blob: f908c6270435911af90ab0b8d3c5f039b6c2fd38 [file] [log] [blame]
/*
* Copyright (C) 2015 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 "profile_saver.h"
#include <fcntl.h>
#include <sys/resource.h>
#include <sys/stat.h>
#include <sys/types.h>
#include "android-base/strings.h"
#include "art_method-inl.h"
#include "base/enums.h"
#include "base/logging.h" // For VLOG.
#include "base/scoped_arena_containers.h"
#include "base/stl_util.h"
#include "base/systrace.h"
#include "base/time_utils.h"
#include "class_table-inl.h"
#include "compiler_filter.h"
#include "dex/dex_file_loader.h"
#include "dex_reference_collection.h"
#include "gc/collector_type.h"
#include "gc/gc_cause.h"
#include "gc/scoped_gc_critical_section.h"
#include "jit/profiling_info.h"
#include "oat_file_manager.h"
#include "profile/profile_compilation_info.h"
#include "scoped_thread_state_change-inl.h"
namespace art {
ProfileSaver* ProfileSaver::instance_ = nullptr;
pthread_t ProfileSaver::profiler_pthread_ = 0U;
static_assert(ProfileCompilationInfo::kIndividualInlineCacheSize ==
InlineCache::kIndividualCacheSize,
"InlineCache and ProfileCompilationInfo do not agree on kIndividualCacheSize");
// At what priority to schedule the saver threads. 9 is the lowest foreground priority on device.
static constexpr int kProfileSaverPthreadPriority = 9;
static void SetProfileSaverThreadPriority(pthread_t thread, int priority) {
#if defined(ART_TARGET_ANDROID)
int result = setpriority(PRIO_PROCESS, pthread_gettid_np(thread), priority);
if (result != 0) {
LOG(ERROR) << "Failed to setpriority to :" << priority;
}
#else
UNUSED(thread);
UNUSED(priority);
#endif
}
static int GetDefaultThreadPriority() {
#if defined(ART_TARGET_ANDROID)
pthread_attr_t attr;
sched_param param;
pthread_attr_init(&attr);
pthread_attr_getschedparam(&attr, &param);
return param.sched_priority;
#else
return 0;
#endif
}
ProfileSaver::ProfileSaver(const ProfileSaverOptions& options,
const std::string& output_filename,
jit::JitCodeCache* jit_code_cache,
const std::vector<std::string>& code_paths)
: jit_code_cache_(jit_code_cache),
shutting_down_(false),
last_time_ns_saver_woke_up_(0),
jit_activity_notifications_(0),
wait_lock_("ProfileSaver wait lock"),
period_condition_("ProfileSaver period condition", wait_lock_),
total_bytes_written_(0),
total_number_of_writes_(0),
total_number_of_code_cache_queries_(0),
total_number_of_skipped_writes_(0),
total_number_of_failed_writes_(0),
total_ms_of_sleep_(0),
total_ns_of_work_(0),
max_number_of_profile_entries_cached_(0),
total_number_of_hot_spikes_(0),
total_number_of_wake_ups_(0),
options_(options) {
DCHECK(options_.IsEnabled());
AddTrackedLocations(output_filename, code_paths);
}
ProfileSaver::~ProfileSaver() {
for (auto& it : profile_cache_) {
delete it.second;
}
}
void ProfileSaver::Run() {
Thread* self = Thread::Current();
// Fetch the resolved classes for the app images after sleeping for
// options_.GetSaveResolvedClassesDelayMs().
// TODO(calin) This only considers the case of the primary profile file.
// Anything that gets loaded in the same VM will not have their resolved
// classes save (unless they started before the initial saving was done).
{
MutexLock mu(self, wait_lock_);
const uint64_t end_time = NanoTime() + MsToNs(options_.GetSaveResolvedClassesDelayMs());
while (true) {
const uint64_t current_time = NanoTime();
if (current_time >= end_time) {
break;
}
period_condition_.TimedWait(self, NsToMs(end_time - current_time), 0);
}
total_ms_of_sleep_ += options_.GetSaveResolvedClassesDelayMs();
}
FetchAndCacheResolvedClassesAndMethods(/*startup=*/ true);
// When we save without waiting for JIT notifications we use a simple
// exponential back off policy bounded by max_wait_without_jit.
uint32_t max_wait_without_jit = options_.GetMinSavePeriodMs() * 16;
uint64_t cur_wait_without_jit = options_.GetMinSavePeriodMs();
// Loop for the profiled methods.
while (!ShuttingDown(self)) {
uint64_t sleep_start = NanoTime();
{
uint64_t sleep_time = 0;
{
MutexLock mu(self, wait_lock_);
if (options_.GetWaitForJitNotificationsToSave()) {
period_condition_.Wait(self);
} else {
period_condition_.TimedWait(self, cur_wait_without_jit, 0);
if (cur_wait_without_jit < max_wait_without_jit) {
cur_wait_without_jit *= 2;
}
}
sleep_time = NanoTime() - sleep_start;
}
// Check if the thread was woken up for shutdown.
if (ShuttingDown(self)) {
break;
}
total_number_of_wake_ups_++;
// We might have been woken up by a huge number of notifications to guarantee saving.
