Changes in libcfa/src/concurrency/kernel.cfa [e3fea42:a7b486b]

File:

• libcfa/src/concurrency/kernel.cfa (modified) (40 diffs)

libcfa/src/concurrency/kernel.cfa

@@ -15 +15 @@
 
 #define __cforall_thread__
+// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
 
 //C Includes
@@ -40 +41 @@
 #include "invoke.h"
 
+
 //-----------------------------------------------------------------------------
 // Some assembly required
@@ -110 +112 @@
 //-----------------------------------------------------------------------------
 //Start and stop routine for the kernel, declared first to make sure they run first
-static void kernel_startup(void)  __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
-static void kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
+static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
+static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
+
+//-----------------------------------------------------------------------------
+// Kernel Scheduling logic
+static $thread * __next_thread(cluster * this);
+static void __run_thread(processor * this, $thread * dst);
+static $thread * __halt(processor * this);
+static bool __wake_one(cluster * cltr, bool was_empty);
+static bool __wake_proc(processor *);
 
 //-----------------------------------------------------------------------------
@@ -117 +127 @@
 KERNEL_STORAGE(cluster,         mainCluster);
 KERNEL_STORAGE(processor,       mainProcessor);
-KERNEL_STORAGE(thread_desc,     mainThread);
+KERNEL_STORAGE($thread, mainThread);
 KERNEL_STORAGE(__stack_t,       mainThreadCtx);
 
 cluster     * mainCluster;
 processor   * mainProcessor;
-thread_desc * mainThread;
+$thread * mainThread;
 
 extern "C" {
@@ -164 +174 @@
 // Main thread construction
 
-void ?{}( coroutine_desc & this, current_stack_info_t * info) with( this ) {
+void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) {
         stack.storage = info->storage;
         with(*stack.storage) {
@@ -179 +189 @@
 }
 
-void ?{}( thread_desc & this, current_stack_info_t * info) with( this ) {
+void ?{}( $thread & this, current_stack_info_t * info) with( this ) {
         state = Start;
         self_cor{ info };
@@ -208 +218 @@
 }
 
-static void start(processor * this);
+static void * __invoke_processor(void * arg);
+
 void ?{}(processor & this, const char name[], cluster & cltr) with( this ) {
         this.name = name;
         this.cltr = &cltr;
         terminated{ 0 };
+        destroyer = 0p;
         do_terminate = false;
         preemption_alarm = 0p;
@@ -218 +230 @@
         runner.proc = &this;
 
-        idleLock{};
-
-        start( &this );
+        idle{};
+
+        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
+
+        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
+
+        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
 }
 
 void ^?{}(processor & this) with( this ){
         if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
-                __cfaabi_dbg_print_safe("Kernel : core %p signaling termination\n", &this);
+                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
 
                 __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
-                wake( &this );
+                __wake_proc( &this );
 
                 P( terminated );
@@ -234 +250 @@
         }
 
-        pthread_join( kernel_thread, 0p );
+        int err = pthread_join( kernel_thread, 0p );
+        if( err != 0 ) abort("KERNEL ERROR: joining processor %p caused error %s\n", &this, strerror(err));
+
         free( this.stack );
 }
 
-void ?{}(cluster & this, const char name[], Duration preemption_rate) with( this ) {
+void ?{}(cluster & this, const char name[], Duration preemption_rate, int io_flags) with( this ) {
         this.name = name;
         this.preemption_rate = preemption_rate;
@@ -244 +262 @@
         ready_queue_lock{};
 
+        #if !defined(__CFA_NO_STATISTICS__)
+                print_stats = false;
+        #endif
+
         procs{ __get };
         idles{ __get };
         threads{ __get };
 
+        __kernel_io_startup( this, io_flags, &this == mainCluster );
+
         doregister(this);
 }
 
 void ^?{}(cluster & this) {
+        __kernel_io_shutdown( this, &this == mainCluster );
+
         unregister(this);
 }
@@ -258 +284 @@
 // Kernel Scheduling logic
 //=============================================================================================
-static void runThread(processor * this, thread_desc * dst);
-static void finishRunning(processor * this);
-static void halt(processor * this);
-
 //Main of the processor contexts
 void main(processorCtx_t & runner) {
@@ -271 +293 @@
         verify(this);
 
