Changes in / [653f2c7:d43cd01]

Location:

src

Files:

• benchmark/CorCtxSwitch.c (modified) (1 diff)
• benchmark/csv-data.c (modified) (1 diff)
• benchmark/interrupt_linux.c (added)
• libcfa/concurrency/alarm.c (modified) (8 diffs)
• libcfa/concurrency/coroutine (modified) (4 diffs)
• libcfa/concurrency/coroutine.c (modified) (2 diffs)
• libcfa/concurrency/invoke.c (modified) (5 diffs)
• libcfa/concurrency/invoke.h (modified) (2 diffs)
• libcfa/concurrency/kernel (modified) (3 diffs)
• libcfa/concurrency/kernel.c (modified) (33 diffs)
• libcfa/concurrency/kernel_private.h (modified) (2 diffs)
• libcfa/concurrency/monitor (modified) (1 diff)
• libcfa/concurrency/monitor.c (modified) (16 diffs)
• libcfa/concurrency/preemption.c (modified) (9 diffs)
• libcfa/concurrency/thread (modified) (1 diff)
• libcfa/concurrency/thread.c (modified) (4 diffs)
• libcfa/libhdr/libdebug.h (modified) (2 diffs)
• tests/sched-int-block.c (modified) (4 diffs)

src/benchmark/CorCtxSwitch.c

@@ -31 +31 @@
 
         StartTime = Time();
-        // for ( volatile unsigned int i = 0; i < NoOfTimes; i += 1 ) {
-        //      resume( this_coroutine() );
-        //      // resume( &s );        
-        // }
         resumer( &s, NoOfTimes );
         EndTime = Time();

src/benchmark/csv-data.c

@@ -38 +38 @@
 
         StartTime = Time();
-        // for ( volatile unsigned int i = 0; i < NoOfTimes; i += 1 ) {
-        //      resume( this_coroutine() );
-        //      // resume( &s );
-        // }
         resumer( &s, NoOfTimes );
         EndTime = Time();

src/libcfa/concurrency/alarm.c

@@ -16 +16 @@
 
 extern "C" {
+#include <errno.h>
+#include <stdio.h>
+#include <string.h>
 #include <time.h>
+#include <unistd.h>
 #include <sys/time.h>
 }
@@ -22 +26 @@
 #include "alarm.h"
 #include "kernel_private.h"
+#include "libhdr.h"
 #include "preemption.h"
 
@@ -31 +36 @@
         timespec curr;
         clock_gettime( CLOCK_REALTIME, &curr );
-        return ((__cfa_time_t)curr.tv_sec * TIMEGRAN) + curr.tv_nsec;
+        __cfa_time_t curr_time = ((__cfa_time_t)curr.tv_sec * TIMEGRAN) + curr.tv_nsec;
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Kernel : current time is %lu\n", curr_time );
+        return curr_time;
 }
 
 void __kernel_set_timer( __cfa_time_t alarm ) {
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Kernel : set timer to %lu\n", (__cfa_time_t)alarm );
         itimerval val;
         val.it_value.tv_sec = alarm / TIMEGRAN;                 // seconds
@@ -71 +79 @@
 }
 
+LIB_DEBUG_DO( bool validate( alarm_list_t * this ) {
+        alarm_node_t ** it = &this->head;
+        while( (*it) ) {
+                it = &(*it)->next;
+        }
+
+        return it == this->tail;
+})
+
 static inline void insert_at( alarm_list_t * this, alarm_node_t * n, __alarm_it_t p ) {
-        assert( !n->next );
+        verify( !n->next );
         if( p == this->tail ) {
                 this->tail = &n->next;
@@ -80 +97 @@
         }
         *p = n;
+
+        verify( validate( this ) );
 }
 
@@ -89 +108 @@
 
         insert_at( this, n, it );
+
+        verify( validate( this ) );
 }
 
@@ -100 +121 @@
                 head->next = NULL;
         }
+        verify( validate( this ) );
         return head;
 }
@@ -105 +127 @@
 static inline void remove_at( alarm_list_t * this, alarm_node_t * n, __alarm_it_t it ) {
         verify( it );
-        verify( (*it)->next == n );
+        verify( (*it) == n );
 
-        (*it)->next = n->next;
+        (*it) = n->next;
         if( !n-> next ) {
                 this->tail = it;
         }
         n->next = NULL;
+
+        verify( validate( this ) );
 }
 
 static inline void remove( alarm_list_t * this, alarm_node_t * n ) {
         alarm_node_t ** it = &this->head;
-        while( (*it) && (*it)->next != n ) {
+        while( (*it) && (*it) != n ) {
                 it = &(*it)->next;
         }
 
+        verify( validate( this ) );
+
         if( *it ) { remove_at( this, n, it ); }
+
+        verify( validate( this ) );
 }
 
 void register_self( alarm_node_t * this ) {
         disable_interrupts();
-        assert( !systemProcessor->pending_alarm );
-        lock( &systemProcessor->alarm_lock );
+        verify( !systemProcessor->pending_alarm );
+        lock( &systemProcessor->alarm_lock, __PRETTY_FUNCTION__ );
         {
+                verify( validate( &systemProcessor->alarms ) );
+                bool first = !systemProcessor->alarms.head;
+
                 insert( &systemProcessor->alarms, this );
                 if( systemProcessor->pending_alarm ) {
                         tick_preemption();
                 }
+                if( first ) {
+                        __kernel_set_timer( systemProcessor->alarms.head->alarm - __kernel_get_time() );
+                }
         }
         unlock( &systemProcessor->alarm_lock );
         this->set = true;
-        enable_interrupts();
+        enable_interrupts( __PRETTY_FUNCTION__ );
 }
 
 void unregister_self( alarm_node_t * this ) {
+        // LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Kernel : unregister %p start\n", this );
         disable_interrupts();
-        lock( &systemProcessor->alarm_lock );
-        remove( &systemProcessor->alarms, this );
+        lock( &systemProcessor->alarm_lock, __PRETTY_FUNCTION__ );
+        {
+                verify( validate( &systemProcessor->alarms ) );
+                remove( &systemProcessor->alarms, this );
+        }
         unlock( &systemProcessor->alarm_lock );
         disable_interrupts();
         this->set = false;
+        // LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Kernel : unregister %p end\n", this );
 }

src/libcfa/concurrency/coroutine

@@ -63 +63 @@
 
 // Get current coroutine
-coroutine_desc * this_coroutine(void);
+extern volatile thread_local coroutine_desc * this_coroutine;
 
 // Private wrappers for context switch and stack creation
@@ -71 +71 @@
 // Suspend implementation inlined for performance
 static inline void suspend() {
-        coroutine_desc * src = this_coroutine();                // optimization
+        coroutine_desc * src = this_coroutine;          // optimization
 
         assertf( src->last != 0,
@@ -88 +88 @@
 forall(dtype T | is_coroutine(T))
 static inline void resume(T * cor) {
-        coroutine_desc * src = this_coroutine();                // optimization
+        coroutine_desc * src = this_coroutine;          // optimization
         coroutine_desc * dst = get_coroutine(cor);
 
@@ -112 +112 @@
 
 static inline void resume(coroutine_desc * dst) {
-        coroutine_desc * src = this_coroutine();                // optimization
+        coroutine_desc * src = this_coroutine;          // optimization
 
         // not resuming self ?

src/libcfa/concurrency/coroutine.c

@@ -32 +32 @@
 #include "invoke.h"
 
-extern thread_local processor * this_processor;
+extern volatile thread_local processor * this_processor;
 
 //-----------------------------------------------------------------------------
@@ -106 +106 @@
 
         // set state of current coroutine to inactive
-        src->state = Inactive;
+        src->state = src->state == Halted ? Halted : Inactive;
 
