Fix Clean Exiting + Optimize Core Queues + Optimize Thread Insertion + Implement HID SendVibrationValue

app/src/main/cpp/emu_jni.cpp

@@ -22,6 +22,7 @@ std::weak_ptr<skyline::gpu::GPU> GpuWeak;
 std::weak_ptr<skyline::input::Input> InputWeak;
 
 extern "C" JNIEXPORT void Java_emu_skyline_EmulationActivity_executeApplication(JNIEnv *env, jobject instance, jstring romUriJstring, jint romType, jint romFd, jint preferenceFd, jstring appFilesPathJstring) {
+    skyline::signal::ScopedStackBlocker stackBlocker;
     Fps = FrameTime = 0;
 
     pthread_setname_np(pthread_self(), "EmuMain");

app/src/main/cpp/skyline/common/signal.cpp

@@ -10,15 +10,19 @@ namespace skyline::signal {
     thread_local std::exception_ptr SignalExceptionPtr;
 
     void ExceptionThrow() {
-        std::rethrow_exception(SignalExceptionPtr);
+        // We need the compiler to not remove the asm at the end of 'std::rethrow_exception' which is a noreturn function
+        volatile bool alwaysTrue{true};
+        if (alwaysTrue)
+            std::rethrow_exception(SignalExceptionPtr);
     }
 
     std::terminate_handler terminateHandler{};
 
+    [[clang::optnone]]
     inline StackFrame *SafeFrameRecurse(size_t depth, StackFrame *frame) {
         if (frame) {
             for (size_t it{}; it < depth; it++) {
-                if (frame->next)
+                if (frame->lr && frame->next)
                     frame = frame->next;
                 else
                     terminateHandler();
@@ -29,6 +33,7 @@ namespace skyline::signal {
         return frame;
     }
 
+    [[clang::optnone]]
     void TerminateHandler() {
         auto exception{std::current_exception()};
         if (terminateHandler && exception && exception == SignalExceptionPtr) {
@@ -36,11 +41,25 @@ namespace skyline::signal {
             asm("MOV %0, FP" : "=r"(frame));
             frame = SafeFrameRecurse(2, frame); // We unroll past 'std::terminate'
 
+            static void *exceptionThrowEnd{};
+            if (!exceptionThrowEnd) {
+                u32 *it{reinterpret_cast<u32 *>(&ExceptionThrow) + 1};
+                while (_Unwind_FindEnclosingFunction(it) == &ExceptionThrow)
+                    it++;
+                exceptionThrowEnd = it - 1;
+            }
+
             auto lookupFrame{frame};
-            while (lookupFrame) {
-                auto function{_Unwind_FindEnclosingFunction(frame->lr)};
-                if (function == &ExceptionThrow)
-                    terminateHandler(); // We have no handler to consume the exception, it's time to quit
+            bool hasAdvanced{};
+            while (lookupFrame && lookupFrame->lr) {
+                if (lookupFrame->lr >= reinterpret_cast<void *>(&ExceptionThrow) && lookupFrame->lr < exceptionThrowEnd) {
+                    if (!hasAdvanced) {
+                        frame = SafeFrameRecurse(2, lookupFrame);
+                        hasAdvanced = true;
+                    } else {
+                        terminateHandler(); // We have no handler to consume the exception, it's time to quit
+                    }
+                }
                 lookupFrame = lookupFrame->next;
             }
 
@@ -68,7 +87,7 @@ namespace skyline::signal {
 
         signalException.frames.push_back(reinterpret_cast<void *>(context->uc_mcontext.pc));
         StackFrame *frame{reinterpret_cast<StackFrame *>(context->uc_mcontext.regs[29])};
-        while (frame) {
+        while (frame && frame->lr) {
             signalException.frames.push_back(frame->lr);
             frame = frame->next;
         }

app/src/main/cpp/skyline/common/signal.h

@@ -14,6 +14,29 @@ namespace skyline::signal {
         void *lr;
     };
 