-        __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this);
+        __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
 
         doregister(this->cltr, this);
@@ -279 +301 @@
                 preemption_scope scope = { this };
 
-                __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);
-
-                thread_desc * readyThread = 0p;
+                __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
+
+                $thread * readyThread = 0p;
                 for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
-                        readyThread = nextThread( this->cltr );
-
-                        if(readyThread) {
-                                verify( ! kernelTLS.preemption_state.enabled );
-
-                                runThread(this, readyThread);
-
-                                verify( ! kernelTLS.preemption_state.enabled );
-
-                                //Some actions need to be taken from the kernel
-                                finishRunning(this);
-
-                                spin_count = 0;
-                        } else {
-                                // spin(this, &spin_count);
-                                halt(this);
+                        // Try to get the next thread
+                        readyThread = __next_thread( this->cltr );
+
+                        // If no ready thread
+                        if( readyThread == 0p ) {
+                                // Block until a thread is ready
+                                readyThread = __halt(this);
+                        }
+
+                        // Check if we actually found a thread
+                        if( readyThread ) {
+                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                                /* paranoid */ verifyf( readyThread->state == Ready || readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
+                                /* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );
+
+                                // We found a thread run it
+                                __run_thread(this, readyThread);
+
+                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
                         }
                 }
 
-                __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this);
+                __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
         }
 
@@ -309 +334 @@
         V( this->terminated );
 
-        __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this);
+        __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
+
+        // HACK : the coroutine context switch expects this_thread to be set
+        // and it make sense for it to be set in all other cases except here
+        // fake it
+        if( this == mainProcessor ) kernelTLS.this_thread = mainThread;
 }
 
@@ -318 +348 @@
 // runThread runs a thread by context switching
 // from the processor coroutine to the target thread
-static void runThread(processor * this, thread_desc * thrd_dst) {
-        coroutine_desc * proc_cor = get_coroutine(this->runner);
-
-        // Reset the terminating actions here
-        this->finish.action_code = No_Action;
+static void __run_thread(processor * this, $thread * thrd_dst) {
+        $coroutine * proc_cor = get_coroutine(this->runner);
 
         // Update global state
         kernelTLS.this_thread = thrd_dst;
 
