使用模擬器進行 Arm 除錯
在使用 V8 程式碼產生時,模擬器和除錯器會很有幫助。
- 它很方便,因為它允許您在沒有實際硬體的情況下測試程式碼產生。
- 不需要交叉或原生編譯。
- 模擬器完全支援產生的程式碼除錯。
請注意,此模擬器是為 V8 目的而設計的。僅實作 V8 使用的功能,您可能會遇到未實作的功能或指令。在這種情況下,請隨時實作它們並提交程式碼!
使用模擬器為 Arm 編譯 #
在 x86 主機上,使用 gm 為 Arm 編譯時,預設會提供模擬器建置
gm arm64.debug # For a 64-bit build or...
gm arm.debug # ... for a 32-bit build.
您也可以建置 optdebug
組態,因為 debug
可能有點慢,特別是如果您想執行 V8 測試套件。
啟動除錯器 #
您可以在 n
指令後立即從命令列啟動除錯器
out/arm64.debug/d8 --stop_sim_at <n> # Or out/arm.debug/d8 for a 32-bit build.
或者,您可以在產生的程式碼中產生中斷點指令
本機中斷點指令會導致程式以 SIGTRAP
訊號暫停,讓您使用 gdb 除錯問題。但是,如果使用模擬器執行,產生的程式碼中的中斷點指令會讓您進入模擬器除錯器。
您可以使用 Torque 中的 DebugBreak()
、CodeStubAssembler、TurboFan 通道中的節點,或直接使用組譯器,以多種方式產生中斷點。
這裡我們專注於除錯低階原生程式碼,因此讓我們看看組譯器方法
TurboAssembler::DebugBreak();
假設我們有一個名為 add
的 jitted 函式,並使用 TurboFan 編譯,而且我們想要在開頭中斷。假設有 test.js
範例
// Our optimized function.
function add(a, b) {
return a + b;
}
// Typical cheat code enabled by --allow-natives-syntax.
%PrepareFunctionForOptimization(add);
// Give the optimizing compiler type feedback so it'll speculate `a` and `b` are
// numbers.
add(1, 3);
// And force it to optimize.
%OptimizeFunctionOnNextCall(add);
add(5, 7);
為執行此操作,我們可以連接到 TurboFan 的 程式碼產生器,並存取組譯器以插入我們的中斷點
void CodeGenerator::AssembleCode() {
// ...
// Check if we're optimizing, then look-up the name of the current function and
// insert a breakpoint.
if (info->IsOptimizing()) {
AllowHandleDereference allow_handle_dereference;
if (info->shared_info()->PassesFilter("add")) {
tasm()->DebugBreak();
}
}
// ...
}
讓我們執行它
$ d8 \
# Enable '%' cheat code JS functions.
--allow-natives-syntax \
# Disassemble our function.
--print-opt-code --print-opt-code-filter="add" --code-comments \
# Disable spectre mitigations for readability.
--no-untrusted-code-mitigations \
test.js
--- Raw source ---
(a, b) {
return a + b;
}
--- Optimized code ---
optimization_id = 0
source_position = 12
kind = OPTIMIZED_FUNCTION
name = add
stack_slots = 6
compiler = turbofan
address = 0x7f0900082ba1
Instructions (size = 504)
0x7f0900082be0 0 d45bd600 constant pool begin (num_const = 6)
0x7f0900082be4 4 00000000 constant
0x7f0900082be8 8 00000001 constant
0x7f0900082bec c 75626544 constant
0x7f0900082bf0 10 65724267 constant
0x7f0900082bf4 14 00006b61 constant
0x7f0900082bf8 18 d45bd7e0 constant
-- Prologue: check code start register --
0x7f0900082bfc 1c 10ffff30 adr x16, #-0x1c (addr 0x7f0900082be0)
0x7f0900082c00 20 eb02021f cmp x16, x2
0x7f0900082c04 24 54000080 b.eq #+0x10 (addr 0x7f0900082c14)
Abort message:
Wrong value in code start register passed
0x7f0900082c08 28 d2800d01 movz x1, #0x68
-- Inlined Trampoline to Abort --
0x7f0900082c0c 2c 58000d70 ldr x16, pc+428 (addr 0x00007f0900082db8) ;; off heap target
0x7f0900082c10 30 d63f0200 blr x16
-- Prologue: check for deoptimization --
[ DecompressTaggedPointer
0x7f0900082c14 34 b85d0050 ldur w16, [x2, #-48]
0x7f0900082c18 38 8b100350 add x16, x26, x16
]
0x7f0900082c1c 3c b8407210 ldur w16, [x16, #7]
0x7f0900082c20 40 36000070 tbz w16, #0, #+0xc (addr 0x7f0900082c2c)
-- Inlined Trampoline to CompileLazyDeoptimizedCode --
0x7f0900082c24 44 58000c31 ldr x17, pc+388 (addr 0x00007f0900082da8) ;; off heap target
0x7f0900082c28 48 d61f0220 br x17
-- B0 start (construct frame) --
(...)
