Support Core Migration for Running External Thread

We did not support migration of threads which were running in a non-cooperative manner, this was partially due to the dependence on per-core conditional variables rather than per-thread which made this harder to do programmatically. This has been fixed by moving to per-thread cvars and therefore the limitation can be removed, this feature is used by Unity games.

.idea/inspectionProfiles/Project_Default.xml

@@ -171,7 +171,7 @@
     </inspection_tool>
     <inspection_tool class="CheckedExceptionClass" enabled="true" level="WARNING" enabled_by_default="true" />
     <inspection_tool class="ClangTidy" enabled="true" level="WARNING" enabled_by_default="true">
-      <option name="clangTidyChecks" value="-*,bugprone-argument-comment,bugprone-assert-side-effect,bugprone-bad-signal-to-kill-thread,bugprone-branch-clone,bugprone-copy-constructor-init,bugprone-dangling-handle,bugprone-dynamic-static-initializers,bugprone-fold-init-type,bugprone-forward-declaration-namespace,bugprone-forwarding-reference-overload,bugprone-inaccurate-erase,bugprone-incorrect-roundings,bugprone-integer-division,bugprone-lambda-function-name,bugprone-macro-parentheses,bugprone-macro-repeated-side-effects,bugprone-misplaced-operator-in-strlen-in-alloc,bugprone-misplaced-pointer-arithmetic-in-alloc,bugprone-misplaced-widening-cast,bugprone-move-forwarding-reference,bugprone-multiple-statement-macro,bugprone-no-escape,bugprone-not-null-terminated-result,bugprone-parent-virtual-call,bugprone-posix-return,bugprone-reserved-identifier,bugprone-sizeof-container,bugprone-sizeof-expression,bugprone-spuriously-wake-up-functions,bugprone-string-constructor,bugprone-string-integer-assignment,bugprone-string-literal-with-embedded-nul,bugprone-suspicious-enum-usage,bugprone-suspicious-include,bugprone-suspicious-memset-usage,bugprone-suspicious-missing-comma,bugprone-suspicious-semicolon,bugprone-suspicious-string-compare,bugprone-swapped-arguments,bugprone-terminating-continue,bugprone-throw-keyword-missing,bugprone-too-small-loop-variable,bugprone-undefined-memory-manipulation,bugprone-undelegated-constructor,bugprone-unhandled-self-assignment,bugprone-unused-raii,bugprone-unused-return-value,bugprone-use-after-move,bugprone-virtual-near-miss,cert-dcl21-cpp,cert-dcl58-cpp,cert-err34-c,cert-err52-cpp,cert-err58-cpp,cert-err60-cpp,cert-flp30-c,cert-msc50-cpp,cert-msc51-cpp,cert-str34-c,cppcoreguidelines-interfaces-global-init,cppcoreguidelines-narrowing-conversions,cppcoreguidelines-pro-type-static-cast-downcast,cppcoreguidelines-slicing,google-default-arguments,google-explicit-constructor,google-runtime-operator,hicpp-exception-baseclass,hicpp-multiway-paths-covered,misc-misplaced-const,misc-new-delete-overloads,misc-no-recursion,misc-non-copyable-objects,misc-throw-by-value-catch-by-reference,misc-unconventional-assign-operator,misc-uniqueptr-reset-release,modernize-avoid-bind,modernize-concat-nested-namespaces,modernize-deprecated-headers,modernize-deprecated-ios-base-aliases,modernize-loop-convert,modernize-make-shared,modernize-make-unique,modernize-pass-by-value,modernize-raw-string-literal,modernize-redundant-void-arg,modernize-replace-auto-ptr,modernize-replace-disallow-copy-and-assign-macro,modernize-replace-random-shuffle,modernize-return-braced-init-list,modernize-shrink-to-fit,modernize-unary-static-assert,modernize-use-auto,modernize-use-bool-literals,modernize-use-emplace,modernize-use-equals-default,modernize-use-equals-delete,modernize-use-nodiscard,modernize-use-noexcept,modernize-use-nullptr,modernize-use-override,modernize-use-transparent-functors,modernize-use-uncaught-exceptions,mpi-buffer-deref,mpi-type-mismatch,openmp-use-default-none,performance-faster-string-find,performance-for-range-copy,performance-implicit-conversion-in-loop,performance-inefficient-algorithm,performance-inefficient-string-concatenation,performance-inefficient-vector-operation,performance-move-const-arg,performance-move-constructor-init,performance-no-automatic-move,performance-noexcept-move-constructor,performance-trivially-destructible,performance-type-promotion-in-math-fn,performance-unnecessary-copy-initialization,performance-unnecessary-value-param,portability-simd-intrinsics,readability-avoid-const-params-in-decls,readability-const-return-type,readability-container-size-empty,readability-convert-member-functions-to-static,readability-delete-null-pointer,readability-deleted-default,readability-inconsistent-declaration-parameter-name,readability-make-member-function-const,readability-misleading-indentation,readability-misplaced-array-index,readability-non-const-parameter,readability-redundant-control-flow,readability-redundant-declaration,readability-redundant-function-ptr-dereference,readability-redundant-smartptr-get,readability-redundant-string-cstr,readability-redundant-string-init,readability-simplify-subscript-expr,readability-static-accessed-through-instance,readability-static-definition-in-anonymous-namespace,readability-string-compare,readability-uniqueptr-delete-release,readability-use-anyofallof" />