         // set new coroutine that task is executing
-        this_processor->current_coroutine = dst;
+        this_coroutine = dst;
 
         // context switch to specified coroutine
+        assert( src->stack.context );
         CtxSwitch( src->stack.context, dst->stack.context );
         // when CtxSwitch returns we are back in the src coroutine              

src/libcfa/concurrency/invoke.c

@@ -29 +29 @@
 
 extern void __suspend_internal(void);
-extern void __leave_monitor_desc( struct monitor_desc * this );
+extern void __leave_thread_monitor( struct thread_desc * this );
+extern void disable_interrupts();
+extern void enable_interrupts( const char * );
 
 void CtxInvokeCoroutine(
-      void (*main)(void *), 
-      struct coroutine_desc *(*get_coroutine)(void *), 
+      void (*main)(void *),
+      struct coroutine_desc *(*get_coroutine)(void *),
       void *this
 ) {
@@ -56 +58 @@
 
 void CtxInvokeThread(
-      void (*dtor)(void *), 
-      void (*main)(void *), 
-      struct thread_desc *(*get_thread)(void *), 
+      void (*dtor)(void *),
+      void (*main)(void *),
+      struct thread_desc *(*get_thread)(void *),
       void *this
 ) {
+      // First suspend, once the thread arrives here,
+      // the function pointer to main can be invalidated without risk
       __suspend_internal();
 
+      // Fetch the thread handle from the user defined thread structure
       struct thread_desc* thrd = get_thread( this );
-      struct coroutine_desc* cor = &thrd->cor;
-      struct monitor_desc* mon = &thrd->mon;
-      cor->state = Active;
 
-      // LIB_DEBUG_PRINTF("Invoke Thread : invoking main %p (args %p)\n", main, this);
+      // Officially start the thread by enabling preemption
+      enable_interrupts( __PRETTY_FUNCTION__ );
+
+      // Call the main of the thread
       main( this );
 
-      __leave_monitor_desc( mon );
+      // To exit a thread we must :
+      // 1 - Mark it as halted
+      // 2 - Leave its monitor
+      // 3 - Disable the interupts
+      // The order of these 3 operations is very important
+      __leave_thread_monitor( thrd );
 
       //Final suspend, should never return
@@ -80 +90 @@
 
 void CtxStart(
-      void (*main)(void *), 
-      struct coroutine_desc *(*get_coroutine)(void *), 
-      void *this, 
+      void (*main)(void *),
+      struct coroutine_desc *(*get_coroutine)(void *),
+      void *this,
       void (*invoke)(void *)
 ) {
@@ -108 +118 @@
         ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->rturn = invoke;
       ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->mxcr = 0x1F80; //Vol. 2A 3-520
-      ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->fcw = 0x037F;  //Vol. 1 8-7 
+      ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->fcw = 0x037F;  //Vol. 1 8-7
 
 #elif defined( __x86_64__ )
@@ -128 +138 @@
       ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->fixedRegisters[1] = invoke;
       ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->mxcr = 0x1F80; //Vol. 2A 3-520
-      ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->fcw = 0x037F;  //Vol. 1 8-7 
+      ((struct FakeStack *)(((struct machine_context_t *)stack->context)->SP))->fcw = 0x037F;  //Vol. 1 8-7
 #else
       #error Only __i386__ and __x86_64__ is supported for threads in cfa

src/libcfa/concurrency/invoke.h

@@ -31 +31 @@
       struct spinlock {
             volatile int lock;
+            #ifdef __CFA_DEBUG__
+                  const char * prev;
+            #endif
       };
 
@@ -83 +86 @@
             struct __thread_queue_t entry_queue;      // queue of threads that are blocked waiting for the monitor
             struct __condition_stack_t signal_stack;  // stack of conditions to run next once we exit the monitor
-            struct monitor_desc * stack_owner;        // if bulk acquiring was used we need to synchronize signals with an other monitor
             unsigned int recursion;                   // monitor routines can be called recursively, we need to keep track of that
       };

src/libcfa/concurrency/kernel

@@ -28 +28 @@
 //-----------------------------------------------------------------------------
 // Locks
-bool try_lock( spinlock * );
-void lock( spinlock * );
+bool try_lock( spinlock *
+        #ifdef __CFA_DEBUG__
+                , const char * caller
+        #endif
+);
+
+void lock( spinlock *
+        #ifdef __CFA_DEBUG__
+                , const char * caller
+        #endif
+);
+
 void unlock( spinlock * );
 
@@ -78 +88 @@
         struct processorCtx_t * runner;
         cluster * cltr;
-        coroutine_desc * current_coroutine;
-        thread_desc * current_thread;
         pthread_t kernel_thread;
         
@@ -90 +98 @@
         unsigned int preemption;
 
-        unsigned short disable_preempt_count;
+        bool pending_preemption;
 
-        bool pending_preemption;
+        char * last_enable;
 };
 

src/libcfa/concurrency/kernel.c

@@ -59 +59 @@
 // Global state
 
-thread_local processor * this_processor;
-
-coroutine_desc * this_coroutine(void) {
-        return this_processor->current_coroutine;
-}
-
-thread_desc * this_thread(void) {
-        return this_processor->current_thread;
-}
+volatile thread_local processor * this_processor;
+volatile thread_local coroutine_desc * this_coroutine;
+volatile thread_local thread_desc * this_thread;
+volatile thread_local unsigned short disable_preempt_count = 1;
 
 //-----------------------------------------------------------------------------
 // Main thread construction
 struct current_stack_info_t {
-        machine_context_t ctx;  
+        machine_context_t ctx;
         unsigned int size;              // size of stack
         void *base;                             // base of stack
@@ -106 +101 @@
 
 void ?{}( coroutine_desc * this, current_stack_info_t * info) {
-        (&this->stack){ info }; 
+        (&this->stack){ info };
         this->name = "Main Thread";
         this->errno_ = 0;
@@ -136 +131 @@
 void ?{}(processor * this, cluster * cltr) {
         this->cltr = cltr;
-        this->current_coroutine = NULL;
-        this->current_thread = NULL;
         (&this->terminated){};
         this->is_terminated = false;
         this->preemption_alarm = NULL;
         this->preemption = default_preemption();
-        this->disable_preempt_count = 1;                //Start with interrupts disabled
         this->pending_preemption = false;
 
@@ -150 +142 @@
 void ?{}(processor * this, cluster * cltr, processorCtx_t * runner) {
         this->cltr = cltr;
-        this->current_coroutine = NULL;
-        this->current_thread = NULL;
         (&this->terminated){};
         this->is_terminated = false;
-        this->disable_preempt_count = 0;
+        this->preemption_alarm = NULL;
+        this->preemption = default_preemption();
         this->pending_preemption = false;
+        this->kernel_thread = pthread_self();
 
         this->runner = runner;
-        LIB_DEBUG_PRINT_SAFE("Kernel : constructing processor context %p\n", runner);
+        LIB_DEBUG_PRINT_SAFE("Kernel : constructing system processor context %p\n", runner);
         runner{ this };
 }
+
+LIB_DEBUG_DO( bool validate( alarm_list_t * this ); )
 
 void ?{}(system_proc_t * this, cluster * cltr, processorCtx_t * runner) {
@@ -168 +162 @@
 
         (&this->proc){ cltr, runner };
+
+        verify( validate( &this->alarms ) );
 }
 
