v8/test/cctest/test-code-pages.cc
Peter Marshall 285e4d69de [unwinder] Add a vector-based code page mechanism for arm32
Add an API on Isolate that returns a sorted vector of code pages allocated
within V8. The implementation is designed to be signal-safe, so that the
user (the UMA sampling profiler) can access this information from a signal
handler, where allocation and taking locks is prohibited.

This CL adds the machinery for maintaining the list of allocated code
pages. Further CLs will modify the Unwinder API itself to accept the code
pages provided by this API.

The unwinder API currently uses the reserved virtual-memory range called
the CodeRange to identify where all V8 code objects live, but this doesn't
exist on arm32 or any 32-bit platform, so this approach adds a way to
expose the location of all valid V8 code objects in a signal-safe way for
use by the UMA sampling profiler.

On 64-bit, this API always gives the code_range and embedded_code_range, and
does not maintain a vector of code pages. This is so that we have a unified
API on 32 and 64-bit that can be used in exactly the same way by embedders.

Design doc:
https://docs.google.com/document/d/1VGwUult5AHLRk658VetwEHMOmDDxA2eDQs9lDFMZTE0

Bug: v8:8116
Change-Id: I732509a45121fc54853182481c24d1083275afce
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/1564068
Commit-Queue: Peter Marshall <petermarshall@chromium.org>
Reviewed-by: Ulan Degenbaev <ulan@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65469}
2019-12-17 08:05:20 +00:00

543 lines
18 KiB
C++

// Copyright 2019 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 "src/api/api-inl.h"
#include "src/codegen/code-desc.h"
#include "src/common/globals.h"
#include "src/execution/isolate.h"
#include "src/handles/handles-inl.h"
#include "src/heap/factory.h"
#include "src/heap/spaces.h"
#include "src/libsampler/sampler.h"
#include "test/cctest/cctest.h"
namespace v8 {
namespace internal {
namespace test_code_pages {
// We have three levels of support which have different behaviors to test.
// 1 - Have code range. ARM64 and x64
// 2 - Have code pages. ARM32 only
// 3 - Nothing - This feature does not work on other platforms.
#if defined(V8_TARGET_ARCH_ARM)
static const bool kHaveCodeRange = false;
static const bool kHaveCodePages = true;
#else
static const bool kHaveCodeRange = kRequiresCodeRange;
static const bool kHaveCodePages = false;
#endif // defined(V8_TARGET_ARCH_ARM)
static const char* foo_source = R"(
function foo%d(a, b) {
let x = a * b;
let y = x ^ b;
let z = y / a;
return x + y - z;
};
%%PrepareFunctionForOptimization(foo%d);
foo%d(1, 2);
foo%d(1, 2);
%%OptimizeFunctionOnNextCall(foo%d);
foo%d(1, 2);
)";
std::string getFooCode(int n) {
constexpr size_t kMaxSize = 512;
char foo_replaced[kMaxSize];
CHECK_LE(n, 999999);
snprintf(foo_replaced, kMaxSize, foo_source, n, n, n, n, n, n);
return std::string(foo_replaced);
}
namespace {
bool PagesHasExactPage(std::vector<MemoryRange>* pages, Address search_page) {
void* addr = reinterpret_cast<void*>(search_page);
auto it =
std::find_if(pages->begin(), pages->end(),
[addr](const MemoryRange& r) { return r.start == addr; });
return it != pages->end();
}
bool PagesHasExactPage(std::vector<MemoryRange>* pages, Address search_page,
size_t size) {
void* addr = reinterpret_cast<void*>(search_page);
auto it = std::find_if(pages->begin(), pages->end(),
[addr, size](const MemoryRange& r) {
return r.start == addr && r.length_in_bytes == size;
});
return it != pages->end();
}
bool PagesContainsAddress(std::vector<MemoryRange>* pages,
Address search_address) {
byte* addr = reinterpret_cast<byte*>(search_address);
auto it =
std::find_if(pages->begin(), pages->end(), [addr](const MemoryRange& r) {
const byte* page_start = reinterpret_cast<const byte*>(r.start);
const byte* page_end = page_start + r.length_in_bytes;
return addr >= page_start && addr < page_end;
});
return it != pages->end();
}
} // namespace
TEST(CodeRangeCorrectContents) {
if (!kHaveCodeRange) return;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
const base::AddressRegion& code_range =
i_isolate->heap()->memory_allocator()->code_range();
CHECK(!code_range.is_empty());
// We should only have the code range and the embedded code range.