// If we didn't meet the minimum saving period go back to sleep (only if missed by
// a reasonable margin).
uint64_t min_save_period_ns = MsToNs(options_.GetMinSavePeriodMs());
while (min_save_period_ns * 0.9 > sleep_time) {
{
MutexLock mu(self, wait_lock_);
period_condition_.TimedWait(self, NsToMs(min_save_period_ns - sleep_time), 0);
sleep_time = NanoTime() - sleep_start;
}
// Check if the thread was woken up for shutdown.
if (ShuttingDown(self)) {
break;
}
total_number_of_wake_ups_++;
}
}
total_ms_of_sleep_ += NsToMs(NanoTime() - sleep_start);
if (ShuttingDown(self)) {
break;
}
uint16_t number_of_new_methods = 0;
uint64_t start_work = NanoTime();
bool profile_saved_to_disk = ProcessProfilingInfo(/*force_save=*/false, &number_of_new_methods);
// Update the notification counter based on result. Note that there might be contention on this
// but we don't care about to be 100% precise.
if (!profile_saved_to_disk) {
// If we didn't save to disk it may be because we didn't have enough new methods.
// Set the jit activity notifications to number_of_new_methods so we can wake up earlier
// if needed.
jit_activity_notifications_ = number_of_new_methods;
}
total_ns_of_work_ += NanoTime() - start_work;
}
}
void ProfileSaver::NotifyJitActivity() {
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
if (instance_ == nullptr || instance_->shutting_down_) {
return;
}
instance_->NotifyJitActivityInternal();
}
void ProfileSaver::WakeUpSaver() {
jit_activity_notifications_ = 0;
last_time_ns_saver_woke_up_ = NanoTime();
period_condition_.Signal(Thread::Current());
}
void ProfileSaver::NotifyJitActivityInternal() {
// Unlikely to overflow but if it happens,
// we would have waken up the saver long before that.
jit_activity_notifications_++;
// Note that we are not as precise as we could be here but we don't want to wake the saver
// every time we see a hot method.
if (jit_activity_notifications_ > options_.GetMinNotificationBeforeWake()) {
MutexLock wait_mutex(Thread::Current(), wait_lock_);
if ((NanoTime() - last_time_ns_saver_woke_up_) > MsToNs(options_.GetMinSavePeriodMs())) {
WakeUpSaver();
} else if (jit_activity_notifications_ > options_.GetMaxNotificationBeforeWake()) {
// Make sure to wake up the saver if we see a spike in the number of notifications.
// This is a precaution to avoid losing a big number of methods in case
// this is a spike with no jit after.
total_number_of_hot_spikes_++;
WakeUpSaver();
}
}
}
class ScopedDefaultPriority {
public:
explicit ScopedDefaultPriority(pthread_t thread) : thread_(thread) {
SetProfileSaverThreadPriority(thread_, GetDefaultThreadPriority());
}
~ScopedDefaultPriority() {
SetProfileSaverThreadPriority(thread_, kProfileSaverPthreadPriority);
}
private:
const pthread_t thread_;
};
// GetClassLoadersVisitor takes a snapshot of the class loaders and stores them in the out
// class_loaders argument. Not affected by class unloading since there are no suspend points in
// the caller.
class GetClassLoadersVisitor : public ClassLoaderVisitor {
public:
explicit GetClassLoadersVisitor(VariableSizedHandleScope* hs,
std::vector<Handle<mirror::ClassLoader>>* class_loaders)
: hs_(hs),
class_loaders_(class_loaders) {}
void Visit(ObjPtr<mirror::ClassLoader> class_loader)
REQUIRES_SHARED(Locks::classlinker_classes_lock_, Locks::mutator_lock_) override {
class_loaders_->push_back(hs_->NewHandle(class_loader));
}
private:
VariableSizedHandleScope* const hs_;
std::vector<Handle<mirror::ClassLoader>>* const class_loaders_;
};
// GetClassesVisitor takes a snapshot of the loaded classes that we may want to visit and stores
// them in the out argument. Not affected by class unloading since there are no suspend points in
// the caller.
