2, #\offset @ pass offset
bl artReadBarrierSlow @ artReadBarrierSlow(ref, rObj, offset)
@ No need to unpoison return value in r0, artReadBarrierSlow() would do the unpoisoning.
.ifnc \rDest, r0
mov \rDest, r0 @ save return value in rDest
.endif
add sp, #8 @ pop padding
.cfi_adjust_cfa_offset -8
POP_REG_NE r0, 0, \rDest @ conditionally restore saved registers
POP_REG_NE r1, 4, \rDest
POP_REG_NE r2, 8, \rDest
POP_REG_NE r3, 12, \rDest
POP_REG_NE ip, 16, \rDest
add sp, #20
.cfi_adjust_cfa_offset -20
pop {lr} @ restore lr
.cfi_adjust_cfa_offset -4
.cfi_restore lr
#else
ldr \rDest, [\rObj, #\offset]
UNPOISON_HEAP_REF \rDest
#endif // USE_READ_BARRIER
.endm
#ifdef USE_READ_BARRIER
.extern artReadBarrierSlow
#endif
.hidden art_quick_aput_obj
ENTRY art_quick_aput_obj
#ifdef USE_READ_BARRIER
@ The offset to .Ldo_aput_null is too large to use cbz due to expansion from READ_BARRIER macro.
tst r2, r2
beq .Ldo_aput_null
#else
cbz r2, .Ldo_aput_null
#endif // USE_READ_BARRIER
READ_BARRIER r3, r0, MIRROR_OBJECT_CLASS_OFFSET
READ_BARRIER ip, r2, MIRROR_OBJECT_CLASS_OFFSET
READ_BARRIER r3, r3, MIRROR_CLASS_COMPONENT_TYPE_OFFSET
cmp r3, ip @ value's type == array's component type - trivial assignability
bne .Lcheck_assignability
.Ldo_aput:
add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET
POISON_HEAP_REF r2
str r2, [r3, r1, lsl #2]
ldr r3, [r9, #THREAD_CARD_TABLE_OFFSET]
lsr r0, r0, #CARD_TABLE_CARD_SHIFT
strb r3, [r3, r0]
blx lr
.Ldo_aput_null:
add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET
str r2, [r3, r1, lsl #2]
blx lr
.Lcheck_assignability:
push {r0-r2, lr} @ save arguments
.cfi_adjust_cfa_offset 16
.cfi_rel_offset r0, 0
.cfi_rel_offset r1, 4
.cfi_rel_offset r2, 8
.cfi_rel_offset lr, 12
mov r1, ip
mov r0, r3
bl artIsAssignableFromCode
cbz r0, .Lthrow_array_store_exception
pop {r0-r2, lr}
.cfi_restore r0
.cfi_restore r1
.cfi_restore r2
.cfi_restore lr
.cfi_adjust_cfa_offset -16
add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET
POISON_HEAP_REF r2
str r2, [r3, r1, lsl #2]
ldr r3, [r9, #THREAD_CARD_TABLE_OFFSET]
lsr r0, r0, #CARD_TABLE_CARD_SHIFT
strb r3, [r3, r0]
blx lr
.Lthrow_array_store_exception:
pop {r0-r2, lr}
/* No need to repeat restore cfi directives, the ones above apply here. */
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r3
mov r1, r2
mov r2, r9 @ pass Thread::Current
bl artThrowArrayStoreException @ (Class*, Class*, Thread*)
bkpt @ unreached
END art_quick_aput_obj
// Macro to facilitate adding new allocation entrypoints.
.macro ONE_ARG_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r1 @ save callee saves in case of GC
mov r1, r9 @ pass Thread::Current
bl \entrypoint @ (uint32_t type_idx, Method* method, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
// Macro to facilitate adding new allocation entrypoints.
.macro TWO_ARG_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r2 @ save callee saves in case of GC
mov r2, r9 @ pass Thread::Current
bl \entrypoint @ (uint32_t type_idx, Method* method, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
// Macro to facilitate adding new array allocation entrypoints.
.macro THREE_ARG_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r3 @ save callee saves in case of GC
mov r3, r9 @ pass Thread::Current
@ (uint32_t type_idx, Method* method, int32_t component_count, Thread*)
bl \entrypoint
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
// Macro to facilitate adding new allocation entrypoints.
.macro FOUR_ARG_DOWNCALL name, entrypoint, return
.extern \entrypoint
ENTRY \name
SETUP_SAVE_REFS_ONLY_FRAME r12 @ save callee saves in case of GC
str r9, [sp, #-16]! @ expand the frame and pass Thread::Current
.cfi_adjust_cfa_offset 16
bl \entrypoint
add sp, #16 @ strip the extra frame
.cfi_adjust_cfa_offset -16
RESTORE_SAVE_REFS_ONLY_FRAME
\return
END \name
.endm
// Macro for string and type resolution and initialization.
.macro ONE_ARG_SAVE_EVERYTHING_DOWNCALL name, entrypoint
.extern \entrypoint
ENTRY \name
SETUP_SAVE_EVERYTHING_FRAME r1 @ save everything in case of GC
mov r1, r9 @ pass Thread::Current
bl \entrypoint @ (uint32_t index, Thread*)
cbz r0, 1f @ If result is null, deliver the OOME.
.cfi_remember_state
RESTORE_SAVE_EVERYTHING_FRAME_KEEP_R0
bx lr
.cfi_restore_state
1:
DELIVER_PENDING_EXCEPTION_FRAME_READY
END \name
.endm
ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode
ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode
ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode
ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_resolve_string, artResolveStringFromCode
/*
* Called by managed code to resolve a static field and load a non-wide value.
*/
ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
/*
* Called by managed code to resolve a static field and load a 64-bit primitive value.
*/
.extern artGet64StaticFromCompiledCode
ENTRY art_quick_get64_static
SETUP_SAVE_REFS_ONLY_FRAME r2 @ save callee saves in case of GC
mov r1, r9 @ pass Thread::Current
bl artGet64StaticFromCompiledCode @ (uint32_t field_idx, Thread*)
ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_
RESTORE_SAVE_REFS_ONLY_FRAME
cbnz r2, 1f @ success if no exception pending
bx lr @ return on success
1:
DELIVER_PENDING_EXCEPTION
END art_quick_get64_static
/*
* Called by managed code to resolve an instance field and load a non-wide value.
*/
TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCompiledCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1
/*
* Called by managed code to resolve an instance field and load a 64-bit primitive value.
*/
.extern artGet64InstanceFromCompiledCode
ENTRY art_quick_get64_instance
SETUP_SAVE_REFS_ONLY_FRAME r2 @ save callee saves in case of GC
mov r2, r9 @ pass Thread::Current
bl artGet64InstanceFromCompiledCode @ (field_idx, Object*, Thread*)
ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_
RESTORE_SAVE_REFS_ONLY_FRAME
cbnz r2, 1f @ success if no exception pending
bx lr @ return on success
1:
DELIVER_PENDING_EXCEPTION
END art_quick_get64_instance
/*
* Called by managed code to resolve a static field and store a value.
*/
TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
/*
* Called by managed code to resolve an instance field and store a non-wide value.
*/
THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCompiledCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER
/*
* Called by managed code to resolve an instance field and store a wide value.
*/
.extern artSet64InstanceFromCompiledCode
ENTRY art_quick_set64_instance
SETUP_SAVE_REFS_ONLY_FRAME r12 @ save callee saves in case of GC
@ r2:r3 contain the wide argument
str r9, [sp, #-16]! @ expand the frame and pass Thread::Current
.cfi_adjust_cfa_offset 16
bl artSet64InstanceFromCompiledCode @ (field_idx, Object*, new_val, Thread*)
add sp, #16 @ release out args
.cfi_adjust_cfa_offset -16
RESTORE_SAVE_REFS_ONLY_FRAME @ TODO: we can clearly save an add here
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_set64_instance
.extern artSet64StaticFromCompiledCode
ENTRY art_quick_set64_static
SETUP_SAVE_REFS_ONLY_FRAME r12 @ save callee saves in case of GC
@ r2:r3 contain the wide argument
str r9, [sp, #-16]! @ expand the frame and pass Thread::Current
.cfi_adjust_cfa_offset 16
bl artSet64StaticFromCompiledCode @ (field_idx, new_val, Thread*)
add sp, #16 @ release out args
.cfi_adjust_cfa_offset -16
RESTORE_SAVE_REFS_ONLY_FRAME @ TODO: we can clearly save an add here
RETURN_IF_RESULT_IS_ZERO
DELIVER_PENDING_EXCEPTION
END art_quick_set64_static
// Generate the allocation entrypoints for each allocator.
GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_TLAB_ALLOCATORS
// Comment out allocators that have arm specific asm.
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab, TLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab, TLAB)
// GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab, TLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab, TLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab, TLAB)
GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab, TLAB)
// A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_RESOLVED_OBJECT(_rosalloc, RosAlloc).
//
// If isInitialized=1 then the compiler assumes the object's class has already been initialized.
// If isInitialized=0 the compiler can only assume it's been at least resolved.
.macro ART_QUICK_ALLOC_OBJECT_ROSALLOC c_name, cxx_name, isInitialized
ENTRY \c_name
// Fast path rosalloc allocation.
// r0: type/return value, r9: Thread::Current
// r1, r2, r3, r12: free.
ldr r3, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local
// allocation stack has room.
// TODO: consider using ldrd.
ldr r12, [r9, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET]
cmp r3, r12
bhs .Lslow_path\c_name
ldr r3, [r0, #MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET] // Load the object size (r3)
cmp r3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread
// local allocation. Also does the
// initialized and finalizable checks.
// When isInitialized == 0, then the class is potentially not yet initialized.
// If the class is not yet initialized, the object size will be very large to force the branch
// below to be taken.
//
// See InitializeClassVisitors in class-inl.h for more details.
bhs .Lslow_path\c_name
// Compute the rosalloc bracket index
// from the size. Since the size is
// already aligned we can combine the
// two shifts together.
add r12, r9, r3, lsr #(ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT - POINTER_SIZE_SHIFT)
// Subtract pointer size since ther
// are no runs for 0 byte allocations
// and the size is already aligned.
// Load the rosalloc run (r12)
ldr r12, [r12, #(THREAD_ROSALLOC_RUNS_OFFSET - __SIZEOF_POINTER__)]
// Load the free list head (r3). This
// will be the return val.
ldr r3, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)]
cbz r3, .Lslow_path\c_name
// "Point of no slow path". Won't go to the slow path from here on. OK to clobber r0 and r1.
ldr r1, [r3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head
// and update the list head with the
// next pointer.
str r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)]
// Store the class pointer in the
// header. This also overwrites the
// next pointer. The offsets are
// asserted to match.
#if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET
#error "Class pointer needs to overwrite next pointer."
#endif
POISON_HEAP_REF r0
str r0, [r3, #MIRROR_OBJECT_CLASS_OFFSET]
// Push the new object onto the thread
// local allocation stack and
// increment the thread local
// allocation stack top.
ldr r1, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET]
str r3, [r1], #COMPRESSED_REFERENCE_SIZE // (Increment r1 as a side effect.)
str r1, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET]
// Decrement the size of the free list
// After this "STR" the object is published to the thread local allocation stack,
// and it will be observable from a runtime internal (eg. Heap::VisitObjects) point of view.
// It is not yet visible to the running (user) compiled code until after the return.
//
// To avoid the memory barrier prior to the "STR", a trick is employed, by differentiating
// the state of the allocation stack slot. It can be a pointer to one of:
// 0) Null entry, because the stack was bumped but the new pointer wasn't written yet.
// (The stack initial state is "null" pointers).
// 1) A partially valid object, with an invalid class pointer to the next free rosalloc slot.
// 2) A fully valid object, with a valid class pointer pointing to a real class.
// Other states are not allowed.
//
// An object that is invalid only temporarily, and will eventually become valid.
// The internal runtime code simply checks if the object is not null or is partial and then
// ignores it.
//
// (Note: The actual check is done by seeing if a non-null object has a class pointer pointing
// to ClassClass, and that the ClassClass's class pointer is self-cyclic. A rosalloc free slot
// "next" pointer is not-cyclic.)
//
// See also b/28790624 for a listing of CLs dealing with this race.
ldr r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)]
sub r1, #1
// TODO: consider combining this store
// and the list head store above using
// strd.
str r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)]
mov r0, r3 // Set the return value and return.
.if \isInitialized == 0
// This barrier is only necessary when the allocation also requires
// a class initialization check.
//
// If the class is already observably initialized, then new-instance allocations are protected
// from publishing by the compiler which inserts its own StoreStore barrier.
dmb ish
// Use a "dmb ish" fence here because if there are later loads of statics (e.g. class size),
// they should happen-after the implicit initialization check.
//
// TODO: Remove this dmb for class initialization checks (b/36692143) by introducing
// a new observably-initialized class state.
.endif
bx lr
.Lslow_path\c_name:
SETUP_SAVE_REFS_ONLY_FRAME r2 @ save callee saves in case of GC
mov r1, r9 @ pass Thread::Current
bl \cxx_name @ (mirror::Class* cls, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
END \c_name
.endm
ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_resolved_rosalloc, artAllocObjectFromCodeResolvedRosAlloc, /* isInitialized */ 0
ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_initialized_rosalloc, artAllocObjectFromCodeInitializedRosAlloc, /* isInitialized */ 1
// The common fast path code for art_quick_alloc_object_resolved/initialized_tlab
// and art_quick_alloc_object_resolved/initialized_region_tlab.
//
// r0: type r9: Thread::Current, r1, r2, r3, r12: free.
// Need to preserve r0 to the slow path.
//
// If isInitialized=1 then the compiler assumes the object's class has already been initialized.
// If isInitialized=0 the compiler can only assume it's been at least resolved.
.macro ALLOC_OBJECT_RESOLVED_TLAB_FAST_PATH slowPathLabel isInitialized
// Load thread_local_pos (r12) and
// thread_local_end (r3) with ldrd.
// Check constraints for ldrd.
#if !((THREAD_LOCAL_POS_OFFSET + 4 == THREAD_LOCAL_END_OFFSET) && (THREAD_LOCAL_POS_OFFSET % 8 == 0))
#error "Thread::thread_local_pos/end must be consecutive and are 8 byte aligned for performance"
#endif
ldrd r12, r3, [r9, #THREAD_LOCAL_POS_OFFSET]
sub r12, r3, r12 // Compute the remaining buf size.
ldr r3, [r0, #MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET] // Load the object size (r3).
cmp r3, r12 // Check if it fits.
// When isInitialized == 0, then the class is potentially not yet initialized.
// If the class is not yet initialized, the object size will be very large to force the branch
// below to be taken.
//
// See InitializeClassVisitors in class-inl.h for more details.
bhi \slowPathLabel
// "Point of no slow path". Won't go to the slow path from here on. OK to clobber r0 and r1.
// Reload old thread_local_pos (r0)
// for the return value.
ldr r2, [r9, #THREAD_LOCAL_POS_OFFSET]
add r1, r2, r3
str r1, [r9, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos.
// After this "STR" the object is published to the thread local allocation stack,
// and it will be observable from a runtime internal (eg. Heap::VisitObjects) point of view.
// It is not yet visible to the running (user) compiled code until after the return.
//
// To avoid the memory barrier prior to the "STR", a trick is employed, by differentiating
// the state of the object. It can be either:
// 1) A partially valid object, with a null class pointer
// (because the initial state of TLAB buffers is all 0s/nulls).
// 2) A fully valid object, with a valid class pointer pointing to a real class.
// Other states are not allowed.
//
// An object that is invalid only temporarily, and will eventually become valid.
// The internal runtime code simply checks if the object is not null or is partial and then
// ignores it.
//
// (Note: The actual check is done by checking that the object's class pointer is non-null.
// Also, unlike rosalloc, the object can never be observed as null).
ldr r1, [r9, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects.
add r1, r1, #1
str r1, [r9, #THREAD_LOCAL_OBJECTS_OFFSET]
POISON_HEAP_REF r0
str r0, [r2, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer.
// Fence. This is "ish" not "ishst" so
// that the code after this allocation
// site will see the right values in
// the fields of the class.
mov r0, r2
.if \isInitialized == 0
// This barrier is only necessary when the allocation also requires
// a class initialization check.
//
// If the class is already observably initialized, then new-instance allocations are protected
// from publishing by the compiler which inserts its own StoreStore barrier.
dmb ish
// Use a "dmb ish" fence here because if there are later loads of statics (e.g. class size),
// they should happen-after the implicit initialization check.
//
// TODO: Remove dmb for class initialization checks (b/36692143)
.endif
bx lr
.endm
// The common code for art_quick_alloc_object_*region_tlab
.macro GENERATE_ALLOC_OBJECT_RESOLVED_TLAB name, entrypoint, isInitialized
ENTRY \name
// Fast path tlab allocation.