--- End code ---
# Debugger hit 0: DebugBreak
0x00007f0900082bfc 10ffff30 adr x16, #-0x1c (addr 0x7f0900082be0)
sim>
我們可以看到我們已在最佳化函式的開頭停止,而且模擬器給了我們一個提示!
請注意,這只是一個範例,而且 V8 會快速變更,因此詳細資料可能會有所不同。但是,您應該可以在任何有組譯器的地方執行此操作。
偵錯指令 #
常見指令 #
在偵錯器提示中輸入 help
以取得可用指令的詳細資料。這些指令包括常見的 gdb 類似指令,例如 stepi
、cont
、disasm
等。如果模擬器在 gdb 下執行,gdb
偵錯器指令會將控制權交給 gdb。然後您可以從 gdb 使用 cont
回到偵錯器。
特定於架構的指令 #
每個目標架構都實作自己的模擬器和偵錯器,因此體驗和詳細資料會有所不同。
printobject $register
(別名 po
) #
描述儲存在暫存器中的 JS 物件。
例如,假設這次我們在 32 位元 Arm 模擬器建置上執行 我們的範例。我們可以檢查以暫存器傳入的輸入引數
$ ./out/arm.debug/d8 --allow-natives-syntax test.js
Simulator hit stop, breaking at the next instruction:
0x26842e24 e24fc00c sub ip, pc, #12
sim> print r1
r1: 0x4b60ffb1 1264648113
# The current function object is passed with r1.
sim> printobject r1
r1:
0x4b60ffb1: [Function] in OldSpace
- map: 0x485801f9 <Map(HOLEY_ELEMENTS)> [FastProperties]
- prototype: 0x4b6010f1 <JSFunction (sfi = 0x42404e99)>
- elements: 0x5b700661 <FixedArray[0]> [HOLEY_ELEMENTS]
- function prototype:
- initial_map:
- shared_info: 0x4b60fe9d <SharedFunctionInfo add>
- name: 0x5b701c5d <String[#3]: add>
- formal_parameter_count: 2
- kind: NormalFunction
- context: 0x4b600c65 <NativeContext[261]>
- code: 0x26842de1 <Code OPTIMIZED_FUNCTION>
- source code: (a, b) {
return a + b;
}
(...)
# Now print the current JS context passed in r7.
sim> printobject r7
r7:
0x449c0c65: [NativeContext] in OldSpace
- map: 0x561000b9 <Map>
- length: 261
- scope_info: 0x34081341 <ScopeInfo SCRIPT_SCOPE [5]>
- previous: 0
- native_context: 0x449c0c65 <NativeContext[261]>
0: 0x34081341 <ScopeInfo SCRIPT_SCOPE [5]>
1: 0
2: 0x449cdaf5 <JSObject>
3: 0x58480c25 <JSGlobal Object>
4: 0x58485499 <Other heap object (EMBEDDER_DATA_ARRAY_TYPE)>
5: 0x561018a1 <Map(HOLEY_ELEMENTS)>
6: 0x3408027d <undefined>
7: 0x449c75c1 <JSFunction ArrayBuffer (sfi = 0x4be8ade1)>
8: 0x561010f9 <Map(HOLEY_ELEMENTS)>
9: 0x449c967d <JSFunction arrayBufferConstructor_DoNotInitialize (sfi = 0x4be8c3ed)>
10: 0x449c8dbd <JSFunction Array (sfi = 0x4be8be59)>
(...)