+    /**
+     * @brief A scoped way to block a stack trace beyond the scope of this object
+     * @note This is used for JNI functions where the stack trace will be determined as they often contain invalid stack frames which'd cause a SIGSEGV
+     */
+    struct ScopedStackBlocker {
+        StackFrame realFrame;
+
+        __attribute__((noinline)) ScopedStackBlocker() {
+            StackFrame *frame;
+            asm("MOV %0, FP" : "=r"(frame));
+            realFrame = *frame;
+            frame->next = nullptr;
+            frame->lr = nullptr;
+        }
+
+        __attribute__((noinline)) ~ScopedStackBlocker() {
+            StackFrame *frame;
+            asm("MOV %0, FP" : "=r"(frame));
+            frame->next = realFrame.next;
+            frame->lr = realFrame.lr;
+        }
+    };
+
     /**
      * @brief An exception object that is designed specifically to hold Linux signals
      * @note This doesn't inherit std::exception as it shouldn't be caught as such

app/src/main/cpp/skyline/gpu/gpfifo.cpp

@@ -105,6 +105,7 @@ namespace skyline::gpu::gpfifo {
             }
         } catch (const std::exception &e) {
             state.logger->Write(Logger::LogLevel::Error, e.what());
+            signal::BlockSignal({SIGINT});
             state.process->Kill(false);
         }
     }

app/src/main/cpp/skyline/kernel/scheduler.cpp

@@ -36,7 +36,7 @@ namespace skyline::kernel {
                     u64 timeslice{};
 
                     if (!candidateCore.queue.empty()) {
-                        std::shared_lock coreLock(candidateCore.mutex);
+                        std::unique_lock coreLock(candidateCore.mutex);
 
                         auto threadIterator{candidateCore.queue.cbegin()};
                         if (threadIterator != candidateCore.queue.cend()) {
@@ -89,8 +89,20 @@ namespace skyline::kernel {
         if (nextThread == core.queue.begin()) {
             if (nextThread != core.queue.end()) {
                 // If the inserted thread has a higher priority than the currently running thread (and the queue isn't empty)
-                // We need to interrupt the currently scheduled thread and put this thread at the front instead
-                core.queue.insert(std::next(core.queue.begin()), thread);
+                if (state.thread == thread) {
+                    // If the current thread is inserting itself then we try to optimize by trying to by forcefully yielding it ourselves now
+                    // We can avoid waiting on it to yield itself on receiving the signal which serializes the entire pipeline
+                    // This isn't done in other cases as this optimization is unsafe when done where serialization is required (Eg: Mutexes)
+                    core.queue.front()->forceYield = true;
+                    core.queue.splice(std::upper_bound(core.queue.begin(), core.queue.end(), thread->priority.load(), type::KThread::IsHigherPriority), core.queue, core.queue.begin());
+                    core.queue.push_front(thread);
+                } else {
+                    // If we're inserting another thread then we just insert it after the thread in line
+                    // It'll automatically be ready to be scheduled when the thread at the front yields
+                    // This enforces strict synchronization for the thread to run and waits till the previous thread has yielded itself
+                    core.queue.insert(std::next(core.queue.begin()), thread);
+                }
+
                 if (state.thread != core.queue.front())
                     core.queue.front()->SendSignal(YieldSignal);
                 else
@@ -109,7 +121,7 @@ namespace skyline::kernel {
         auto &thread{state.thread};
         auto *core{&cores.at(thread->coreId)};
 
-        std::shared_lock lock(core->mutex);
+        std::unique_lock lock(core->mutex);
         if (loadBalance && thread->affinityMask.count() > 1) {
             std::chrono::milliseconds loadBalanceThreshold{PreemptiveTimeslice * 2}; //!< The amount of time that needs to pass unscheduled for a thread to attempt load balancing
             while (!thread->wakeCondition.wait_for(lock, loadBalanceThreshold, [&]() { return !core->queue.empty() && core->queue.front() == thread; })) {
@@ -119,7 +131,7 @@ namespace skyline::kernel {
                     lock.lock();
                 } else {
                     core = &cores.at(thread->coreId);
-                    lock = std::shared_lock(core->mutex);
+                    lock = std::unique_lock(core->mutex);
                 }
 
                 loadBalanceThreshold *= 2; // We double the duration required for future load balancing for this invocation to minimize pointless load balancing
@@ -141,7 +153,7 @@ namespace skyline::kernel {
         auto &thread{state.thread};
         auto *core{&cores.at(thread->coreId)};
 