// r0: type, r9: Thread::Current
// r1, r2, r3, r12: free.
ALLOC_OBJECT_RESOLVED_TLAB_FAST_PATH .Lslow_path\name, \isInitialized
.Lslow_path\name:
SETUP_SAVE_REFS_ONLY_FRAME r2 // Save callee saves in case of GC.
mov r1, r9 // Pass Thread::Current.
bl \entrypoint // (mirror::Class* klass, Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
END \name
.endm
GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB, /* isInitialized */ 0
GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB, /* isInitialized */ 1
GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_resolved_tlab, artAllocObjectFromCodeResolvedTLAB, /* isInitialized */ 0
GENERATE_ALLOC_OBJECT_RESOLVED_TLAB art_quick_alloc_object_initialized_tlab, artAllocObjectFromCodeInitializedTLAB, /* isInitialized */ 1
// The common fast path code for art_quick_alloc_array_resolved/initialized_tlab
// and art_quick_alloc_array_resolved/initialized_region_tlab.
//
// r0: type r1: component_count r2: total_size r9: Thread::Current, r3, r12: free.
// Need to preserve r0 and r1 to the slow path.
.macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE slowPathLabel
and r2, r2, #OBJECT_ALIGNMENT_MASK_TOGGLED // Apply alignment mask
// (addr + 7) & ~7.
// Load thread_local_pos (r3) and
// thread_local_end (r12) with ldrd.
// Check constraints for ldrd.
#if !((THREAD_LOCAL_POS_OFFSET + 4 == THREAD_LOCAL_END_OFFSET) && (THREAD_LOCAL_POS_OFFSET % 8 == 0))
#error "Thread::thread_local_pos/end must be consecutive and are 8 byte aligned for performance"
#endif
ldrd r3, r12, [r9, #THREAD_LOCAL_POS_OFFSET]
sub r12, r12, r3 // Compute the remaining buf size.
cmp r2, r12 // Check if the total_size fits.
// The array class is always initialized here. Unlike new-instance,
// this does not act as a double test.
bhi \slowPathLabel
// "Point of no slow path". Won't go to the slow path from here on. OK to clobber r0 and r1.
add r2, r2, r3
str r2, [r9, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos.
ldr r2, [r9, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects.
add r2, r2, #1
str r2, [r9, #THREAD_LOCAL_OBJECTS_OFFSET]
POISON_HEAP_REF r0
str r0, [r3, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer.
str r1, [r3, #MIRROR_ARRAY_LENGTH_OFFSET] // Store the array length.
// Fence. This is "ish" not "ishst" so
// that the code after this allocation
// site will see the right values in
// the fields of the class.
mov r0, r3
// new-array is special. The class is loaded and immediately goes to the Initialized state
// before it is published. Therefore the only fence needed is for the publication of the object.
// See ClassLinker::CreateArrayClass() for more details.
// For publication of the new array, we don't need a 'dmb ishst' here.
// The compiler generates 'dmb ishst' for all new-array insts.
bx lr
.endm
.macro GENERATE_ALLOC_ARRAY_TLAB name, entrypoint, size_setup
ENTRY \name
// Fast path array allocation for region tlab allocation.
// r0: mirror::Class* type
// r1: int32_t component_count
// r9: thread
// r2, r3, r12: free.
\size_setup .Lslow_path\name
ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE .Lslow_path\name
.Lslow_path\name:
// r0: mirror::Class* klass
// r1: int32_t component_count
// r2: Thread* self
SETUP_SAVE_REFS_ONLY_FRAME r2 // save callee saves in case of GC
mov r2, r9 // pass Thread::Current
bl \entrypoint
RESTORE_SAVE_REFS_ONLY_FRAME
RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER
END \name
.endm
.macro COMPUTE_ARRAY_SIZE_UNKNOWN slow_path
bkpt // We should never enter here.
// Code below is for reference.
// Possibly a large object, go slow.
// Also does negative array size check.
movw r2, #((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_WIDE_ARRAY_DATA_OFFSET) / 8)
cmp r1, r2
bhi \slow_path
// Array classes are never finalizable
// or uninitialized, no need to check.
ldr r3, [r0, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Load component type
UNPOISON_HEAP_REF r3
ldr r3, [r3, #MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET]
lsr r3, r3, #PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT // Component size shift is in high 16
// bits.
lsl r2, r1, r3 // Calculate data size
// Add array data offset and alignment.
add r2, r2, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
#if MIRROR_WIDE_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4
#error Long array data offset must be 4 greater than int array data offset.
#endif
add r3, r3, #1 // Add 4 to the length only if the
// component size shift is 3
// (for 64 bit alignment).
and r3, r3, #4
add r2, r2, r3
.endm
.macro COMPUTE_ARRAY_SIZE_8 slow_path
// Possibly a large object, go slow.
// Also does negative array size check.
movw r2, #(MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET)
cmp r1, r2
bhi \slow_path
// Add array data offset and alignment.
add r2, r1, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
.endm
.macro COMPUTE_ARRAY_SIZE_16 slow_path
// Possibly a large object, go slow.
// Also does negative array size check.
movw r2, #((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET) / 2)
cmp r1, r2
bhi \slow_path
lsl r2, r1, #1
// Add array data offset and alignment.
add r2, r2, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
.endm
.macro COMPUTE_ARRAY_SIZE_32 slow_path
// Possibly a large object, go slow.
// Also does negative array size check.
movw r2, #((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET) / 4)
cmp r1, r2
bhi \slow_path
lsl r2, r1, #2
// Add array data offset and alignment.
add r2, r2, #(MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
.endm
.macro COMPUTE_ARRAY_SIZE_64 slow_path
// Possibly a large object, go slow.
// Also does negative array size check.
movw r2, #((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_LONG_ARRAY_DATA_OFFSET) / 8)
cmp r1, r2
bhi \slow_path
lsl r2, r1, #3
// Add array data offset and alignment.
add r2, r2, #(MIRROR_WIDE_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK)
.endm
# TODO(ngeoffray): art_quick_alloc_array_resolved_region_tlab is not used for arm, remove
# the entrypoint once all backends have been updated to use the size variants.
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_8
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_16
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_32
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_64
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_8
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_16
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_32
GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_64
/*
* Called by managed code when the value in rSUSPEND has been decremented to 0.
*/
.extern artTestSuspendFromCode
ENTRY art_quick_test_suspend
#ifdef ARM_R4_SUSPEND_FLAG
ldrh rSUSPEND, [rSELF, #THREAD_FLAGS_OFFSET]
cbnz rSUSPEND, 1f @ check Thread::Current()->suspend_count_ == 0
mov rSUSPEND, #SUSPEND_CHECK_INTERVAL @ reset rSUSPEND to SUSPEND_CHECK_INTERVAL
bx lr @ return if suspend_count_ == 0
1:
mov rSUSPEND, #SUSPEND_CHECK_INTERVAL @ reset rSUSPEND to SUSPEND_CHECK_INTERVAL
#endif
SETUP_SAVE_EVERYTHING_FRAME r0 @ save everything for GC stack crawl
mov r0, rSELF
bl artTestSuspendFromCode @ (Thread*)
RESTORE_SAVE_EVERYTHING_FRAME
bx lr
END art_quick_test_suspend
ENTRY art_quick_implicit_suspend
mov r0, rSELF
SETUP_SAVE_REFS_ONLY_FRAME r1 @ save callee saves for stack crawl
bl artTestSuspendFromCode @ (Thread*)
RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN
END art_quick_implicit_suspend
/*
* Called by managed code that is attempting to call a method on a proxy class. On entry
* r0 holds the proxy method and r1 holds the receiver; r2 and r3 may contain arguments. The
* frame size of the invoked proxy method agrees with a ref and args callee save frame.
*/
.extern artQuickProxyInvokeHandler
ENTRY art_quick_proxy_invoke_handler
SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_R0
mov r2, r9 @ pass Thread::Current
mov r3, sp @ pass SP
blx artQuickProxyInvokeHandler @ (Method* proxy method, receiver, Thread*, SP)
ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_
// Tear down the callee-save frame. Skip arg registers.
add sp, #(FRAME_SIZE_SAVE_REFS_AND_ARGS - FRAME_SIZE_SAVE_REFS_ONLY)
.cfi_adjust_cfa_offset -(FRAME_SIZE_SAVE_REFS_AND_ARGS - FRAME_SIZE_SAVE_REFS_ONLY)
RESTORE_SAVE_REFS_ONLY_FRAME
cbnz r2, 1f @ success if no exception is pending
vmov d0, r0, r1 @ store into fpr, for when it's a fpr return...
bx lr @ return on success
1:
DELIVER_PENDING_EXCEPTION
END art_quick_proxy_invoke_handler
/*
* Called to resolve an imt conflict.