class GetClassesVisitor : public ClassVisitor {
public:
explicit GetClassesVisitor(bool profile_boot_class_path,
ScopedArenaVector<ObjPtr<mirror::Class>>* out)
: profile_boot_class_path_(profile_boot_class_path),
out_(out) {}
bool operator()(ObjPtr<mirror::Class> klass) override REQUIRES_SHARED(Locks::mutator_lock_) {
if (klass->IsProxyClass() ||
klass->IsArrayClass() ||
klass->IsPrimitive() ||
!klass->IsResolved() ||
klass->IsErroneousResolved() ||
(!profile_boot_class_path_ && klass->GetClassLoader() == nullptr)) {
return true;
}
out_->push_back(klass);
return true;
}
private:
const bool profile_boot_class_path_;
ScopedArenaVector<ObjPtr<mirror::Class>>* const out_;
};
using MethodReferenceCollection = DexReferenceCollection<uint16_t, ScopedArenaAllocatorAdapter>;
using TypeReferenceCollection = DexReferenceCollection<dex::TypeIndex,
ScopedArenaAllocatorAdapter>;
// Iterate over all of the loaded classes and visit each one. For each class, add it to the
// resolved_classes out argument if startup is true.
// Add methods to the hot_methods out argument if the number of samples is greater or equal to
// hot_method_sample_threshold, add it to sampled_methods if it has at least one sample.
static void SampleClassesAndExecutedMethods(pthread_t profiler_pthread,
bool profile_boot_class_path,
ScopedArenaAllocator* allocator,
uint32_t hot_method_sample_threshold,
bool startup,
TypeReferenceCollection* resolved_classes,
MethodReferenceCollection* hot_methods,
MethodReferenceCollection* sampled_methods) {
Thread* const self = Thread::Current();
ClassLinker* const class_linker = Runtime::Current()->GetClassLinker();
// Restore profile saver thread priority during the GC critical section. This helps prevent
// priority inversions blocking the GC for long periods of time.
std::unique_ptr<ScopedDefaultPriority> sdp;
// Only restore default priority if we are the profile saver thread. Other threads that call this
// are threads calling Stop and the signal catcher (for SIGUSR1).
if (pthread_self() == profiler_pthread) {
sdp.reset(new ScopedDefaultPriority(profiler_pthread));
}
// Do ScopedGCCriticalSection before acquiring mutator lock to prevent the GC running and
// blocking threads during thread root flipping. Since the GC is a background thread, blocking it
// is not a problem.
ScopedObjectAccess soa(self);
gc::ScopedGCCriticalSection sgcs(self,
gc::kGcCauseProfileSaver,
gc::kCollectorTypeCriticalSection);
VariableSizedHandleScope hs(soa.Self());
std::vector<Handle<mirror::ClassLoader>> class_loaders;
if (profile_boot_class_path) {
// First add the boot class loader since visit classloaders doesn't visit it.
class_loaders.push_back(hs.NewHandle<mirror::ClassLoader>(nullptr));
}
GetClassLoadersVisitor class_loader_visitor(&hs, &class_loaders);
{
// Read the class loaders into a temporary array to prevent contention problems on the
// class_linker_classes_lock.
ScopedTrace trace2("Get class loaders");
ReaderMutexLock mu(soa.Self(), *Locks::classlinker_classes_lock_);
class_linker->VisitClassLoaders(&class_loader_visitor);
}
ScopedArenaVector<ObjPtr<mirror::Class>> classes(allocator->Adapter());
for (Handle<mirror::ClassLoader> class_loader : class_loaders) {
ClassTable* table = class_linker->ClassTableForClassLoader(class_loader.Get());
if (table == nullptr) {
// If the class loader has not loaded any classes, it may have a null table.
continue;
}
GetClassesVisitor get_classes_visitor(profile_boot_class_path, &classes);
{
// Collect the classes into a temporary array to prevent lock contention on the class
// table lock. We want to avoid blocking class loading in other threads as much as
// possible.
