source: libcfa/src/concurrency/kernel.cfa@ 3381ed7

//
// Cforall Version 1.0.0 Copyright (C) 2016 University of Waterloo
//
// The contents of this file are covered under the licence agreement in the
// file "LICENCE" distributed with Cforall.
//
// kernel.c --
//
// Author           : Thierry Delisle
// Created On       : Tue Jan 17 12:27:26 2017
// Last Modified By : Peter A. Buhr
// Last Modified On : Thu Jan 30 22:55:50 2020
// Update Count     : 56
//

#define __cforall_thread__

//C Includes
#include <stddef.h>
#include <errno.h>
#include <string.h>
extern "C" {
#include <stdio.h>
#include <fenv.h>
#include <sys/resource.h>
#include <signal.h>
#include <unistd.h>
#include <limits.h>                                                                             // PTHREAD_STACK_MIN
#include <sys/mman.h>                                                                   // mprotect
}

//CFA Includes
#include "time.hfa"
#include "kernel_private.hfa"
#include "preemption.hfa"
#include "startup.hfa"

//Private includes
#define __CFA_INVOKE_PRIVATE__
#include "invoke.h"

//-----------------------------------------------------------------------------
// Some assembly required
#if defined( __i386 )
        #define CtxGet( ctx )        \
                __asm__ volatile (     \
                        "movl %%esp,%0\n"\
                        "movl %%ebp,%1\n"\
                        : "=rm" (ctx.SP),\
                                "=rm" (ctx.FP) \
                )

        // mxcr : SSE Status and Control bits (control bits are preserved across function calls)
        // fcw  : X87 FPU control word (preserved across function calls)
        #define __x87_store         \
                uint32_t __mxcr;      \
                uint16_t __fcw;       \
                __asm__ volatile (    \
                        "stmxcsr %0\n"  \
                        "fnstcw  %1\n"  \
                        : "=m" (__mxcr),\
                                "=m" (__fcw)  \
                )

        #define __x87_load         \
                __asm__ volatile (   \
                        "fldcw  %1\n"  \
                        "ldmxcsr %0\n" \
                        ::"m" (__mxcr),\
                                "m" (__fcw)  \
                )

#elif defined( __x86_64 )
        #define CtxGet( ctx )        \
                __asm__ volatile (     \
                        "movq %%rsp,%0\n"\
                        "movq %%rbp,%1\n"\
                        : "=rm" (ctx.SP),\
                                "=rm" (ctx.FP) \
                )

        #define __x87_store         \
                uint32_t __mxcr;      \
                uint16_t __fcw;       \
                __asm__ volatile (    \
                        "stmxcsr %0\n"  \
                        "fnstcw  %1\n"  \
                        : "=m" (__mxcr),\
                                "=m" (__fcw)  \
                )

        #define __x87_load          \
                __asm__ volatile (    \
                        "fldcw  %1\n"   \
                        "ldmxcsr %0\n"  \
                        :: "m" (__mxcr),\
                                "m" (__fcw)  \
                )


#elif defined( __ARM_ARCH )
#define CtxGet( ctx ) __asm__ ( \
                "mov %0,%%sp\n"   \
                "mov %1,%%r11\n"   \
        : "=rm" (ctx.SP), "=rm" (ctx.FP) )
#else
        #error unknown hardware architecture
#endif

//-----------------------------------------------------------------------------
//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 ) ));

//-----------------------------------------------------------------------------
// Kernel storage
KERNEL_STORAGE(cluster,         mainCluster);
KERNEL_STORAGE(processor,       mainProcessor);
KERNEL_STORAGE(thread_desc,     mainThread);
KERNEL_STORAGE(__stack_t,       mainThreadCtx);

cluster     * mainCluster;
processor   * mainProcessor;
thread_desc * mainThread;

extern "C" {
        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
}

size_t __page_size = 0;

//-----------------------------------------------------------------------------
// Global state
thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = {
        NULL,                                                                                           // cannot use 0p
        NULL,
        { 1, false, false },
        6u //this should be seeded better but due to a bug calling rdtsc doesn't work
};