* r0 is the conflict ArtMethod.
* r12 is a hidden argument that holds the target interface method's dex method index.
*
* Note that this stub writes to r0, r4, and r12.
*/
ENTRY art_quick_imt_conflict_trampoline
ldr r4, [sp, #0] // Load referrer
ldr r4, [r4, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_32] // Load dex cache methods array
ldr r12, [r4, r12, lsl #POINTER_SIZE_SHIFT] // Load interface method
ldr r0, [r0, #ART_METHOD_JNI_OFFSET_32] // Load ImtConflictTable
ldr r4, [r0] // Load first entry in ImtConflictTable.
.Limt_table_iterate:
cmp r4, r12
// Branch if found. Benchmarks have shown doing a branch here is better.
beq .Limt_table_found
// If the entry is null, the interface method is not in the ImtConflictTable.
cbz r4, .Lconflict_trampoline
// Iterate over the entries of the ImtConflictTable.
ldr r4, [r0, #(2 * __SIZEOF_POINTER__)]!
b .Limt_table_iterate
.Limt_table_found:
// We successfully hit an entry in the table. Load the target method
// and jump to it.
ldr r0, [r0, #__SIZEOF_POINTER__]
ldr pc, [r0, #ART_METHOD_QUICK_CODE_OFFSET_32]
.Lconflict_trampoline:
// Call the runtime stub to populate the ImtConflictTable and jump to the
// resolved method.
mov r0, r12 // Load interface method
INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline
END art_quick_imt_conflict_trampoline
.extern artQuickResolutionTrampoline
ENTRY art_quick_resolution_trampoline
SETUP_SAVE_REFS_AND_ARGS_FRAME r2
mov r2, r9 @ pass Thread::Current
mov r3, sp @ pass SP
blx artQuickResolutionTrampoline @ (Method* called, receiver, Thread*, SP)
cbz r0, 1f @ is code pointer null? goto exception
mov r12, r0
ldr r0, [sp, #0] @ load resolved method in r0
RESTORE_SAVE_REFS_AND_ARGS_FRAME
bx r12 @ tail-call into actual code
1:
RESTORE_SAVE_REFS_AND_ARGS_FRAME
DELIVER_PENDING_EXCEPTION
END art_quick_resolution_trampoline
/*
* Called to do a generic JNI down-call
*/
ENTRY art_quick_generic_jni_trampoline
SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_R0
// Save rSELF
mov r11, rSELF
// Save SP , so we can have static CFI info. r10 is saved in ref_and_args.
mov r10, sp
.cfi_def_cfa_register r10
sub sp, sp, #5120
// prepare for artQuickGenericJniTrampoline call
// (Thread*, SP)
// r0 r1 <= C calling convention
// rSELF r10 <= where they are
mov r0, rSELF // Thread*
mov r1, r10
blx artQuickGenericJniTrampoline // (Thread*, sp)
// The C call will have registered the complete save-frame on success.
// The result of the call is:
// r0: pointer to native code, 0 on error.
// r1: pointer to the bottom of the used area of the alloca, can restore stack till there.
// Check for error = 0.
cbz r0, .Lexception_in_native
// Release part of the alloca.
mov sp, r1
// Save the code pointer
mov r12, r0
// Load parameters from frame into registers.
pop {r0-r3}
// Softfloat.
// TODO: Change to hardfloat when supported.
blx r12 // native call.
// result sign extension is handled in C code
// prepare for artQuickGenericJniEndTrampoline call
// (Thread*, result, result_f)
// r0 r2,r3 stack <= C calling convention
// r11 r0,r1 r0,r1 <= where they are
sub sp, sp, #8 // Stack alignment.
push {r0-r1}
mov r3, r1
mov r2, r0
mov r0, r11
blx artQuickGenericJniEndTrampoline
// Restore self pointer.
mov r9, r11
// Pending exceptions possible.
ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_
cbnz r2, .Lexception_in_native
// Tear down the alloca.
mov sp, r10
.cfi_def_cfa_register sp
// Tear down the callee-save frame. Skip arg registers.
add sp, #FRAME_SIZE_SAVE_REFS_AND_ARGS-FRAME_SIZE_SAVE_REFS_ONLY
.cfi_adjust_cfa_offset -(FRAME_SIZE_SAVE_REFS_AND_ARGS-FRAME_SIZE_SAVE_REFS_ONLY)
RESTORE_SAVE_REFS_ONLY_FRAME
// store into fpr, for when it's a fpr return...
vmov d0, r0, r1
bx lr // ret
// Undo the unwinding information from above since it doesn't apply below.
.cfi_def_cfa_register r10
.cfi_adjust_cfa_offset FRAME_SIZE_SAVE_REFS_AND_ARGS-FRAME_SIZE_SAVE_REFS_ONLY
.Lexception_in_native:
ldr sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET]
.cfi_def_cfa_register sp
# This will create a new save-all frame, required by the runtime.
DELIVER_PENDING_EXCEPTION
END art_quick_generic_jni_trampoline
.extern artQuickToInterpreterBridge
ENTRY art_quick_to_interpreter_bridge
SETUP_SAVE_REFS_AND_ARGS_FRAME r1
mov r1, r9 @ pass Thread::Current
mov r2, sp @ pass SP
blx artQuickToInterpreterBridge @ (Method* method, Thread*, SP)
ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_
// Tear down the callee-save frame. Skip arg registers.
add sp, #(FRAME_SIZE_SAVE_REFS_AND_ARGS - FRAME_SIZE_SAVE_REFS_ONLY)
.cfi_adjust_cfa_offset -(FRAME_SIZE_SAVE_REFS_AND_ARGS - FRAME_SIZE_SAVE_REFS_ONLY)
RESTORE_SAVE_REFS_ONLY_FRAME
cbnz r2, 1f @ success if no exception is pending
vmov d0, r0, r1 @ store into fpr, for when it's a fpr return...
bx lr @ return on success
1:
DELIVER_PENDING_EXCEPTION
END art_quick_to_interpreter_bridge
/*
* Called to attempt to execute an obsolete method.
*/
ONE_ARG_RUNTIME_EXCEPTION art_invoke_obsolete_method_stub, artInvokeObsoleteMethod
/*
* Routine that intercepts method calls and returns.
*/
.extern artInstrumentationMethodEntryFromCode
.extern artInstrumentationMethodExitFromCode
ENTRY art_quick_instrumentation_entry
@ Make stack crawlable and clobber r2 and r3 (post saving)
SETUP_SAVE_REFS_AND_ARGS_FRAME r2
@ preserve r0 (not normally an arg) knowing there is a spare slot in kSaveRefsAndArgs.
str r0, [sp, #4]
mov r2, r9 @ pass Thread::Current
mov r3, sp @ pass SP
blx artInstrumentationMethodEntryFromCode @ (Method*, Object*, Thread*, SP)
cbz r0, .Ldeliver_instrumentation_entry_exception
@ Deliver exception if we got nullptr as function.
mov r12, r0 @ r12 holds reference to code
ldr r0, [sp, #4] @ restore r0
RESTORE_SAVE_REFS_AND_ARGS_FRAME
blx r12 @ call method with lr set to art_quick_instrumentation_exit
@ Deliberate fall-through into art_quick_instrumentation_exit.
.type art_quick_instrumentation_exit, #function
.global art_quick_instrumentation_exit
art_quick_instrumentation_exit:
mov lr, #0 @ link register is to here, so clobber with 0 for later checks
SETUP_SAVE_REFS_ONLY_FRAME r2 @ set up frame knowing r2 and r3 must be dead on exit
mov r12, sp @ remember bottom of caller's frame
push {r0-r1} @ save return value
.cfi_adjust_cfa_offset 8
.cfi_rel_offset r0, 0
.cfi_rel_offset r1, 4
mov r2, sp @ store gpr_res pointer.
vpush {d0} @ save fp return value
.cfi_adjust_cfa_offset 8
mov r3, sp @ store fpr_res pointer
mov r1, r12 @ pass SP
mov r0, r9 @ pass Thread::Current
blx artInstrumentationMethodExitFromCode @ (Thread*, SP, gpr_res*, fpr_res*)
mov r2, r0 @ link register saved by instrumentation
mov lr, r1 @ r1 is holding link register if we're to bounce to deoptimize
vpop {d0} @ restore fp return value
.cfi_adjust_cfa_offset -8
pop {r0, r1} @ restore return value
.cfi_adjust_cfa_offset -8
.cfi_restore r0
.cfi_restore r1
RESTORE_SAVE_REFS_ONLY_FRAME
cbz r2, .Ldo_deliver_instrumentation_exception
@ Deliver exception if we got nullptr as function.
bx r2 @ Otherwise, return
.Ldeliver_instrumentation_entry_exception:
@ Deliver exception for art_quick_instrumentation_entry placed after
@ art_quick_instrumentation_exit so that the fallthrough works.