@@ -184 +180 @@
 
 void ^?{}(cluster * this) {
-        
+
 }
 
@@ -203 +199 @@
 
                 thread_desc * readyThread = NULL;
-                for( unsigned int spin_count = 0; ! this->is_terminated; spin_count++ ) 
+                for( unsigned int spin_count = 0; ! this->is_terminated; spin_count++ )
                 {
                         readyThread = nextThread( this->cltr );
@@ -209 +205 @@
                         if(readyThread)
                         {
+                                verify( disable_preempt_count > 0 );
+
                                 runThread(this, readyThread);
+
+                                verify( disable_preempt_count > 0 );
 
                                 //Some actions need to be taken from the kernel
@@ -229 +229 @@
 }
 
-// runThread runs a thread by context switching 
-// from the processor coroutine to the target thread 
+// runThread runs a thread by context switching
+// from the processor coroutine to the target thread
 void runThread(processor * this, thread_desc * dst) {
         coroutine_desc * proc_cor = get_coroutine(this->runner);
         coroutine_desc * thrd_cor = get_coroutine(dst);
-        
+
         //Reset the terminating actions here
         this->finish.action_code = No_Action;
 
         //Update global state
-        this->current_thread = dst;
+        this_thread = dst;
 
         // Context Switch to the thread
@@ -246 +246 @@
 }
 
-// Once a thread has finished running, some of 
+// Once a thread has finished running, some of
 // its final actions must be executed from the kernel
 void finishRunning(processor * this) {
@@ -256 +256 @@
         }
         else if( this->finish.action_code == Release_Schedule ) {
-                unlock( this->finish.lock );            
+                unlock( this->finish.lock );
                 ScheduleThread( this->finish.thrd );
         }
@@ -289 +289 @@
         processor * proc = (processor *) arg;
         this_processor = proc;
+        this_coroutine = NULL;
+        this_thread = NULL;
+        disable_preempt_count = 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. 
+        // to waste the perfectly valid stack create by pthread.
         current_stack_info_t info;
         machine_context_t ctx;
@@ -300 +303 @@
 
         //Set global state
-        proc->current_coroutine = &proc->runner->__cor;
-        proc->current_thread = NULL;
+        this_coroutine = &proc->runner->__cor;
+        this_thread = NULL;
 
         //We now have a proper context from which to schedule threads
         LIB_DEBUG_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 
+        // 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.
         proc_cor_storage.__cor.state = Active;
@@ -315 +318 @@
 
         // Main routine of the core returned, the core is now fully terminated
-        LIB_DEBUG_PRINT_SAFE("Kernel : core %p main ended (%p)\n", proc, proc->runner); 
+        LIB_DEBUG_PRINT_SAFE("Kernel : core %p main ended (%p)\n", proc, proc->runner);
 
         return NULL;
@@ -322 +325 @@
 void start(processor * this) {
         LIB_DEBUG_PRINT_SAFE("Kernel : Starting core %p\n", this);
-        
+
+        // SIGALRM must only be caught by the system processor
+        sigset_t old_mask;
+        bool is_system_proc = this_processor == &systemProcessor->proc;
+        if ( is_system_proc ) {
+                // Child kernel-thread inherits the signal mask from the parent kernel-thread. So one special case for the
+                // system processor creating the user processor => toggle the blocking SIGALRM on system processor, create user
+                // processor, and toggle back (below) previous signal mask of the system processor.
+
+                sigset_t new_mask;
+                sigemptyset( &new_mask );
+                sigemptyset( &old_mask );
+                sigaddset( &new_mask, SIGALRM );
+
+                if ( sigprocmask( SIG_BLOCK, &new_mask, &old_mask ) == -1 ) {
+                        abortf( "internal error, sigprocmask" );
+                }
+
+                assert( ! sigismember( &old_mask, SIGALRM ) );
+        }
+
         pthread_create( &this->kernel_thread, NULL, CtxInvokeProcessor, (void*)this );
 
-        LIB_DEBUG_PRINT_SAFE("Kernel : core %p started\n", this);       
+        // Toggle back previous signal mask of system processor.
+        if ( is_system_proc ) {
+                if ( sigprocmask( SIG_SETMASK, &old_mask, NULL ) == -1 ) {
+                        abortf( "internal error, sigprocmask" );
+                } // if
+        } // if
+
+        LIB_DEBUG_PRINT_SAFE("Kernel : core %p started\n", this);
 }
 
@@ -331 +361 @@
 // Scheduler routines
 void ScheduleThread( thread_desc * thrd ) {
-        if( !thrd ) return;
+        // if( !thrd ) return;
+        assert( thrd );
+        assert( thrd->cor.state != Halted );
+
+        verify( disable_preempt_count > 0 );
 
         verifyf( thrd->next == NULL, "Expected null got %p", thrd->next );
-        
-        lock( &systemProcessor->proc.cltr->lock );
+
+        lock( &systemProcessor->proc.cltr->lock, __PRETTY_FUNCTION__ );
         append( &systemProcessor->proc.cltr->ready_queue, thrd );
         unlock( &systemProcessor->proc.cltr->lock );
+
+        verify( disable_preempt_count > 0 );
 }
 
 thread_desc * nextThread(cluster * this) {
-        lock( &this->lock );
+        verify( disable_preempt_count > 0 );
+        lock( &this->lock, __PRETTY_FUNCTION__ );
         thread_desc * head = pop_head( &this->ready_queue );
         unlock( &this->lock );
+        verify( disable_preempt_count > 0 );
         return head;
 }
 
-void ScheduleInternal() {
+void BlockInternal() {
+        disable_interrupts();
+        verify( disable_preempt_count > 0 );
         suspend();
-}
-
-void ScheduleInternal( spinlock * lock ) {
+        verify( disable_preempt_count > 0 );
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
+
+void BlockInternal( spinlock * lock ) {
+        disable_interrupts();
         this_processor->finish.action_code = Release;
         this_processor->finish.lock = lock;
+
+        verify( disable_preempt_count > 0 );
         suspend();
-}
-
-void ScheduleInternal( thread_desc * thrd ) {
+        verify( disable_preempt_count > 0 );
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
+
+void BlockInternal( thread_desc * thrd ) {
+        disable_interrupts();
+        assert( thrd->cor.state != Halted );
         this_processor->finish.action_code = Schedule;
         this_processor->finish.thrd = thrd;
+
+        verify( disable_preempt_count > 0 );
         suspend();
-}
-
-void ScheduleInternal( spinlock * lock, thread_desc * thrd ) {
+        verify( disable_preempt_count > 0 );
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
+
+void BlockInternal( spinlock * lock, thread_desc * thrd ) {
+        disable_interrupts();
         this_processor->finish.action_code = Release_Schedule;
         this_processor->finish.lock = lock;
         this_processor->finish.thrd = thrd;
+
+        verify( disable_preempt_count > 0 );
         suspend();
-}
-
-void ScheduleInternal(spinlock ** locks, unsigned short count) {
+        verify( disable_preempt_count > 0 );
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
+
+void BlockInternal(spinlock ** locks, unsigned short count) {
+        disable_interrupts();
         this_processor->finish.action_code = Release_Multi;
         this_processor->finish.locks = locks;
         this_processor->finish.lock_count = count;
+
+        verify( disable_preempt_count > 0 );
         suspend();
-}
-
-void ScheduleInternal(spinlock ** locks, unsigned short lock_count, thread_desc ** thrds, unsigned short thrd_count) {
+        verify( disable_preempt_count > 0 );
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
+
+void BlockInternal(spinlock ** locks, unsigned short lock_count, thread_desc ** thrds, unsigned short thrd_count) {
+        disable_interrupts();
         this_processor->finish.action_code = Release_Multi_Schedule;
         this_processor->finish.locks = locks;
@@ -383 +451 @@
         this_processor->finish.thrds = thrds;
         this_processor->finish.thrd_count = thrd_count;
+
+        verify( disable_preempt_count > 0 );
         suspend();
+        verify( disable_preempt_count > 0 );
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
 }
 