trace
(別名 t
) #
啟用或停用追蹤執行的指令。
啟用時,模擬器會在執行指令時印出組譯後的指令。如果您正在執行 64 位元 Arm 建置,模擬器也能追蹤暫存器值的變更。
您也可以使用 --trace-sim
旗標從命令列啟用此功能,以從一開始就啟用追蹤。
使用相同的 範例
$ out/arm64.debug/d8 --allow-natives-syntax \
# --debug-sim is required on 64-bit Arm to enable disassembly
# when tracing.
--debug-sim test.js
# Debugger hit 0: DebugBreak
0x00007f1e00082bfc 10ffff30 adr x16, #-0x1c (addr 0x7f1e00082be0)
sim> trace
0x00007f1e00082bfc 10ffff30 adr x16, #-0x1c (addr 0x7f1e00082be0)
Enabling disassembly, registers and memory write tracing
# Break on the return address stored in the lr register.
sim> break lr
Set a breakpoint at 0x7f1f880abd28
0x00007f1e00082bfc 10ffff30 adr x16, #-0x1c (addr 0x7f1e00082be0)
# Continuing will trace the function's execution until we return, allowing
# us to make sense of what is happening.
sim> continue
# x0: 0x00007f1e00082ba1
# x1: 0x00007f1e08250125
# x2: 0x00007f1e00082be0
(...)
# We first load the 'a' and 'b' arguments from the stack and check if they
# are tagged numbers. This is indicated by the least significant bit being 0.
0x00007f1e00082c90 f9401fe2 ldr x2, [sp, #56]
# x2: 0x000000000000000a <- 0x00007f1f821f0278
0x00007f1e00082c94 7200005f tst w2, #0x1
# NZCV: N:0 Z:1 C:0 V:0
0x00007f1e00082c98 54000ac1 b.ne #+0x158 (addr 0x7f1e00082df0)
0x00007f1e00082c9c f9401be3 ldr x3, [sp, #48]
# x3: 0x000000000000000e <- 0x00007f1f821f0270
0x00007f1e00082ca0 7200007f tst w3, #0x1
# NZCV: N:0 Z:1 C:0 V:0
0x00007f1e00082ca4 54000a81 b.ne #+0x150 (addr 0x7f1e00082df4)
# Then we untag and add 'a' and 'b' together.
0x00007f1e00082ca8 13017c44 asr w4, w2, #1
# x4: 0x0000000000000005
0x00007f1e00082cac 2b830484 adds w4, w4, w3, asr #1
# NZCV: N:0 Z:0 C:0 V:0
# x4: 0x000000000000000c
# That's 5 + 7 == 12, all good!
# Then we check for overflows and tag the result again.
0x00007f1e00082cb0 54000a46 b.vs #+0x148 (addr 0x7f1e00082df8)
0x00007f1e00082cb4 2b040082 adds w2, w4, w4
# NZCV: N:0 Z:0 C:0 V:0
# x2: 0x0000000000000018
0x00007f1e00082cb8 54000466 b.vs #+0x8c (addr 0x7f1e00082d44)
# And finally we place the result in x0.
0x00007f1e00082cbc aa0203e0 mov x0, x2
# x0: 0x0000000000000018
(...)
0x00007f1e00082cec d65f03c0 ret
Hit and disabled a breakpoint at 0x7f1f880abd28.
