Kernel: Move some time related code from Scheduler into TimeManagement

Use the TimerQueue to expire blocking operations, which is one less thing
the Scheduler needs to check on every iteration.

Also, add a BlockTimeout class that will automatically handle relative or
absolute timeouts as well as overriding timeouts (e.g. socket timeouts)
more consistently.

Also, rework the TimerQueue class to be able to fire events from
any processor, which requires Timer to be RefCounted. Also allow
creating id-less timers for use by blocking operations.

1 files changed, 39 insertions, 19 deletions

diff --git a/Kernel/Thread.cpp b/Kernel/Thread.cpp
index 29536a2fed..fa433fdfbd 100644
--- a/Kernel/Thread.cpp
+++ b/Kernel/Thread.cpp
@@ -26,6 +26,7 @@
 
 #include <AK/Demangle.h>
 #include <AK/StringBuilder.h>
+#include <AK/Time.h>
 #include <Kernel/Arch/i386/CPU.h>
 #include <Kernel/FileSystem/FileDescription.h>
 #include <Kernel/KSyms.h>
@@ -247,15 +248,16 @@ void Thread::relock_process(bool did_unlock)
     Processor::current().restore_critical(prev_crit, prev_flags);
 }
 
-u64 Thread::sleep(u64 ticks)
+auto Thread::sleep(const timespec& duration, timespec* remaining_time) -> BlockResult
 {
     ASSERT(state() == Thread::Running);
-    u64 wakeup_time = g_uptime + ticks;
-    auto ret = Thread::current()->block<Thread::SleepBlocker>(nullptr, wakeup_time);
-    if (wakeup_time > g_uptime) {
-        ASSERT(ret.was_interrupted());
-    }
-    return wakeup_time;
+    return Thread::current()->block<Thread::SleepBlocker>(nullptr, Thread::BlockTimeout(false, &duration), remaining_time);
+}
+
+auto Thread::sleep_until(const timespec& deadline) -> BlockResult
+{
+    ASSERT(state() == Thread::Running);
+    return Thread::current()->block<Thread::SleepBlocker>(nullptr, Thread::BlockTimeout(true, &deadline));
 }
 
 const char* Thread::state_string() const
@@ -990,10 +992,12 @@ const LogStream& operator<<(const LogStream& stream, const Thread& value)
     return stream << value.process().name() << "(" << value.pid().value() << ":" << value.tid().value() << ")";
 }
 
-Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeval* timeout, Atomic<bool>* lock, RefPtr<Thread> beneficiary)
+Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, const BlockTimeout& timeout, Atomic<bool>* lock, RefPtr<Thread> beneficiary)
 {
     auto* current_thread = Thread::current();
-    TimerId timer_id {};
+    RefPtr<Timer> timer;
+    bool block_finished = false;
+    bool did_timeout = false;
     bool did_unlock;
 
     {
@@ -1004,6 +1008,25 @@ Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeva
         // we need to wait until the scheduler lock is released again
         {
             ScopedSpinLock sched_lock(g_scheduler_lock);
+            if (!timeout.is_infinite()) {
+                timer = TimerQueue::the().add_timer_without_id(timeout.absolute_time(), [&]() {
+                    // NOTE: this may execute on the same or any other processor!
+                    ScopedSpinLock lock(g_scheduler_lock);
+                    if (!block_finished) {
+                        did_timeout = true;
+                        wake_from_queue();
+                    }
+                });
+                if (!timer) {
+                    dbg() << "wait_on timed out before blocking";
+                    // We timed out already, don't block
+                    // The API contract guarantees we return with interrupts enabled,
+                    // regardless of how we got called
+                    critical.set_interrupt_flag_on_destruction(true);
+                    return BlockResult::InterruptedByTimeout;
+                }
+            }
+
             // m_queue can only be accessed safely if g_scheduler_lock is held!
             m_queue = &queue;
             if (!queue.enqueue(*current_thread)) {
@@ -1014,7 +1037,6 @@ Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeva
                 // The API contract guarantees we return with interrupts enabled,
                 // regardless of how we got called
                 critical.set_interrupt_flag_on_destruction(true);
-
                 return BlockResult::NotBlocked;
             }
 
@@ -1024,12 +1046,6 @@ Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeva
             set_state(State::Queued);
             m_wait_reason = reason;
 
-            if (timeout) {
-                timer_id = TimerQueue::the().add_timer(*timeout, [&]() {
-                    wake_from_queue();
-                });
-            }
-
             // Yield and wait for the queue to wake us up again.
             if (beneficiary)
                 Scheduler::donate_to(beneficiary, reason);
@@ -1058,6 +1074,7 @@ Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeva
         // To be able to look at m_wait_queue_node we once again need the
         // scheduler lock, which is held when we insert into the queue
         ScopedSpinLock sched_lock(g_scheduler_lock);
+        block_finished = true;
 
         if (m_queue) {
             ASSERT(m_queue == &queue);
@@ -1071,10 +1088,13 @@ Thread::BlockResult Thread::wait_on(WaitQueue& queue, const char* reason, timeva
             // In this case, the queue should not contain us anymore.
             result = BlockResult::InterruptedByDeath;
         }
+    }
 
-        // Make sure we cancel the timer if woke normally.
-        if (timeout && !result.was_interrupted())
-            TimerQueue::the().cancel_timer(timer_id);
+    if (timer && !did_timeout) {
+        // Cancel the timer while not holding any locks. This allows
+        // the timer function to complete before we remove it
+        // (e.g. if it's on another processor)
+        TimerQueue::the().cancel_timer(timer.release_nonnull());
     }
 
     // The API contract guarantees we return with interrupts enabled,