@@ -392 +465 @@
 // Kernel boot procedures
 void kernel_startup(void) {
-        LIB_DEBUG_PRINT_SAFE("Kernel : Starting\n");    
+        LIB_DEBUG_PRINT_SAFE("Kernel : Starting\n");
 
         // Start by initializing the main thread
-        // SKULLDUGGERY: the mainThread steals the process main thread 
+        // SKULLDUGGERY: the mainThread steals the process main thread
         // which will then be scheduled by the systemProcessor normally
         mainThread = (thread_desc *)&mainThread_storage;
@@ -403 +476 @@
         LIB_DEBUG_PRINT_SAFE("Kernel : Main thread ready\n");
 
-        // Enable preemption
-        kernel_start_preemption();
-
         // Initialize the system cluster
         systemCluster = (cluster *)&systemCluster_storage;
@@ -417 +487 @@
         systemProcessor{ systemCluster, (processorCtx_t *)&systemProcessorCtx_storage };
 
-        // Add the main thread to the ready queue 
+        // Add the main thread to the ready queue
         // once resume is called on systemProcessor->runner the mainThread needs to be scheduled like any normal thread
         ScheduleThread(mainThread);
@@ -423 +493 @@
         //initialize the global state variables
         this_processor = &systemProcessor->proc;
-        this_processor->current_thread = mainThread;
-        this_processor->current_coroutine = &mainThread->cor;
+        this_thread = mainThread;
+        this_coroutine = &mainThread->cor;
+        disable_preempt_count = 1;
+
+        // Enable preemption
+        kernel_start_preemption();
 
         // SKULLDUGGERY: Force a context switch to the system 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. 
+        // mainThread is on the ready queue when this call is made.
         resume( systemProcessor->proc.runner );
 
@@ -435 +509 @@
         // THE SYSTEM IS NOW COMPLETELY RUNNING
         LIB_DEBUG_PRINT_SAFE("Kernel : Started\n--------------------------------------------------\n\n");
+
+        enable_interrupts( __PRETTY_FUNCTION__ );
 }
 
 void kernel_shutdown(void) {
         LIB_DEBUG_PRINT_SAFE("\n--------------------------------------------------\nKernel : Shutting down\n");
+
+        disable_interrupts();
 
         // SKULLDUGGERY: Notify the systemProcessor it needs to terminates.
@@ -448 +526 @@
         // THE SYSTEM IS NOW COMPLETELY STOPPED
 
+        // Disable preemption
+        kernel_stop_preemption();
+
         // Destroy the system processor and its context in reverse order of construction
         // These were manually constructed so we need manually destroy them
@@ -457 +538 @@
         ^(mainThread){};
 
-        LIB_DEBUG_PRINT_SAFE("Kernel : Shutdown complete\n");   
+        LIB_DEBUG_PRINT_SAFE("Kernel : Shutdown complete\n");
 }
 
@@ -467 +548 @@
         // 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 );
+        lock( &kernel_abort_lock, __PRETTY_FUNCTION__ );
 
         // first task to abort ?
@@ -473 +554 @@
                 kernel_abort_called = true;
                 unlock( &kernel_abort_lock );
-        } 
+        }
         else {
                 unlock( &kernel_abort_lock );
-                
+
                 sigset_t mask;
                 sigemptyset( &mask );
@@ -482 +563 @@
                 sigaddset( &mask, SIGUSR1 );                    // block SIGUSR1 signals
                 sigsuspend( &mask );                            // block the processor to prevent further damage during abort
-                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it              
-        }
-
-        return this_thread();
+                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it
+        }
+
+        return this_thread;
 }
 
@@ -494 +575 @@
         __lib_debug_write( STDERR_FILENO, abort_text, len );
 
-        if ( thrd != this_coroutine() ) {
-                len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", this_coroutine()->name, this_coroutine() );
+        if ( thrd != this_coroutine ) {
+                len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", this_coroutine->name, this_coroutine );
                 __lib_debug_write( STDERR_FILENO, abort_text, len );
-        } 
+        }
         else {
                 __lib_debug_write( STDERR_FILENO, ".\n", 2 );
@@ -505 +586 @@
 extern "C" {
         void __lib_debug_acquire() {
-                lock(&kernel_debug_lock);
+                lock(&kernel_debug_lock, __PRETTY_FUNCTION__);
         }
 
@@ -525 +606 @@
 }
 
-bool try_lock( spinlock * this ) {
-        return this->lock == 0 && __sync_lock_test_and_set_4( &this->lock, 1 ) == 0;
-}
-
-void lock( spinlock * this ) {
+bool try_lock( spinlock * this, const char * caller ) {
+        bool ret = this->lock == 0 && __sync_lock_test_and_set_4( &this->lock, 1 ) == 0;
+        this->prev = caller;
+        return ret;
+}
+
+void lock( spinlock * this, const char * caller ) {
         for ( unsigned int i = 1;; i += 1 ) {
                 if ( this->lock == 0 && __sync_lock_test_and_set_4( &this->lock, 1 ) == 0 ) break;
         }
+        this->prev = caller;
 }
 
@@ -547 +631 @@
 
 void wait( signal_once * this ) {
-        lock( &this->lock );
+        lock( &this->lock, __PRETTY_FUNCTION__ );
         if( !this->cond ) {
-                append( &this->blocked, this_thread() );
-                ScheduleInternal( &this->lock );
-                lock( &this->lock );
-        }
-        unlock( &this->lock );
+                append( &this->blocked, (thread_desc*)this_thread );
+                BlockInternal( &this->lock );
+        }
+        else {
+                unlock( &this->lock );
+        }
 }
 
 void signal( signal_once * this ) {
-        lock( &this->lock );
+        lock( &this->lock, __PRETTY_FUNCTION__ );
         {
                 this->cond = true;
 
+                disable_interrupts();
                 thread_desc * it;
                 while( it = pop_head( &this->blocked) ) {
                         ScheduleThread( it );
                 }
+                enable_interrupts( __PRETTY_FUNCTION__ );
         }
         unlock( &this->lock );
@@ -590 +677 @@
                 }
                 head->next = NULL;
-        }       
+        }
         return head;
 }
@@ -609 +696 @@
                 this->top = top->next;
                 top->next = NULL;
-        }       
+        }
         return top;
 }

src/libcfa/concurrency/kernel_private.h

@@ -27 +27 @@
 //-----------------------------------------------------------------------------
 // Scheduler
+
+extern "C" {
+        void disable_interrupts();
+        void enable_interrupts_noRF();
+        void enable_interrupts( const char * );
+}
+
 void ScheduleThread( thread_desc * );
+static inline void WakeThread( thread_desc * thrd ) {
+        if( !thrd ) return;
+
+        disable_interrupts();
+        ScheduleThread( thrd );
+        enable_interrupts( __PRETTY_FUNCTION__ );
+}
 thread_desc * nextThread(cluster * this);
 
-void ScheduleInternal(void);
-void ScheduleInternal(spinlock * lock);
-void ScheduleInternal(thread_desc * thrd);
-void ScheduleInternal(spinlock * lock, thread_desc * thrd);
-void ScheduleInternal(spinlock ** locks, unsigned short count);
-void ScheduleInternal(spinlock ** locks, unsigned short count, thread_desc ** thrds, unsigned short thrd_count);
+void BlockInternal(void);
+void BlockInternal(spinlock * lock);
+void BlockInternal(thread_desc * thrd);
+void BlockInternal(spinlock * lock, thread_desc * thrd);
+void BlockInternal(spinlock ** locks, unsigned short count);
+void BlockInternal(spinlock ** locks, unsigned short count, thread_desc ** thrds, unsigned short thrd_count);
 
