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.

1 files changed, 103 insertions, 78 deletions

diff --git a/src/core/hle/svc.cpp b/src/core/hle/svc.cpp
index c6b80dc50..061692af8 100644
--- a/src/core/hle/svc.cpp
+++ b/src/core/hle/svc.cpp
@@ -249,27 +249,30 @@ static ResultCode WaitSynchronization1(Handle handle, s64 nano_seconds) {
     auto object = Kernel::g_handle_table.GetWaitObject(handle);
     Kernel::Thread* thread = Kernel::GetCurrentThread();
 
-    thread->waitsynch_waited = false;
-
     if (object == nullptr)
         return ERR_INVALID_HANDLE;
 
     LOG_TRACE(Kernel_SVC, "called handle=0x%08X(%s:%s), nanoseconds=%lld", handle,
               object->GetTypeName().c_str(), object->GetName().c_str(), nano_seconds);
 
-    HLE::Reschedule(__func__);
-
-    // Check for next thread to schedule
     if (object->ShouldWait()) {
 
+        if (nano_seconds == 0)
+            return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                              ErrorSummary::StatusChanged,
+                              ErrorLevel::Info);
+
         object->AddWaitingThread(thread);
-        Kernel::WaitCurrentThread_WaitSynchronization({object}, false, false);
+        thread->status = THREADSTATUS_WAIT_SYNCH;
 
         // Create an event to wake the thread up after the specified nanosecond delay has passed
         thread->WakeAfterDelay(nano_seconds);
 
-        // NOTE: output of this SVC will be set later depending on how the thread resumes
-        return HLE::RESULT_INVALID;
+        // Note: The output of this SVC will be set to RESULT_SUCCESS if the thread resumes due to a signal in one of its wait objects.
+        // Otherwise we retain the default value of timeout.
+        return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                               ErrorSummary::StatusChanged,
+                               ErrorLevel::Info);
     }
 
     object->Acquire();
@@ -283,8 +286,6 @@ static ResultCode WaitSynchronizationN(s32* out, Handle* handles, s32 handle_cou
     bool wait_thread = !wait_all;
     int handle_index = 0;
     Kernel::Thread* thread = Kernel::GetCurrentThread();
-    bool was_waiting = thread->waitsynch_waited;
-    thread->waitsynch_waited = false;
 
     // Check if 'handles' is invalid
     if (handles == nullptr)
@@ -300,90 +301,113 @@ static ResultCode WaitSynchronizationN(s32* out, Handle* handles, s32 handle_cou
         return ResultCode(ErrorDescription::OutOfRange, ErrorModule::OS,
                           ErrorSummary::InvalidArgument, ErrorLevel::Usage);
 
-    // If 'handle_count' is non-zero, iterate through each handle and wait the current thread if
-    // necessary
-    if (handle_count != 0) {
-        bool selected = false; // True once an object has been selected
-
-        Kernel::SharedPtr<Kernel::WaitObject> wait_object;
-
-        for (int i = 0; i < handle_count; ++i) {
-            auto object = Kernel::g_handle_table.GetWaitObject(handles[i]);
-            if (object == nullptr)
-                return ERR_INVALID_HANDLE;
-
-            // Check if the current thread should wait on this object...
-            if (object->ShouldWait()) {
-
-                // Check we are waiting on all objects...
-                if (wait_all)
-                    // Wait the thread
-                    wait_thread = true;
-            } else {
-                // Do not wait on this object, check if this object should be selected...
-                if (!wait_all && (!selected || (wait_object == object && was_waiting))) {
-                    // Do not wait the thread
-                    wait_thread = false;
-                    handle_index = i;
-                    wait_object = object;
-                    selected = true;
-                }
-            }
-        }
-    } else {
-        // If no handles were passed in, put the thread to sleep only when 'wait_all' is false
-        // NOTE: This should deadlock the current thread if no timeout was specified
-        if (!wait_all) {
-            wait_thread = true;
-        }
+    using ObjectPtr = Kernel::SharedPtr<Kernel::WaitObject>;
+
+    std::vector<ObjectPtr> objects(handle_count);
+
+    for (int i = 0; i < handle_count; ++i) {
+        auto object = Kernel::g_handle_table.GetWaitObject(handles[i]);
+        if (object == nullptr)
+            return ERR_INVALID_HANDLE;
+        objects[i] = object;
     }
 
-    SCOPE_EXIT({
-        HLE::Reschedule("WaitSynchronizationN");
-    }); // Reschedule after putting the threads to sleep.
+    // Clear the mapping of wait object indices
+    thread->wait_objects_index.clear();
+
+    if (!wait_all) {
+        // Find the first object that is acquireable in the provided list of objects
+        auto itr = std::find_if(objects.begin(), objects.end(), [](const ObjectPtr& object) {
+            return !object->ShouldWait();
+        });
+
+        if (itr != objects.end()) {
+            // We found a ready object, acquire it and set the result value
+            ObjectPtr object = *itr;
+            object->Acquire();
+            *out = std::distance(objects.begin(), itr);
+            return RESULT_SUCCESS;
+        }
+
+        // No objects were ready to be acquired, prepare to suspend the thread.
+
+        // If a timeout value of 0 was provided, just return the Timeout error code instead of suspending the thread.
+        if (nano_seconds == 0) {
+            return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                              ErrorSummary::StatusChanged,
+                              ErrorLevel::Info);
+        }
 