-        // set state of processor coroutine to inactive and the thread to active
-        proc_cor->state = proc_cor->state == Halted ? Halted : Inactive;
-        thrd_dst->state = Active;
-
-        // set context switch to the thread that the processor is executing
-        verify( thrd_dst->context.SP );
-        CtxSwitch( &proc_cor->context, &thrd_dst->context );
-        // when CtxSwitch returns we are back in the processor coroutine
-
-        // set state of processor coroutine to active and the thread to inactive
-        thrd_dst->state = thrd_dst->state == Halted ? Halted : Inactive;
+        // set state of processor coroutine to inactive
+        verify(proc_cor->state == Active);
+        proc_cor->state = Blocked;
+
+        // Actually run the thread
+        RUNNING:  while(true) {
+                if(unlikely(thrd_dst->preempted)) {
+                        thrd_dst->preempted = __NO_PREEMPTION;
+                        verify(thrd_dst->state == Active  || thrd_dst->state == Rerun);
+                } else {
+                        verify(thrd_dst->state == Blocked || thrd_dst->state == Ready); // Ready means scheduled normally, blocked means rerun
+                        thrd_dst->state = Active;
+                }
+
+                __cfaabi_dbg_debug_do(
+                        thrd_dst->park_stale   = true;
+                        thrd_dst->unpark_stale = true;
+                )
+
+                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
+                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) || thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst ); // add escape condition if we are setting up the processor
+                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) || thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst ); // add escape condition if we are setting up the processor
+
+                // set context switch to the thread that the processor is executing
+                verify( thrd_dst->context.SP );
+                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
+                // when __cfactx_switch returns we are back in the processor coroutine
+
+                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst );
+                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
+                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
+                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+
+
+                // We just finished running a thread, there are a few things that could have happened.
+                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
+                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
+                // 4 - Preempted
+                // In case 1, we may have won a race so we can't write to the state again.
+                // In case 2, we lost the race so we now own the thread.
+
+                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
+                        // The thread was preempted, reschedule it and reset the flag
+                        __schedule_thread( thrd_dst );
+                        break RUNNING;
+                }
+
+                // set state of processor coroutine to active and the thread to inactive
+                static_assert(sizeof(thrd_dst->state) == sizeof(int));
+                enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Blocked, __ATOMIC_SEQ_CST);
+                __cfaabi_dbg_debug_do( thrd_dst->park_result = old_state; )
+                switch(old_state) {
+                        case Halted:
+                                // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
+                                thrd_dst->state = Halted;
+
+                                // We may need to wake someone up here since
+                                unpark( this->destroyer __cfaabi_dbg_ctx2 );
+                                this->destroyer = 0p;
+                                break RUNNING;
+                        case Active:
+                                // This is case 1, the regular case, nothing more is needed
+                                break RUNNING;
+                        case Rerun:
+                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
+                                // In this case, just run it again.
+                                continue RUNNING;
+                        default:
+                                // This makes no sense, something is wrong abort
+                                abort("Finished running a thread that was Blocked/Start/Primed %d\n", old_state);
+                }
+        }
+
+        // Just before returning to the processor, set the processor coroutine to active
         proc_cor->state = Active;
+        kernelTLS.this_thread = 0p;
 }
 
 // KERNEL_ONLY
-static void returnToKernel() {
-        coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
-        thread_desc * thrd_src = kernelTLS.this_thread;
-
-        // set state of current coroutine to inactive
-        thrd_src->state = thrd_src->state == Halted ? Halted : Inactive;
-        proc_cor->state = Active;
-        int local_errno = *__volatile_errno();
-        #if defined( __i386 ) || defined( __x86_64 )
-                __x87_store;
-        #endif
-
-        // set new coroutine that the processor is executing
-        // and context switch to it
-        verify( proc_cor->context.SP );
-        CtxSwitch( &thrd_src->context, &proc_cor->context );
-
-        // set state of new coroutine to active
-        proc_cor->state = proc_cor->state == Halted ? Halted : Inactive;
-        thrd_src->state = Active;
-
-        #if defined( __i386 ) || defined( __x86_64 )
-                __x87_load;
-        #endif
-        *__volatile_errno() = local_errno;
-}
-
-// KERNEL_ONLY
-// Once a thread has finished running, some of
-// its final actions must be executed from the kernel
-static void finishRunning(processor * this) with( this->finish ) {
-        verify( ! kernelTLS.preemption_state.enabled );
-        choose( action_code ) {
-        case No_Action:
-                break;
-        case Release:
-                unlock( *lock );
-        case Schedule:
-                ScheduleThread( thrd );
-        case Release_Schedule:
-                unlock( *lock );
-                ScheduleThread( thrd );
-        case Release_Multi:
-                for(int i = 0; i < lock_count; i++) {
-                        unlock( *locks[i] );
-                }
-        case Release_Multi_Schedule:
-                for(int i = 0; i < lock_count; i++) {
-                        unlock( *locks[i] );
-                }
-                for(int i = 0; i < thrd_count; i++) {
-                        ScheduleThread( thrds[i] );
-                }
-        case Callback:
-                callback();
-        default:
-                abort("KERNEL ERROR: Unexpected action to run after thread");
-        }
+void returnToKernel() {
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        $coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
+        $thread * thrd_src = kernelTLS.this_thread;
+
+        // Run the thread on this processor
+        {
+                int local_errno = *__volatile_errno();
+                #if defined( __i386 ) || defined( __x86_64 )
+                        __x87_store;
+                #endif
+                verify( proc_cor->context.SP );
+                __cfactx_switch( &thrd_src->context, &proc_cor->context );
+                #if defined( __i386 ) || defined( __x86_64 )
+                        __x87_load;
+                #endif
+                *__volatile_errno() = local_errno;
+        }
+
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too small.\n", thrd_src );
+        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too large.\n", thrd_src );
 }
 