ScopedTrace trace3("Visiting class table");
table->Visit(get_classes_visitor);
}
for (ObjPtr<mirror::Class> klass : classes) {
if (startup) {
// We only record classes for the startup case. This may change in the future.
resolved_classes->AddReference(&klass->GetDexFile(), klass->GetDexTypeIndex());
}
// Visit all of the methods in the class to see which ones were executed.
for (ArtMethod& method : klass->GetMethods(kRuntimePointerSize)) {
if (!method.IsNative() && !method.IsAbstract()) {
DCHECK(!method.IsProxyMethod());
const uint16_t counter = method.GetCounter();
// Mark startup methods as hot if they have more than hot_method_sample_threshold
// samples. This means they will get compiled by the compiler driver.
if (method.GetProfilingInfo(kRuntimePointerSize) != nullptr ||
method.PreviouslyWarm() ||
counter >= hot_method_sample_threshold) {
hot_methods->AddReference(method.GetDexFile(), method.GetDexMethodIndex());
} else if (counter != 0) {
sampled_methods->AddReference(method.GetDexFile(), method.GetDexMethodIndex());
}
} else {
// We do not record native methods. Once we AOT-compile the app, all native
// methods shall have their thunks compiled.
}
}
}
classes.clear();
}
}
void ProfileSaver::FetchAndCacheResolvedClassesAndMethods(bool startup) {
ScopedTrace trace(__PRETTY_FUNCTION__);
const uint64_t start_time = NanoTime();
// Resolve any new registered locations.
ResolveTrackedLocations();
Thread* const self = Thread::Current();
Runtime* const runtime = Runtime::Current();
ArenaStack stack(runtime->GetArenaPool());
ScopedArenaAllocator allocator(&stack);
MethodReferenceCollection hot_methods(allocator.Adapter(), allocator.Adapter());
MethodReferenceCollection sampled_methods(allocator.Adapter(), allocator.Adapter());
TypeReferenceCollection resolved_classes(allocator.Adapter(), allocator.Adapter());
const bool is_low_ram = Runtime::Current()->GetHeap()->IsLowMemoryMode();
pthread_t profiler_pthread;
{
MutexLock mu(self, *Locks::profiler_lock_);
profiler_pthread = profiler_pthread_;
}
const uint32_t hot_method_sample_threshold = startup ?
options_.GetHotStartupMethodSamples(is_low_ram) :
std::numeric_limits<uint32_t>::max();
SampleClassesAndExecutedMethods(profiler_pthread,
options_.GetProfileBootClassPath(),
&allocator,
hot_method_sample_threshold,
startup,
&resolved_classes,
&hot_methods,
&sampled_methods);
MutexLock mu(self, *Locks::profiler_lock_);
uint64_t total_number_of_profile_entries_cached = 0;
using Hotness = ProfileCompilationInfo::MethodHotness;
for (const auto& it : tracked_dex_base_locations_) {
std::set<DexCacheResolvedClasses> resolved_classes_for_location;
const std::string& filename = it.first;
auto info_it = profile_cache_.find(filename);
if (info_it == profile_cache_.end()) {
info_it = profile_cache_.Put(
filename,
new ProfileCompilationInfo(Runtime::Current()->GetArenaPool()));
}
ProfileCompilationInfo* cached_info = info_it->second;
const std::set<std::string>& locations = it.second;
VLOG(profiler) << "Locations for " << it.first << " " << android::base::Join(locations, ':');
for (const auto& pair : hot_methods.GetMap()) {
const DexFile* const dex_file = pair.first;
const std::string base_location = DexFileLoader::GetBaseLocation(dex_file->GetLocation());
const MethodReferenceCollection::IndexVector& indices = pair.second;
VLOG(profiler) << "Location " << dex_file->GetLocation()
<< " base_location=" << base_location
<< " found=" << (locations.find(base_location) != locations.end())
<< " indices size=" << indices.size();
if (locations.find(base_location) != locations.end()) {
uint8_t flags = Hotness::kFlagHot;
flags |= startup ? Hotness::kFlagStartup : Hotness::kFlagPostStartup;
cached_info->AddMethodsForDex(
static_cast<Hotness::Flag>(flags),
dex_file,
indices.begin(),
indices.end());
}
}
for (const auto& pair : sampled_methods.GetMap()) {
const DexFile* const dex_file = pair.first;
const std::string base_location = DexFileLoader::GetBaseLocation(dex_file->GetLocation());
const MethodReferenceCollection::IndexVector& indices = pair.second;
VLOG(profiler) << "Location " << base_location
<< " found=" << (locations.find(base_location) != locations.end())
<< " indices size=" << indices.size();
if (locations.find(base_location) != locations.end()) {
cached_info->AddMethodsForDex(startup ? Hotness::kFlagStartup : Hotness::kFlagPostStartup,
dex_file,
indices.begin(),
indices.end());
}
}
for (const auto& pair : resolved_classes.GetMap()) {
const DexFile* const dex_file = pair.first;
const std::string base_location = DexFileLoader::GetBaseLocation(dex_file->GetLocation());
if (locations.find(base_location) != locations.end()) {
const TypeReferenceCollection::IndexVector& classes = pair.second;
VLOG(profiler) << "Added " << classes.size() << " classes for location "
<< base_location
<< " (" << dex_file->GetLocation() << ")";
cached_info->AddClassesForDex(dex_file, classes.begin(), classes.end());
} else {
VLOG(profiler) << "Location not found " << base_location;
}
}
total_number_of_profile_entries_cached += resolved_classes_for_location.size();
}
max_number_of_profile_entries_cached_ = std::max(
max_number_of_profile_entries_cached_,
total_number_of_profile_entries_cached);
VLOG(profiler) << "Profile saver recorded " << hot_methods.NumReferences() << " hot methods and "
<< sampled_methods.NumReferences() << " sampled methods with threshold "
<< hot_method_sample_threshold << " in "
<< PrettyDuration(NanoTime() - start_time);
}
bool ProfileSaver::ProcessProfilingInfo(bool force_save, /*out*/uint16_t* number_of_new_methods) {
ScopedTrace trace(__PRETTY_FUNCTION__);
// Resolve any new registered locations.