//-----------------------------------------------------------------------------
// Struct to steal stack
struct current_stack_info_t {
        __stack_t * storage;                                                            // pointer to stack object
        void * base;                                                                            // base of stack
        void * limit;                                                                           // stack grows towards stack limit
        void * context;                                                                         // address of cfa_context_t
};

void ?{}( current_stack_info_t & this ) {
        __stack_context_t ctx;
        CtxGet( ctx );
        this.base = ctx.FP;

        rlimit r;
        getrlimit( RLIMIT_STACK, &r);
        size_t size = r.rlim_cur;

        this.limit = (void *)(((intptr_t)this.base) - size);
        this.context = &storage_mainThreadCtx;
}

//-----------------------------------------------------------------------------
// Main thread construction

void ?{}( coroutine_desc & this, current_stack_info_t * info) with( this ) {
        stack.storage = info->storage;
        with(*stack.storage) {
                limit     = info->limit;
                base      = info->base;
        }
        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
        *istorage |= 0x1;
        name = "Main Thread";
        state = Start;
        starter = 0p;
        last = 0p;
        cancellation = 0p;
}

void ?{}( thread_desc & this, current_stack_info_t * info) with( this ) {
        state = Start;
        self_cor{ info };
        curr_cor = &self_cor;
        curr_cluster = mainCluster;
        self_mon.owner = &this;
        self_mon.recursion = 1;
        self_mon_p = &self_mon;
        next = 0p;

        node.next = 0p;
        node.prev = 0p;
        doregister(curr_cluster, this);

        monitors{ &self_mon_p, 1, (fptr_t)0 };
}

//-----------------------------------------------------------------------------
// Processor coroutine
void ?{}(processorCtx_t & this) {

}

// Construct the processor context of non-main processors
static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
        (this.__cor){ info };
        this.proc = proc;
}

static void start(processor * this);
void ?{}(processor & this, const char * name, cluster & cltr) with( this ) {
        this.name = name;
        this.cltr = &cltr;
        terminated{ 0 };
        do_terminate = false;
        preemption_alarm = 0p;
        pending_preemption = false;
        runner.proc = &this;

        idleLock{};

        start( &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);

                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
                wake( &this );

                P( terminated );
                verify( kernelTLS.this_processor != &this);
        }

        pthread_join( kernel_thread, 0p );
        free( this.stack );
}

void ?{}(cluster & this, const char * name, Duration preemption_rate) with( this ) {
        this.name = name;
        this.preemption_rate = preemption_rate;
        ready_queue{};
        ready_queue_lock{};

        procs{ __get };
        idles{ __get };
        threads{ __get };

        doregister(this);
}

void ^?{}(cluster & this) {
        unregister(this);
}

//=============================================================================================
// Kernel Scheduling logic
//=============================================================================================
static thread_desc * nextThread(cluster * this);
static void runThread(processor * this, thread_desc * dst);
static void halt(processor * this);

//Main of the processor contexts
void main(processorCtx_t & runner) {
        // Because of a bug, we couldn't initialized the seed on construction
        // Do it here
        kernelTLS.rand_seed ^= rdtscl();

        processor * this = runner.proc;
        verify(this);

        __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this);

        doregister(this->cltr, this);

        {
                // Setup preemption data
                preemption_scope scope = { this };

                __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);

                thread_desc * 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) {
                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
                                /* paranoid */ verifyf( readyThread->state == Inactive || readyThread->state == Start || readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
                                /* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );

                                runThread(this, readyThread);

                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );

                                spin_count = 0;
                        } else {
                                // spin(this, &spin_count);
                                // halt(this);
                        }
                }

                __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this);
        }

        unregister(this->cltr, this);

        V( this->terminated );

        __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this);
}

static int * __volatile_errno() __attribute__((noinline));
static int * __volatile_errno() { asm(""); return &errno; }

// KERNEL ONLY
// 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;

        // Update global state
        kernelTLS.this_thread = thrd_dst;

        // set state of processor coroutine to inactive
        verify(proc_cor->state == Active);
        proc_cor->state = Inactive;

        // 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 || thrd_dst->state == Reschedule);
                } else {
                        verify(thrd_dst->state == Start || thrd_dst->state == Primed || thrd_dst->state == Inactive);
                        thrd_dst->state = Active;
                }

                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );

                // 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

                /* 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.
                // 3 - Polite Racy case : the thread has blocked, someone has already tried to schedule it, but the thread is nice and wants to go through the ready-queue any way
                // 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.
                // In case 3, we lost the race but can just reschedule the thread.

