Threading: Reworked the way our scheduler works.

Threads will now be awakened when the objects they're waiting on are signaled, instead of repeating the WaitSynchronization call every now and then.

The scheduler is now called once after every SVC call, and once after a thread is awakened from sleep by its timeout callback.

This new implementation is based off reverse-engineering of the real kernel.

See https://gist.github.com/Subv/02f29bd9f1e5deb7aceea1e8f019c8f4 for a more detailed description of how the real kernel handles rescheduling.

6 files changed, 76 insertions, 111 deletions

diff --git a/src/core/hle/kernel/address_arbiter.cpp b/src/core/hle/kernel/address_arbiter.cpp
index 37eec4c84..b5a0cc3a3 100644
--- a/src/core/hle/kernel/address_arbiter.cpp
+++ b/src/core/hle/kernel/address_arbiter.cpp
@@ -79,8 +79,6 @@ ResultCode AddressArbiter::ArbitrateAddress(ArbitrationType type, VAddr address,
                           ErrorSummary::WrongArgument, ErrorLevel::Usage);
     }
 
-    HLE::Reschedule(__func__);
-
     // The calls that use a timeout seem to always return a Timeout error even if they did not put
     // the thread to sleep
     if (type == ArbitrationType::WaitIfLessThanWithTimeout ||
diff --git a/src/core/hle/kernel/kernel.cpp b/src/core/hle/kernel/kernel.cpp
index 0c8752670..be7a5a6d8 100644
--- a/src/core/hle/kernel/kernel.cpp
+++ b/src/core/hle/kernel/kernel.cpp
@@ -31,13 +31,62 @@ void WaitObject::RemoveWaitingThread(Thread* thread) {
         waiting_threads.erase(itr);
 }
 
-void WaitObject::WakeupAllWaitingThreads() {
-    for (auto thread : waiting_threads)
-        thread->ResumeFromWait();
+SharedPtr<Thread> WaitObject::GetHighestPriorityReadyThread() {
+    // Remove the threads that are ready or already running from our waitlist
+    waiting_threads.erase(std::remove_if(waiting_threads.begin(), waiting_threads.end(), [](SharedPtr<Thread> thread) -> bool {
+        return thread->status == THREADSTATUS_RUNNING || thread->status == THREADSTATUS_READY;
+    }), waiting_threads.end());
+
+    if (waiting_threads.empty())
+        return nullptr;
 
-    waiting_threads.clear();
+    auto candidate_threads = waiting_threads;
 
-    HLE::Reschedule(__func__);
+    // Eliminate all threads that are waiting on more than one object, and not all of them are ready
+    candidate_threads.erase(std::remove_if(candidate_threads.begin(), candidate_threads.end(), [](SharedPtr<Thread> thread) -> bool {
+        for (auto object : thread->wait_objects)
+            if (object->ShouldWait())
+                return true;
+        return false;
+    }), candidate_threads.end());
+
+    // Return the thread with the lowest priority value (The one with the highest priority)
+    auto thread_itr = std::min_element(candidate_threads.begin(), candidate_threads.end(), [](const SharedPtr<Thread>& lhs, const SharedPtr<Thread>& rhs) {
+        return lhs->current_priority < rhs->current_priority;
+    });
+
+    if (thread_itr == candidate_threads.end())
+        return nullptr;
+
+    return *thread_itr;
+}
+
+void WaitObject::WakeupAllWaitingThreads() {
+    // Wake up all threads that can be awoken, in priority order
+    while (auto thread = GetHighestPriorityReadyThread()) {
+        if (thread->wait_objects.empty()) {
+            Acquire();
+            // Set the output index of the WaitSynchronizationN call to the index of this object.
+            if (thread->wait_set_output) {
+                thread->SetWaitSynchronizationOutput(thread->GetWaitObjectIndex(this));
+                thread->wait_set_output = false;
+            }
+        } else {
+            for (auto object : thread->wait_objects) {
+                object->Acquire();
+                // Remove the thread from the object's waitlist
+                object->RemoveWaitingThread(thread.get());
+            }
+            // Note: This case doesn't update the output index of WaitSynchronizationN.
+            // Clear the thread's waitlist
+            thread->wait_objects.clear();
+        }
+
+        // Set the result of the call to WaitSynchronization to RESULT_SUCCESS
+        thread->SetWaitSynchronizationResult(RESULT_SUCCESS);
+        thread->ResumeFromWait();
+        // Note: Removing the thread from the object's waitlist will be done by GetHighestPriorityReadyThread
+    }
 }
 