ResolveTrackedLocations();
SafeMap<std::string, std::set<std::string>> tracked_locations;
{
// Make a copy so that we don't hold the lock while doing I/O.
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
tracked_locations = tracked_dex_base_locations_;
}
bool profile_file_saved = false;
if (number_of_new_methods != nullptr) {
*number_of_new_methods = 0;
}
// We only need to do this once, not once per dex location.
// TODO: Figure out a way to only do it when stuff has changed? It takes 30-50ms.
FetchAndCacheResolvedClassesAndMethods(/*startup=*/ false);
for (const auto& it : tracked_locations) {
if (!force_save && ShuttingDown(Thread::Current())) {
// The ProfileSaver is in shutdown mode, meaning a stop request was made and
// we need to exit cleanly (by waiting for the saver thread to finish). Unless
// we have a request for a forced save, do not do any processing so that we
// speed up the exit.
return true;
}
const std::string& filename = it.first;
const std::set<std::string>& locations = it.second;
VLOG(profiler) << "Tracked filename " << filename << " locations "
<< android::base::Join(locations, ":");
std::vector<ProfileMethodInfo> profile_methods;
{
ScopedObjectAccess soa(Thread::Current());
jit_code_cache_->GetProfiledMethods(locations, profile_methods);
total_number_of_code_cache_queries_++;
}
{
ProfileCompilationInfo info(Runtime::Current()->GetArenaPool());
if (!info.Load(filename, /*clear_if_invalid=*/ true)) {
LOG(WARNING) << "Could not forcefully load profile " << filename;
continue;
}
uint64_t last_save_number_of_methods = info.GetNumberOfMethods();
uint64_t last_save_number_of_classes = info.GetNumberOfResolvedClasses();
VLOG(profiler) << "last_save_number_of_methods=" << last_save_number_of_methods
<< " last_save_number_of_classes=" << last_save_number_of_classes
<< " number of profiled methods=" << profile_methods.size();
// Try to add the method data. Note this may fail is the profile loaded from disk contains
// outdated data (e.g. the previous profiled dex files might have been updated).
// If this happens we clear the profile data and for the save to ensure the file is cleared.
if (!info.AddMethods(profile_methods,
ProfileCompilationInfo::MethodHotness::kFlagPostStartup)) {
LOG(WARNING) << "Could not add methods to the existing profiler. "
<< "Clearing the profile data.";
info.ClearData();
force_save = true;
}
auto profile_cache_it = profile_cache_.find(filename);
if (profile_cache_it != profile_cache_.end()) {
if (!info.MergeWith(*(profile_cache_it->second))) {
LOG(WARNING) << "Could not merge the profile. Clearing the profile data.";
info.ClearData();
force_save = true;
}
} else if (VLOG_IS_ON(profiler)) {
LOG(INFO) << "Failed to find cached profile for " << filename;
for (auto&& pair : profile_cache_) {
LOG(INFO) << "Cached profile " << pair.first;
}
}
int64_t delta_number_of_methods =
info.GetNumberOfMethods() - last_save_number_of_methods;
int64_t delta_number_of_classes =
info.GetNumberOfResolvedClasses() - last_save_number_of_classes;
if (!force_save &&
delta_number_of_methods < options_.GetMinMethodsToSave() &&
delta_number_of_classes < options_.GetMinClassesToSave()) {
VLOG(profiler) << "Not enough information to save to: " << filename
<< " Number of methods: " << delta_number_of_methods
<< " Number of classes: " << delta_number_of_classes;
total_number_of_skipped_writes_++;
continue;
}
if (number_of_new_methods != nullptr) {
*number_of_new_methods =
std::max(static_cast<uint16_t>(delta_number_of_methods),
*number_of_new_methods);
}
uint64_t bytes_written;
// Force the save. In case the profile data is corrupted or the the profile
// has the wrong version this will "fix" the file to the correct format.