-    // If thread should wait, then set its state to waiting
-    if (wait_thread) {
+        // Put the thread to sleep
+        thread->status = THREADSTATUS_WAIT_SYNCH;
 
-        // Actually wait the current thread on each object if we decided to wait...
-        std::vector<SharedPtr<Kernel::WaitObject>> wait_objects;
-        wait_objects.reserve(handle_count);
+        // Clear the thread's waitlist, we won't use it for wait_all = false
+        thread->wait_objects.clear();
 
-        for (int i = 0; i < handle_count; ++i) {
-            auto object = Kernel::g_handle_table.GetWaitObject(handles[i]);
-            object->AddWaitingThread(Kernel::GetCurrentThread());
-            wait_objects.push_back(object);
+        // Add the thread to each of the objects' waiting threads.
+        for (int i = 0; i < objects.size(); ++i) {
+            ObjectPtr object = objects[i];
+            // Set the index of this object in the mapping of Objects -> index for this thread.
+            thread->wait_objects_index[object->GetObjectId()] = i;
+            object->AddWaitingThread(thread);
+            // TODO(Subv): Perform things like update the mutex lock owner's priority to prevent priority inversion.
         }
 
-        Kernel::WaitCurrentThread_WaitSynchronization(std::move(wait_objects), true, wait_all);
+        // Note: If no handles and no timeout were given, then the thread will deadlock, this is consistent with hardware behavior.
 
         // Create an event to wake the thread up after the specified nanosecond delay has passed
-        Kernel::GetCurrentThread()->WakeAfterDelay(nano_seconds);
+        thread->WakeAfterDelay(nano_seconds);
 
-        // NOTE: output of this SVC will be set later depending on how the thread resumes
-        return HLE::RESULT_INVALID;
-    }
+        // Note: The output of this SVC will be set to RESULT_SUCCESS if the thread resumes due to a signal in one of its wait objects.
+        // Otherwise we retain the default value of timeout, and -1 in the out parameter
+        thread->wait_set_output = true;
+        *out = -1;
+        return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                          ErrorSummary::StatusChanged,
+                          ErrorLevel::Info);
+    } else {
+        bool all_available = std::all_of(objects.begin(), objects.end(), [](const ObjectPtr& object) {
+            return !object->ShouldWait();
+        });
+        if (all_available) {
+            // We can acquire all objects right now, do so.
+            for (auto object : objects)
+                object->Acquire();
+            // Note: In this case, the `out` parameter is not set, and retains whatever value it had before.
+            return RESULT_SUCCESS;
+        }
 
-    // Acquire objects if we did not wait...
-    for (int i = 0; i < handle_count; ++i) {
-        auto object = Kernel::g_handle_table.GetWaitObject(handles[i]);
+        // Not all objects were available right now, prepare to suspend the thread.
 
-        // Acquire the object if it is not waiting...
-        if (!object->ShouldWait()) {
-            object->Acquire();
+        // If a timeout value of 0 was provided, just return the Timeout error code instead of suspending the thread.
+        if (nano_seconds == 0) {
+            return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                              ErrorSummary::StatusChanged,
+                              ErrorLevel::Info);
+        }
+
+        // Put the thread to sleep
+        thread->status = THREADSTATUS_WAIT_SYNCH;
 
-            // If this was the first non-waiting object and 'wait_all' is false, don't acquire
-            // any other objects
-            if (!wait_all)
-                break;
+        // Set the thread's waitlist to the list of objects passed to WaitSynchronizationN
+        thread->wait_objects = objects;
+
+        // Add the thread to each of the objects' waiting threads.
+        for (auto object : objects) {
+            object->AddWaitingThread(thread);
+            // TODO(Subv): Perform things like update the mutex lock owner's priority to prevent priority inversion.
         }
-    }
 
-    // TODO(bunnei): If 'wait_all' is true, this is probably wrong. However, real hardware does
-    // not seem to set it to any meaningful value.
-    *out = handle_count != 0 ? (wait_all ? -1 : handle_index) : 0;
+        // Create an event to wake the thread up after the specified nanosecond delay has passed
+        thread->WakeAfterDelay(nano_seconds);
 
-    return RESULT_SUCCESS;
+        // This value gets set to -1 by default in this case, it is not modified after this.
+        *out = -1;
+        // Note: The output of this SVC will be set to RESULT_SUCCESS if the thread resumes due to a signal in one of its wait objects.
+        return ResultCode(ErrorDescription::Timeout, ErrorModule::OS,
+                          ErrorSummary::StatusChanged,
+                          ErrorLevel::Info);
+    }
 }
 
 /// Create an address arbiter (to allocate access to shared resources)
@@ -1148,6 +1172,7 @@ void CallSVC(u32 immediate) {
     if (info) {
         if (info->func) {
             info->func();
+            HLE::Reschedule(__func__);
         } else {
             LOG_ERROR(Kernel_SVC, "unimplemented SVC function %s(..)", info->name);
         }