                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
                        // The thread was preempted, reschedule it and reset the flag
                        ScheduleThread( 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, Inactive, __ATOMIC_SEQ_CST);
                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;
                                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;
                        case Reschedule:
                                // This is case 3, someone tried to run this before it finished blocking
                                // but it must go through the ready-queue
                                thrd_dst->state = Inactive;  /*restore invariant */
                                ScheduleThread( thrd_dst );
                                break RUNNING;
                        default:
                                // This makes no sense, something is wrong abort
                                abort("Finished running a thread that was Inactive/Start/Primed %d\n", old_state);
                }
        }

        // Just before returning to the processor, set the processor coroutine to active
        proc_cor->state = Active;
}

// KERNEL_ONLY
static void returnToKernel() {
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
        thread_desc * 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 );
                CtxSwitch( &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 );
}

// KERNEL_ONLY
// Context invoker for processors
// This is the entry point for processors (kernel threads)
// It effectively constructs a coroutine by stealing the pthread stack
static void * CtxInvokeProcessor(void * arg) {
        processor * proc = (processor *) arg;
        kernelTLS.this_processor = proc;
        kernelTLS.this_thread    = 0p;
        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
        // SKULLDUGGERY: We want to create a context for the processor coroutine
        // which is needed for the 2-step context switch. However, there is no reason
        // to waste the perfectly valid stack create by pthread.
        current_stack_info_t info;
        __stack_t ctx;
        info.storage = &ctx;
        (proc->runner){ proc, &info };

        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);

        //Set global state
        kernelTLS.this_thread = 0p;

        //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);

        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
        // resume it to start it like it normally would, it will just context switch
        // back to here. Instead directly call the main since we already are on the
        // appropriate stack.
        get_coroutine(proc->runner)->state = Active;
        main( proc->runner );
        get_coroutine(proc->runner)->state = Halted;

        // 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);

        return 0p;
}

static void Abort( int ret, const char * func ) {
        if ( ret ) {                                                                            // pthread routines return errno values
                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
        } // if
} // Abort

void * create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
        pthread_attr_t attr;

        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute

        size_t stacksize;
        // default stack size, normally defined by shell limit
        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
        assert( stacksize >= PTHREAD_STACK_MIN );

        void * stack;
        __cfaabi_dbg_debug_do(
                stack = memalign( __page_size, stacksize + __page_size );
                // pthread has no mechanism to create the guard page in user supplied stack.
                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
                } // if
        );
        __cfaabi_dbg_no_debug_do(
                stack = malloc( stacksize );
        );

        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );

        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
        return stack;
}

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);

        verify( ! kernelTLS.preemption_state.enabled );

        kernelTLS.this_thread->curr_cor = dst;
        __stack_prepare( &dst->stack, 65000 );
        CtxStart(main, dst, this->runner, CtxInvokeCoroutine);

        verify( ! kernelTLS.preemption_state.enabled );

        dst->last = &src->self_cor;
        dst->starter = dst->starter ? dst->starter : &src->self_cor;

        // set state of current coroutine to inactive
        src->state = src->state == Halted ? Halted : Inactive;

        // context switch to specified coroutine
        verify( dst->context.SP );
        CtxSwitch( &src->context, &dst->context );
        // when CtxSwitch returns we are back in the src coroutine

        mainThread->curr_cor = &mainThread->self_cor;

        // set state of new coroutine to active
        src->state = Active;

        verify( ! kernelTLS.preemption_state.enabled );
}

// KERNEL_ONLY
void kernel_last_resume( processor * this ) {
        coroutine_desc * src = &mainThread->self_cor;
        coroutine_desc * dst = get_coroutine(this->runner);

        verify( ! kernelTLS.preemption_state.enabled );
        verify( dst->starter == src );
        verify( dst->context.SP );

        // context switch to the processor
        CtxSwitch( &src->context, &dst->context );
}

//-----------------------------------------------------------------------------
// Scheduler routines
// KERNEL ONLY
void ScheduleThread( thread_desc * thrd ) with( *thrd->curr_cluster ) {
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
        /* paranoid */ if( thrd->state == Inactive || 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 || thrd->state == Reschedule,
                          "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 );