RESTORE_SAVE_REFS_AND_ARGS_FRAME
.Ldo_deliver_instrumentation_exception:
DELIVER_PENDING_EXCEPTION
END art_quick_instrumentation_entry
/*
* Instrumentation has requested that we deoptimize into the interpreter. The deoptimization
* will long jump to the upcall with a special exception of -1.
*/
.extern artDeoptimize
ENTRY art_quick_deoptimize
SETUP_SAVE_ALL_CALLEE_SAVES_FRAME r0
mov r0, r9 @ pass Thread::Current
blx artDeoptimize @ (Thread*)
END art_quick_deoptimize
/*
* Compiled code has requested that we deoptimize into the interpreter. The deoptimization
* will long jump to the interpreter bridge.
*/
.extern artDeoptimizeFromCompiledCode
ENTRY art_quick_deoptimize_from_compiled_code
SETUP_SAVE_EVERYTHING_FRAME r1
mov r1, r9 @ pass Thread::Current
blx artDeoptimizeFromCompiledCode @ (DeoptimizationKind, Thread*)
END art_quick_deoptimize_from_compiled_code
/*
* Signed 64-bit integer multiply.
*
* Consider WXxYZ (r1r0 x r3r2) with a long multiply:
* WX
* x YZ
* --------
* ZW ZX
* YW YX
*
* The low word of the result holds ZX, the high word holds
* (ZW+YX) + (the high overflow from ZX). YW doesn't matter because
* it doesn't fit in the low 64 bits.
*
* Unlike most ARM math operations, multiply instructions have
* restrictions on using the same register more than once (Rd and Rm
* cannot be the same).
*/
/* mul-long vAA, vBB, vCC */
ENTRY art_quick_mul_long
push {r9 - r10}
.cfi_adjust_cfa_offset 8
.cfi_rel_offset r9, 0
.cfi_rel_offset r10, 4
mul ip, r2, r1 @ ip<- ZxW
umull r9, r10, r2, r0 @ r9/r10 <- ZxX
mla r2, r0, r3, ip @ r2<- YxX + (ZxW)
add r10, r2, r10 @ r10<- r10 + low(ZxW + (YxX))
mov r0,r9
mov r1,r10
pop {r9 - r10}
.cfi_adjust_cfa_offset -8
.cfi_restore r9
.cfi_restore r10
bx lr
END art_quick_mul_long
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to ignore all but the low
* 6 bits.
* On entry:
* r0: low word
* r1: high word
* r2: shift count
*/
/* shl-long vAA, vBB, vCC */
ARM_ENTRY art_quick_shl_long @ ARM code as thumb code requires spills
and r2, r2, #63 @ r2<- r2 & 0x3f
mov r1, r1, asl r2 @ r1<- r1 << r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r1, r1, r0, lsr r3 @ r1<- r1 | (r0 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r1, r0, asl ip @ if r2 >= 32, r1<- r0 << (r2-32)
mov r0, r0, asl r2 @ r0<- r0 << r2
bx lr
END art_quick_shl_long
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to ignore all but the low
* 6 bits.
* On entry:
* r0: low word
* r1: high word
* r2: shift count
*/
/* shr-long vAA, vBB, vCC */
ARM_ENTRY art_quick_shr_long @ ARM code as thumb code requires spills
and r2, r2, #63 @ r0<- r0 & 0x3f
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r0, r1, asr ip @ if r2 >= 32, r0<-r1 >> (r2-32)
mov r1, r1, asr r2 @ r1<- r1 >> r2
bx lr
END art_quick_shr_long
/*
* Long integer shift. This is different from the generic 32/64-bit
* binary operations because vAA/vBB are 64-bit but vCC (the shift
* distance) is 32-bit. Also, Dalvik requires us to ignore all but the low
* 6 bits.
* On entry:
* r0: low word
* r1: high word
* r2: shift count
*/
/* ushr-long vAA, vBB, vCC */
ARM_ENTRY art_quick_ushr_long @ ARM code as thumb code requires spills
and r2, r2, #63 @ r0<- r0 & 0x3f
mov r0, r0, lsr r2 @ r0<- r2 >> r2
rsb r3, r2, #32 @ r3<- 32 - r2
orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2))
subs ip, r2, #32 @ ip<- r2 - 32
movpl r0, r1, lsr ip @ if r2 >= 32, r0<-r1 >>> (r2-32)
mov r1, r1, lsr r2 @ r1<- r1 >>> r2
bx lr
END art_quick_ushr_long
/*
* String's indexOf.
*
* On entry:
* r0: string object (known non-null)
* r1: char to match (known <= 0xFFFF)
* r2: Starting offset in string data
*/
ENTRY art_quick_indexof
push {r4, r10-r11, lr} @ 4 words of callee saves
.cfi_adjust_cfa_offset 16
.cfi_rel_offset r4, 0
.cfi_rel_offset r10, 4
.cfi_rel_offset r11, 8
.cfi_rel_offset lr, 12
#if (STRING_COMPRESSION_FEATURE)
ldr r4, [r0, #MIRROR_STRING_COUNT_OFFSET]
#else
ldr r3, [r0, #MIRROR_STRING_COUNT_OFFSET]
#endif
add r0, #MIRROR_STRING_VALUE_OFFSET
#if (STRING_COMPRESSION_FEATURE)
/* r4 count (with flag) and r3 holds actual length */
lsr r3, r4, #1
#endif
/* Clamp start to [0..count] */
cmp r2, #0
it lt
movlt r2, #0
cmp r2, r3
it gt
movgt r2, r3
/* Save a copy in r12 to later compute result */
mov r12, r0
/* Build pointer to start of data to compare and pre-bias */
#if (STRING_COMPRESSION_FEATURE)
lsrs r4, r4, #1
bcc .Lstring_indexof_compressed
#endif
add r0, r0, r2, lsl #1
sub r0, #2
/* Compute iteration count */
sub r2, r3, r2
/*
* At this point we have:
* r0: start of data to test
* r1: char to compare
* r2: iteration count
* r4: compression style (used temporarily)
* r12: original start of string data
* r3, r4, r10, r11 available for loading string data
*/
subs r2, #4
blt .Lindexof_remainder
.Lindexof_loop4:
ldrh r3, [r0, #2]!
ldrh r4, [r0, #2]!
ldrh r10, [r0, #2]!
ldrh r11, [r0, #2]!
cmp r3, r1
beq .Lmatch_0
cmp r4, r1
beq .Lmatch_1
cmp r10, r1
beq .Lmatch_2
cmp r11, r1
beq .Lmatch_3
subs r2, #4
bge .Lindexof_loop4
.Lindexof_remainder:
adds r2, #4
beq .Lindexof_nomatch
.Lindexof_loop1:
ldrh r3, [r0, #2]!
cmp r3, r1
beq .Lmatch_3
subs r2, #1
bne .Lindexof_loop1
.Lindexof_nomatch:
mov r0, #-1
pop {r4, r10-r11, pc}
.Lmatch_0:
sub r0, #6
sub r0, r12
asr r0, r0, #1
pop {r4, r10-r11, pc}
.Lmatch_1:
sub r0, #4
sub r0, r12
asr r0, r0, #1
pop {r4, r10-r11, pc}
.Lmatch_2:
sub r0, #2
sub r0, r12
asr r0, r0, #1
pop {r4, r10-r11, pc}
.Lmatch_3:
sub r0, r12
asr r0, r0, #1
pop {r4, r10-r11, pc}
#if (STRING_COMPRESSION_FEATURE)
.Lstring_indexof_compressed:
add r0, r0, r2
sub r0, #1
sub r2, r3, r2
.Lstring_indexof_compressed_loop:
subs r2, #1
blt .Lindexof_nomatch
ldrb r3, [r0, #1]!