@@ -406 +462 @@
 // This is the entry point for processors (kernel threads)
 // It effectively constructs a coroutine by stealing the pthread stack
-static void * CtxInvokeProcessor(void * arg) {
+static void * __invoke_processor(void * arg) {
         processor * proc = (processor *) arg;
         kernelTLS.this_processor = proc;
@@ -425 +481 @@
 
         //We now have a proper context from which to schedule threads
-        __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
+        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
 
         // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
@@ -436 +492 @@
 
         // Main routine of the core returned, the core is now fully terminated
-        __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner);
+        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
 
         return 0p;
@@ -447 +503 @@
 } // Abort
 
-void * create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
+void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
         pthread_attr_t attr;
 
@@ -469 +525 @@
         );
 
-        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" ); 
+        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
 
         Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
@@ -475 +531 @@
 }
 
-static void start(processor * this) {
-        __cfaabi_dbg_print_safe("Kernel : Starting core %p\n", this);
-
-        this->stack = create_pthread( &this->kernel_thread, CtxInvokeProcessor, (void *)this );
-
-        __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);
-}
-
 // KERNEL_ONLY
-void kernel_first_resume( processor * this ) {
-        thread_desc * src = mainThread;
-        coroutine_desc * dst = get_coroutine(this->runner);
+static void __kernel_first_resume( processor * this ) {
+        $thread * src = mainThread;
+        $coroutine * dst = get_coroutine(this->runner);
 
         verify( ! kernelTLS.preemption_state.enabled );
 
+        kernelTLS.this_thread->curr_cor = dst;
         __stack_prepare( &dst->stack, 65000 );
-        CtxStart(&this->runner, CtxInvokeCoroutine);
+        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
 
         verify( ! kernelTLS.preemption_state.enabled );
@@ -498 +547 @@
         dst->starter = dst->starter ? dst->starter : &src->self_cor;
 
-        // set state of current coroutine to inactive
-        src->state = src->state == Halted ? Halted : Inactive;
+        // make sure the current state is still correct
+        /* paranoid */ verify(src->state == Ready);
 
         // context switch to specified coroutine
         verify( dst->context.SP );
-        CtxSwitch( &src->context, &dst->context );
-        // when CtxSwitch returns we are back in the src coroutine
-
-        // set state of new coroutine to active
-        src->state = Active;
+        __cfactx_switch( &src->context, &dst->context );
+        // when __cfactx_switch returns we are back in the src coroutine
+
+        mainThread->curr_cor = &mainThread->self_cor;
+
+        // make sure the current state has been update
+        /* paranoid */ verify(src->state == Active);
 
         verify( ! kernelTLS.preemption_state.enabled );
@@ -513 +564 @@
 
 // KERNEL_ONLY
-void kernel_last_resume( processor * this ) {
-        coroutine_desc * src = &mainThread->self_cor;
-        coroutine_desc * dst = get_coroutine(this->runner);
+static void __kernel_last_resume( processor * this ) {
+        $coroutine * src = &mainThread->self_cor;
+        $coroutine * dst = get_coroutine(this->runner);
 
         verify( ! kernelTLS.preemption_state.enabled );
@@ -521 +572 @@
         verify( dst->context.SP );
 
+        // SKULLDUGGERY in debug the processors check that the
+        // stack is still within the limit of the stack limits after running a thread.
+        // that check doesn't make sense if we context switch to the processor using the
+        // coroutine semantics. Since this is a special case, use the current context
+        // info to populate these fields.
+        __cfaabi_dbg_debug_do(
+                __stack_context_t ctx;
+                CtxGet( ctx );
+                mainThread->context.SP = ctx.SP;
+                mainThread->context.FP = ctx.FP;
+        )
+
         // context switch to the processor
-        CtxSwitch( &src->context, &dst->context );
+        __cfactx_switch( &src->context, &dst->context );
 }
 