+      <option name="clangTidyChecks" value="-*,bugprone-argument-comment,bugprone-assert-side-effect,bugprone-bad-signal-to-kill-thread,bugprone-branch-clone,bugprone-copy-constructor-init,bugprone-dangling-handle,bugprone-dynamic-static-initializers,bugprone-fold-init-type,bugprone-forward-declaration-namespace,bugprone-forwarding-reference-overload,bugprone-inaccurate-erase,bugprone-incorrect-roundings,bugprone-integer-division,bugprone-lambda-function-name,bugprone-macro-parentheses,bugprone-macro-repeated-side-effects,bugprone-misplaced-operator-in-strlen-in-alloc,bugprone-misplaced-pointer-arithmetic-in-alloc,bugprone-misplaced-widening-cast,bugprone-move-forwarding-reference,bugprone-multiple-statement-macro,bugprone-no-escape,bugprone-not-null-terminated-result,bugprone-parent-virtual-call,bugprone-posix-return,bugprone-reserved-identifier,bugprone-sizeof-container,bugprone-sizeof-expression,bugprone-spuriously-wake-up-functions,bugprone-string-constructor,bugprone-string-integer-assignment,bugprone-string-literal-with-embedded-nul,bugprone-suspicious-enum-usage,bugprone-suspicious-include,bugprone-suspicious-memset-usage,bugprone-suspicious-missing-comma,bugprone-suspicious-semicolon,bugprone-suspicious-string-compare,bugprone-swapped-arguments,bugprone-terminating-continue,bugprone-throw-keyword-missing,bugprone-too-small-loop-variable,bugprone-undefined-memory-manipulation,bugprone-undelegated-constructor,bugprone-unhandled-self-assignment,bugprone-unused-raii,bugprone-unused-return-value,bugprone-use-after-move,bugprone-virtual-near-miss,cert-dcl21-cpp,cert-dcl58-cpp,cert-err34-c,cert-err52-cpp,cert-err58-cpp,cert-err60-cpp,cert-flp30-c,cert-msc50-cpp,cert-msc51-cpp,cppcoreguidelines-interfaces-global-init,cppcoreguidelines-narrowing-conversions,cppcoreguidelines-pro-type-static-cast-downcast,cppcoreguidelines-slicing,google-default-arguments,google-explicit-constructor,google-runtime-operator,hicpp-exception-baseclass,hicpp-multiway-paths-covered,misc-misplaced-const,misc-new-delete-overloads,misc-no-recursion,misc-non-copyable-objects,misc-throw-by-value-catch-by-reference,misc-unconventional-assign-operator,misc-uniqueptr-reset-release,modernize-avoid-bind,modernize-concat-nested-namespaces,modernize-deprecated-headers,modernize-deprecated-ios-base-aliases,modernize-loop-convert,modernize-make-shared,modernize-make-unique,modernize-pass-by-value,modernize-raw-string-literal,modernize-redundant-void-arg,modernize-replace-auto-ptr,modernize-replace-disallow-copy-and-assign-macro,modernize-replace-random-shuffle,modernize-return-braced-init-list,modernize-shrink-to-fit,modernize-unary-static-assert,modernize-use-auto,modernize-use-bool-literals,modernize-use-emplace,modernize-use-equals-default,modernize-use-equals-delete,modernize-use-nodiscard,modernize-use-noexcept,modernize-use-nullptr,modernize-use-override,modernize-use-transparent-functors,modernize-use-uncaught-exceptions,mpi-buffer-deref,mpi-type-mismatch,openmp-use-default-none,performance-faster-string-find,performance-for-range-copy,performance-implicit-conversion-in-loop,performance-inefficient-algorithm,performance-inefficient-string-concatenation,performance-inefficient-vector-operation,performance-move-const-arg,performance-move-constructor-init,performance-no-automatic-move,performance-noexcept-move-constructor,performance-trivially-destructible,performance-type-promotion-in-math-fn,performance-unnecessary-copy-initialization,performance-unnecessary-value-param,portability-simd-intrinsics,readability-avoid-const-params-in-decls,readability-const-return-type,readability-container-size-empty,readability-convert-member-functions-to-static,readability-delete-null-pointer,readability-deleted-default,readability-inconsistent-declaration-parameter-name,readability-make-member-function-const,readability-misleading-indentation,readability-misplaced-array-index,readability-non-const-parameter,readability-redundant-control-flow,readability-redundant-declaration,readability-redundant-function-ptr-dereference,readability-redundant-smartptr-get,readability-redundant-string-cstr,readability-redundant-string-init,readability-simplify-subscript-expr,readability-static-accessed-through-instance,readability-static-definition-in-anonymous-namespace,readability-string-compare,readability-uniqueptr-delete-release,readability-use-anyofallof" />
     </inspection_tool>
     <inspection_tool class="ClangTidyInspection" enabled="false" level="WARNING" enabled_by_default="false">
       <option name="useCustomListOfClangTidyChecks" value="false" />