-        std::shared_lock lock(core->mutex);
+        std::unique_lock lock(core->mutex);
         if (thread->wakeCondition.wait_for(lock, timeout, [&]() { return !core->queue.empty() && core->queue.front() == thread; })) {
             if (thread->priority == core->preemptionPriority) {
                 struct itimerspec spec{.it_value = {.tv_nsec = std::chrono::duration_cast<std::chrono::nanoseconds>(PreemptiveTimeslice).count()}};
@@ -177,10 +189,36 @@ namespace skyline::kernel {
                 struct itimerspec spec{};
                 timer_settime(thread->preemptionTimer, 0, &spec, nullptr);
             }
+
             thread->isPreempted = false;
+        } else if (thread->forceYield) {
+            // We need to check if this thread was yielded by another thread on behalf of this thread
+            // If it is then we just need to disarm the preemption timer and update the average timeslice
+            // If it isn't then we need to throw an exception as it's indicative of an invalid state
+            struct ThreadComparision {
+                constexpr bool operator()(const i8 priority, const std::shared_ptr<type::KThread> &it) { return priority < it->priority; }
+
+                constexpr bool operator()(const std::shared_ptr<type::KThread> &it, const i8 priority) { return it->priority < priority; }
+            };
+            auto bounds{std::equal_range(core.queue.begin(), core.queue.end(), thread->priority.load(), ThreadComparision{})};
+
+            auto iterator{std::find(bounds.first, bounds.second, thread)};
+            if (iterator != bounds.second) {
+                thread->averageTimeslice = (thread->averageTimeslice / 4) + (3 * (util::GetTimeTicks() - thread->timesliceStart / 4));
+
+                if (cooperative && thread->isPreempted) {
+                    struct itimerspec spec{};
+                    timer_settime(thread->preemptionTimer, 0, &spec, nullptr);
+                }
+
+                thread->isPreempted = false;
+            } else {
+                throw exception("T{} called Rotate while not being in C{}'s queue after being forcefully yielded", thread->id, thread->coreId);
+            }
         } else {
-            throw exception("T{} called Rotate while not being at the front of C{}'s queue", thread->id, thread->coreId);
+            throw exception("T{} called Rotate while not being in C{}'s queue", thread->id, thread->coreId);
         }
+        thread->forceYield = false;
     }
 
     void Scheduler::UpdatePriority(const std::shared_ptr<type::KThread> &thread) {

app/src/main/cpp/skyline/kernel/scheduler.h

@@ -46,7 +46,7 @@ namespace skyline {
             struct CoreContext {
                 u8 id;
                 u8 preemptionPriority; //!< The priority at which this core becomes preemptive as opposed to cooperative
-                std::shared_mutex mutex; //!< Synchronizes all operations on the queue
+                std::mutex mutex; //!< Synchronizes all operations on the queue
                 std::list<std::shared_ptr<type::KThread>> queue; //!< A queue of threads which are running or to be run on this core
 
                 CoreContext(u8 id, u8 preemptionPriority);

app/src/main/cpp/skyline/kernel/svc.cpp

@@ -647,7 +647,7 @@ namespace skyline::kernel::svc {
         }
 
         if (wakeObject) {
-            state.logger->Debug("svcWaitSynchronization: Signalled handle: 0x{:X}", waitHandles[index]);
+            state.logger->Debug("svcWaitSynchronization: Signalled handle: 0x{:X}", waitHandles[wakeIndex]);
             state.ctx->gpr.w0 = Result{};
             state.ctx->gpr.w1 = wakeIndex;
         } else if (state.thread->cancelSync) {

app/src/main/cpp/skyline/kernel/types/KThread.h

@@ -6,6 +6,7 @@
 #include <csetjmp>
 #include <nce/guest.h>
 #include <kernel/scheduler.h>
+#include <common/signal.h>
 #include "KSyncObject.h"
 #include "KPrivateMemory.h"
 #include "KSharedMemory.h"
@@ -40,7 +41,7 @@ namespace skyline {
             u64 entryArgument;
             void *stackTop;
 