0x00007f1f880abd28 f85e83b4 ldur x20, [fp, #-24]
sim>
break $address
#
在指定的位址插入中斷點。
請注意,在 32 位元 Arm 上,您只能有一個中斷點,而且您需要停用程式碼頁面的寫入保護才能插入中斷點。64 位元 Arm 模擬器沒有這些限制。
再次使用我們的 範例
$ out/arm.debug/d8 --allow-natives-syntax \
# This is useful to know which address to break to.
--print-opt-code --print-opt-code-filter="add" \
test.js
(...)
Simulator hit stop, breaking at the next instruction:
0x488c2e20 e24fc00c sub ip, pc, #12
# Break on a known interesting address, where we start
# loading 'a' and 'b'.
sim> break 0x488c2e9c
sim> continue
0x488c2e9c e59b200c ldr r2, [fp, #+12]
# We can look-ahead with 'disasm'.
sim> disasm 10
0x488c2e9c e59b200c ldr r2, [fp, #+12]
0x488c2ea0 e3120001 tst r2, #1
0x488c2ea4 1a000037 bne +228 -> 0x488c2f88
0x488c2ea8 e59b3008 ldr r3, [fp, #+8]
0x488c2eac e3130001 tst r3, #1
0x488c2eb0 1a000037 bne +228 -> 0x488c2f94
0x488c2eb4 e1a040c2 mov r4, r2, asr #1
0x488c2eb8 e09440c3 adds r4, r4, r3, asr #1
0x488c2ebc 6a000037 bvs +228 -> 0x488c2fa0
0x488c2ec0 e0942004 adds r2, r4, r4
# And try and break on the result of the first `adds` instructions.
sim> break 0x488c2ebc
setting breakpoint failed
# Ah, we need to delete the breakpoint first.
sim> del
sim> break 0x488c2ebc
sim> cont
0x488c2ebc 6a000037 bvs +228 -> 0x488c2fa0
sim> print r4
r4: 0x0000000c 12
# That's 5 + 7 == 12, all good!
產生中斷點指令,並具有一些額外功能 #
你可以使用具有相同效果但附加額外功能的較低層級指令,取代 TurboAssembler::DebugBreak()
。
stop()
(32 位元 Arm)#
Assembler::stop(Condition cond = al, int32_t code = kDefaultStopCode);
第一個參數是條件,第二個是停止碼。如果指定了碼,且小於 256,則停止會被標示為「已監控」,且可以停用/啟用;計數器也會追蹤模擬器觸發此碼的次數。
假設我們正在處理這段 V8 C++ 程式碼
__ stop(al, 123);
__ mov(r0, r0);
__ mov(r0, r0);
__ mov(r0, r0);
__ mov(r0, r0);
__ mov(r0, r0);
__ stop(al, 0x1);
__ mov(r1, r1);
__ mov(r1, r1);
__ mov(r1, r1);
__ mov(r1, r1);
__ mov(r1, r1);
以下是範例除錯階段
我們觸發了第一個停止點。
Simulator hit stop 123, breaking at the next instruction:
0xb53559e8 e1a00000 mov r0, r0
我們可以使用 disasm
來查看以下停止點。
sim> disasm
0xb53559e8 e1a00000 mov r0, r0
0xb53559ec e1a00000 mov r0, r0
0xb53559f0 e1a00000 mov r0, r0
0xb53559f4 e1a00000 mov r0, r0
0xb53559f8 e1a00000 mov r0, r0
0xb53559fc ef800001 stop 1 - 0x1
0xb5355a00 e1a00000 mov r1, r1
0xb5355a04 e1a00000 mov r1, r1
0xb5355a08 e1a00000 mov r1, r1
可以列印所有(已監控)至少觸發過一次的停止點資訊。
sim> stop info all
Stop information:
stop 123 - 0x7b: Enabled, counter = 1
sim> cont
Simulator hit stop 1, breaking at the next instruction:
0xb5355a04 e1a00000 mov r1, r1
sim> stop info all
Stop information:
stop 1 - 0x1: Enabled, counter = 1
stop 123 - 0x7b: Enabled, counter = 1
可以停用或啟用停止點。(僅適用於已監控的停止點。)
sim> stop disable 1
sim> cont
Simulator hit stop 123, breaking at the next instruction:
0xb5356808 e1a00000 mov r0, r0
sim> cont
Simulator hit stop 123, breaking at the next instruction:
0xb5356c28 e1a00000 mov r0, r0
sim> stop info all
Stop information:
stop 1 - 0x1: Disabled, counter = 2
stop 123 - 0x7b: Enabled, counter = 3
sim> stop enable 1
sim> cont
Simulator hit stop 1, breaking at the next instruction:
0xb5356c44 e1a00000 mov r1, r1
sim> stop disable all
sim> con
Debug()
(64 位元 Arm)#
MacroAssembler::Debug(const char* message, uint32_t code, Instr params = BREAK);
此指令預設為中斷點,但也可以啟用和停用追蹤,就像你在除錯器中使用 trace
指令一樣。你也可以提供訊息和碼作為識別碼。
假設我們正在處理這段 V8 C++ 程式碼,取自準備框架以呼叫 JS 函式的原生內建函式。
int64_t bad_frame_pointer = -1L; // Bad frame pointer, should fail if it is used.