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

@@ -61,8 +61,9 @@ namespace skyline::kernel {
             if (optimalCore != currentCore) {
                 if (!alwaysInsert && thread == state.thread)
                     RemoveThread();
-                else if (!alwaysInsert && thread != state.thread) [[unlikely]]
+                else if (!alwaysInsert && thread != state.thread)
-                    throw exception("Migrating an external thread (T{}) without 'alwaysInsert' isn't supported", thread->id);
+                    [[unlikely]]
+                        throw exception("Migrating an external thread (T{}) without 'alwaysInsert' isn't supported", thread->id);
                 thread->coreId = optimalCore->id;
                 InsertThread(thread);
                 state.logger->Debug("Load Balancing T{}: C{} -> C{}", thread->id, currentCore->id, optimalCore->id);
@@ -121,12 +122,26 @@ namespace skyline::kernel {
 
     void Scheduler::WaitSchedule(bool loadBalance) {
         auto &thread{state.thread};
-        auto *core{&cores.at(thread->coreId)};
+        CoreContext *core{&cores.at(thread->coreId)};
-
         std::unique_lock lock(core->mutex);
+
+        auto wakeFunction{[&]() {
+            if (!thread->affinityMask.test(thread->coreId)) [[unlikely]] {
+                lock.unlock();
+
+                RemoveThread();
+                thread->coreId = thread->idealCore;
+                InsertThread(thread);
+
+                core = &cores.at(thread->coreId);
+                lock = std::unique_lock(core->mutex);
+            }
+            return !core->queue.empty() && core->queue.front() == thread;
+        }};
+
         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; })) {
+            while (!thread->wakeCondition.wait_for(lock, loadBalanceThreshold, wakeFunction)) {
                 lock.unlock();
                 LoadBalance(state.thread);
                 if (thread->coreId == core->id) {
@@ -139,7 +154,7 @@ namespace skyline::kernel {
                 loadBalanceThreshold *= 2; // We double the duration required for future load balancing for this invocation to minimize pointless load balancing
             }
         } else {
-            thread->wakeCondition.wait(lock, [&]() { return !core->queue.empty() && core->queue.front() == thread; });
+            thread->wakeCondition.wait(lock, wakeFunction);
         }
 