cmp r3, r1
beq .Lstring_indexof_compressed_matched
b .Lstring_indexof_compressed_loop
.Lstring_indexof_compressed_matched:
sub r0, r12
pop {r4, r10-r11, pc}
#endif
END art_quick_indexof
/* Assembly routines used to handle ABI differences. */
/* double fmod(double a, double b) */
.extern fmod
ENTRY art_quick_fmod
push {lr}
.cfi_adjust_cfa_offset 4
.cfi_rel_offset lr, 0
sub sp, #4
.cfi_adjust_cfa_offset 4
vmov r0, r1, d0
vmov r2, r3, d1
bl fmod
vmov d0, r0, r1
add sp, #4
.cfi_adjust_cfa_offset -4
pop {pc}
END art_quick_fmod
/* float fmodf(float a, float b) */
.extern fmodf
ENTRY art_quick_fmodf
push {lr}
.cfi_adjust_cfa_offset 4
.cfi_rel_offset lr, 0
sub sp, #4
.cfi_adjust_cfa_offset 4
vmov r0, r1, d0
bl fmodf
vmov s0, r0
add sp, #4
.cfi_adjust_cfa_offset -4
pop {pc}
END art_quick_fmodf
/* int64_t art_d2l(double d) */
.extern art_d2l
ENTRY art_quick_d2l
vmov r0, r1, d0
b art_d2l
END art_quick_d2l
/* int64_t art_f2l(float f) */
.extern art_f2l
ENTRY art_quick_f2l
vmov r0, s0
b art_f2l
END art_quick_f2l
/* float art_l2f(int64_t l) */
.extern art_l2f
ENTRY art_quick_l2f
push {lr}
.cfi_adjust_cfa_offset 4
.cfi_rel_offset lr, 0
sub sp, #4
.cfi_adjust_cfa_offset 4
bl art_l2f
vmov s0, r0
add sp, #4
.cfi_adjust_cfa_offset -4
pop {pc}
END art_quick_l2f
.macro CONDITIONAL_CBZ reg, reg_if, dest
.ifc \reg, \reg_if
cbz \reg, \dest
.endif
.endm
.macro CONDITIONAL_CMPBZ reg, reg_if, dest
.ifc \reg, \reg_if
cmp \reg, #0
beq \dest
.endif
.endm
// Use CBZ if the register is in {r0, r7} otherwise compare and branch.
.macro SMART_CBZ reg, dest
CONDITIONAL_CBZ \reg, r0, \dest
CONDITIONAL_CBZ \reg, r1, \dest
CONDITIONAL_CBZ \reg, r2, \dest
CONDITIONAL_CBZ \reg, r3, \dest
CONDITIONAL_CBZ \reg, r4, \dest
CONDITIONAL_CBZ \reg, r5, \dest
CONDITIONAL_CBZ \reg, r6, \dest
CONDITIONAL_CBZ \reg, r7, \dest
CONDITIONAL_CMPBZ \reg, r8, \dest
CONDITIONAL_CMPBZ \reg, r9, \dest
CONDITIONAL_CMPBZ \reg, r10, \dest
CONDITIONAL_CMPBZ \reg, r11, \dest
CONDITIONAL_CMPBZ \reg, r12, \dest
CONDITIONAL_CMPBZ \reg, r13, \dest
CONDITIONAL_CMPBZ \reg, r14, \dest
CONDITIONAL_CMPBZ \reg, r15, \dest
.endm
/*
* Create a function `name` calling the ReadBarrier::Mark routine,
* getting its argument and returning its result through register
* `reg`, saving and restoring all caller-save registers.
*
* IP is clobbered; `reg` must not be IP.
*
* If `reg` is different from `r0`, the generated function follows a
* non-standard runtime calling convention:
* - register `reg` is used to pass the (sole) argument of this
* function (instead of R0);
* - register `reg` is used to return the result of this function
* (instead of R0);
* - R0 is treated like a normal (non-argument) caller-save register;
* - everything else is the same as in the standard runtime calling
* convention (e.g. standard callee-save registers are preserved).
*/
.macro READ_BARRIER_MARK_REG name, reg
ENTRY \name
// Null check so that we can load the lock word.
SMART_CBZ \reg, .Lret_rb_\name
// Check lock word for mark bit, if marked return. Use IP for scratch since it is blocked.
ldr ip, [\reg, MIRROR_OBJECT_LOCK_WORD_OFFSET]
tst ip, #LOCK_WORD_MARK_BIT_MASK_SHIFTED
beq .Lnot_marked_rb_\name
// Already marked, return right away.
.Lret_rb_\name:
bx lr
.Lnot_marked_rb_\name:
// Test that both the forwarding state bits are 1.
#if (LOCK_WORD_STATE_SHIFT != 30) || (LOCK_WORD_STATE_FORWARDING_ADDRESS != 3)
// To use "CMP ip, #modified-immediate; BHS", we need the lock word state in
// the highest bits and the "forwarding address" state to have all bits set.
#error "Unexpected lock word state shift or forwarding address state value."
#endif
cmp ip, #(LOCK_WORD_STATE_FORWARDING_ADDRESS << LOCK_WORD_STATE_SHIFT)
bhs .Lret_forwarding_address\name
.Lslow_rb_\name:
// Save IP: The kSaveEverything entrypoint art_quick_resolve_string used to
// make a tail call here. Currently, it serves only for stack alignment but
// we may reintroduce kSaveEverything calls here in the future.
push {r0-r4, r9, ip, lr} @ save return address, core caller-save registers and ip
.cfi_adjust_cfa_offset 32
.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 r9, 20
.cfi_rel_offset ip, 24
.cfi_rel_offset lr, 28
.ifnc \reg, r0
mov r0, \reg @ pass arg1 - obj from `reg`
.endif
vpush {s0-s15} @ save floating-point caller-save registers
.cfi_adjust_cfa_offset 64
bl artReadBarrierMark @ r0 <- artReadBarrierMark(obj)
vpop {s0-s15} @ restore floating-point registers
.cfi_adjust_cfa_offset -64
.ifc \reg, r0 @ Save result to the stack slot or destination register.
str r0, [sp, #0]
.else
.ifc \reg, r1
str r0, [sp, #4]
.else
.ifc \reg, r2
str r0, [sp, #8]
.else
.ifc \reg, r3
str r0, [sp, #12]
.else
.ifc \reg, r4
str r0, [sp, #16]
.else
.ifc \reg, r9
str r0, [sp, #20]
.else
mov \reg, r0
.endif
.endif
.endif
.endif
.endif
.endif
pop {r0-r4, r9, ip, lr} @ restore caller-save registers
.cfi_adjust_cfa_offset -32
.cfi_restore r0
.cfi_restore r1
.cfi_restore r2
.cfi_restore r3
.cfi_restore r4
.cfi_restore r9
.cfi_restore ip
.cfi_restore lr
bx lr
.Lret_forwarding_address\name:
// Shift left by the forwarding address shift. This clears out the state bits since they are
// in the top 2 bits of the lock word.
lsl \reg, ip, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT
bx lr
END \name
.endm
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg00, r0
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, r1
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, r2
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, r3
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, r4
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, r5
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, r6
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, r7
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, r8
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, r9
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, r10
READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, r11
// Helper macros for Baker CC read barrier mark introspection (BRBMI).
.macro BRBMI_FOR_12_REGISTERS macro_for_register, macro_for_reserved_register
\macro_for_register r0
\macro_for_register r1
\macro_for_register r2
\macro_for_register r3
\macro_for_reserved_register // R4 is reserved for the entrypoint address.
\macro_for_register r5
\macro_for_register r6
\macro_for_register r7
\macro_for_register r8
\macro_for_register r9
\macro_for_register r10
\macro_for_register r11
.endm
.macro BRBMI_FOR_REGISTERS macro_for_register, macro_for_reserved_register
BRBMI_FOR_12_REGISTERS \macro_for_register, \macro_for_reserved_register
\macro_for_reserved_register // IP is reserved.
\macro_for_reserved_register // SP is reserved.
\macro_for_reserved_register // LR is reserved.
\macro_for_reserved_register // PC is reserved.