CHECK_EQ(2, pages->size());
CHECK(PagesHasExactPage(pages, code_range.begin(), code_range.size()));
CHECK(PagesHasExactPage(pages,
reinterpret_cast<Address>(i_isolate->embedded_blob()),
i_isolate->embedded_blob_size()));
}
TEST(CodePagesCorrectContents) {
if (!kHaveCodePages) return;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
// There might be other pages already.
CHECK_GE(pages->size(), 1);
const base::AddressRegion& code_range =
i_isolate->heap()->memory_allocator()->code_range();
CHECK(code_range.is_empty());
// We should have the embedded code range even when there is no regular code
// range.
CHECK(PagesHasExactPage(pages,
reinterpret_cast<Address>(i_isolate->embedded_blob()),
i_isolate->embedded_blob_size()));
}
TEST(OptimizedCodeWithCodeRange) {
if (!kHaveCodeRange) return;
FLAG_allow_natives_syntax = true;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
HandleScope scope(i_isolate);
std::string foo_str = getFooCode(1);
CompileRun(foo_str.c_str());
v8::Local<v8::Function> local_foo = v8::Local<v8::Function>::Cast(
env.local()->Global()->Get(env.local(), v8_str("foo1")).ToLocalChecked());
Handle<JSFunction> foo =
Handle<JSFunction>::cast(v8::Utils::OpenHandle(*local_foo));
AbstractCode abstract_code = foo->abstract_code();
// We don't produce optimized code when run with --no-opt.
if (!abstract_code.IsCode() && FLAG_opt == false) return;
CHECK(abstract_code.IsCode());
Code foo_code = abstract_code.GetCode();
CHECK(i_isolate->heap()->InSpace(foo_code, CODE_SPACE));
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
CHECK(PagesContainsAddress(pages, foo_code.address()));
}
TEST(OptimizedCodeWithCodePages) {
if (!kHaveCodePages) return;
// We don't want incremental marking to start which could cause the code to
// not be collected on the CollectGarbage() call.
ManualGCScope manual_gc_scope;
FLAG_allow_natives_syntax = true;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
const void* created_page = nullptr;
int num_foos_created = 0;
{
HandleScope scope(i_isolate);
size_t num_code_pages = 0;
size_t initial_num_code_pages = 0;
// Keep generating new code until a new code page is added to the list.
for (int n = 0; n < 999999; n++) {
// Compile and optimize the code and get a reference to it.
std::string foo_str = getFooCode(n);
char foo_name[10];
snprintf(foo_name, sizeof(foo_name), "foo%d", n);
CompileRun(foo_str.c_str());
v8::Local<v8::Function> local_foo =
v8::Local<v8::Function>::Cast(env.local()
->Global()
->Get(env.local(), v8_str(foo_name))
.ToLocalChecked());
Handle<JSFunction> foo =
Handle<JSFunction>::cast(v8::Utils::OpenHandle(*local_foo));
AbstractCode abstract_code = foo->abstract_code();
// We don't produce optimized code when run with --no-opt.
if (!abstract_code.IsCode() && FLAG_opt == false) return;
CHECK(abstract_code.IsCode());
Code foo_code = abstract_code.GetCode();
CHECK(i_isolate->heap()->InSpace(foo_code, CODE_SPACE));
// Check that the generated code ended up in one of the code pages
// returned by GetCodePages().
byte* foo_code_ptr = reinterpret_cast<byte*>(foo_code.address());
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
// Wait until after we have created the first function to take the initial
// number of pages so that this test isn't brittle to irrelevant
// implementation details.
