v8/src/platform-linux.cc
mikhail.naganov@gmail.com 8318f5e5f1 Reimplement profiler sampler on Mac OS X to get it working under Chromium.
Previous implementation of sampler for OS X was copied from the Linux one. But BSD (OS X) and Linux has a very important difference in signal handling. LinuxThreads doesn't support the notion of process-directed signals. So, the SIGPROF signal was directed to the thread that installed the handler---the V8 thread. But on BSD, signal handling is implemented according to POSIX spec, where process-directed signal is to be handled by an arbitrary selected thread. By a coincidence, in V8's sample shell and in Chromium's test shell, V8's thread was picked almost every time, so sampling seemed working. But not in case of Chromium.

So, I've changed the implementation of profiler sampler to use the same scheme as on Windows---a dedicated thread with high priority is used to periodically pause and sample V8's thread.

Review URL: http://codereview.chromium.org/147150

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@2315 ce2b1a6d-e550-0410-aec6-3dcde31c8c00
2009-07-01 08:46:59 +00:00

718 lines
19 KiB
C++

// Copyright 2006-2008 the V8 project authors. All rights reserved.
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
// * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following
// disclaimer in the documentation and/or other materials provided
// with the distribution.
// * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived
// from this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
// Platform specific code for Linux goes here. For the POSIX comaptible parts
// the implementation is in platform-posix.cc.
#include <pthread.h>
#include <semaphore.h>
#include <signal.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/types.h>
#include <stdlib.h>
// Ubuntu Dapper requires memory pages to be marked as
// executable. Otherwise, OS raises an exception when executing code
// in that page.
#include <sys/types.h> // mmap & munmap
#include <sys/mman.h> // mmap & munmap
#include <sys/stat.h> // open
#include <fcntl.h> // open
#include <unistd.h> // sysconf
#ifdef __GLIBC__
#include <execinfo.h> // backtrace, backtrace_symbols
#endif // def __GLIBC__
#include <strings.h> // index
#include <errno.h>
#include <stdarg.h>
#undef MAP_TYPE
#include "v8.h"
#include "platform.h"
namespace v8 {
namespace internal {
// 0 is never a valid thread id on Linux since tids and pids share a
// name space and pid 0 is reserved (see man 2 kill).
static const pthread_t kNoThread = (pthread_t) 0;
double ceiling(double x) {
return ceil(x);
}
void OS::Setup() {
// Seed the random number generator.
// Convert the current time to a 64-bit integer first, before converting it
// to an unsigned. Going directly can cause an overflow and the seed to be
// set to all ones. The seed will be identical for different instances that
// call this setup code within the same millisecond.
uint64_t seed = static_cast<uint64_t>(TimeCurrentMillis());
srandom(static_cast<unsigned int>(seed));
}
double OS::nan_value() {
return NAN;
}
int OS::ActivationFrameAlignment() {
#ifdef V8_TARGET_ARCH_ARM
// On EABI ARM targets this is required for fp correctness in the
// runtime system.
return 8;
#else
// With gcc 4.4 the tree vectorization optimiser can generate code
// that requires 16 byte alignment such as movdqa on x86.
return 16;
#endif
}
// We keep the lowest and highest addresses mapped as a quick way of
// determining that pointers are outside the heap (used mostly in assertions
// and verification). The estimate is conservative, ie, not all addresses in
// 'allocated' space are actually allocated to our heap. The range is
// [lowest, highest), inclusive on the low and and exclusive on the high end.
static void* lowest_ever_allocated = reinterpret_cast<void*>(-1);
static void* highest_ever_allocated = reinterpret_cast<void*>(0);
static void UpdateAllocatedSpaceLimits(void* address, int size) {
lowest_ever_allocated = Min(lowest_ever_allocated, address);
highest_ever_allocated =
Max(highest_ever_allocated,
reinterpret_cast<void*>(reinterpret_cast<char*>(address) + size));
}
bool OS::IsOutsideAllocatedSpace(void* address) {
return address < lowest_ever_allocated || address >= highest_ever_allocated;
}
size_t OS::AllocateAlignment() {
return sysconf(_SC_PAGESIZE);
}
void* OS::Allocate(const size_t requested,
size_t* allocated,
bool is_executable) {
const size_t msize = RoundUp(requested, sysconf(_SC_PAGESIZE));
int prot = PROT_READ | PROT_WRITE | (is_executable ? PROT_EXEC : 0);
void* mbase = mmap(NULL, msize, prot, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
if (mbase == MAP_FAILED) {
LOG(StringEvent("OS::Allocate", "mmap failed"));
return NULL;
}
*allocated = msize;
UpdateAllocatedSpaceLimits(mbase, msize);
return mbase;
}
void OS::Free(void* address, const size_t size) {
// TODO(1240712): munmap has a return value which is ignored here.