 //-----------------------------------------------------------------------------
@@ -60 +74 @@
 extern cluster * systemCluster;
 extern system_proc_t * systemProcessor;
-extern thread_local processor * this_processor;
-
-static inline void disable_interrupts() {
-        __attribute__((unused)) unsigned short prev = __atomic_fetch_add_2( &this_processor->disable_preempt_count, 1, __ATOMIC_SEQ_CST );
-        assert( prev != (unsigned short) -1 );
-}
-
-static inline void enable_interrupts_noRF() {
-        __attribute__((unused)) unsigned short prev = __atomic_fetch_add_2( &this_processor->disable_preempt_count, -1, __ATOMIC_SEQ_CST );
-        verify( prev != (unsigned short) 0 );
-}
-
-static inline void enable_interrupts() {
-        __attribute__((unused)) unsigned short prev = __atomic_fetch_add_2( &this_processor->disable_preempt_count, -1, __ATOMIC_SEQ_CST );
-        verify( prev != (unsigned short) 0 );
-        if( prev == 1 && this_processor->pending_preemption ) {
-                ScheduleInternal( this_processor->current_thread );
-                this_processor->pending_preemption = false;
-        }
-}
+extern volatile thread_local processor * this_processor;
+extern volatile thread_local coroutine_desc * this_coroutine;
+extern volatile thread_local thread_desc * this_thread;
+extern volatile thread_local unsigned short disable_preempt_count;
 
 //-----------------------------------------------------------------------------

src/libcfa/concurrency/monitor

@@ -26 +26 @@
 static inline void ?{}(monitor_desc * this) {
         this->owner = NULL;
-        this->stack_owner = NULL;
         this->recursion = 0;
 }

src/libcfa/concurrency/monitor.c

@@ -44 +44 @@
 
 extern "C" {
-        void __enter_monitor_desc(monitor_desc * this) {
-                lock( &this->lock );
-                thread_desc * thrd = this_thread();
-
-                LIB_DEBUG_PRINT_SAFE("%p Entering %p (o: %p, r: %i)\n", thrd, this, this->owner, this->recursion);
+        void __enter_monitor_desc( monitor_desc * this ) {
+                lock( &this->lock, __PRETTY_FUNCTION__ );
+                thread_desc * thrd = this_thread;
+
+                // LIB_DEBUG_PRINT_SAFE("%p Entering %p (o: %p, r: %i)\n", thrd, this, this->owner, this->recursion);
 
                 if( !this->owner ) {
@@ -62 +62 @@
                         //Some one else has the monitor, wait in line for it
                         append( &this->entry_queue, thrd );
-                        LIB_DEBUG_PRINT_SAFE("%p Blocking on entry\n", thrd);
-                        ScheduleInternal( &this->lock );
-
-                        //ScheduleInternal will unlock spinlock, no need to unlock ourselves
+                        // LIB_DEBUG_PRINT_SAFE("%p Blocking on entry\n", thrd);
+                        BlockInternal( &this->lock );
+
+                        //BlockInternal will unlock spinlock, no need to unlock ourselves
                         return; 
                 }
@@ -75 +75 @@
         // leave pseudo code :
         //      TODO
-        void __leave_monitor_desc(monitor_desc * this) {
-                lock( &this->lock );
-
-                LIB_DEBUG_PRINT_SAFE("%p Leaving %p (o: %p, r: %i)\n", thrd, this, this->owner, this->recursion);
-                verifyf( this_thread() == this->owner, "Expected owner to be %p, got %p (r: %i)", this_thread(), this->owner, this->recursion );
+        void __leave_monitor_desc( monitor_desc * this ) {
+                lock( &this->lock, __PRETTY_FUNCTION__ );
+
+                // LIB_DEBUG_PRINT_SAFE("%p Leaving %p (o: %p, r: %i). ", this_thread, this, this->owner, this->recursion);
+                verifyf( this_thread == this->owner, "Expected owner to be %p, got %p (r: %i)", this_thread, this->owner, this->recursion );
 
                 //Leaving a recursion level, decrement the counter
@@ -96 +96 @@
                 unlock( &this->lock );
 
-                LIB_DEBUG_PRINT_SAFE("Next owner is %p\n", new_owner);
+                // LIB_DEBUG_PRINT_SAFE("Next owner is %p\n", new_owner);
 
                 //We need to wake-up the thread
-                ScheduleThread( new_owner );
+                WakeThread( new_owner );
+        }
+
+        void __leave_thread_monitor( thread_desc * thrd ) {
+                monitor_desc * this = &thrd->mon;
+                lock( &this->lock, __PRETTY_FUNCTION__ );
+
+                disable_interrupts();
+
+                thrd->cor.state = Halted;
+
+                verifyf( thrd == this->owner, "Expected owner to be %p, got %p (r: %i)", thrd, this->owner, this->recursion );
+
+                //Leaving a recursion level, decrement the counter
+                this->recursion -= 1;
+
+                //If we haven't left the last level of recursion
+                //it means we don't need to do anything
+                if( this->recursion != 0) {
+                        unlock( &this->lock );
+                        return;
+                }
+
+                thread_desc * new_owner = next_thread( this );
+
+                //We can now let other threads in safely
+                unlock( &this->lock );
+
+                //We need to wake-up the thread
+                if( new_owner) ScheduleThread( new_owner );
         }
 }
@@ -121 +150 @@
         enter( this->m, this->count );
 
-        this->prev_mntrs = this_thread()->current_monitors;
-        this->prev_count = this_thread()->current_monitor_count;
-
-        this_thread()->current_monitors      = m;
-        this_thread()->current_monitor_count = count;
+        this->prev_mntrs = this_thread->current_monitors;
+        this->prev_count = this_thread->current_monitor_count;
+
+        this_thread->current_monitors      = m;
+        this_thread->current_monitor_count = count;
 }
 
@@ -131 +160 @@
         leave( this->m, this->count );
 
-        this_thread()->current_monitors      = this->prev_mntrs;
-        this_thread()->current_monitor_count = this->prev_count;
+        this_thread->current_monitors      = this->prev_mntrs;
+        this_thread->current_monitor_count = this->prev_count;
 }
 
@@ -170 +199 @@
         unsigned short count = this->monitor_count;
         unsigned int recursions[ count ];               //Save the current recursion levels to restore them later
-        spinlock *   locks     [ count ];               //We need to pass-in an array of locks to ScheduleInternal
+        spinlock *   locks     [ count ];               //We need to pass-in an array of locks to BlockInternal
 
         LIB_DEBUG_PRINT_SAFE("count %i\n", count);
 
-        __condition_node_t waiter = { this_thread(), count, user_info };
+        __condition_node_t waiter = { (thread_desc*)this_thread, count, user_info };
 
         __condition_criterion_t criteria[count];
@@ -208 +237 @@
 
         // Everything is ready to go to sleep
-        ScheduleInternal( locks, count, threads, thread_count );
+        BlockInternal( locks, count, threads, thread_count );
 
 
@@ -234 +263 @@
         //Some more checking in debug
         LIB_DEBUG_DO(
-                thread_desc * this_thrd = this_thread();
+                thread_desc * this_thrd = this_thread;
                 if ( this->monitor_count != this_thrd->current_monitor_count ) {
                         abortf( "Signal on condition %p made with different number of monitor(s), expected %i got %i", this, this->monitor_count, this_thrd->current_monitor_count );
@@ -281 +310 @@
         unsigned short count = this->monitor_count;
         unsigned int recursions[ count ];               //Save the current recursion levels to restore them later
-        spinlock *   locks     [ count ];               //We need to pass-in an array of locks to ScheduleInternal
+        spinlock *   locks     [ count ];               //We need to pass-in an array of locks to BlockInternal
 
         lock_all( this->monitors, locks, count );
 
