RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-25 11:07:45 +00:00
Code Issues Activity Actions

Kernel: Move block condition evaluation out of the Scheduler

Browse source 
This makes the Scheduler a lot leaner by not having to evaluate
block conditions every time it is invoked. Instead evaluate them as
the states change, and unblock threads at that point.

This also implements some more waitid/waitpid/wait features and
behavior. For example, WUNTRACED and WNOWAIT are now supported. And
wait will now not return EINTR when SIGCHLD is delivered at the
same time.
This commit is contained in:
Tom 2020-11-29 16:05:27 -07:00 • committed by Andreas Kling
parent 6a620562cc
commit 046d6855f5
53 changed files with 2027 additions and 930 deletions
Download patch file Download diff file Expand all files Collapse all files

260
Kernel/FileSystem/Plan9FileSystem.cpp
View file

@ -36,6 +36,7 @@ NonnullRefPtr<Plan9FS> Plan9FS::create(FileDescription& file_description)
Plan9FS::Plan9FS(FileDescription& file_description)
: FileBackedFS(file_description)
, m_completion_blocker(*this)
{
}
@ -216,6 +217,8 @@ private:
bool Plan9FS::initialize()
{
ensure_thread();
Message version_message { *this, Message::Type::Tversion };
version_message << (u32)m_max_message_size << "9P2000.L";
@ -412,7 +415,97 @@ const KBuffer& Plan9FS::Message::build()
return m_built.buffer;
}
KResult Plan9FS::post_message(Message& message)
Plan9FS::ReceiveCompletion::ReceiveCompletion(u16 tag)
: tag(tag)
{
}
Plan9FS::ReceiveCompletion::~ReceiveCompletion()
{
}
bool Plan9FS::Blocker::unblock(u16 tag)
{
{
ScopedSpinLock lock(m_lock);
if (m_did_unblock)
return false;
m_did_unblock = true;
if (m_completion->tag != tag)
return false;
if (!m_completion->result.is_error())
m_message = move(*m_completion->message);
}
return unblock();
}
void Plan9FS::Blocker::not_blocking(bool)
{
{
ScopedSpinLock lock(m_lock);
if (m_did_unblock)
return;
}
m_fs.m_completion_blocker.try_unblock(*this);
}
bool Plan9FS::Blocker::is_completed() const
{
ScopedSpinLock lock(m_completion->lock);
return m_completion->completed;
}
bool Plan9FS::Plan9FSBlockCondition::should_add_blocker(Thread::Blocker& b, void*)
{
// NOTE: m_lock is held already!
auto& blocker = static_cast<Blocker&>(b);
return !blocker.is_completed();
}
void Plan9FS::Plan9FSBlockCondition::unblock_completed(u16 tag)
{
unblock([&](Thread::Blocker& b, void*) {
ASSERT(b.blocker_type() == Thread::Blocker::Type::Plan9FS);
auto& blocker = static_cast<Blocker&>(b);
return blocker.unblock(tag);
});
}
void Plan9FS::Plan9FSBlockCondition::unblock_all()
{
BlockCondition::unblock_all([&](Thread::Blocker& b, void*) {
ASSERT(b.blocker_type() == Thread::Blocker::Type::Plan9FS);
auto& blocker = static_cast<Blocker&>(b);
return blocker.unblock();
});
}
void Plan9FS::Plan9FSBlockCondition::try_unblock(Plan9FS::Blocker& blocker)
{
if (m_fs.is_complete(*blocker.completion())) {
ScopedSpinLock lock(m_lock);
blocker.unblock(blocker.completion()->tag);
}
}
bool Plan9FS::is_complete(const ReceiveCompletion& completion)
{
LOCKER(m_lock);
if (m_completions.contains(completion.tag)) {
// If it's still in the map then it can't be complete
ASSERT(!completion.completed);
return false;
}
// if it's not in the map anymore, it must be complete. But we MUST
// hold m_lock to be able to check completion.completed!
ASSERT(completion.completed);
return true;
}
KResult Plan9FS::post_message(Message& message, RefPtr<ReceiveCompletion> completion)
{
auto& buffer = message.build();
const u8* data = buffer.data();
@ -421,9 +514,21 @@ KResult Plan9FS::post_message(Message& message)
LOCKER(m_send_lock);
if (completion) {
// Save the completion record *before* we send the message. This
// ensures that it exists when the thread reads the response
LOCKER(m_lock);
auto tag = completion->tag;
m_completions.set(tag, completion.release_nonnull());
// TODO: What if there is a collision? Do we need to wait until
// the existing record with the tag completes before queueing
// this one?
}
while (size > 0) {
if (!description.can_write()) {
if (Thread::current()->block<Thread::WriteBlocker>(nullptr, description).was_interrupted())
auto unblock_flags = Thread::FileBlocker::BlockFlags::None;
if (Thread::current()->block<Thread::WriteBlocker>(nullptr, description, unblock_flags).was_interrupted())
return KResult(-EINTR);
}
auto data_buffer = UserOrKernelBuffer::for_kernel_buffer(const_cast<u8*>(data));
@ -443,7 +548,8 @@ KResult Plan9FS::do_read(u8* data, size_t size)
auto& description = file_description();
while (size > 0) {
if (!description.can_read()) {
if (Thread::current()->block<Thread::ReadBlocker>(nullptr, description).was_interrupted())
auto unblock_flags = Thread::FileBlocker::BlockFlags::None;
if (Thread::current()->block<Thread::ReadBlocker>(nullptr, description, unblock_flags).was_interrupted())
return KResult(-EINTR);
}
auto data_buffer = UserOrKernelBuffer::for_kernel_buffer(data);