munmap(address, size);
}
#ifdef ENABLE_HEAP_PROTECTION
void OS::Protect(void* address, size_t size) {
// TODO(1240712): mprotect has a return value which is ignored here.
mprotect(address, size, PROT_READ);
}
void OS::Unprotect(void* address, size_t size, bool is_executable) {
// TODO(1240712): mprotect has a return value which is ignored here.
int prot = PROT_READ | PROT_WRITE | (is_executable ? PROT_EXEC : 0);
mprotect(address, size, prot);
}
#endif
void OS::Sleep(int milliseconds) {
unsigned int ms = static_cast<unsigned int>(milliseconds);
usleep(1000 * ms);
}
void OS::Abort() {
// Redirect to std abort to signal abnormal program termination.
abort();
}
void OS::DebugBreak() {
// TODO(lrn): Introduce processor define for runtime system (!= V8_ARCH_x,
// which is the architecture of generated code).
#if defined(__arm__) || defined(__thumb__)
asm("bkpt 0");
#else
asm("int $3");
#endif
}
class PosixMemoryMappedFile : public OS::MemoryMappedFile {
public:
PosixMemoryMappedFile(FILE* file, void* memory, int size)
: file_(file), memory_(memory), size_(size) { }
virtual ~PosixMemoryMappedFile();
virtual void* memory() { return memory_; }
private:
FILE* file_;
void* memory_;
int size_;
};
OS::MemoryMappedFile* OS::MemoryMappedFile::create(const char* name, int size,
void* initial) {
FILE* file = fopen(name, "w+");
if (file == NULL) return NULL;
int result = fwrite(initial, size, 1, file);
if (result < 1) {
fclose(file);
return NULL;
}
void* memory =
mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED, fileno(file), 0);
return new PosixMemoryMappedFile(file, memory, size);
}
PosixMemoryMappedFile::~PosixMemoryMappedFile() {
if (memory_) munmap(memory_, size_);
fclose(file_);
}
#ifdef ENABLE_LOGGING_AND_PROFILING
static uintptr_t StringToULong(char* buffer) {
return strtoul(buffer, NULL, 16); // NOLINT
}
#endif
void OS::LogSharedLibraryAddresses() {
#ifdef ENABLE_LOGGING_AND_PROFILING
static const int MAP_LENGTH = 1024;
int fd = open("/proc/self/maps", O_RDONLY);
if (fd < 0) return;
while (true) {
char addr_buffer[11];
addr_buffer[0] = '0';
addr_buffer[1] = 'x';
addr_buffer[10] = 0;
int result = read(fd, addr_buffer + 2, 8);
if (result < 8) break;
uintptr_t start = StringToULong(addr_buffer);
result = read(fd, addr_buffer + 2, 1);
if (result < 1) break;
if (addr_buffer[2] != '-') break;
result = read(fd, addr_buffer + 2, 8);
if (result < 8) break;
uintptr_t end = StringToULong(addr_buffer);
char buffer[MAP_LENGTH];
int bytes_read = -1;
do {
bytes_read++;
if (bytes_read >= MAP_LENGTH - 1)
break;
result = read(fd, buffer + bytes_read, 1);
if (result < 1) break;
} while (buffer[bytes_read] != '\n');
buffer[bytes_read] = 0;
// Ignore mappings that are not executable.
if (buffer[3] != 'x') continue;
char* start_of_path = index(buffer, '/');
// If there is no filename for this line then log it as an anonymous
// mapping and use the address as its name.
if (start_of_path == NULL) {
// 40 is enough to print a 64 bit address range.