if (n == 0) {
initial_num_code_pages = pages->size();
}
num_code_pages = pages->size();
// Check that the code object was allocation on any of the pages returned
// by GetCodePages().
auto it = std::find_if(
pages->begin(), pages->end(), [foo_code_ptr](const MemoryRange& r) {
const byte* page_start = reinterpret_cast<const byte*>(r.start);
const byte* page_end = page_start + r.length_in_bytes;
return foo_code_ptr >= page_start && foo_code_ptr < page_end;
});
CHECK_NE(it, pages->end());
// Store the page that was created just for our functions - we expect it
// to be removed later.
if (num_code_pages > initial_num_code_pages) {
created_page = it->start;
num_foos_created = n + 1;
break;
}
}
CHECK_NOT_NULL(created_page);
}
// Now delete all our foos and force a GC and check that the page is removed
// from the list.
{
HandleScope scope(i_isolate);
for (int n = 0; n < num_foos_created; n++) {
char foo_name[10];
snprintf(foo_name, sizeof(foo_name), "foo%d", n);
env.local()
->Global()
->Set(env.local(), v8_str(foo_name), Undefined(isolate))
.Check();
}
}
CcTest::CollectGarbage(CODE_SPACE);
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
auto it = std::find_if(
pages->begin(), pages->end(),
[created_page](const MemoryRange& r) { return r.start == created_page; });
CHECK_EQ(it, pages->end());
}
TEST(LargeCodeObject) {
if (!kHaveCodeRange && !kHaveCodePages) return;
// We don't want incremental marking to start which could cause the code to
// not be collected on the CollectGarbage() call.
ManualGCScope manual_gc_scope;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
// Create a big function that ends up in CODE_LO_SPACE.
const int instruction_size = Page::kPageSize + 1;
STATIC_ASSERT(instruction_size > kMaxRegularHeapObjectSize);
std::unique_ptr<byte[]> instructions(new byte[instruction_size]);
CodeDesc desc;
desc.buffer = instructions.get();
desc.buffer_size = instruction_size;
desc.instr_size = instruction_size;
desc.reloc_size = 0;
desc.constant_pool_size = 0;
desc.unwinding_info = nullptr;
desc.unwinding_info_size = 0;
desc.origin = nullptr;
Address stale_code_address;
{
HandleScope scope(i_isolate);
Handle<Code> foo_code =
Factory::CodeBuilder(i_isolate, desc, Code::WASM_FUNCTION).Build();
CHECK(i_isolate->heap()->InSpace(*foo_code, CODE_LO_SPACE));
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
if (kHaveCodeRange) {
CHECK(PagesContainsAddress(pages, foo_code->address()));
} else {
CHECK(PagesHasExactPage(pages, foo_code->address()));
}
stale_code_address = foo_code->address();
}
// Delete the large code object.
CcTest::CollectGarbage(CODE_LO_SPACE);
CHECK(!i_isolate->heap()->InSpaceSlow(stale_code_address, CODE_LO_SPACE));
// Check that it was removed from CodePages.
std::vector<MemoryRange>* pages = i_isolate->GetCodePages();
CHECK(!PagesHasExactPage(pages, stale_code_address));
}
static constexpr size_t kBufSize = v8::Isolate::kMinCodePagesBufferSize;
class SignalSender : public sampler::Sampler {
public:
explicit SignalSender(v8::Isolate* isolate) : sampler::Sampler(isolate) {}
// Called during the signal/thread suspension.
void SampleStack(const v8::RegisterState& regs) override {
MemoryRange* code_pages_copy = code_pages_copy_.load();
CHECK_NOT_NULL(code_pages_copy);
size_t num_pages = isolate_->CopyCodePages(kBufSize, code_pages_copy);
CHECK_LE(num_pages, kBufSize);
sample_semaphore_.Signal();
}
// Called on the sampling thread to trigger a sample. Blocks until the sample
// is finished.