-            std::condition_variable_any wakeCondition; //!< A conditional variable which is signalled to wake the current thread while it's sleeping
+            std::condition_variable wakeCondition; //!< A conditional variable which is signalled to wake the current thread while it's sleeping
             std::atomic<u8> basePriority; //!< The priority of the thread for the scheduler without any priority-inheritance
             std::atomic<u8> priority; //!< The priority of the thread for the scheduler
             i8 idealCore; //!< The ideal CPU core for this thread to run on
@@ -51,6 +52,7 @@ namespace skyline {
             u64 averageTimeslice{}; //!< A weighted average of the timeslice duration for this thread
             timer_t preemptionTimer{}; //!< A kernel timer used for preemption interrupts
             bool isPreempted{}; //!< If the preemption timer has been armed and will fire
+            bool forceYield{}; //!< If the thread has been forcefully yielded by another thread
             std::mutex waiterMutex; //!< Synchronizes operations on mutation of the waiter members
             u32* waitKey; //!< The key of the mutex which this thread is waiting on
             KHandle waitTag; //!< The handle of the thread which requested the mutex lock

app/src/main/cpp/skyline/services/hid/IHidServer.cpp

@@ -142,6 +142,18 @@ namespace skyline::service::hid {
         return {};
     }
 
+    Result IHidServer::SendVibrationValue(type::KSession &session, ipc::IpcRequest &request, ipc::IpcResponse &response) {
+        const auto &handle{request.Pop<NpadDeviceHandle>()};
+        auto &device{state.input->npad.at(handle.id)};
+        if (device.type == handle.GetType()) {
+            const auto &value{request.Pop<NpadVibrationValue>()};
+            state.logger->Debug("Vibration - Handle: 0x{:02X} (0b{:05b}), Vibration: {:.2f}@{:.2f}Hz, {:.2f}@{:.2f}Hz", static_cast<u8>(handle.id), static_cast<u8>(handle.type), value.amplitudeLow, value.frequencyLow, value.amplitudeHigh, value.frequencyHigh);
+            device.Vibrate(handle.isRight, value);
+        }
+
+        return {};
+    }
+
     Result IHidServer::SendVibrationValues(type::KSession &session, ipc::IpcRequest &request, ipc::IpcResponse &response) {
         request.Skip<u64>(); // appletResourceUserId
 

app/src/main/cpp/skyline/services/hid/IHidServer.h

@@ -115,7 +115,13 @@ namespace skyline::service::hid {
         Result CreateActiveVibrationDeviceList(type::KSession &session, ipc::IpcRequest &request, ipc::IpcResponse &response);
 
         /**
-         * @brief Send vibration values to an NPad
+         * @brief Send a single vibration value to a HID device
+         * @url https://switchbrew.org/wiki/HID_services#SendVibrationValue
+         */
+        Result SendVibrationValue(type::KSession &session, ipc::IpcRequest &request, ipc::IpcResponse &response);
+
+        /**
+         * @brief Send vibration values to a HID device
          * @url https://switchbrew.org/wiki/HID_services#SendVibrationValues
          */
         Result SendVibrationValues(type::KSession &session, ipc::IpcRequest &request, ipc::IpcResponse &response);
@@ -138,6 +144,7 @@ namespace skyline::service::hid {
             SFUNC(0x7B, IHidServer, SetNpadJoyAssignmentModeSingle),
             SFUNC(0x7C, IHidServer, SetNpadJoyAssignmentModeDual),
             SFUNC(0xCB, IHidServer, CreateActiveVibrationDeviceList),
+            SFUNC(0xC9, IHidServer, SendVibrationValue),
             SFUNC(0xCE, IHidServer, SendVibrationValues)
         )
     };