ASSERT(sizeof(buffer) > 40);
snprintf(buffer,
sizeof(buffer),
"%08" V8PRIxPTR "-%08" V8PRIxPTR,
start,
end);
LOG(SharedLibraryEvent(buffer, start, end));
} else {
buffer[bytes_read] = 0;
LOG(SharedLibraryEvent(start_of_path, start, end));
}
}
close(fd);
#endif
}
int OS::StackWalk(Vector<OS::StackFrame> frames) {
// backtrace is a glibc extension.
#ifdef __GLIBC__
int frames_size = frames.length();
void** addresses = NewArray<void*>(frames_size);
int frames_count = backtrace(addresses, frames_size);
char** symbols;
symbols = backtrace_symbols(addresses, frames_count);
if (symbols == NULL) {
DeleteArray(addresses);
return kStackWalkError;
}
for (int i = 0; i < frames_count; i++) {
frames[i].address = addresses[i];
// Format a text representation of the frame based on the information
// available.
SNPrintF(MutableCStrVector(frames[i].text, kStackWalkMaxTextLen),
"%s",
symbols[i]);
// Make sure line termination is in place.
frames[i].text[kStackWalkMaxTextLen - 1] = '\0';
}
DeleteArray(addresses);
free(symbols);
return frames_count;
#else // ndef __GLIBC__
return 0;
#endif // ndef __GLIBC__
}
// Constants used for mmap.
static const int kMmapFd = -1;
static const int kMmapFdOffset = 0;
VirtualMemory::VirtualMemory(size_t size) {
address_ = mmap(NULL, size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE,
kMmapFd, kMmapFdOffset);
size_ = size;
}
VirtualMemory::~VirtualMemory() {
if (IsReserved()) {
if (0 == munmap(address(), size())) address_ = MAP_FAILED;
}
}
bool VirtualMemory::IsReserved() {
return address_ != MAP_FAILED;
}
bool VirtualMemory::Commit(void* address, size_t size, bool is_executable) {
int prot = PROT_READ | PROT_WRITE | (is_executable ? PROT_EXEC : 0);
if (MAP_FAILED == mmap(address, size, prot,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_FIXED,
kMmapFd, kMmapFdOffset)) {
return false;
}
UpdateAllocatedSpaceLimits(address, size);
return true;
}
bool VirtualMemory::Uncommit(void* address, size_t size) {
return mmap(address, size, PROT_NONE,
MAP_PRIVATE | MAP_ANONYMOUS | MAP_NORESERVE,
kMmapFd, kMmapFdOffset) != MAP_FAILED;
}
class ThreadHandle::PlatformData : public Malloced {
public:
explicit PlatformData(ThreadHandle::Kind kind) {
Initialize(kind);
}
void Initialize(ThreadHandle::Kind kind) {
switch (kind) {
case ThreadHandle::SELF: thread_ = pthread_self(); break;
case ThreadHandle::INVALID: thread_ = kNoThread; break;
}
}
pthread_t thread_; // Thread handle for pthread.
};
ThreadHandle::ThreadHandle(Kind kind) {
data_ = new PlatformData(kind);
}
void ThreadHandle::Initialize(ThreadHandle::Kind kind) {
data_->Initialize(kind);
}
ThreadHandle::~ThreadHandle() {
delete data_;
}
bool ThreadHandle::IsSelf() const {
return pthread_equal(data_->thread_, pthread_self());
}
bool ThreadHandle::IsValid() const {
return data_->thread_ != kNoThread;
}
Thread::Thread() : ThreadHandle(ThreadHandle::INVALID) {
}
Thread::~Thread() {
}
static void* ThreadEntry(void* arg) {
Thread* thread = reinterpret_cast<Thread*>(arg);
// This is also initialized by the first argument to pthread_create() but we
// don't know which thread will run first (the original thread or the new
// one) so we initialize it here too.