 const std::vector<SharedPtr<Thread>>& WaitObject::GetWaitingThreads() const {
diff --git a/src/core/hle/kernel/kernel.h b/src/core/hle/kernel/kernel.h
index 231cf7b75..eb5a3bf7e 100644
--- a/src/core/hle/kernel/kernel.h
+++ b/src/core/hle/kernel/kernel.h
@@ -155,6 +155,9 @@ public:
     /// Wake up all threads waiting on this object
     void WakeupAllWaitingThreads();
 
+    /// Obtains the highest priority thread that is ready to run from this object's waiting list.
+    SharedPtr<Thread> GetHighestPriorityReadyThread();
+
     /// Get a const reference to the waiting threads list for debug use
     const std::vector<SharedPtr<Thread>>& GetWaitingThreads() const;
 
diff --git a/src/core/hle/kernel/thread.cpp b/src/core/hle/kernel/thread.cpp
index 84d6d24c6..49ed9d899 100644
--- a/src/core/hle/kernel/thread.cpp
+++ b/src/core/hle/kernel/thread.cpp
@@ -120,8 +120,6 @@ void Thread::Stop() {
     u32 tls_slot =
         ((tls_address - Memory::TLS_AREA_VADDR) % Memory::PAGE_SIZE) / Memory::TLS_ENTRY_SIZE;
     Kernel::g_current_process->tls_slots[tls_page].reset(tls_slot);
-
-    HLE::Reschedule(__func__);
 }
 
 Thread* ArbitrateHighestPriorityThread(u32 address) {
@@ -181,50 +179,6 @@ static void PriorityBoostStarvedThreads() {
 }
 
 /**
- * Gets the registers for timeout parameter of the next WaitSynchronization call.
- * @param thread a pointer to the thread that is ready to call WaitSynchronization
- * @returns a tuple of two register pointers to low and high part of the timeout parameter
- */
-static std::tuple<u32*, u32*> GetWaitSynchTimeoutParameterRegister(Thread* thread) {
-    bool thumb_mode = (thread->context.cpsr & TBIT) != 0;
-    u16 thumb_inst = Memory::Read16(thread->context.pc & 0xFFFFFFFE);
-    u32 inst = Memory::Read32(thread->context.pc & 0xFFFFFFFC) & 0x0FFFFFFF;
-
-    if ((thumb_mode && thumb_inst == 0xDF24) || (!thumb_mode && inst == 0x0F000024)) {
-        // svc #0x24 (WaitSynchronization1)
-        return std::make_tuple(&thread->context.cpu_registers[2],
-                               &thread->context.cpu_registers[3]);
-    } else if ((thumb_mode && thumb_inst == 0xDF25) || (!thumb_mode && inst == 0x0F000025)) {
-        // svc #0x25 (WaitSynchronizationN)
-        return std::make_tuple(&thread->context.cpu_registers[0],
-                               &thread->context.cpu_registers[4]);
-    }
-
-    UNREACHABLE();
-}
-
-/**
- * Updates the WaitSynchronization timeout parameter according to the difference
- * between ticks of the last WaitSynchronization call and the incoming one.
- * @param timeout_low a pointer to the register for the low part of the timeout parameter
- * @param timeout_high a pointer to the register for the high part of the timeout parameter
- * @param last_tick tick of the last WaitSynchronization call
- */
-static void UpdateTimeoutParameter(u32* timeout_low, u32* timeout_high, u64 last_tick) {
-    s64 timeout = ((s64)*timeout_high << 32) | *timeout_low;
-
-    if (timeout != -1) {
-        timeout -= cyclesToUs(CoreTiming::GetTicks() - last_tick) * 1000; // in nanoseconds
-
-        if (timeout < 0)
-            timeout = 0;
-
-        *timeout_low = timeout & 0xFFFFFFFF;
-        *timeout_high = timeout >> 32;
-    }
-}
-
-/**
  * Switches the CPU's active thread context to that of the specified thread
  * @param new_thread The thread to switch to
  */
@@ -254,32 +208,6 @@ static void SwitchContext(Thread* new_thread) {
 
         current_thread = new_thread;
 