.endm
.macro BRBMI_RETURN_SWITCH_CASE reg
.Lmark_introspection_return_switch_case_\reg:
mov \reg, ip
bx lr
.endm
.macro BRBMI_BAD_RETURN_SWITCH_CASE
.Lmark_introspection_return_switch_case_bad:
BRBMI_BKPT_FILL_4B
.endm
.macro BRBMI_RETURN_SWITCH_CASE_OFFSET reg
.byte (.Lmark_introspection_return_switch_case_\reg - .Lmark_introspection_return_table) / 2
.endm
.macro BRBMI_BAD_RETURN_SWITCH_CASE_OFFSET
.byte (.Lmark_introspection_return_switch_case_bad - .Lmark_introspection_return_table) / 2
.endm
#if BAKER_MARK_INTROSPECTION_FIELD_LDR_WIDE_OFFSET != BAKER_MARK_INTROSPECTION_ARRAY_LDR_OFFSET
#error "Array and field introspection code sharing requires same LDR offset."
#endif
.macro BRBMI_ARRAY_LOAD index_reg
ldr ip, [ip, \index_reg, lsl #2] // 4 bytes.
b art_quick_read_barrier_mark_introspection // Should be 2 bytes, encoding T2.
.balign 8 // Add padding to 8 bytes.
.endm
.macro BRBMI_BKPT_FILL_4B
bkpt 0
bkpt 0
.endm
.macro BRBMI_BKPT_FILL_8B
BRBMI_BKPT_FILL_4B
BRBMI_BKPT_FILL_4B
.endm
.macro BRBMI_RUNTIME_CALL
// Note: This macro generates exactly 22 bytes of code. The core register
// PUSH and the MOVs are 16-bit instructions, the rest is 32-bit instructions.
push {r0-r3, r7, lr} // Save return address and caller-save registers.
.cfi_adjust_cfa_offset 24
.cfi_rel_offset r0, 0
.cfi_rel_offset r1, 4
.cfi_rel_offset r2, 8
.cfi_rel_offset r3, 12
.cfi_rel_offset r7, 16
.cfi_rel_offset lr, 20
mov r0, ip // Pass the reference.
vpush {s0-s15} // save floating-point caller-save registers
.cfi_adjust_cfa_offset 64
bl artReadBarrierMark // r0 <- artReadBarrierMark(obj)
vpop {s0-s15} // restore floating-point registers
.cfi_adjust_cfa_offset -64
mov ip, r0 // Move reference to ip in preparation for return switch.
pop {r0-r3, r7, lr} // Restore registers.
.cfi_adjust_cfa_offset -24
.cfi_restore r0
.cfi_restore r1
.cfi_restore r2
.cfi_restore r3
.cfi_restore r7
.cfi_restore lr
.endm
.macro BRBMI_CHECK_NULL_AND_MARKED label_suffix
// If reference is null, just return it in the right register.
cmp ip, #0
beq .Lmark_introspection_return\label_suffix
// Use R4 as temp and check the mark bit of the reference.
ldr r4, [ip, #MIRROR_OBJECT_LOCK_WORD_OFFSET]
tst r4, #LOCK_WORD_MARK_BIT_MASK_SHIFTED
beq .Lmark_introspection_unmarked\label_suffix
.Lmark_introspection_return\label_suffix:
.endm
.macro BRBMI_UNMARKED_FORWARDING_ADDRESS_CHECK label_suffix
.Lmark_introspection_unmarked\label_suffix:
// Check if the top two bits are one, if this is the case it is a forwarding address.
#if (LOCK_WORD_STATE_SHIFT != 30) || (LOCK_WORD_STATE_FORWARDING_ADDRESS != 3)
// To use "CMP ip, #modified-immediate; BHS", we need the lock word state in
// the highest bits and the "forwarding address" state to have all bits set.
#error "Unexpected lock word state shift or forwarding address state value."
#endif
cmp r4, #(LOCK_WORD_STATE_FORWARDING_ADDRESS << LOCK_WORD_STATE_SHIFT)
bhs .Lmark_introspection_forwarding_address\label_suffix
.endm
.macro BRBMI_EXTRACT_FORWARDING_ADDRESS label_suffix
.Lmark_introspection_forwarding_address\label_suffix:
// Note: This macro generates exactly 22 bytes of code, the branch is near.
// Shift left by the forwarding address shift. This clears out the state bits since they are
// in the top 2 bits of the lock word.
lsl ip, r4, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT
b .Lmark_introspection_return\label_suffix
.endm
.macro BRBMI_LOAD_RETURN_REG_FROM_CODE_wide ldr_offset
// Load the half of the instruction that contains Rt. Adjust for the thumb state in LR.
ldrh r4, [lr, #(-1 + \ldr_offset + 2)]
.endm
.macro BRBMI_LOAD_RETURN_REG_FROM_CODE_narrow ldr_offset
// Load the 16-bit instruction. Adjust for the thumb state in LR.
ldrh r4, [lr, #(-1 + \ldr_offset)]
.endm
.macro BRBMI_GC_ROOT_AND_FIELD_SLOW_PATH gc_root_ldr_offset, label_suffix
.balign 64
.thumb_func
.type art_quick_read_barrier_mark_introspection_gc_roots\label_suffix, #function
.hidden art_quick_read_barrier_mark_introspection_gc_roots\label_suffix
.global art_quick_read_barrier_mark_introspection_gc_roots\label_suffix
art_quick_read_barrier_mark_introspection_gc_roots\label_suffix:
BRBMI_RUNTIME_CALL
// Load the LDR (or the half of it) that contains Rt.
BRBMI_LOAD_RETURN_REG_FROM_CODE\label_suffix \gc_root_ldr_offset
b .Lmark_introspection_extract_register_and_return\label_suffix
// We've used 28 bytes since the "gc_roots" entrypoint (22 bytes for
// BRBMI_RUNTIME_CALL, 4 bytes for LDRH and 2 bytes for the branch). Squeeze
// the 6 byte forwarding address extraction here across the 32-byte boundary.
BRBMI_EXTRACT_FORWARDING_ADDRESS \label_suffix
// And the slow path taking exactly 30 bytes (6 bytes for the forwarding
// address check, 22 bytes for BRBMI_RUNTIME_CALL and 2 bytes for the near
// branch) shall take the rest of the 32-byte section (within a cache line).
BRBMI_UNMARKED_FORWARDING_ADDRESS_CHECK \label_suffix
BRBMI_RUNTIME_CALL
b .Lmark_introspection_return\label_suffix
.endm
/*
* Use introspection to load a reference from the same address as the LDR
* instruction in generated code would load (unless loaded by the thunk,
* see below), call ReadBarrier::Mark() with that reference if needed
* and return it in the same register as the LDR instruction would load.
*
* The entrypoint is called through a thunk that differs across load kinds.
* For field and array loads the LDR instruction in generated code follows
* the branch to the thunk, i.e. the LDR is (ignoring the heap poisoning)
* at [LR, #(-4 - 1)] (encoding T3) or [LR, #(-2 - 1)] (encoding T1) where
* the -1 is an adjustment for the Thumb mode bit in LR, and the thunk
* knows the holder and performs the gray bit check, returning to the LDR
* instruction if the object is not gray, so this entrypoint no longer
* needs to know anything about the holder. For GC root loads, the LDR
* instruction in generated code precedes the branch to the thunk, i.e. the
* LDR is at [LR, #(-8 - 1)] (encoding T3) or [LR, #(-6 - 1)] (encoding T1)
* where the -1 is again the Thumb mode bit adjustment, and the thunk does
* not do the gray bit check.
*
* For field accesses and array loads with a constant index the thunk loads
* the reference into IP using introspection and calls the main entrypoint,
* art_quick_read_barrier_mark_introspection. With heap poisoning enabled,
* the passed reference is poisoned.
*
* For array accesses with non-constant index, the thunk inserts the bits
* 0-5 of the LDR instruction to the entrypoint address, effectively
* calculating a switch case label based on the index register (bits 0-3)
* and adding an extra offset (bits 4-5 hold the shift which is always 2
* for reference loads) to differentiate from the main entrypoint, then
* moves the base register to IP and jumps to the switch case. Therefore
* we need to align the main entrypoint to 512 bytes, accounting for
* a 256-byte offset followed by 16 array entrypoints starting at
* art_quick_read_barrier_mark_introspection_arrays, each containing an LDR
* (register) and a branch to the main entrypoint.
*
* For GC root accesses we cannot use the main entrypoint because of the
* different offset where the LDR instruction in generated code is located.
* (And even with heap poisoning enabled, GC roots are not poisoned.)