if (info.Save(filename, &bytes_written)) {
// We managed to save the profile. Clear the cache stored during startup.
if (profile_cache_it != profile_cache_.end()) {
ProfileCompilationInfo *cached_info = profile_cache_it->second;
profile_cache_.erase(profile_cache_it);
delete cached_info;
}
if (bytes_written > 0) {
total_number_of_writes_++;
total_bytes_written_ += bytes_written;
profile_file_saved = true;
} else {
// At this point we could still have avoided the write.
// We load and merge the data from the file lazily at its first ever
// save attempt. So, whatever we are trying to save could already be
// in the file.
total_number_of_skipped_writes_++;
}
} else {
LOG(WARNING) << "Could not save profiling info to " << filename;
total_number_of_failed_writes_++;
}
}
}
// Trim the maps to madvise the pages used for profile info.
// It is unlikely we will need them again in the near feature.
Runtime::Current()->GetArenaPool()->TrimMaps();
return profile_file_saved;
}
void* ProfileSaver::RunProfileSaverThread(void* arg) {
Runtime* runtime = Runtime::Current();
bool attached = runtime->AttachCurrentThread("Profile Saver",
/*as_daemon=*/true,
runtime->GetSystemThreadGroup(),
/*create_peer=*/true);
if (!attached) {
CHECK(runtime->IsShuttingDown(Thread::Current()));
return nullptr;
}
ProfileSaver* profile_saver = reinterpret_cast<ProfileSaver*>(arg);
profile_saver->Run();
runtime->DetachCurrentThread();
VLOG(profiler) << "Profile saver shutdown";
return nullptr;
}
static bool ShouldProfileLocation(const std::string& location, bool profile_aot_code) {
if (profile_aot_code) {
// If we have to profile all the code, irrespective of its compilation state, return true
// right away.
return true;
}
OatFileManager& oat_manager = Runtime::Current()->GetOatFileManager();
const OatFile* oat_file = oat_manager.FindOpenedOatFileFromDexLocation(location);
if (oat_file == nullptr) {
// This can happen if we fallback to run code directly from the APK.
// Profile it with the hope that the background dexopt will get us back into
// a good state.
VLOG(profiler) << "Asked to profile a location without an oat file:" << location;
return true;
}
CompilerFilter::Filter filter = oat_file->GetCompilerFilter();
if ((filter == CompilerFilter::kSpeed) || (filter == CompilerFilter::kEverything)) {
VLOG(profiler)
<< "Skip profiling oat file because it's already speed|everything compiled: "
<< location << " oat location: " << oat_file->GetLocation();
return false;
}
return true;
}
void ProfileSaver::Start(const ProfileSaverOptions& options,
const std::string& output_filename,
jit::JitCodeCache* jit_code_cache,
const std::vector<std::string>& code_paths) {
Runtime* const runtime = Runtime::Current();
DCHECK(options.IsEnabled());
DCHECK(runtime->GetJit() != nullptr);
DCHECK(!output_filename.empty());
DCHECK(jit_code_cache != nullptr);
std::vector<std::string> code_paths_to_profile;
for (const std::string& location : code_paths) {
if (ShouldProfileLocation(location, options.GetProfileAOTCode())) {
VLOG(profiler) << "Code path to profile " << location;
code_paths_to_profile.push_back(location);
}
}
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
// Support getting profile samples for the boot class path. This will be used to generate the boot
// image profile. The intention is to use this code to generate to boot image but not use it in
// production. b/37966211
if (options.GetProfileBootClassPath()) {
std::set<std::string> code_paths_keys;
for (const std::string& location : code_paths) {
code_paths_keys.insert(ProfileCompilationInfo::GetProfileDexFileKey(location));
}
for (const DexFile* dex_file : runtime->GetClassLinker()->GetBootClassPath()) {
// Don't check ShouldProfileLocation since the boot class path may be speed compiled.
const std::string& location = dex_file->GetLocation();
const std::string key = ProfileCompilationInfo::GetProfileDexFileKey(location);
VLOG(profiler) << "Registering boot dex file " << location;
if (code_paths_keys.find(key) != code_paths_keys.end()) {
LOG(WARNING) << "Boot class path location key conflicts with code path " << location;
} else if (instance_ == nullptr) {
// Only add the boot class path once since Start may be called multiple times for secondary
// dexes.