thread->thread_handle_data()->thread_ = pthread_self();
ASSERT(thread->IsValid());
thread->Run();
return NULL;
}
void Thread::Start() {
pthread_create(&thread_handle_data()->thread_, NULL, ThreadEntry, this);
ASSERT(IsValid());
}
void Thread::Join() {
pthread_join(thread_handle_data()->thread_, NULL);
}
Thread::LocalStorageKey Thread::CreateThreadLocalKey() {
pthread_key_t key;
int result = pthread_key_create(&key, NULL);
USE(result);
ASSERT(result == 0);
return static_cast<LocalStorageKey>(key);
}
void Thread::DeleteThreadLocalKey(LocalStorageKey key) {
pthread_key_t pthread_key = static_cast<pthread_key_t>(key);
int result = pthread_key_delete(pthread_key);
USE(result);
ASSERT(result == 0);
}
void* Thread::GetThreadLocal(LocalStorageKey key) {
pthread_key_t pthread_key = static_cast<pthread_key_t>(key);
return pthread_getspecific(pthread_key);
}
void Thread::SetThreadLocal(LocalStorageKey key, void* value) {
pthread_key_t pthread_key = static_cast<pthread_key_t>(key);
pthread_setspecific(pthread_key, value);
}
void Thread::YieldCPU() {
sched_yield();
}
class LinuxMutex : public Mutex {
public:
LinuxMutex() {
pthread_mutexattr_t attrs;
int result = pthread_mutexattr_init(&attrs);
ASSERT(result == 0);
result = pthread_mutexattr_settype(&attrs, PTHREAD_MUTEX_RECURSIVE);
ASSERT(result == 0);
result = pthread_mutex_init(&mutex_, &attrs);
ASSERT(result == 0);
}
virtual ~LinuxMutex() { pthread_mutex_destroy(&mutex_); }
virtual int Lock() {
int result = pthread_mutex_lock(&mutex_);
return result;
}
virtual int Unlock() {
int result = pthread_mutex_unlock(&mutex_);
return result;
}
private:
pthread_mutex_t mutex_; // Pthread mutex for POSIX platforms.
};
Mutex* OS::CreateMutex() {
return new LinuxMutex();
}
class LinuxSemaphore : public Semaphore {
public:
explicit LinuxSemaphore(int count) { sem_init(&sem_, 0, count); }
virtual ~LinuxSemaphore() { sem_destroy(&sem_); }
virtual void Wait();
virtual bool Wait(int timeout);
virtual void Signal() { sem_post(&sem_); }
private:
sem_t sem_;
};
void LinuxSemaphore::Wait() {
while (true) {
int result = sem_wait(&sem_);
if (result == 0) return; // Successfully got semaphore.
CHECK(result == -1 && errno == EINTR); // Signal caused spurious wakeup.
}
}
#ifndef TIMEVAL_TO_TIMESPEC
#define TIMEVAL_TO_TIMESPEC(tv, ts) do { \
(ts)->tv_sec = (tv)->tv_sec; \
(ts)->tv_nsec = (tv)->tv_usec * 1000; \
} while (false)
#endif
bool LinuxSemaphore::Wait(int timeout) {
const long kOneSecondMicros = 1000000; // NOLINT
// Split timeout into second and nanosecond parts.
struct timeval delta;
delta.tv_usec = timeout % kOneSecondMicros;
delta.tv_sec = timeout / kOneSecondMicros;
struct timeval current_time;
// Get the current time.
if (gettimeofday(&current_time, NULL) == -1) {
return false;
}
// Calculate time for end of timeout.
struct timeval end_time;
timeradd(&current_time, &delta, &end_time);
struct timespec ts;
TIMEVAL_TO_TIMESPEC(&end_time, &ts);
// Wait for semaphore signalled or timeout.
while (true) {
int result = sem_timedwait(&sem_, &ts);
if (result == 0) return true; // Successfully got semaphore.
if (result > 0) {
// For glibc prior to 2.3.4 sem_timedwait returns the error instead of -1.
errno = result;
result = -1;
}
if (result == -1 && errno == ETIMEDOUT) return false; // Timeout.
CHECK(result == -1 && errno == EINTR); // Signal caused spurious wakeup.