        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);
        }

        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
}

// KERNEL ONLY
static thread_desc * nextThread(cluster * this) with( *this ) {
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );

        lock( ready_queue_lock __cfaabi_dbg_ctx2 );
        thread_desc * head = pop_head( ready_queue );
        unlock( ready_queue_lock );

        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        return head;
}

void unpark( thread_desc * thrd, bool must_yield ) {
        if( !thrd ) return;

        enum coroutine_state new_state = must_yield ? Reschedule : Rerun;

        disable_interrupts();
        static_assert(sizeof(thrd->state) == sizeof(int));
        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, new_state, __ATOMIC_SEQ_CST);
        switch(old_state) {
                case Active:
                        // Wake won the race, the thread will reschedule/rerun itself
                        break;
                case Inactive:
                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );

                        // Wake lost the race,
                        thrd->state = Inactive;
                        ScheduleThread( thrd );
                        break;
                case Rerun:
                case Reschedule:
                        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();
        }
        enable_interrupts( __cfaabi_dbg_ctx );
}

void park( void ) {
        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
        disable_interrupts();
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );

        returnToKernel();

        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        enable_interrupts( __cfaabi_dbg_ctx );
        /* paranoid */ verify( kernelTLS.preemption_state.enabled );

}

// KERNEL ONLY
void LeaveThread() {
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
        returnToKernel();
}

// KERNEL ONLY
bool force_yield( __Preemption_Reason reason ) {
        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
        disable_interrupts();
        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );

        thread_desc * thrd = kernelTLS.this_thread;
        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun || thrd->state == Reschedule);

        // 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;
}

//=============================================================================================
// Kernel Setup logic
//=============================================================================================
//-----------------------------------------------------------------------------
// Kernel boot procedures
static void kernel_startup(void) {
        verify( ! kernelTLS.preemption_state.enabled );
        __cfaabi_dbg_print_safe("Kernel : Starting\n");

        __page_size = sysconf( _SC_PAGESIZE );

        __cfa_dbg_global_clusters.list{ __get };
        __cfa_dbg_global_clusters.lock{};

        // Initialize the main cluster
        mainCluster = (cluster *)&storage_mainCluster;
        (*mainCluster){"Main Cluster"};

        __cfaabi_dbg_print_safe("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;
        current_stack_info_t info;
        info.storage = (__stack_t*)&storage_mainThreadCtx;
        (*mainThread){ &info };

        __cfaabi_dbg_print_safe("Kernel : Main thread ready\n");



        // Construct the processor context of the main processor
        void ?{}(processorCtx_t & this, processor * proc) {
                (this.__cor){ "Processor" };
                this.__cor.starter = 0p;
                this.proc = proc;
        }

        void ?{}(processor & this) with( this ) {
                name = "Main Processor";
                cltr = mainCluster;
                terminated{ 0 };
                do_terminate = false;
                preemption_alarm = 0p;
                pending_preemption = false;
                kernel_thread = pthread_self();

                runner{ &this };
                __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner);
        }

        // Initialize the main processor and the main processor ctx
        // (the coroutine that contains the processing control flow)
        mainProcessor = (processor *)&storage_mainProcessor;
        (*mainProcessor){};

        //initialize the global state variables
        kernelTLS.this_processor = mainProcessor;
        kernelTLS.this_thread    = mainThread;

        // Enable preemption
        kernel_start_preemption();

        // 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);

        // 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
        // mainThread is on the ready queue when this call is made.
        kernel_first_resume( kernelTLS.this_processor );



        // THE SYSTEM IS NOW COMPLETELY RUNNING
        __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n");

        verify( ! kernelTLS.preemption_state.enabled );
        enable_interrupts( __cfaabi_dbg_ctx );
        verify( TL_GET( preemption_state.enabled ) );
}

static void kernel_shutdown(void) {
        __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n");

        verify( TL_GET( preemption_state.enabled ) );
        disable_interrupts();
        verify( ! kernelTLS.preemption_state.enabled );

        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
        // When its coroutine terminates, it return control to the mainThread
        // which is currently here
        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
        kernel_last_resume( kernelTLS.this_processor );
        mainThread->self_cor.state = Halted;

        // THE SYSTEM IS NOW COMPLETELY STOPPED

        // Disable preemption
        kernel_stop_preemption();

        // 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){};

        // 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){};