         //create creteria
-        __condition_node_t waiter = { this_thread(), count, 0 };
+        __condition_node_t waiter = { (thread_desc*)this_thread, count, 0 };
 
         __condition_criterion_t criteria[count];
@@ -309 +338 @@
 
         //Everything is ready to go to sleep
-        ScheduleInternal( locks, count, &signallee, 1 );
+        BlockInternal( locks, count, &signallee, 1 );
 
 
@@ -335 +364 @@
 // Internal scheduling
 void __accept_internal( unsigned short count, __acceptable_t * acceptables, void (*func)(void) ) {
-        // thread_desc * this = this_thread();
+        // thread_desc * this = this_thread;
 
         // unsigned short count = this->current_monitor_count;
         // unsigned int recursions[ count ];            //Save the current recursion levels to restore them later
-        // spinlock *   locks     [ count ];            //We need to pass-in an array of locks to ScheduleInternal
+        // spinlock *   locks     [ count ];            //We need to pass-in an array of locks to BlockInternal
 
         // lock_all( this->current_monitors, locks, count );
@@ -348 +377 @@
 
         // // // Everything is ready to go to sleep
-        // // ScheduleInternal( locks, count, threads, thread_count );
+        // // BlockInternal( locks, count, threads, thread_count );
 
 
@@ -393 +422 @@
 static inline void lock_all( spinlock ** locks, unsigned short count ) {
         for( int i = 0; i < count; i++ ) {
-                lock( locks[i] );
+                lock( locks[i], __PRETTY_FUNCTION__ );
         }
 }
@@ -400 +429 @@
         for( int i = 0; i < count; i++ ) {
                 spinlock * l = &source[i]->lock;
-                lock( l );
+                lock( l, __PRETTY_FUNCTION__ );
                 if(locks) locks[i] = l;
         }
@@ -457 +486 @@
 
 static inline void brand_condition( condition * this ) {
-        thread_desc * thrd = this_thread();
+        thread_desc * thrd = this_thread;
         if( !this->monitors ) {
                 LIB_DEBUG_PRINT_SAFE("Branding\n");

src/libcfa/concurrency/preemption.c

@@ -17 +17 @@
 #include "preemption.h"
 
+
 extern "C" {
+#include <errno.h>
+#include <execinfo.h>
+#define __USE_GNU
 #include <signal.h>
-}
-
-#define __CFA_DEFAULT_PREEMPTION__ 10
+#undef __USE_GNU
+#include <stdio.h>
+#include <string.h>
+#include <unistd.h>
+}
+
+
+#ifdef __USE_STREAM__
+#include "fstream"
+#endif
+#include "libhdr.h"
+
+#define __CFA_DEFAULT_PREEMPTION__ 10000
 
 __attribute__((weak)) unsigned int default_preemption() {
@@ -27 +41 @@
 }
 
+#define __CFA_SIGCXT__ ucontext_t *
+#define __CFA_SIGPARMS__ __attribute__((unused)) int sig, __attribute__((unused)) siginfo_t *sfp, __attribute__((unused)) __CFA_SIGCXT__ cxt
+
 static void preempt( processor   * this );
 static void timeout( thread_desc * this );
 
+void sigHandler_ctxSwitch( __CFA_SIGPARMS__ );
+void sigHandler_alarm    ( __CFA_SIGPARMS__ );
+void sigHandler_segv     ( __CFA_SIGPARMS__ );
+void sigHandler_abort    ( __CFA_SIGPARMS__ );
+
+static void __kernel_sigaction( int sig, void (*handler)(__CFA_SIGPARMS__), int flags );
+
 //=============================================================================================
 // Kernel Preemption logic
@@ -35 +59 @@
 
 void kernel_start_preemption() {
-
-}
+        LIB_DEBUG_PRINT_SAFE("Kernel : Starting preemption\n");
+        __kernel_sigaction( SIGUSR1, sigHandler_ctxSwitch, SA_SIGINFO );
+        __kernel_sigaction( SIGALRM, sigHandler_alarm    , SA_SIGINFO );
+        __kernel_sigaction( SIGSEGV, sigHandler_segv     , SA_SIGINFO );
+        __kernel_sigaction( SIGBUS , sigHandler_segv     , SA_SIGINFO );
+        // __kernel_sigaction( SIGABRT, sigHandler_abort    , SA_SIGINFO );
+}
+
+void kernel_stop_preemption() {
+        //Block all signals, we are no longer in a position to handle them
+        sigset_t mask;
+        sigfillset( &mask );
+        sigprocmask( SIG_BLOCK, &mask, NULL );
+        LIB_DEBUG_PRINT_SAFE("Kernel : Preemption stopped\n");
+
+        // assert( !systemProcessor->alarms.head );
+        // assert( systemProcessor->alarms.tail == &systemProcessor->alarms.head );
+}
+
+LIB_DEBUG_DO( bool validate( alarm_list_t * this ); )
 
 void tick_preemption() {
+        // LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Ticking preemption\n" );
+
         alarm_list_t * alarms = &systemProcessor->alarms;
         __cfa_time_t currtime = __kernel_get_time();
         while( alarms->head && alarms->head->alarm < currtime ) {
                 alarm_node_t * node = pop(alarms);
+                // LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Ticking %p\n", node );
+
                 if( node->kernel_alarm ) {
                         preempt( node->proc );
@@ -50 +96 @@
                 }
 
+                verify( validate( alarms ) );
+
                 if( node->period > 0 ) {
-                        node->alarm += node->period;
+                        node->alarm = currtime + node->period;
                         insert( alarms, node );
                 }
@@ -62 +110 @@
                 __kernel_set_timer( alarms->head->alarm - currtime );
         }
+
+        verify( validate( alarms ) );
+        // LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Ticking preemption done\n" );
 }
 
 void update_preemption( processor * this, __cfa_time_t duration ) {
-        //     assert( THREAD_GETMEM( disableInt ) && THREAD_GETMEM( disableIntCnt ) == 1 );
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Processor : %p updating preemption to %lu\n", this, duration );
+
         alarm_node_t * alarm = this->preemption_alarm;
+        duration *= 1000;
 
         // Alarms need to be enabled
@@ -97 +150 @@
 
 void ^?{}( preemption_scope * this ) {
+        disable_interrupts();
+
         update_preemption( this->proc, 0 );
 }
@@ -104 +159 @@
 //=============================================================================================
 
+extern "C" {
+        void disable_interrupts() {
+                __attribute__((unused)) unsigned short new_val = __atomic_add_fetch_2( &disable_preempt_count, 1, __ATOMIC_SEQ_CST );
+                verify( new_val < (unsigned short)65_000 );
+                verify( new_val != (unsigned short) 0 );
+        }
+
+        void enable_interrupts_noRF() {
+                unsigned short prev = __atomic_fetch_add_2( &disable_preempt_count, -1, __ATOMIC_SEQ_CST );
+                verify( prev != (unsigned short) 0 );
+        }
+
+        void enable_interrupts( const char * func ) {
+                processor * proc   = this_processor;
+                thread_desc * thrd = this_thread;
+                unsigned short prev = __atomic_fetch_add_2( &disable_preempt_count, -1, __ATOMIC_SEQ_CST );
+                verify( prev != (unsigned short) 0 );
+                if( prev == 1 && proc->pending_preemption ) {
+                        proc->pending_preemption = false;
+                        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Executing deferred CtxSwitch on %p\n", this_processor );
+                        BlockInternal( thrd );
+                        LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Executing deferred back\n" );
+                }
+
+                proc->last_enable = func;
+        }
+}
+
+static inline void signal_unblock( int sig ) {
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Processor : %p unblocking sig %i\n", this_processor, sig );
+
+        sigset_t mask;
+        sigemptyset( &mask );
+        sigaddset( &mask, sig );
+
+        if ( sigprocmask( SIG_UNBLOCK, &mask, NULL ) == -1 ) {
+            abortf( "internal error, sigprocmask" );
+        } // if
+}
+
 static inline bool preemption_ready() {
-        return this_processor->disable_preempt_count == 0;
+        return disable_preempt_count == 0;
 }
 