}
}
Semaphore* OS::CreateSemaphore(int count) {
return new LinuxSemaphore(count);
}
#ifdef ENABLE_LOGGING_AND_PROFILING
static Sampler* active_sampler_ = NULL;
#if !defined(__GLIBC__) && (defined(__arm__) || defined(__thumb__))
// Android runs a fairly new Linux kernel, so signal info is there,
// but the C library doesn't have the structs defined.
struct sigcontext {
uint32_t trap_no;
uint32_t error_code;
uint32_t oldmask;
uint32_t gregs[16];
uint32_t arm_cpsr;
uint32_t fault_address;
};
typedef uint32_t __sigset_t;
typedef struct sigcontext mcontext_t;
typedef struct ucontext {
uint32_t uc_flags;
struct ucontext *uc_link;
stack_t uc_stack;
mcontext_t uc_mcontext;
__sigset_t uc_sigmask;
} ucontext_t;
enum ArmRegisters {R15 = 15, R13 = 13, R11 = 11};
#endif
static void ProfilerSignalHandler(int signal, siginfo_t* info, void* context) {
USE(info);
if (signal != SIGPROF) return;
if (active_sampler_ == NULL) return;
TickSample sample;
// If profiling, we extract the current pc and sp.
if (active_sampler_->IsProfiling()) {
// Extracting the sample from the context is extremely machine dependent.
ucontext_t* ucontext = reinterpret_cast<ucontext_t*>(context);
mcontext_t& mcontext = ucontext->uc_mcontext;
#if V8_HOST_ARCH_IA32
sample.pc = mcontext.gregs[REG_EIP];
sample.sp = mcontext.gregs[REG_ESP];
sample.fp = mcontext.gregs[REG_EBP];
#elif V8_HOST_ARCH_X64
sample.pc = mcontext.gregs[REG_RIP];
sample.sp = mcontext.gregs[REG_RSP];
sample.fp = mcontext.gregs[REG_RBP];
#elif V8_HOST_ARCH_ARM
// An undefined macro evaluates to 0, so this applies to Android's Bionic also.
#if (__GLIBC__ < 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ <= 3))
sample.pc = mcontext.gregs[R15];
sample.sp = mcontext.gregs[R13];
sample.fp = mcontext.gregs[R11];
#else
sample.pc = mcontext.arm_pc;
sample.sp = mcontext.arm_sp;
sample.fp = mcontext.arm_fp;
#endif
#endif
active_sampler_->SampleStack(&sample);
}
// We always sample the VM state.
sample.state = Logger::state();
active_sampler_->Tick(&sample);
}
class Sampler::PlatformData : public Malloced {
public:
PlatformData() {
signal_handler_installed_ = false;
}
bool signal_handler_installed_;
struct sigaction old_signal_handler_;
struct itimerval old_timer_value_;
};
Sampler::Sampler(int interval, bool profiling)
: interval_(interval), profiling_(profiling), active_(false) {
data_ = new PlatformData();
}
Sampler::~Sampler() {
delete data_;
}
void Sampler::Start() {
// There can only be one active sampler at the time on POSIX
// platforms.
if (active_sampler_ != NULL) return;
// Request profiling signals.
struct sigaction sa;
sa.sa_sigaction = ProfilerSignalHandler;
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_SIGINFO;
if (sigaction(SIGPROF, &sa, &data_->old_signal_handler_) != 0) return;
data_->signal_handler_installed_ = true;
// Set the itimer to generate a tick for each interval.
itimerval itimer;
itimer.it_interval.tv_sec = interval_ / 1000;
itimer.it_interval.tv_usec = (interval_ % 1000) * 1000;
itimer.it_value.tv_sec = itimer.it_interval.tv_sec;
itimer.it_value.tv_usec = itimer.it_interval.tv_usec;
setitimer(ITIMER_PROF, &itimer, &data_->old_timer_value_);
// Set this sampler as the active sampler.
active_sampler_ = this;
active_ = true;
}
void Sampler::Stop() {
// Restore old signal handler
if (data_->signal_handler_installed_) {
setitimer(ITIMER_PROF, &data_->old_timer_value_, NULL);
sigaction(SIGPROF, &data_->old_signal_handler_, 0);
data_->signal_handler_installed_ = false;
}
// This sampler is no longer the active sampler.
active_sampler_ = NULL;
active_ = false;
}
#endif // ENABLE_LOGGING_AND_PROFILING
} } // namespace v8::internal