        ^(__cfa_dbg_global_clusters.list){};
        ^(__cfa_dbg_global_clusters.lock){};

        __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n");
}

//=============================================================================================
// Kernel Quiescing
//=============================================================================================
static void halt(processor * this) with( *this ) {
        // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) );

        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);
        }
}

//=============================================================================================
// Unexpected Terminating logic
//=============================================================================================
static __spinlock_t kernel_abort_lock;
static bool kernel_abort_called = false;

void * kernel_abort(void) __attribute__ ((__nothrow__)) {
        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
        // the globalAbort flag is true.
        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );

        // first task to abort ?
        if ( kernel_abort_called ) {                    // not first task to abort ?
                unlock( kernel_abort_lock );

                sigset_t mask;
                sigemptyset( &mask );
                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
                sigaddset( &mask, SIGUSR1 );            // block SIGALRM signals
                sigsuspend( &mask );                            // block the processor to prevent further damage during abort
                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it
        }
        else {
                kernel_abort_called = true;
                unlock( kernel_abort_lock );
        }

        return kernelTLS.this_thread;
}

void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
        thread_desc * thrd = kernel_data;

        if(thrd) {
                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );

                if ( &thrd->self_cor != thrd->curr_cor ) {
                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
                }
                else {
                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
                }
        }
        else {
                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
        }
}

int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
}

static __spinlock_t kernel_debug_lock;

extern "C" {
        void __cfaabi_bits_acquire() {
                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
        }

        void __cfaabi_bits_release() {
                unlock( kernel_debug_lock );
        }
}

//=============================================================================================
// Kernel Utilities
//=============================================================================================
//-----------------------------------------------------------------------------
// Locks
void  ?{}( semaphore & this, int count = 1 ) {
        (this.lock){};
        this.count = count;
        (this.waiting){};
}
void ^?{}(semaphore & this) {}

void P(semaphore & this) with( this ){
        lock( lock __cfaabi_dbg_ctx2 );
        count -= 1;
        if ( count < 0 ) {
                // queue current task
                append( waiting, kernelTLS.this_thread );

                // atomically release spin lock and block
                unlock( lock );
                park();
        }
        else {
            unlock( lock );
        }
}

void V(semaphore & this) with( this ) {
        thread_desc * thrd = 0p;
        lock( lock __cfaabi_dbg_ctx2 );
        count += 1;
        if ( count <= 0 ) {
                // remove task at head of waiting list
                thrd = pop_head( waiting );
        }

        unlock( lock );

        // make new owner
        unpark( thrd );
}

//-----------------------------------------------------------------------------
// Global Queues
void doregister( cluster     & cltr ) {
        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
        push_front( __cfa_dbg_global_clusters.list, cltr );
        unlock    ( __cfa_dbg_global_clusters.lock );
}

void unregister( cluster     & cltr ) {
        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
        remove( __cfa_dbg_global_clusters.list, cltr );
        unlock( __cfa_dbg_global_clusters.lock );
}

void doregister( cluster * cltr, thread_desc & thrd ) {
        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
        cltr->nthreads += 1;
        push_front(cltr->threads, thrd);
        unlock    (cltr->thread_list_lock);
}

void unregister( cluster * cltr, thread_desc & thrd ) {
        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
        remove(cltr->threads, thrd );
        cltr->nthreads -= 1;
        unlock(cltr->thread_list_lock);
}

void doregister( cluster * cltr, processor * proc ) {
        lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2);
        cltr->nprocessors += 1;
        push_front(cltr->procs, *proc);
        unlock    (cltr->proc_list_lock);
}

void unregister( cluster * cltr, processor * proc ) {
        lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2);
        remove(cltr->procs, *proc );
        cltr->nprocessors -= 1;
        unlock(cltr->proc_list_lock);
}

//-----------------------------------------------------------------------------
// Debug
__cfaabi_dbg_debug_do(
        extern "C" {
                void __cfaabi_dbg_record(__spinlock_t & this, const char * prev_name) {
                        this.prev_name = prev_name;
                        this.prev_thrd = kernelTLS.this_thread;
                }
        }
)

//-----------------------------------------------------------------------------
// Debug
bool threading_enabled(void) {
        return true;
}
// Local Variables: //
// mode: c //
// tab-width: 4 //
// End: //