 //-----------------------------------------------------------------------------
 // Scheduler routines
-
 // KERNEL ONLY
-void ScheduleThread( thread_desc * thrd ) {
-        verify( thrd );
-        verify( thrd->state != Halted );
-
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
-
-        with( *thrd->curr_cluster ) {
-                lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
-                bool was_empty = !(ready_queue != 0);
-                append( ready_queue, thrd );
-                unlock( ready_queue_lock );
-
-                if(was_empty) {
-                        lock      (proc_list_lock __cfaabi_dbg_ctx2);
-                        if(idles) {
-                                wake_fast(idles.head);
-                        }
-                        unlock    (proc_list_lock);
-                }
-                else if( struct processor * idle = idles.head ) {
-                        wake_fast(idle);
-                }
-
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
+void __schedule_thread( $thread * thrd ) with( *thrd->curr_cluster ) {
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
+        /* paranoid */ if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
+                          "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
+        /* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active || thrd->state == Rerun,
+                          "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
+        /* paranoid */ #endif
+        /* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
+
+        if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
+
+        lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
+        bool was_empty = !(ready_queue != 0);
+        append( ready_queue, thrd );
+        unlock( ready_queue_lock );
+
+        __wake_one(thrd->curr_cluster, was_empty);
+
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
 }
 
 // KERNEL ONLY
-thread_desc * nextThread(cluster * this) with( *this ) {
-        verify( ! kernelTLS.preemption_state.enabled );
+static $thread * __next_thread(cluster * this) with( *this ) {
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+
         lock( ready_queue_lock __cfaabi_dbg_ctx2 );
-        thread_desc * head = pop_head( ready_queue );
+        $thread * head = pop_head( ready_queue );
         unlock( ready_queue_lock );
-        verify( ! kernelTLS.preemption_state.enabled );
+
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
         return head;
 }
 