void SampleIntoVector(MemoryRange output_buffer[]) {
code_pages_copy_.store(output_buffer);
DoSample();
sample_semaphore_.Wait();
code_pages_copy_.store(nullptr);
}
private:
base::Semaphore sample_semaphore_{0};
std::atomic<MemoryRange*> code_pages_copy_{nullptr};
};
class SamplingThread : public base::Thread {
public:
explicit SamplingThread(SignalSender* signal_sender)
: base::Thread(base::Thread::Options("SamplingThread")),
signal_sender_(signal_sender) {}
// Blocks until a sample is taken.
void TriggerSample() { signal_sender_->SampleIntoVector(code_pages_copy_); }
void Run() override {
while (running_.load()) {
TriggerSample();
}
}
// Called from the main thread. Blocks until a sample is taken. Not
// thread-safe so do not call while this thread is running.
static std::vector<MemoryRange> DoSynchronousSample(v8::Isolate* isolate) {
MemoryRange code_pages_copy[kBufSize];
size_t num_pages = isolate->CopyCodePages(kBufSize, code_pages_copy);
DCHECK_LE(num_pages, kBufSize);
return std::vector<MemoryRange>{code_pages_copy,
&code_pages_copy[num_pages]};
}
void Stop() { running_.store(false); }
private:
std::atomic_bool running_{true};
SignalSender* signal_sender_;
MemoryRange code_pages_copy_[kBufSize];
};
TEST(LargeCodeObjectWithSignalHandler) {
if (!kHaveCodeRange && !kHaveCodePages) return;
// We don't want incremental marking to start which could cause the code to
// not be collected on the CollectGarbage() call.
ManualGCScope manual_gc_scope;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
// Create a big function that ends up in CODE_LO_SPACE.
const int instruction_size = Page::kPageSize + 1;
STATIC_ASSERT(instruction_size > kMaxRegularHeapObjectSize);
std::unique_ptr<byte[]> instructions(new byte[instruction_size]);
CodeDesc desc;
desc.buffer = instructions.get();
desc.buffer_size = instruction_size;
desc.instr_size = instruction_size;
desc.reloc_size = 0;
desc.constant_pool_size = 0;
desc.unwinding_info = nullptr;
desc.unwinding_info_size = 0;
desc.origin = nullptr;
Address stale_code_address;
SignalSender signal_sender(isolate);
signal_sender.Start();
// Take an initial sample.
std::vector<MemoryRange> initial_pages =
SamplingThread::DoSynchronousSample(isolate);
SamplingThread sampling_thread(&signal_sender);
sampling_thread.StartSynchronously();
{
HandleScope scope(i_isolate);
Handle<Code> foo_code =
Factory::CodeBuilder(i_isolate, desc, Code::WASM_FUNCTION).Build();
CHECK(i_isolate->heap()->InSpace(*foo_code, CODE_LO_SPACE));
// Do a synchronous sample to ensure that we capture the state with the
// extra code page.
sampling_thread.Stop();
sampling_thread.Join();
// Check that the page was added.
std::vector<MemoryRange> pages =
SamplingThread::DoSynchronousSample(isolate);
if (kHaveCodeRange) {
CHECK(PagesContainsAddress(&pages, foo_code->address()));
} else {
CHECK(PagesHasExactPage(&pages, foo_code->address()));
}
stale_code_address = foo_code->address();
}
// Start async sampling again to detect threading issues.
sampling_thread.StartSynchronously();
// Delete the large code object.
CcTest::CollectGarbage(CODE_LO_SPACE);
CHECK(!i_isolate->heap()->InSpaceSlow(stale_code_address, CODE_LO_SPACE));
sampling_thread.Stop();
sampling_thread.Join();
std::vector<MemoryRange> pages = SamplingThread::DoSynchronousSample(isolate);
CHECK(!PagesHasExactPage(&pages, stale_code_address));
signal_sender.Stop();
}
TEST(Sorted) {
if (!kHaveCodeRange && !kHaveCodePages) return;
// We don't want incremental marking to start which could cause the code to
// not be collected on the CollectGarbage() call.