__ Mov(x13, bad_frame_pointer);
__ Mov(x12, StackFrame::TypeToMarker(type));
__ Mov(x11, ExternalReference::Create(IsolateAddressId::kCEntryFPAddress,
masm->isolate()));
__ Ldr(x10, MemOperand(x11));
__ Push(x13, x12, xzr, x10);
使用 DebugBreak()
插入中斷點可能很有用,這樣我們就可以在執行時檢查目前的狀態。但如果我們改用 Debug()
,我們可以更進一步追蹤這段程式碼
// Start tracing and log disassembly and register values.
__ Debug("start tracing", 42, TRACE_ENABLE | LOG_ALL);
int64_t bad_frame_pointer = -1L; // Bad frame pointer, should fail if it is used.
__ Mov(x13, bad_frame_pointer);
__ Mov(x12, StackFrame::TypeToMarker(type));
__ Mov(x11, ExternalReference::Create(IsolateAddressId::kCEntryFPAddress,
masm->isolate()));
__ Ldr(x10, MemOperand(x11));
__ Push(x13, x12, xzr, x10);
// Stop tracing.
__ Debug("stop tracing", 42, TRACE_DISABLE);
它讓我們可以追蹤我們正在處理的程式碼片段的暫存器值
$ d8 --allow-natives-syntax --debug-sim test.js
# NZCV: N:0 Z:0 C:0 V:0
# FPCR: AHP:0 DN:0 FZ:0 RMode:0b00 (Round to Nearest)
# x0: 0x00007fbf00000000
# x1: 0x00007fbf0804030d
# x2: 0x00007fbf082500e1
(...)
0x00007fc039d31cb0 9280000d movn x13, #0x0
# x13: 0xffffffffffffffff
0x00007fc039d31cb4 d280004c movz x12, #0x2
# x12: 0x0000000000000002
0x00007fc039d31cb8 d2864110 movz x16, #0x3208
# ip0: 0x0000000000003208
0x00007fc039d31cbc 8b10034b add x11, x26, x16
# x11: 0x00007fbf00003208
0x00007fc039d31cc0 f940016a ldr x10, [x11]
# x10: 0x0000000000000000 <- 0x00007fbf00003208
0x00007fc039d31cc4 a9be7fea stp x10, xzr, [sp, #-32]!
# sp: 0x00007fc033e81340
# x10: 0x0000000000000000 -> 0x00007fc033e81340
# xzr: 0x0000000000000000 -> 0x00007fc033e81348
0x00007fc039d31cc8 a90137ec stp x12, x13, [sp, #16]
# x12: 0x0000000000000002 -> 0x00007fc033e81350
# x13: 0xffffffffffffffff -> 0x00007fc033e81358
0x00007fc039d31ccc 910063fd add fp, sp, #0x18 (24)
# fp: 0x00007fc033e81358
0x00007fc039d31cd0 d45bd600 hlt #0xdeb0