1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
|
// Copyright 2018 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <bitset>
#include "src/codegen/assembler-inl.h"
#include "src/codegen/macro-assembler-inl.h"
#include "src/execution/simulator.h"
#include "src/utils/utils.h"
#include "src/wasm/jump-table-assembler.h"
#include "test/cctest/cctest.h"
#include "test/common/assembler-tester.h"
namespace v8 {
namespace internal {
namespace wasm {
#if 0
#define TRACE(...) PrintF(__VA_ARGS__)
#else
#define TRACE(...)
#endif
#define __ masm.
namespace {
static volatile int global_stop_bit = 0;
constexpr int kJumpTableSlotCount = 128;
constexpr uint32_t kJumpTableSize =
JumpTableAssembler::SizeForNumberOfSlots(kJumpTableSlotCount);
#if V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_X64
constexpr uint32_t kAvailableBufferSlots =
(kMaxWasmCodeMemory - kJumpTableSize) / AssemblerBase::kMinimalBufferSize;
constexpr uint32_t kBufferSlotStartOffset =
RoundUp<AssemblerBase::kMinimalBufferSize>(kJumpTableSize);
#else
constexpr uint32_t kAvailableBufferSlots = 0;
#endif
Address GenerateJumpTableThunk(
Address jump_target, byte* thunk_slot_buffer,
std::bitset<kAvailableBufferSlots>* used_slots,
std::vector<std::unique_ptr<TestingAssemblerBuffer>>* thunk_buffers) {
#if V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_X64
// To guarantee that the branch range lies within the near-call range,
// generate the thunk in the same (kMaxWasmCodeMemory-sized) buffer as the
// jump_target itself.
//
// Allocate a slot that we haven't already used. This is necessary because
// each test iteration expects to generate two unique addresses and we leave
// each slot executable (and not writable).
base::RandomNumberGenerator* rng =
CcTest::i_isolate()->random_number_generator();
// Ensure a chance of completion without too much thrashing.
DCHECK(used_slots->count() < (used_slots->size() / 2));
int buffer_index;
do {
buffer_index = rng->NextInt(kAvailableBufferSlots);
} while (used_slots->test(buffer_index));
used_slots->set(buffer_index);
byte* buffer =
thunk_slot_buffer + buffer_index * AssemblerBase::kMinimalBufferSize;
#else
USE(thunk_slot_buffer);
USE(used_slots);
thunk_buffers->emplace_back(AllocateAssemblerBuffer(
AssemblerBase::kMinimalBufferSize, GetRandomMmapAddr()));
byte* buffer = thunk_buffers->back()->start();
#endif
MacroAssembler masm(
nullptr, AssemblerOptions{}, CodeObjectRequired::kNo,
ExternalAssemblerBuffer(buffer, AssemblerBase::kMinimalBufferSize));
Label exit;
Register scratch = kReturnRegister0;
Address stop_bit_address = reinterpret_cast<Address>(&global_stop_bit);
#if V8_TARGET_ARCH_X64
__ Move(scratch, stop_bit_address, RelocInfo::NONE);
__ testl(MemOperand(scratch, 0), Immediate(1));
__ j(not_zero, &exit);
__ Jump(jump_target, RelocInfo::NONE);
#elif V8_TARGET_ARCH_IA32
__ Move(scratch, Immediate(stop_bit_address, RelocInfo::NONE));
__ test(MemOperand(scratch, 0), Immediate(1));
__ j(not_zero, &exit);
__ jmp(jump_target, RelocInfo::NONE);
#elif V8_TARGET_ARCH_ARM
__ mov(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ ldr(scratch, MemOperand(scratch, 0));
__ tst(scratch, Operand(1));
__ b(ne, &exit);
__ Jump(jump_target, RelocInfo::NONE);
#elif V8_TARGET_ARCH_ARM64
__ Mov(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ Ldr(scratch, MemOperand(scratch, 0));
__ Tbnz(scratch, 0, &exit);
__ Mov(scratch, Immediate(jump_target, RelocInfo::NONE));
__ Br(scratch);
#elif V8_TARGET_ARCH_PPC64
__ mov(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ LoadP(scratch, MemOperand(scratch));
__ cmpi(scratch, Operand::Zero());
__ bne(&exit);
__ mov(scratch, Operand(jump_target, RelocInfo::NONE));
__ Jump(scratch);
#elif V8_TARGET_ARCH_S390X
__ mov(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ LoadP(scratch, MemOperand(scratch));
__ CmpP(scratch, Operand(0));
__ bne(&exit);
__ mov(scratch, Operand(jump_target, RelocInfo::NONE));
__ Jump(scratch);
#elif V8_TARGET_ARCH_MIPS64
__ li(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ Lw(scratch, MemOperand(scratch, 0));
__ Branch(&exit, ne, scratch, Operand(zero_reg));
__ Jump(jump_target, RelocInfo::NONE);
#elif V8_TARGET_ARCH_MIPS
__ li(scratch, Operand(stop_bit_address, RelocInfo::NONE));
__ lw(scratch, MemOperand(scratch, 0));
__ Branch(&exit, ne, scratch, Operand(zero_reg));
__ Jump(jump_target, RelocInfo::NONE);
#else
#error Unsupported architecture
#endif
__ bind(&exit);
__ Ret();
CodeDesc desc;
masm.GetCode(nullptr, &desc);
return reinterpret_cast<Address>(buffer);