ManualGCScope manual_gc_scope;
LocalContext env;
v8::Isolate* isolate = env->GetIsolate();
Isolate* i_isolate = reinterpret_cast<Isolate*>(isolate);
// Create a big function that ends up in CODE_LO_SPACE.
const int instruction_size = Page::kPageSize + 1;
STATIC_ASSERT(instruction_size > kMaxRegularHeapObjectSize);
std::unique_ptr<byte[]> instructions(new byte[instruction_size]);
CodeDesc desc;
desc.buffer = instructions.get();
desc.buffer_size = instruction_size;
desc.instr_size = instruction_size;
desc.reloc_size = 0;
desc.constant_pool_size = 0;
desc.unwinding_info = nullptr;
desc.unwinding_info_size = 0;
desc.origin = nullptr;
// Take an initial sample.
std::vector<MemoryRange> initial_pages =
SamplingThread::DoSynchronousSample(isolate);
size_t initial_num_pages = initial_pages.size();
auto compare = [](const MemoryRange& a, const MemoryRange& b) {
return a.start < b.start;
};
{
HandleScope outer_scope(i_isolate);
Handle<Code> code1, code3;
Address code2_address;
code1 = Factory::CodeBuilder(i_isolate, desc, Code::WASM_FUNCTION).Build();
CHECK(i_isolate->heap()->InSpace(*code1, CODE_LO_SPACE));
{
HandleScope scope(i_isolate);
// Create three large code objects, we'll delete the middle one and check
// everything is still sorted.
Handle<Code> code2 =
Factory::CodeBuilder(i_isolate, desc, Code::WASM_FUNCTION).Build();
CHECK(i_isolate->heap()->InSpace(*code2, CODE_LO_SPACE));
code3 =
Factory::CodeBuilder(i_isolate, desc, Code::WASM_FUNCTION).Build();
CHECK(i_isolate->heap()->InSpace(*code3, CODE_LO_SPACE));
code2_address = code2->address();
CHECK(i_isolate->heap()->InSpaceSlow(code1->address(), CODE_LO_SPACE));
CHECK(i_isolate->heap()->InSpaceSlow(code2->address(), CODE_LO_SPACE));
CHECK(i_isolate->heap()->InSpaceSlow(code3->address(), CODE_LO_SPACE));
// Check that the pages were added.
std::vector<MemoryRange> pages =
SamplingThread::DoSynchronousSample(isolate);
if (kHaveCodeRange) {
CHECK_EQ(pages.size(), initial_num_pages);
} else {
CHECK_EQ(pages.size(), initial_num_pages + 3);
}
CHECK(std::is_sorted(pages.begin(), pages.end(), compare));
code3 = scope.CloseAndEscape(code3);
}
CHECK(i_isolate->heap()->InSpaceSlow(code1->address(), CODE_LO_SPACE));
CHECK(i_isolate->heap()->InSpaceSlow(code2_address, CODE_LO_SPACE));
CHECK(i_isolate->heap()->InSpaceSlow(code3->address(), CODE_LO_SPACE));
// Delete code2.
CcTest::CollectGarbage(CODE_LO_SPACE);
CHECK(i_isolate->heap()->InSpaceSlow(code1->address(), CODE_LO_SPACE));
CHECK(!i_isolate->heap()->InSpaceSlow(code2_address, CODE_LO_SPACE));
CHECK(i_isolate->heap()->InSpaceSlow(code3->address(), CODE_LO_SPACE));
std::vector<MemoryRange> pages =
SamplingThread::DoSynchronousSample(isolate);
if (kHaveCodeRange) {
CHECK_EQ(pages.size(), initial_num_pages);
} else {
CHECK_EQ(pages.size(), initial_num_pages + 2);
}
CHECK(std::is_sorted(pages.begin(), pages.end(), compare));
}
}
} // namespace test_code_pages
} // namespace internal
} // namespace v8