-void BlockInternal() {
+// KERNEL ONLY unpark with out disabling interrupts
+void __unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
+        static_assert(sizeof(thrd->state) == sizeof(int));
+
+        // record activity
+        __cfaabi_dbg_record_thrd( *thrd, false, caller );
+
+        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
+        __cfaabi_dbg_debug_do( thrd->unpark_result = old_state; )
+        switch(old_state) {
+                case Active:
+                        // Wake won the race, the thread will reschedule/rerun itself
+                        break;
+                case Blocked:
+                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
+
+                        // Wake lost the race,
+                        thrd->state = Blocked;
+                        __schedule_thread( thrd );
+                        break;
+                case Rerun:
+                        abort("More than one thread attempted to schedule thread %p\n", thrd);
+                        break;
+                case Halted:
+                case Start:
+                case Primed:
+                default:
+                        // This makes no sense, something is wrong abort
+                        abort();
+        }
+}
+
+void unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
+        if( !thrd ) return;
+
         disable_interrupts();
-        verify( ! kernelTLS.preemption_state.enabled );
+        __unpark( thrd __cfaabi_dbg_ctx_fwd2 );
+        enable_interrupts( __cfaabi_dbg_ctx );
+}
+
+void park( __cfaabi_dbg_ctx_param ) {
+        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+        disable_interrupts();
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
+
+        // record activity
+        __cfaabi_dbg_record_thrd( *kernelTLS.this_thread, true, caller );
+
         returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
+
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
         enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal( __spinlock_t * lock ) {
+        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+
+}
+
+// KERNEL ONLY
+void __leave_thread() {
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        returnToKernel();
+        abort();
+}
+
+// KERNEL ONLY
+bool force_yield( __Preemption_Reason reason ) {
+        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
         disable_interrupts();
-        with( *kernelTLS.this_processor ) {
-                finish.action_code = Release;
-                finish.lock        = lock;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal( thread_desc * thrd ) {
-        disable_interrupts();
-        with( * kernelTLS.this_processor ) {
-                finish.action_code = Schedule;
-                finish.thrd        = thrd;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal( __spinlock_t * lock, thread_desc * thrd ) {
-        assert(thrd);
-        disable_interrupts();
-        with( * kernelTLS.this_processor ) {
-                finish.action_code = Release_Schedule;
-                finish.lock        = lock;
-                finish.thrd        = thrd;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal(__spinlock_t * locks [], unsigned short count) {
-        disable_interrupts();
-        with( * kernelTLS.this_processor ) {
-                finish.action_code = Release_Multi;
-                finish.locks       = locks;
-                finish.lock_count  = count;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal(__spinlock_t * locks [], unsigned short lock_count, thread_desc * thrds [], unsigned short thrd_count) {
-        disable_interrupts();
-        with( *kernelTLS.this_processor ) {
-                finish.action_code = Release_Multi_Schedule;
-                finish.locks       = locks;
-                finish.lock_count  = lock_count;
-                finish.thrds       = thrds;
-                finish.thrd_count  = thrd_count;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-void BlockInternal(__finish_callback_fptr_t callback) {
-        disable_interrupts();
-        with( *kernelTLS.this_processor ) {
-                finish.action_code = Callback;
-                finish.callback    = callback;
-        }
-
-        verify( ! kernelTLS.preemption_state.enabled );
-        returnToKernel();
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        enable_interrupts( __cfaabi_dbg_ctx );
-}
-
-// KERNEL ONLY
-void LeaveThread(__spinlock_t * lock, thread_desc * thrd) {
-        verify( ! kernelTLS.preemption_state.enabled );
-        with( * kernelTLS.this_processor ) {
-                finish.action_code = thrd ? Release_Schedule : Release;
-                finish.lock        = lock;
-                finish.thrd        = thrd;
-        }
-
-        returnToKernel();
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+
+        $thread * thrd = kernelTLS.this_thread;
+        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun);
+
+        // SKULLDUGGERY: It is possible that we are preempting this thread just before
+        // it was going to park itself. If that is the case and it is already using the
+        // intrusive fields then we can't use them to preempt the thread
+        // If that is the case, abandon the preemption.
+        bool preempted = false;
+        if(thrd->next == 0p) {
+                preempted = true;
+                thrd->preempted = reason;
+                returnToKernel();
+        }
+
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        enable_interrupts_noPoll();
+        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+
+        return preempted;
 }
 
@@ -684 +721 @@
 //-----------------------------------------------------------------------------
 // Kernel boot procedures
-static void kernel_startup(void) {
+static void __kernel_startup(void) {
         verify( ! kernelTLS.preemption_state.enabled );
-        __cfaabi_dbg_print_safe("Kernel : Starting\n");
+        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
 
         __page_size = sysconf( _SC_PAGESIZE );
@@ -697 +734 @@
         (*mainCluster){"Main Cluster"};
 
-        __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n");
+        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
 
         // Start by initializing the main thread
         // SKULLDUGGERY: the mainThread steals the process main thread
         // which will then be scheduled by the mainProcessor normally
-        mainThread = (thread_desc *)&storage_mainThread;
+        mainThread = ($thread *)&storage_mainThread;
         current_stack_info_t info;
         info.storage = (__stack_t*)&storage_mainThreadCtx;
         (*mainThread){ &info };
 
-        __cfaabi_dbg_print_safe("Kernel : Main thread ready\n");
+        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
 
 
@@ -728 +765 @@
 
                 runner{ &this };
-                __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner);
+                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
         }
 
@@ -745 +782 @@
         // Add the main thread to the ready queue
         // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
-        ScheduleThread(mainThread);
+        __schedule_thread(mainThread);
 
         // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
-        // context. Hence, the main thread does not begin through CtxInvokeThread, like all other threads. The trick here is that
+        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
         // mainThread is on the ready queue when this call is made.
-        kernel_first_resume( kernelTLS.this_processor );
-
+        __kernel_first_resume( kernelTLS.this_processor );
 