@ -461,9 +567,6 @@ KResult Plan9FS::do_read(u8* data, size_t size)
KResult Plan9FS::read_and_dispatch_one_message()
{
ASSERT(m_someone_is_reading);
// That someone is us.
struct [[gnu::packed]] Header
{
u32 size;
@ -485,112 +588,42 @@ KResult Plan9FS::read_and_dispatch_one_message()
LOCKER(m_lock);
auto optional_completion = m_completions.get(header.tag);
if (!optional_completion.has_value()) {
if (m_tags_to_ignore.contains(header.tag)) {
m_tags_to_ignore.remove(header.tag);
} else {
dbg() << "Received a 9p message of type " << header.type << " with an unexpected tag " << header.tag << ", dropping";
}
return KSuccess;
if (optional_completion.has_value()) {
auto completion = optional_completion.value();
ScopedSpinLock lock(completion->lock);
completion->result = KSuccess;
completion->message = new Message { move(buffer) };
completion->completed = true;
m_completions.remove(header.tag);
m_completion_blocker.unblock_completed(header.tag);
} else {
dbg() << "Received a 9p message of type " << header.type << " with an unexpected tag " << header.tag << ", dropping";
}
ReceiveCompletion& completion = *optional_completion.value();
completion.result = KSuccess;
completion.message = Message { move(buffer) };
completion.completed = true;
m_completions.remove(header.tag);
return KSuccess;
}
bool Plan9FS::Blocker::should_unblock(Thread&)
{
if (m_completion.completed)
return true;
bool someone_else_is_reading = m_completion.fs.m_someone_is_reading.exchange(true);
if (!someone_else_is_reading) {
// We're gonna start reading ourselves; unblock.
return true;
}
return false;
}
KResult Plan9FS::wait_for_specific_message(u16 tag, Message& out_message)
{
KResult result = KSuccess;
ReceiveCompletion completion { *this, out_message, result, false };
{
LOCKER(m_lock);
m_completions.set(tag, &completion);
}
// Block until either:
// * Someone else reads the message we're waiting for, and hands it to us;
// * Or we become the one to read and dispatch messages.
if (Thread::current()->block<Plan9FS::Blocker>(nullptr, completion).was_interrupted()) {
LOCKER(m_lock);
m_completions.remove(tag);
return KResult(-EINTR);
}
// See for which reason we woke up.
if (completion.completed) {
// Somebody else completed it for us; nothing further to do.
return result;
}
while (!completion.completed && result.is_success()) {
result = read_and_dispatch_one_message();
}
if (result.is_error()) {
// If we fail to read, wake up everyone with an error.
LOCKER(m_lock);
for (auto& it : m_completions) {
it.value->result = result;
it.value->completed = true;
}
m_completions.clear();
}
// Wake up someone else, if anyone is interested...
m_someone_is_reading = false;
// ...and return.
return result;
}
KResult Plan9FS::post_message_and_explicitly_ignore_reply(Message& message)
{
auto tag = message.tag();
{
LOCKER(m_lock);
m_tags_to_ignore.set(tag);
}
auto result = post_message(message);
if (result.is_error()) {
LOCKER(m_lock);
m_tags_to_ignore.remove(tag);
}
return result;
return post_message(message, {});
}
KResult Plan9FS::post_message_and_wait_for_a_reply(Message& message, bool auto_convert_error_reply_to_error)
KResult Plan9FS::post_message_and_wait_for_a_reply(Message& message)
{
auto request_type = message.type();
auto tag = message.tag();
auto result = post_message(message);
if (result.is_error())
return result;
result = wait_for_specific_message(tag, message);
auto completion = adopt(*new ReceiveCompletion(tag));
auto result = post_message(message, completion);
if (result.is_error())
return result;
if (Thread::current()->block<Plan9FS::Blocker>(nullptr, *this, message, completion).was_interrupted())
return KResult(-EINTR);
if (!auto_convert_error_reply_to_error)
return KSuccess;
if (completion->result.is_error()) {
dbg() << "Plan9FS: Message was aborted with error " << completion->result;
return KResult(-EIO);
}
auto reply_type = message.type();
@ -624,6 +657,39 @@ ssize_t Plan9FS::adjust_buffer_size(ssize_t size) const
return min(size, max_size);
}
void Plan9FS::thread_main()
{
dbg() << "Plan9FS: Thread running";
do {
auto result = read_and_dispatch_one_message();
if (result.is_error()) {
// If we fail to read, wake up everyone with an error.
LOCKER(m_lock);
for (auto& it : m_completions) {
it.value->result = result;
it.value->completed = true;
}
m_completions.clear();
m_completion_blocker.unblock_all();
dbg() << "Plan9FS: Thread terminating, error reading";
return;
}
} while (!m_thread_shutdown.load(AK::MemoryOrder::memory_order_relaxed));
dbg() << "Plan9FS: Thread terminating";
}
void Plan9FS::ensure_thread()
{
ScopedSpinLock lock(m_thread_lock);
if (!m_thread_running.exchange(true, AK::MemoryOrder::memory_order_acq_rel)) {
Process::create_kernel_process(m_thread, "Plan9FS", [&]() {
thread_main();
m_thread_running.store(false, AK::MemoryOrder::memory_order_release);
});
}
}
Plan9FSInode::Plan9FSInode(Plan9FS& fs, u32 fid)
: Inode(fs, fid)
{