// We still do the collision check above. This handles any secondary dexes that conflict
// with the boot class path dex files.
code_paths_to_profile.push_back(location);
}
}
}
if (code_paths_to_profile.empty()) {
VLOG(profiler) << "No code paths should be profiled.";
return;
}
if (instance_ != nullptr) {
// If we already have an instance, make sure it uses the same jit_code_cache.
// This may be called multiple times via Runtime::registerAppInfo (e.g. for
// apps which share the same runtime).
DCHECK_EQ(instance_->jit_code_cache_, jit_code_cache);
// Add the code_paths to the tracked locations.
instance_->AddTrackedLocations(output_filename, code_paths_to_profile);
return;
}
VLOG(profiler) << "Starting profile saver using output file: " << output_filename
<< ". Tracking: " << android::base::Join(code_paths_to_profile, ':');
instance_ = new ProfileSaver(options,
output_filename,
jit_code_cache,
code_paths_to_profile);
// Create a new thread which does the saving.
CHECK_PTHREAD_CALL(
pthread_create,
(&profiler_pthread_, nullptr, &RunProfileSaverThread, reinterpret_cast<void*>(instance_)),
"Profile saver thread");
SetProfileSaverThreadPriority(profiler_pthread_, kProfileSaverPthreadPriority);
}
void ProfileSaver::Stop(bool dump_info) {
ProfileSaver* profile_saver = nullptr;
pthread_t profiler_pthread = 0U;
{
MutexLock profiler_mutex(Thread::Current(), *Locks::profiler_lock_);
VLOG(profiler) << "Stopping profile saver thread";
profile_saver = instance_;
profiler_pthread = profiler_pthread_;
if (instance_ == nullptr) {
DCHECK(false) << "Tried to stop a profile saver which was not started";
return;
}
if (instance_->shutting_down_) {
DCHECK(false) << "Tried to stop the profile saver twice";
return;
}
instance_->shutting_down_ = true;
}
{
// Wake up the saver thread if it is sleeping to allow for a clean exit.
MutexLock wait_mutex(Thread::Current(), profile_saver->wait_lock_);
profile_saver->period_condition_.Signal(Thread::Current());
}
// Force save everything before destroying the thread since we want profiler_pthread_ to remain
// valid.
instance_->ProcessProfilingInfo(/*force_save=*/true, /*number_of_new_methods=*/nullptr);
// Wait for the saver thread to stop.
CHECK_PTHREAD_CALL(pthread_join, (profiler_pthread, nullptr), "profile saver thread shutdown");
{
MutexLock profiler_mutex(Thread::Current(), *Locks::profiler_lock_);
if (dump_info) {
instance_->DumpInfo(LOG_STREAM(INFO));
}
instance_ = nullptr;
profiler_pthread_ = 0U;
}
delete profile_saver;
}
bool ProfileSaver::ShuttingDown(Thread* self) {
MutexLock mu(self, *Locks::profiler_lock_);
return shutting_down_;
}
bool ProfileSaver::IsStarted() {
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
return instance_ != nullptr;
}
static void AddTrackedLocationsToMap(const std::string& output_filename,
const std::vector<std::string>& code_paths,
SafeMap<std::string, std::set<std::string>>* map) {
std::vector<std::string> code_paths_and_filenames;
// The dex locations are sometimes set to the filename instead of the full path.
// So make sure we have both "locations" when tracking what needs to be profiled.
// - apps + system server have filenames
// - boot classpath elements have full paths
// TODO(calin, ngeoffray, vmarko) This is an workaround for using filanames as
// dex locations - needed to prebuilt with a partial boot image
// (commit: c4a924d8c74241057d957d360bf31cd5cd0e4f9c).
// We should find a better way which allows us to do the tracking based on full paths.
for (const std::string& path : code_paths) {
size_t last_sep_index = path.find_last_of('/');
if (last_sep_index == path.size() - 1) {
// Should not happen, but anyone can register code paths so better be prepared and ignore
// such locations.
continue;
}
std::string filename = last_sep_index == std::string::npos
? path
: path.substr(last_sep_index + 1);
code_paths_and_filenames.push_back(path);
code_paths_and_filenames.push_back(filename);
}
auto it = map->find(output_filename);
if (it == map->end()) {
map->Put(
output_filename,
std::set<std::string>(code_paths_and_filenames.begin(), code_paths_and_filenames.end()));
} else {
it->second.insert(code_paths_and_filenames.begin(), code_paths_and_filenames.end());
}
}
void ProfileSaver::AddTrackedLocations(const std::string& output_filename,
const std::vector<std::string>& code_paths) {
// Add the code paths to the list of tracked location.