 
         // THE SYSTEM IS NOW COMPLETELY RUNNING
-        __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n");
+
+
+        // Now that the system is up, finish creating systems that need threading
+        __kernel_io_finish_start( *mainCluster );
+
+
+        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
 
         verify( ! kernelTLS.preemption_state.enabled );
@@ -762 +804 @@
 }
 
-static void kernel_shutdown(void) {
-        __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n");
-
-        verify( TL_GET( preemption_state.enabled ) );
+static void __kernel_shutdown(void) {
+        //Before we start shutting things down, wait for systems that need threading to shutdown
+        __kernel_io_prepare_stop( *mainCluster );
+
+        /* paranoid */ verify( TL_GET( preemption_state.enabled ) );
         disable_interrupts();
-        verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+
+        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
 
         // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
@@ -773 +818 @@
         // which is currently here
         __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
-        kernel_last_resume( kernelTLS.this_processor );
+        __kernel_last_resume( kernelTLS.this_processor );
         mainThread->self_cor.state = Halted;
 
@@ -783 +828 @@
         // Destroy the main processor and its context in reverse order of construction
         // These were manually constructed so we need manually destroy them
-        ^(mainProcessor->runner){};
-        ^(mainProcessor){};
+        void ^?{}(processor & this) with( this ){
+                /* paranoid */ verify( this.do_terminate == true );
+        }
+
+        ^(*mainProcessor){};
 
         // Final step, destroy the main thread since it is no longer needed
         // Since we provided a stack to this taxk it will not destroy anything
-        ^(mainThread){};
+        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
+        ^(*mainThread){};
+
+        ^(*mainCluster){};
 
         ^(__cfa_dbg_global_clusters.list){};
         ^(__cfa_dbg_global_clusters.lock){};
 
-        __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n");
+        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
 }
 
 //=============================================================================================
-// Kernel Quiescing
+// Kernel Idle Sleep
 //=============================================================================================
-static void halt(processor * this) with( *this ) {
-        // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) );
-
+static $thread * __halt(processor * this) with( *this ) {
+        if( do_terminate ) return 0p;
+
+        // First, lock the cluster idle
+        lock( cltr->idle_lock __cfaabi_dbg_ctx2 );
+
+        // Check if we can find a thread
+        if( $thread * found = __next_thread( cltr ) ) {
+                unlock( cltr->idle_lock );
+                return found;
+        }
+
+        // Move this processor from the active list to the idle list
+        move_to_front(cltr->procs, cltr->idles, *this);
+
+        // Unlock the idle lock so we don't go to sleep with a lock
+        unlock    (cltr->idle_lock);
+
+        // We are ready to sleep
+        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p ready to sleep\n", this);
+        wait( idle );
+
+        // We have woken up
+        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p woke up and ready to run\n", this);
+
+        // Get ourself off the idle list
         with( *cltr ) {
-                lock      (proc_list_lock __cfaabi_dbg_ctx2);
-                remove    (procs, *this);
-                push_front(idles, *this);
-                unlock    (proc_list_lock);
-        }
-
-        __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this);
-
-        wait( idleLock );
-
-        __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this);
-
-        with( *cltr ) {
-                lock      (proc_list_lock __cfaabi_dbg_ctx2);
-                remove    (idles, *this);
-                push_front(procs, *this);
-                unlock    (proc_list_lock);
-        }
+                lock  (idle_lock __cfaabi_dbg_ctx2);
+                move_to_front(idles, procs, *this);
+                unlock(idle_lock);
+        }
+
+        // Don't check the ready queue again, we may not be in a position to run a thread
+        return 0p;
+}
+
+// Wake a thread from the front if there are any
+static bool __wake_one(cluster * this, __attribute__((unused)) bool force) {
+        // if we don't want to force check if we know it's false
+        // if( !this->idles.head && !force ) return false;
+
+        // First, lock the cluster idle
+        lock( this->idle_lock __cfaabi_dbg_ctx2 );
+
+        // Check if there is someone to wake up
+        if( !this->idles.head ) {
+                // Nope unlock and return false
+                unlock( this->idle_lock );
+                return false;
+        }
+
+        // Wake them up
+        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this->idles.head);
+        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        post( this->idles.head->idle );
+
+        // Unlock and return true
+        unlock( this->idle_lock );
+        return true;
+}
+
+// Unconditionnaly wake a thread
+static bool __wake_proc(processor * this) {
+        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
+
+        disable_interrupts();
+                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                bool ret = post( this->idle );
+        enable_interrupts( __cfaabi_dbg_ctx );
+
+        return ret;
 }
 