@@ -116 +211 @@
 }
 
-void sigHandler_ctxSwitch( __attribute__((unused)) int sig ) {
+extern "C" {
+        __attribute__((noinline)) void __debug_break() {
+                pthread_kill( pthread_self(), SIGTRAP );
+        }
+}
+
+void sigHandler_ctxSwitch( __CFA_SIGPARMS__ ) {
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Ctx Switch IRH %p running %p @ %p\n", this_processor, this_thread, (void *)(cxt->uc_mcontext.gregs[REG_RIP]) );
+
         if( preemption_ready() ) {
-                ScheduleInternal( this_processor->current_thread );
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Ctx Switch IRH : Blocking thread %p on %p\n", this_thread, this_processor );
+                signal_unblock( SIGUSR1 );
+                BlockInternal( (thread_desc*)this_thread );
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Ctx Switch IRH : Back\n\n");
+        }
+        else {
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Ctx Switch IRH : Defering\n" );
+                defer_ctxSwitch();
+                signal_unblock( SIGUSR1 );
+        }
+}
+
+void sigHandler_alarm( __CFA_SIGPARMS__ ) {
+        LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "\nAlarm IRH %p running %p @ %p\n", this_processor, this_thread, (void *)(cxt->uc_mcontext.gregs[REG_RIP]) );
+
+        // if( ((intptr_t)cxt->uc_mcontext.gregs[REG_RIP]) > 0xFFFFFF ) __debug_break();
+
+        if( try_lock( &systemProcessor->alarm_lock, __PRETTY_FUNCTION__ ) ) {
+                tick_preemption();
+                unlock( &systemProcessor->alarm_lock );
+        }
+        else {
+                defer_alarm();
+        }
+
+        signal_unblock( SIGALRM );
+
+        if( preemption_ready() && this_processor->pending_preemption ) {
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Alarm IRH : Blocking thread %p on %p\n", this_thread, this_processor );
+                this_processor->pending_preemption = false;
+                BlockInternal( (thread_desc*)this_thread );
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( STDERR_FILENO, "Alarm Switch IRH : Back\n\n");
+        }
+}
+
+static void preempt( processor * this ) {
+        // LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO, "Processor : signalling %p\n", this );
+
+        if( this != systemProcessor ) {
+                pthread_kill( this->kernel_thread, SIGUSR1 );
         }
         else {
@@ -125 +267 @@
 }
 
-void sigHandler_alarm( __attribute__((unused)) int sig ) {
-        if( try_lock( &systemProcessor->alarm_lock ) ) {
-                tick_preemption();
-                unlock( &systemProcessor->alarm_lock );
-        }
-        else {
-                defer_alarm();
-        }
-}
-
-static void preempt( processor * this ) {
-        pthread_kill( this->kernel_thread, SIGUSR1 );
-}
-
 static void timeout( thread_desc * this ) {
         //TODO : implement waking threads
 }
+
+static void __kernel_sigaction( int sig, void (*handler)(__CFA_SIGPARMS__), int flags ) {
+        struct sigaction act;
+
+        act.sa_sigaction = (void (*)(int, siginfo_t *, void *))handler;
+        act.sa_flags = flags;
+
+        // disabled during signal handler
+        sigemptyset( &act.sa_mask );
+        sigaddset( &act.sa_mask, sig );
+
+        if ( sigaction( sig, &act, NULL ) == -1 ) {
+                LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO,
+                        " __kernel_sigaction( sig:%d, handler:%p, flags:%d ), problem installing signal handler, error(%d) %s.\n",
+                        sig, handler, flags, errno, strerror( errno )
+                );
+                _exit( EXIT_FAILURE );
+        }
+}
+
+typedef void (*sa_handler_t)(int);
+
+static void __kernel_sigdefault( int sig ) {
+        struct sigaction act;
+
+        // act.sa_handler = SIG_DFL;
+        act.sa_flags = 0;
+        sigemptyset( &act.sa_mask );
+
+        if ( sigaction( sig, &act, NULL ) == -1 ) {
+                LIB_DEBUG_PRINT_BUFFER_DECL( STDERR_FILENO,
+                        " __kernel_sigdefault( sig:%d ), problem reseting signal handler, error(%d) %s.\n",
+                        sig, errno, strerror( errno )
+                );
+                _exit( EXIT_FAILURE );
+        }
+}
+
+//=============================================================================================
+// Terminating Signals logic
+//=============================================================================================
+
+LIB_DEBUG_DO(
+        static void __kernel_backtrace( int start ) {
+                // skip first N stack frames
+
+                enum { Frames = 50 };
+                void * array[Frames];
+                int size = backtrace( array, Frames );
+                char ** messages = backtrace_symbols( array, size );
+
+                // find executable name
+                *index( messages[0], '(' ) = '\0';
+                #ifdef __USE_STREAM__
+                serr | "Stack back trace for:" | messages[0] | endl;
+                #else
+                fprintf( stderr, "Stack back trace for: %s\n", messages[0]);
+                #endif
+
+                // skip last 2 stack frames after main
+                for ( int i = start; i < size && messages != NULL; i += 1 ) {
+                        char * name = NULL;
+                        char * offset_begin = NULL;
+                        char * offset_end = NULL;
+
+                        for ( char *p = messages[i]; *p; ++p ) {
+                                // find parantheses and +offset
+                                if ( *p == '(' ) {
+                                        name = p;
+                                }
+                                else if ( *p == '+' ) {
+                                        offset_begin = p;
+                                }
+                                else if ( *p == ')' ) {
+                                        offset_end = p;
+                                        break;
+                                }
+                        }
+
+                        // if line contains symbol print it
+                        int frameNo = i - start;
+                        if ( name && offset_begin && offset_end && name < offset_begin ) {
+                                // delimit strings
+                                *name++ = '\0';
+                                *offset_begin++ = '\0';
+                                *offset_end++ = '\0';
+
+                                #ifdef __USE_STREAM__
+                                serr    | "("  | frameNo | ")" | messages[i] | ":"
+                                        | name | "+" | offset_begin | offset_end | endl;
+                                #else
+                                fprintf( stderr, "(%i) %s : %s + %s %s\n", frameNo, messages[i], name, offset_begin, offset_end);
+                                #endif
+                        }
+                        // otherwise, print the whole line
+                        else {
+                                #ifdef __USE_STREAM__
+                                serr | "(" | frameNo | ")" | messages[i] | endl;
+                                #else
+                                fprintf( stderr, "(%i) %s\n", frameNo, messages[i] );
+                                #endif
+                        }
+                }
+
+                free( messages );
+        }
+)
+
+void sigHandler_segv( __CFA_SIGPARMS__ ) {
+        LIB_DEBUG_DO(
+                #ifdef __USE_STREAM__
+                serr    | "*CFA runtime error* program cfa-cpp terminated with"
+                        | (sig == SIGSEGV ? "segment fault." : "bus error.")
+                        | endl;
+                #else
+                fprintf( stderr, "*CFA runtime error* program cfa-cpp terminated with %s\n", sig == SIGSEGV ? "segment fault." : "bus error." );
+                #endif
+
+                // skip first 2 stack frames
+                __kernel_backtrace( 1 );
+        )
+        exit( EXIT_FAILURE );
+}
+
+// void sigHandler_abort( __CFA_SIGPARMS__ ) {
+//      // skip first 6 stack frames
+//      LIB_DEBUG_DO( __kernel_backtrace( 6 ); )
+
+//      // reset default signal handler
+//      __kernel_sigdefault( SIGABRT );
+
+//      raise( SIGABRT );
+// }

src/libcfa/concurrency/thread

@@ -54 +54 @@
 }
 
-thread_desc * this_thread(void);
+extern volatile thread_local thread_desc * this_thread;
 
 forall( dtype T | is_thread(T) )

src/libcfa/concurrency/thread.c

@@ -28 +28 @@
 }
 
-extern thread_local processor * this_processor;
+extern volatile thread_local processor * this_processor;
 