* To re-use the same entrypoint pointer in generated code, we make sure
* that the gc root entrypoint (a copy of the entrypoint with a different
* offset for introspection loads) is located at a known offset (128 bytes,
* or BAKER_MARK_INTROSPECTION_GC_ROOT_ENTRYPOINT_OFFSET) from the main
* entrypoint and the GC root thunk adjusts the entrypoint pointer, moves
* the root register to IP and jumps to the customized entrypoint,
* art_quick_read_barrier_mark_introspection_gc_roots. The thunk also
* performs all the fast-path checks, so we need just the slow path.
*
* The code structure is
* art_quick_read_barrier_mark_introspection:
* Up to 32 bytes code for main entrypoint fast-path code for fields
* (and array elements with constant offset) with LDR encoding T3;
* jumps to the switch in the "narrow" entrypoint.
* Padding to 32 bytes if needed.
* art_quick_read_barrier_mark_introspection_narrow:
* Up to 48 bytes code for fast path code for fields (and array
* elements with constant offset) with LDR encoding T1, ending in the
* return switch instruction TBB and the table with switch offsets.
* Padding to 80 bytes if needed.
* .Lmark_introspection_return_switch_case_r0:
* Exactly 48 bytes of code for the return switch cases (12 cases,
* including BKPT for the reserved registers).
* Ends at 128 bytes total.
* art_quick_read_barrier_mark_introspection_gc_roots_wide:
* GC root entrypoint code for LDR encoding T3 (28 bytes).
* Forwarding address extraction for LDR encoding T3 (6 bytes).
* Slow path for main entrypoint for LDR encoding T3 (30 bytes).
* Ends at 192 bytes total.
* art_quick_read_barrier_mark_introspection_gc_roots_narrow:
* GC root entrypoint code for LDR encoding T1 (28 bytes).
* Forwarding address extraction for LDR encoding T1 (6 bytes).
* Slow path for main entrypoint for LDR encoding T1 (30 bytes).
* Ends at 256 bytes total.
* art_quick_read_barrier_mark_introspection_arrays:
* Exactly 128 bytes for array load switch cases (16x2 instructions).
*/
.balign 512
ENTRY art_quick_read_barrier_mark_introspection
// At this point, IP contains the reference, R4 can be freely used.
// (R4 is reserved for the entrypoint address.)
// For heap poisoning, the reference is poisoned, so unpoison it first.
UNPOISON_HEAP_REF ip
// Check for null or marked, lock word is loaded into IP.
BRBMI_CHECK_NULL_AND_MARKED _wide
// Load the half of the instruction that contains Rt.
BRBMI_LOAD_RETURN_REG_FROM_CODE_wide BAKER_MARK_INTROSPECTION_FIELD_LDR_WIDE_OFFSET
.Lmark_introspection_extract_register_and_return_wide:
lsr r4, r4, #12 // Extract `ref_reg`.
b .Lmark_introspection_return_switch
.balign 32
.thumb_func
.type art_quick_read_barrier_mark_introspection_narrow, #function
.hidden art_quick_read_barrier_mark_introspection_narrow
.global art_quick_read_barrier_mark_introspection_narrow
art_quick_read_barrier_mark_introspection_narrow:
// At this point, IP contains the reference, R4 can be freely used.
// (R4 is reserved for the entrypoint address.)
// For heap poisoning, the reference is poisoned, so unpoison it first.
UNPOISON_HEAP_REF ip
// Check for null or marked, lock word is loaded into R4.
BRBMI_CHECK_NULL_AND_MARKED _narrow
// Load the 16-bit instruction.
BRBMI_LOAD_RETURN_REG_FROM_CODE_narrow BAKER_MARK_INTROSPECTION_FIELD_LDR_NARROW_OFFSET
.Lmark_introspection_extract_register_and_return_narrow:
and r4, r4, #7 // Extract `ref_reg`.
.Lmark_introspection_return_switch:
tbb [pc, r4] // Jump to the switch case.
.Lmark_introspection_return_table:
BRBMI_FOR_REGISTERS BRBMI_RETURN_SWITCH_CASE_OFFSET, BRBMI_BAD_RETURN_SWITCH_CASE_OFFSET
.balign 16
BRBMI_FOR_12_REGISTERS BRBMI_RETURN_SWITCH_CASE, BRBMI_BAD_RETURN_SWITCH_CASE
BRBMI_GC_ROOT_AND_FIELD_SLOW_PATH BAKER_MARK_INTROSPECTION_GC_ROOT_LDR_WIDE_OFFSET, _wide
BRBMI_GC_ROOT_AND_FIELD_SLOW_PATH BAKER_MARK_INTROSPECTION_GC_ROOT_LDR_NARROW_OFFSET, _narrow
.balign 256
.thumb_func
.type art_quick_read_barrier_mark_introspection_arrays, #function
.hidden art_quick_read_barrier_mark_introspection_arrays
.global art_quick_read_barrier_mark_introspection_arrays
art_quick_read_barrier_mark_introspection_arrays:
BRBMI_FOR_REGISTERS BRBMI_ARRAY_LOAD, BRBMI_BKPT_FILL_8B
END art_quick_read_barrier_mark_introspection
.extern artInvokePolymorphic
ENTRY art_quick_invoke_polymorphic
SETUP_SAVE_REFS_AND_ARGS_FRAME r2
mov r2, r9 @ pass Thread::Current
mov r3, sp @ pass SP
mov r0, #0 @ initialize 64-bit JValue as zero.
str r0, [sp, #-4]!
.cfi_adjust_cfa_offset 4
str r0, [sp, #-4]!
.cfi_adjust_cfa_offset 4
mov r0, sp @ pass JValue for return result as first argument.
bl artInvokePolymorphic @ artInvokePolymorphic(JValue, receiver, Thread*, SP)
sub r0, 'A' @ return value is descriptor of handle's return type.
cmp r0, 'Z' - 'A' @ check if value is in bounds of handler table
bgt .Lcleanup_and_return @ and clean-up if not.
adr r1, .Lhandler_table
tbb [r0, r1] @ branch to handler for return value based on return type.
.Lstart_of_handlers:
.Lstore_boolean_result:
ldrb r0, [sp] @ Copy boolean value to return value of this function.
b .Lcleanup_and_return
.Lstore_char_result:
ldrh r0, [sp] @ Copy char value to return value of this function.
b .Lcleanup_and_return
.Lstore_float_result:
vldr s0, [sp] @ Copy float value from JValue result to the context restored by
vstr s0, [sp, #16] @ RESTORE_SAVE_REFS_AND_ARGS_FRAME.
b .Lcleanup_and_return
.Lstore_double_result:
vldr d0, [sp] @ Copy double value from JValue result to the context restored by
vstr d0, [sp, #16] @ RESTORE_SAVE_REFS_AND_ARGS_FRAME.
b .Lcleanup_and_return
.Lstore_long_result:
ldr r1, [sp, #4] @ Copy the upper bits from JValue result to the context restored by
str r1, [sp, #80] @ RESTORE_SAVE_REFS_AND_ARGS_FRAME.
// Fall-through for lower bits.
.Lstore_int_result:
ldr r0, [sp] @ Copy int value to return value of this function.
// Fall-through to clean up and return.
.Lcleanup_and_return:
add sp, #8
.cfi_adjust_cfa_offset -8
RESTORE_SAVE_REFS_AND_ARGS_FRAME
RETURN_OR_DELIVER_PENDING_EXCEPTION_REG r2
.macro HANDLER_TABLE_OFFSET handler_label
.byte (\handler_label - .Lstart_of_handlers) / 2
.endm
.Lhandler_table:
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // A
HANDLER_TABLE_OFFSET(.Lstore_int_result) // B (byte)
HANDLER_TABLE_OFFSET(.Lstore_char_result) // C (char)
HANDLER_TABLE_OFFSET(.Lstore_double_result) // D (double)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // E
HANDLER_TABLE_OFFSET(.Lstore_float_result) // F (float)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // G
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // H
HANDLER_TABLE_OFFSET(.Lstore_int_result) // I (int)
HANDLER_TABLE_OFFSET(.Lstore_long_result) // J (long)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // K
HANDLER_TABLE_OFFSET(.Lstore_int_result) // L (object)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // M
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // N
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // O
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // P
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // Q
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // R
HANDLER_TABLE_OFFSET(.Lstore_int_result) // S (short)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // T
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // U
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // V (void)
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // W
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // X
HANDLER_TABLE_OFFSET(.Lcleanup_and_return) // Y
HANDLER_TABLE_OFFSET(.Lstore_boolean_result) // Z (boolean)
.purgem HANDLER_TABLE_OFFSET
END art_quick_invoke_polymorphic