@@ -854 +954 @@
 
 void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
-        thread_desc * thrd = kernel_data;
+        $thread * thrd = kernel_data;
 
         if(thrd) {
@@ -902 +1002 @@
 void ^?{}(semaphore & this) {}
 
-void P(semaphore & this) with( this ){
+bool P(semaphore & this) with( this ){
         lock( lock __cfaabi_dbg_ctx2 );
         count -= 1;
@@ -910 +1010 @@
 
                 // atomically release spin lock and block
-                BlockInternal( &lock );
+                unlock( lock );
+                park( __cfaabi_dbg_ctx );
+                return true;
         }
         else {
             unlock( lock );
-        }
-}
-
-void V(semaphore & this) with( this ) {
-        thread_desc * thrd = 0p;
+            return false;
+        }
+}
+
+bool V(semaphore & this) with( this ) {
+        $thread * thrd = 0p;
         lock( lock __cfaabi_dbg_ctx2 );
         count += 1;
@@ -929 +1032 @@
 
         // make new owner
-        WakeThread( thrd );
+        unpark( thrd __cfaabi_dbg_ctx2 );
+
+        return thrd != 0p;
+}
+
+bool V(semaphore & this, unsigned diff) with( this ) {
+        $thread * thrd = 0p;
+        lock( lock __cfaabi_dbg_ctx2 );
+        int release = max(-count, (int)diff);
+        count += diff;
+        for(release) {
+                unpark( pop_head( waiting ) __cfaabi_dbg_ctx2 );
+        }
+
+        unlock( lock );
+
+        return thrd != 0p;
 }
 
@@ -946 +1065 @@
 }
 
-void doregister( cluster * cltr, thread_desc & thrd ) {
+void doregister( cluster * cltr, $thread & thrd ) {
         lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
         cltr->nthreads += 1;
@@ -953 +1072 @@
 }
 
-void unregister( cluster * cltr, thread_desc & thrd ) {
+void unregister( cluster * cltr, $thread & thrd ) {
         lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
         remove(cltr->threads, thrd );
@@ -961 +1080 @@
 
 void doregister( cluster * cltr, processor * proc ) {
-        lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2);
+        lock      (cltr->idle_lock __cfaabi_dbg_ctx2);
         cltr->nprocessors += 1;
         push_front(cltr->procs, *proc);
-        unlock    (cltr->proc_list_lock);
+        unlock    (cltr->idle_lock);
 }
 
 void unregister( cluster * cltr, processor * proc ) {
-        lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2);
+        lock  (cltr->idle_lock __cfaabi_dbg_ctx2);
         remove(cltr->procs, *proc );
         cltr->nprocessors -= 1;
-        unlock(cltr->proc_list_lock);
+        unlock(cltr->idle_lock);
 }
 
@@ -978 +1097 @@
 __cfaabi_dbg_debug_do(
         extern "C" {
-                void __cfaabi_dbg_record(__spinlock_t & this, const char prev_name[]) {
+                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
                         this.prev_name = prev_name;
                         this.prev_thrd = kernelTLS.this_thread;
                 }
+
+                void __cfaabi_dbg_record_thrd($thread & this, bool park, const char prev_name[]) {
+                        if(park) {
+                                this.park_caller   = prev_name;
+                                this.park_stale    = false;
+                        }
+                        else {
+                                this.unpark_caller = prev_name;
+                                this.unpark_stale  = false;
+                        }
+                }
         }
 )
@@ -987 +1117 @@
 //-----------------------------------------------------------------------------
 // Debug
-bool threading_enabled(void) {
+bool threading_enabled(void) __attribute__((const)) {
         return true;
 }