 //-----------------------------------------------------------------------------
@@ -71 +71 @@
         coroutine_desc* thrd_c = get_coroutine(this);
         thread_desc*  thrd_h = get_thread   (this);
-        thrd_c->last = this_coroutine();
-        this_processor->current_coroutine = thrd_c;
+        thrd_c->last = this_coroutine;
 
-        LIB_DEBUG_PRINT_SAFE("Thread start : %p (t %p, c %p)\n", this, thrd_c, thrd_h);
+        // LIB_DEBUG_PRINT_SAFE("Thread start : %p (t %p, c %p)\n", this, thrd_c, thrd_h);
 
+        disable_interrupts();
         create_stack(&thrd_c->stack, thrd_c->stack.size);
+        this_coroutine = thrd_c;
         CtxStart(this, CtxInvokeThread);
+        assert( thrd_c->last->stack.context );
         CtxSwitch( thrd_c->last->stack.context, thrd_c->stack.context );
 
         ScheduleThread(thrd_h);
+        enable_interrupts( __PRETTY_FUNCTION__ );
 }
 
 void yield( void ) {
-        ScheduleInternal( this_processor->current_thread );
+        BlockInternal( (thread_desc *)this_thread );
 }
 
@@ -95 +98 @@
 void ThreadCtxSwitch(coroutine_desc* src, coroutine_desc* dst) {
         // set state of current coroutine to inactive
-        src->state = Inactive;
+        src->state = src->state == Halted ? Halted : Inactive;
         dst->state = Active;
 
@@ -103 +106 @@
         // set new coroutine that the processor is executing
         // and context switch to it
-        this_processor->current_coroutine = dst;
+        this_coroutine = dst;
+        assert( src->stack.context );
         CtxSwitch( src->stack.context, dst->stack.context );
-        this_processor->current_coroutine = src;
+        this_coroutine = src;
 
         // set state of new coroutine to active
-        dst->state = Inactive;
+        dst->state = dst->state == Halted ? Halted : Inactive;
         src->state = Active;
 }

src/libcfa/libhdr/libdebug.h

@@ -18 +18 @@
 
 #ifdef __CFA_DEBUG__
-        #define LIB_DEBUG_DO(x) x
-        #define LIB_NO_DEBUG_DO(x) ((void)0)
+        #define LIB_DEBUG_DO(...) __VA_ARGS__
+        #define LIB_NO_DEBUG_DO(...)
 #else
-        #define LIB_DEBUG_DO(x) ((void)0)
-        #define LIB_NO_DEBUG_DO(x) x      
+        #define LIB_DEBUG_DO(...)
+        #define LIB_NO_DEBUG_DO(...) __VA_ARGS__
 #endif
 
@@ -51 +51 @@
 
 #ifdef __CFA_DEBUG_PRINT__
-      #define LIB_DEBUG_WRITE( fd, buffer, len )  __lib_debug_write( fd, buffer, len )
-      #define LIB_DEBUG_ACQUIRE()                 __lib_debug_acquire()
-      #define LIB_DEBUG_RELEASE()                 __lib_debug_release()
-      #define LIB_DEBUG_PRINT_SAFE(...)           __lib_debug_print_safe   (__VA_ARGS__)
-      #define LIB_DEBUG_PRINT_NOLOCK(...)         __lib_debug_print_nolock (__VA_ARGS__)
-      #define LIB_DEBUG_PRINT_BUFFER(...)         __lib_debug_print_buffer (__VA_ARGS__)
+        #define LIB_DEBUG_WRITE( fd, buffer, len )     __lib_debug_write( fd, buffer, len )
+        #define LIB_DEBUG_ACQUIRE()                    __lib_debug_acquire()
+        #define LIB_DEBUG_RELEASE()                    __lib_debug_release()
+        #define LIB_DEBUG_PRINT_SAFE(...)              __lib_debug_print_safe   (__VA_ARGS__)
+        #define LIB_DEBUG_PRINT_NOLOCK(...)            __lib_debug_print_nolock (__VA_ARGS__)
+        #define LIB_DEBUG_PRINT_BUFFER(...)            __lib_debug_print_buffer (__VA_ARGS__)
+        #define LIB_DEBUG_PRINT_BUFFER_DECL(fd, ...)   char text[256]; int len = snprintf( text, 256, __VA_ARGS__ ); __lib_debug_write( fd, text, len );
+        #define LIB_DEBUG_PRINT_BUFFER_LOCAL(fd, ...)  len = snprintf( text, 256, __VA_ARGS__ ); __lib_debug_write( fd, text, len );
 #else
-      #define LIB_DEBUG_WRITE(...)          ((void)0)
-      #define LIB_DEBUG_ACQUIRE()           ((void)0)
-      #define LIB_DEBUG_RELEASE()           ((void)0)
-      #define LIB_DEBUG_PRINT_SAFE(...)     ((void)0)
-      #define LIB_DEBUG_PRINT_NOLOCK(...)   ((void)0)
-      #define LIB_DEBUG_PRINT_BUFFER(...)   ((void)0)
+        #define LIB_DEBUG_WRITE(...)               ((void)0)
+        #define LIB_DEBUG_ACQUIRE()                ((void)0)
+        #define LIB_DEBUG_RELEASE()                ((void)0)
+        #define LIB_DEBUG_PRINT_SAFE(...)          ((void)0)
+        #define LIB_DEBUG_PRINT_NOLOCK(...)        ((void)0)
+        #define LIB_DEBUG_PRINT_BUFFER(...)        ((void)0)
+        #define LIB_DEBUG_PRINT_BUFFER_DECL(...)   ((void)0)
+        #define LIB_DEBUG_PRINT_BUFFER_LOCAL(...)  ((void)0)
 #endif
 

src/tests/sched-int-block.c

@@ -29 +29 @@
 //------------------------------------------------------------------------------
 void wait_op( global_data_t * mutex a, global_data_t * mutex b, unsigned i ) {
-        wait( &cond, (uintptr_t)this_thread() );
+        wait( &cond, (uintptr_t)this_thread );
 
         yield( ((unsigned)rand48()) % 10 );
@@ -38 +38 @@
         }
 
-        a->last_thread = b->last_thread = this_thread();
+        a->last_thread = b->last_thread = this_thread;
 
         yield( ((unsigned)rand48()) % 10 );
@@ -54 +54 @@
         yield( ((unsigned)rand48()) % 10 );
 
-        a->last_thread = b->last_thread = a->last_signaller = b->last_signaller = this_thread();
+        a->last_thread = b->last_thread = a->last_signaller = b->last_signaller = this_thread;
 
         if( !is_empty( &cond ) ) {
@@ -84 +84 @@
 //------------------------------------------------------------------------------
 void barge_op( global_data_t * mutex a ) {
-        a->last_thread = this_thread();
+        a->last_thread = this_thread;
 }