-        // If the thread was waited by a svcWaitSynch call, step back PC by one instruction to rerun
-        // the SVC when the thread wakes up. This is necessary to ensure that the thread can acquire
-        // the requested wait object(s) before continuing.
-        if (new_thread->waitsynch_waited) {
-            // CPSR flag indicates CPU mode
-            bool thumb_mode = (new_thread->context.cpsr & TBIT) != 0;
-
-            // SVC instruction is 2 bytes for THUMB, 4 bytes for ARM
-            new_thread->context.pc -= thumb_mode ? 2 : 4;
-
-            // Get the register for timeout parameter
-            u32 *timeout_low, *timeout_high;
-            std::tie(timeout_low, timeout_high) = GetWaitSynchTimeoutParameterRegister(new_thread);
-
-            // Update the timeout parameter
-            UpdateTimeoutParameter(timeout_low, timeout_high, new_thread->last_running_ticks);
-        }
-
-        // Clean up the thread's wait_objects, they'll be restored if needed during
-        // the svcWaitSynchronization call
-        for (size_t i = 0; i < new_thread->wait_objects.size(); ++i) {
-            SharedPtr<WaitObject> object = new_thread->wait_objects[i];
-            object->RemoveWaitingThread(new_thread);
-        }
-        new_thread->wait_objects.clear();
-
         ready_queue.remove(new_thread->current_priority, new_thread);
         new_thread->status = THREADSTATUS_RUNNING;
 
@@ -319,17 +247,13 @@ static Thread* PopNextReadyThread() {
 void WaitCurrentThread_Sleep() {
     Thread* thread = GetCurrentThread();
     thread->status = THREADSTATUS_WAIT_SLEEP;
-
-    HLE::Reschedule(__func__);
 }
 
 void WaitCurrentThread_WaitSynchronization(std::vector<SharedPtr<WaitObject>> wait_objects,
-                                           bool wait_set_output, bool wait_all) {
+                                           bool wait_set_output) {
     Thread* thread = GetCurrentThread();
     thread->wait_set_output = wait_set_output;
-    thread->wait_all = wait_all;
     thread->wait_objects = std::move(wait_objects);
-    thread->waitsynch_waited = true;
     thread->status = THREADSTATUS_WAIT_SYNCH;
 }
 
@@ -351,15 +275,11 @@ static void ThreadWakeupCallback(u64 thread_handle, int cycles_late) {
         return;
     }
 
-    thread->waitsynch_waited = false;
-
     if (thread->status == THREADSTATUS_WAIT_SYNCH || thread->status == THREADSTATUS_WAIT_ARB) {
+        thread->wait_set_output = false;
         thread->SetWaitSynchronizationResult(ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
                                                         ErrorSummary::StatusChanged,
                                                         ErrorLevel::Info));
-
-        if (thread->wait_set_output)
-            thread->SetWaitSynchronizationOutput(-1);
     }
 
     thread->ResumeFromWait();
@@ -399,6 +319,7 @@ void Thread::ResumeFromWait() {
 
     ready_queue.push_back(current_priority, this);
     status = THREADSTATUS_READY;
+    HLE::Reschedule(__func__);
 }
 
 /**
@@ -494,13 +415,11 @@ ResultVal<SharedPtr<Thread>> Thread::Create(std::string name, VAddr entry_point,
     thread->last_running_ticks = CoreTiming::GetTicks();
     thread->processor_id = processor_id;
     thread->wait_set_output = false;
-    thread->wait_all = false;
     thread->wait_objects.clear();
     thread->wait_address = 0;
     thread->name = std::move(name);
     thread->callback_handle = wakeup_callback_handle_table.Create(thread).MoveFrom();
     thread->owner_process = g_current_process;
-    thread->waitsynch_waited = false;
 
     // Find the next available TLS index, and mark it as used
     auto& tls_slots = Kernel::g_current_process->tls_slots;
@@ -555,8 +474,6 @@ ResultVal<SharedPtr<Thread>> Thread::Create(std::string name, VAddr entry_point,
     ready_queue.push_back(thread->current_priority, thread.get());
     thread->status = THREADSTATUS_READY;
 
-    HLE::Reschedule(__func__);
-
     return MakeResult<SharedPtr<Thread>>(std::move(thread));
 }
 