         if (thread->priority == core->preemptionPriority) {
@@ -156,7 +171,19 @@ namespace skyline::kernel {
         auto *core{&cores.at(thread->coreId)};
 
         std::unique_lock lock(core->mutex);
-        if (thread->wakeCondition.wait_for(lock, timeout, [&]() { return !core->queue.empty() && core->queue.front() == thread; })) {
+        if (thread->wakeCondition.wait_for(lock, timeout, [&]() {
+            if (!thread->affinityMask.test(thread->coreId)) [[unlikely]] {
+                lock.unlock();
+
+                RemoveThread();
+                thread->coreId = thread->idealCore;
+                InsertThread(thread);
+
+                core = &cores.at(thread->coreId);
+                lock = std::unique_lock(core->mutex);
+            }
+            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()}};
                 timer_settime(thread->preemptionTimer, 0, &spec, nullptr);
@@ -182,11 +209,12 @@ namespace skyline::kernel {
             // Splice the linked element from the beginning of the queue to where it's priority is present
             core.queue.splice(std::upper_bound(core.queue.begin(), core.queue.end(), thread->priority.load(), type::KThread::IsHigherPriority), core.queue, core.queue.begin());
 
-            auto& front{core.queue.front()};
+            auto &front{core.queue.front()};
             if (front != thread)
                 front->wakeCondition.notify_one(); // If we aren't at the front of the queue, only then should we wake the thread at the front up
-        } else if (!thread->forceYield) { [[unlikely]]
+        } else if (!thread->forceYield) {
-            throw exception("T{} called Rotate while not being in C{}'s queue", thread->id, thread->coreId);
+            [[unlikely]]
+                throw exception("T{} called Rotate while not being in C{}'s queue", thread->id, thread->coreId);
         }
 
         thread->averageTimeslice = (thread->averageTimeslice / 4) + (3 * (util::GetTimeTicks() - thread->timesliceStart / 4));
@@ -202,6 +230,34 @@ namespace skyline::kernel {
         thread->forceYield = false;
     }
 
+    void Scheduler::RemoveThread() {
+        auto &thread{state.thread};
+        auto &core{cores.at(thread->coreId)};
+        {
+            std::unique_lock lock(core.mutex);
+            auto it{std::find(core.queue.begin(), core.queue.end(), thread)};
+            if (it != core.queue.end()) {
+                it = core.queue.erase(it);
+                if (it == core.queue.begin()) {
+                    // We need to update the averageTimeslice accordingly, if we've been unscheduled by this
+                    if (thread->timesliceStart)
+                        thread->averageTimeslice = (thread->averageTimeslice / 4) + (3 * (util::GetTimeTicks() - thread->timesliceStart / 4));
+
+                    if (it != core.queue.end())
+                        (*it)->wakeCondition.notify_one(); // We need to wake the thread at the front of the queue, if we were at the front previously
+                }
+            }
+        }
+
+        if (thread->isPreempted) {
+            struct itimerspec spec{};
+            timer_settime(thread->preemptionTimer, 0, &spec, nullptr);
+            thread->isPreempted = false;
+        }
+
+        YieldPending = false;
+    }
+
     void Scheduler::UpdatePriority(const std::shared_ptr<type::KThread> &thread) {
         std::lock_guard migrationLock(thread->coreMigrationMutex);
         auto *core{&cores.at(thread->coreId)};
@@ -254,6 +310,15 @@ namespace skyline::kernel {
         }
     }
 