}
class JumpTableRunner : public v8::base::Thread {
public:
JumpTableRunner(Address slot_address, int runner_id)
: Thread(Options("JumpTableRunner")),
slot_address_(slot_address),
runner_id_(runner_id) {}
void Run() override {
TRACE("Runner #%d is starting ...\n", runner_id_);
GeneratedCode<void>::FromAddress(CcTest::i_isolate(), slot_address_).Call();
TRACE("Runner #%d is stopping ...\n", runner_id_);
USE(runner_id_);
}
private:
Address slot_address_;
int runner_id_;
};
class JumpTablePatcher : public v8::base::Thread {
public:
JumpTablePatcher(Address slot_start, uint32_t slot_index, Address thunk1,
Address thunk2)
: Thread(Options("JumpTablePatcher")),
slot_start_(slot_start),
slot_index_(slot_index),
thunks_{thunk1, thunk2} {}
void Run() override {
TRACE("Patcher is starting ...\n");
constexpr int kNumberOfPatchIterations = 64;
for (int i = 0; i < kNumberOfPatchIterations; ++i) {
TRACE(" patch slot " V8PRIxPTR_FMT " to thunk #%d\n",
slot_start_ + JumpTableAssembler::SlotIndexToOffset(slot_index_),
i % 2);
JumpTableAssembler::PatchJumpTableSlot(
slot_start_, slot_index_, thunks_[i % 2], WasmCode::kFlushICache);
}
TRACE("Patcher is stopping ...\n");
}
private:
Address slot_start_;
uint32_t slot_index_;
Address thunks_[2];
};
} // namespace
// This test is intended to stress concurrent patching of jump-table slots. It
// uses the following setup:
// 1) Picks a particular slot of the jump-table. Slots are iterated over to
// ensure multiple entries (at different offset alignments) are tested.
// 2) Starts multiple runners that spin through the above slot. The runners
// use thunk code that will jump to the same jump-table slot repeatedly
// until the {global_stop_bit} indicates a test-end condition.
// 3) Start a patcher that repeatedly patches the jump-table slot back and
// forth between two thunk. If there is a race then chances are high that
// one of the runners is currently executing the jump-table slot.
TEST(JumpTablePatchingStress) {
constexpr int kNumberOfRunnerThreads = 5;
#if V8_TARGET_ARCH_ARM64 || V8_TARGET_ARCH_X64
// We need the branches (from GenerateJumpTableThunk) to be within near-call
// range of the jump table slots. The address hint to AllocateAssemblerBuffer
// is not reliable enough to guarantee that we can always achieve this with
// separate allocations, so for Arm64 we generate all code in a single
// kMaxMasmCodeMemory-sized chunk.
//
// TODO(wasm): Currently {kMaxWasmCodeMemory} limits code sufficiently, so
// that the jump table only supports {near_call} distances.
STATIC_ASSERT(kMaxWasmCodeMemory >= kJumpTableSize);
auto buffer = AllocateAssemblerBuffer(kMaxWasmCodeMemory);
byte* thunk_slot_buffer = buffer->start() + kBufferSlotStartOffset;
#else
auto buffer = AllocateAssemblerBuffer(kJumpTableSize);
byte* thunk_slot_buffer = nullptr;
#endif
std::bitset<kAvailableBufferSlots> used_thunk_slots;
buffer->MakeWritableAndExecutable();
// Iterate through jump-table slots to hammer at different alignments within
// the jump-table, thereby increasing stress for variable-length ISAs.
Address slot_start = reinterpret_cast<Address>(buffer->start());
for (int slot = 0; slot < kJumpTableSlotCount; ++slot) {
TRACE("Hammering on jump table slot #%d ...\n", slot);
uint32_t slot_offset = JumpTableAssembler::JumpSlotIndexToOffset(slot);
std::vector<std::unique_ptr<TestingAssemblerBuffer>> thunk_buffers;
Address thunk1 =
GenerateJumpTableThunk(slot_start + slot_offset, thunk_slot_buffer,
&used_thunk_slots, &thunk_buffers);
Address thunk2 =
GenerateJumpTableThunk(slot_start + slot_offset, thunk_slot_buffer,
&used_thunk_slots, &thunk_buffers);
TRACE(" generated thunk1: " V8PRIxPTR_FMT "\n", thunk1);
TRACE(" generated thunk2: " V8PRIxPTR_FMT "\n", thunk2);
JumpTableAssembler::PatchJumpTableSlot(slot_start, slot, thunk1,
WasmCode::kFlushICache);
for (auto& buf : thunk_buffers) buf->MakeExecutable();
// Start multiple runner threads and a patcher thread that hammer on the
// same jump-table slot concurrently.
std::list<JumpTableRunner> runners;
for (int runner = 0; runner < kNumberOfRunnerThreads; ++runner) {
runners.emplace_back(slot_start + slot_offset, runner);
}
JumpTablePatcher patcher(slot_start, slot, thunk1, thunk2);
global_stop_bit = 0; // Signal runners to keep going.
for (auto& runner : runners) runner.Start();
patcher.Start();
patcher.Join();
global_stop_bit = -1; // Signal runners to stop.
for (auto& runner : runners) runner.Join();
}
}
#undef __
#undef TRACE
} // namespace wasm
} // namespace internal
} // namespace v8
|