@@ -619,14 +536,6 @@ void Reschedule() {
 
     HLE::DoneRescheduling();
 
-    // Don't bother switching to the same thread.
-    // But if the thread was waiting on objects, we still need to switch it
-    // to perform PC modification, change state to RUNNING, etc.
-    // This occurs in the case when an object the thread is waiting on immediately wakes up
-    // the current thread before Reschedule() is called.
-    if (next == cur && (next == nullptr || next->waitsynch_waited == false))
-        return;
-
     if (cur && next) {
         LOG_TRACE(Kernel, "context switch %u -> %u", cur->GetObjectId(), next->GetObjectId());
     } else if (cur) {
diff --git a/src/core/hle/kernel/thread.h b/src/core/hle/kernel/thread.h
index e0ffcea8a..63b97b74f 100644
--- a/src/core/hle/kernel/thread.h
+++ b/src/core/hle/kernel/thread.h
@@ -5,6 +5,7 @@
 #pragma once
 
 #include <string>
+#include <unordered_map>
 #include <vector>
 #include <boost/container/flat_set.hpp>
 #include "common/common_types.h"
@@ -125,6 +126,16 @@ public:
     void SetWaitSynchronizationOutput(s32 output);
 
     /**
+     * Retrieves the index that this particular object occupies in the list of objects
+     * that the thread passed to WaitSynchronizationN.
+     * It is used to set the output value of WaitSynchronizationN when the thread is awakened.
+     * @param object Object to query the index of.
+     */
+    s32 GetWaitObjectIndex(WaitObject* object) {
+        return wait_objects_index[object->GetObjectId()];
+    }
+
+    /**
      * Stops a thread, invalidating it from further use
      */
     void Stop();
@@ -154,16 +165,16 @@ public:
 
     VAddr tls_address; ///< Virtual address of the Thread Local Storage of the thread
 
-    bool waitsynch_waited; ///< Set to true if the last svcWaitSynch call caused the thread to wait
-
     /// Mutexes currently held by this thread, which will be released when it exits.
     boost::container::flat_set<SharedPtr<Mutex>> held_mutexes;
 
     SharedPtr<Process> owner_process;                ///< Process that owns this thread
     std::vector<SharedPtr<WaitObject>> wait_objects; ///< Objects that the thread is waiting on
+    std::unordered_map<int, s32> wait_objects_index; ///< Mapping of Object ids to their position in the last waitlist that this object waited on.
+
     VAddr wait_address;   ///< If waiting on an AddressArbiter, this is the arbitration address
-    bool wait_all;        ///< True if the thread is waiting on all objects before resuming
-    bool wait_set_output; ///< True if the output parameter should be set on thread wakeup
+
+    bool wait_set_output; ///< True if the WaitSynchronizationN output parameter should be set on thread wakeup
 
     std::string name;
 
@@ -215,10 +226,9 @@ void WaitCurrentThread_Sleep();
  * @param wait_objects Kernel objects that we are waiting on
  * @param wait_set_output If true, set the output parameter on thread wakeup (for
  * WaitSynchronizationN only)
- * @param wait_all If true, wait on all objects before resuming (for WaitSynchronizationN only)
  */
 void WaitCurrentThread_WaitSynchronization(std::vector<SharedPtr<WaitObject>> wait_objects,
-                                           bool wait_set_output, bool wait_all);
+                                           bool wait_set_output);
 
 /**
  * Waits the current thread from an ArbitrateAddress call
diff --git a/src/core/hle/kernel/timer.cpp b/src/core/hle/kernel/timer.cpp
index eac181f4e..b50cf520d 100644
--- a/src/core/hle/kernel/timer.cpp
+++ b/src/core/hle/kernel/timer.cpp
@@ -60,14 +60,10 @@ void Timer::Set(s64 initial, s64 interval) {
     u64 initial_microseconds = initial / 1000;
     CoreTiming::ScheduleEvent(usToCycles(initial_microseconds), timer_callback_event_type,
                               callback_handle);
-
-    HLE::Reschedule(__func__);
 }
 
 void Timer::Cancel() {
     CoreTiming::UnscheduleEvent(timer_callback_event_type, callback_handle);
-
-    HLE::Reschedule(__func__);
 }
 
 void Timer::Clear() {