AddTrackedLocationsToMap(output_filename, code_paths, &tracked_dex_base_locations_);
// The code paths may contain symlinks which could fool the profiler.
// If the dex file is compiled with an absolute location but loaded with symlink
// the profiler could skip the dex due to location mismatch.
// To avoid this, we add the code paths to the temporary cache of 'to_be_resolved'
// locations. When the profiler thread executes we will resolve the paths to their
// real paths.
// Note that we delay taking the realpath to avoid spending more time than needed
// when registering location (as it is done during app launch).
AddTrackedLocationsToMap(output_filename,
code_paths,
&tracked_dex_base_locations_to_be_resolved_);
}
void ProfileSaver::DumpInstanceInfo(std::ostream& os) {
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
if (instance_ != nullptr) {
instance_->DumpInfo(os);
}
}
void ProfileSaver::DumpInfo(std::ostream& os) {
os << "ProfileSaver total_bytes_written=" << total_bytes_written_ << '\n'
<< "ProfileSaver total_number_of_writes=" << total_number_of_writes_ << '\n'
<< "ProfileSaver total_number_of_code_cache_queries="
<< total_number_of_code_cache_queries_ << '\n'
<< "ProfileSaver total_number_of_skipped_writes=" << total_number_of_skipped_writes_ << '\n'
<< "ProfileSaver total_number_of_failed_writes=" << total_number_of_failed_writes_ << '\n'
<< "ProfileSaver total_ms_of_sleep=" << total_ms_of_sleep_ << '\n'
<< "ProfileSaver total_ms_of_work=" << NsToMs(total_ns_of_work_) << '\n'
<< "ProfileSaver max_number_profile_entries_cached="
<< max_number_of_profile_entries_cached_ << '\n'
<< "ProfileSaver total_number_of_hot_spikes=" << total_number_of_hot_spikes_ << '\n'
<< "ProfileSaver total_number_of_wake_ups=" << total_number_of_wake_ups_ << '\n';
}
void ProfileSaver::ForceProcessProfiles() {
ProfileSaver* saver = nullptr;
{
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
saver = instance_;
}
// TODO(calin): this is not actually thread safe as the instance_ may have been deleted,
// but we only use this in testing when we now this won't happen.
// Refactor the way we handle the instance so that we don't end up in this situation.
if (saver != nullptr) {
saver->ProcessProfilingInfo(/*force_save=*/true, /*number_of_new_methods=*/nullptr);
}
}
bool ProfileSaver::HasSeenMethod(const std::string& profile, bool hot, MethodReference ref) {
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
if (instance_ != nullptr) {
ProfileCompilationInfo info(Runtime::Current()->GetArenaPool());
if (!info.Load(profile, /*clear_if_invalid=*/false)) {
return false;
}
ProfileCompilationInfo::MethodHotness hotness = info.GetMethodHotness(ref);
// Ignore hot parameter for now since it was causing test 595 to be flaky. TODO: Investigate.
// b/63635729
UNUSED(hot);
return hotness.IsInProfile();
}
return false;
}
void ProfileSaver::ResolveTrackedLocations() {
SafeMap<std::string, std::set<std::string>> locations_to_be_resolved;
{
// Make a copy so that we don't hold the lock while doing I/O.
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
locations_to_be_resolved = tracked_dex_base_locations_to_be_resolved_;
tracked_dex_base_locations_to_be_resolved_.clear();
}
// Resolve the locations.
SafeMap<std::string, std::vector<std::string>> resolved_locations_map;
for (const auto& it : locations_to_be_resolved) {
const std::string& filename = it.first;
const std::set<std::string>& locations = it.second;
auto resolved_locations_it = resolved_locations_map.Put(
filename,
std::vector<std::string>(locations.size()));
for (const auto& location : locations) {
UniqueCPtr<const char[]> location_real(realpath(location.c_str(), nullptr));
// Note that it's ok if we cannot get the real path.
if (location_real != nullptr) {
resolved_locations_it->second.emplace_back(location_real.get());
}
}
}
// Add the resolved locations to the tracked collection.
MutexLock mu(Thread::Current(), *Locks::profiler_lock_);
for (const auto& it : resolved_locations_map) {
AddTrackedLocationsToMap(it.first, it.second, &tracked_dex_base_locations_);
}
}
} // namespace art