+    void Scheduler::UpdateCore(const std::shared_ptr<type::KThread> &thread) {
+        auto *core{&cores.at(thread->coreId)};
+        std::unique_lock coreLock(core->mutex);
+        if (core->queue.front() == thread)
+            thread->SendSignal(YieldSignal);
+        else
+            thread->wakeCondition.notify_one();
+    }
+
     void Scheduler::ParkThread() {
         auto &thread{state.thread};
         std::lock_guard migrationLock(thread->coreMigrationMutex);
@@ -293,32 +358,4 @@ namespace skyline::kernel {
             }
         }
     }
-
-    void Scheduler::RemoveThread() {
-        auto &thread{state.thread};
-        auto &core{cores.at(thread->coreId)};
-        {
-            std::unique_lock lock(core.mutex);
-            auto it{std::find(core.queue.begin(), core.queue.end(), thread)};
-            if (it != core.queue.end()) {
-                it = core.queue.erase(it);
-                if (it == core.queue.begin()) {
-                    // We need to update the averageTimeslice accordingly, if we've been unscheduled by this
-                    if (thread->timesliceStart)
-                        thread->averageTimeslice = (thread->averageTimeslice / 4) + (3 * (util::GetTimeTicks() - thread->timesliceStart / 4));
-
-                    if (it != core.queue.end())
-                        (*it)->wakeCondition.notify_one(); // We need to wake the thread at the front of the queue, if we were at the front previously
-                }
-            }
-        }
-
-        if (thread->isPreempted) {
-            struct itimerspec spec{};
-            timer_settime(thread->preemptionTimer, 0, &spec, nullptr);
-            thread->isPreempted = false;
-        }
-
-        YieldPending = false;
-    }
 }

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

@@ -103,11 +103,22 @@ namespace skyline {
              */
             void Rotate(bool cooperative = true);
 
+            /**
+             * @brief Removes the calling thread from it's resident core queue
+             */
+            void RemoveThread();
+
             /**
              * @brief Updates the placement of the supplied thread in it's resident core's queue according to it's new priority
              */
             void UpdatePriority(const std::shared_ptr<type::KThread>& thread);
 
+            /**
+             * @brief Updates the core that the supplied thread is resident to according to it's new affinity mask and ideal core
+             * @note This supports changing the core of a thread which is currently running
+             */
+            void UpdateCore(const std::shared_ptr<type::KThread>& thread);
+
             /**
              * @brief Parks the calling thread after removing it from it's resident core's queue and inserts it on the core it's been awoken on
              * @note This will not handle waiting for the thread to be scheduled, this should be followed with a call to WaitSchedule/TimedWaitSchedule
@@ -119,11 +130,6 @@ namespace skyline {
              * @note We will only wake a thread if it is determined to be a better pick than the thread which would be run on this core next
              */
             void WakeParkedThread();
-
-            /**
-             * @brief Removes the calling thread from it's resident core queue
-             */
-            void RemoveThread();
         };
 
         /**

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

@@ -412,10 +412,11 @@ namespace skyline::kernel::svc {
 
             state.logger->Debug("svcSetThreadCoreMask: Setting thread #{}'s Ideal Core ({}) + Affinity Mask ({})", thread->id, idealCore, affinityMask);
 
+            std::lock_guard guard(thread->coreMigrationMutex);
             thread->idealCore = idealCore;
             thread->affinityMask = affinityMask;
 
-            if (!affinityMask.test(thread->coreId)) {
+            if (!affinityMask.test(thread->coreId) && thread->coreId != constant::ParkedCoreId) {
                 state.logger->Debug("svcSetThreadCoreMask: Migrating thread #{} to Ideal Core C{} -> C{}", thread->id, thread->coreId, idealCore);
 
                 if (thread == state.thread) {
@@ -425,7 +426,7 @@ namespace skyline::kernel::svc {
                 } else if (!thread->running) {
                     thread->coreId = idealCore;
                 } else {
-                    throw exception("svcSetThreadCoreMask: Migrating a running thread due to a new core mask is not supported");
+                    state.scheduler->UpdateCore(thread);
                 }
             }
 

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

@@ -41,7 +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