Changeset 3f7e12cb for src/libcfa/concurrency/monitor.c

Timestamp:

Nov 8, 2017, 5:43:33 PM (8 years ago)

Author:

Aaron Moss <a3moss@…>

Branches:

ADT, aaron-thesis, arm-eh, ast-experimental, cleanup-dtors, deferred_resn, demangler, enum, forall-pointer-decay, jacob/cs343-translation, jenkins-sandbox, master, new-ast, new-ast-unique-expr, new-env, no_list, persistent-indexer, pthread-emulation, qualifiedEnum, resolv-new, with_gc

Children:

954908d

Parents:

78315272 (diff), e35f30a (diff)
Note: this is a merge changeset, the changes displayed below correspond to the merge itself.
Use the (diff) links above to see all the changes relative to each parent.

Message:

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc

File:

• src/libcfa/concurrency/monitor.c (modified) (31 diffs)

src/libcfa/concurrency/monitor.c

@@ -17 +17 @@
 
 #include <stdlib>
+#include <inttypes.h>
 
 #include "libhdr.h"
 #include "kernel_private.h"
 
+#include "bits/algorithms.h"
+
 //-----------------------------------------------------------------------------
 // Forward declarations
-static inline void set_owner( monitor_desc * this, thread_desc * owner );
+static inline void set_owner ( monitor_desc * this, thread_desc * owner );
+static inline void set_owner ( monitor_desc * storage [], __lock_size_t count, thread_desc * owner );
+static inline void set_mask  ( monitor_desc * storage [], __lock_size_t count, const __waitfor_mask_t & mask );
+static inline void reset_mask( monitor_desc * this );
+
 static inline thread_desc * next_thread( monitor_desc * this );
-static inline int is_accepted( thread_desc * owner, monitor_desc * this, monitor_desc ** group, int group_cnt, void (*func)() );
-
-static inline void lock_all( spinlock ** locks, unsigned short count );
-static inline void lock_all( monitor_desc ** source, spinlock ** /*out*/ locks, unsigned short count );
-static inline void unlock_all( spinlock ** locks, unsigned short count );
-static inline void unlock_all( monitor_desc ** locks, unsigned short count );
-
-static inline void save_recursion   ( monitor_desc ** ctx, unsigned int * /*out*/ recursions, unsigned short count );
-static inline void restore_recursion( monitor_desc ** ctx, unsigned int * /*in */ recursions, unsigned short count );
-
-static inline void init     ( int count, monitor_desc ** monitors, __condition_node_t * waiter, __condition_criterion_t * criteria );
-static inline void init_push( int count, monitor_desc ** monitors, __condition_node_t * waiter, __condition_criterion_t * criteria );
-
-static inline thread_desc * check_condition( __condition_criterion_t * );
-static inline void brand_condition( condition * );
-static inline unsigned short insert_unique( thread_desc ** thrds, unsigned short end, thread_desc * val );
-
-static inline thread_desc * search_entry_queue( __acceptable_t * acceptables, int acc_count, monitor_desc ** monitors, int count );
+static inline bool is_accepted( monitor_desc * this, const __monitor_group_t & monitors );
+
+static inline void lock_all  ( spinlock * locks [], __lock_size_t count );
+static inline void lock_all  ( monitor_desc * source [], spinlock * /*out*/ locks [], __lock_size_t count );
+static inline void unlock_all( spinlock * locks [], __lock_size_t count );
+static inline void unlock_all( monitor_desc * locks [], __lock_size_t count );
+
+static inline void save   ( monitor_desc * ctx [], __lock_size_t count, spinlock * locks [], unsigned int /*out*/ recursions [], __waitfor_mask_t /*out*/ masks [] );
+static inline void restore( monitor_desc * ctx [], __lock_size_t count, spinlock * locks [], unsigned int /*in */ recursions [], __waitfor_mask_t /*in */ masks [] );
+
+static inline void init     ( __lock_size_t count, monitor_desc * monitors [], __condition_node_t & waiter, __condition_criterion_t criteria [] );
+static inline void init_push( __lock_size_t count, monitor_desc * monitors [], __condition_node_t & waiter, __condition_criterion_t criteria [] );
+
+static inline thread_desc *        check_condition   ( __condition_criterion_t * );
+static inline void                 brand_condition   ( condition & );
+static inline [thread_desc *, int] search_entry_queue( const __waitfor_mask_t &, monitor_desc * monitors [], __lock_size_t count );
+
+forall(dtype T | sized( T ))
+static inline __lock_size_t insert_unique( T * array [], __lock_size_t & size, T * val );
+static inline __lock_size_t count_max    ( const __waitfor_mask_t & mask );
+static inline __lock_size_t aggregate    ( monitor_desc * storage [], const __waitfor_mask_t & mask );
 
 //-----------------------------------------------------------------------------
 // Useful defines
-#define wait_ctx(thrd, user_info)                               /* Create the necessary information to use the signaller stack       */ \
-        __condition_node_t waiter = { thrd, count, user_info };   /* Create the node specific to this wait operation                   */ \
-        __condition_criterion_t criteria[count];                  /* Create the creteria this wait operation needs to wake up          */ \
-        init( count, monitors, &waiter, criteria );               /* Link everything together                                          */ \
-
-#define wait_ctx_primed(thrd, user_info)                        /* Create the necessary information to use the signaller stack       */ \
-        __condition_node_t waiter = { thrd, count, user_info };   /* Create the node specific to this wait operation                   */ \
-        __condition_criterion_t criteria[count];                  /* Create the creteria this wait operation needs to wake up          */ \
-        init_push( count, monitors, &waiter, criteria );          /* Link everything together and push it to the AS-Stack              */ \
-
-#define monitor_ctx( mons, cnt )              /* Define that create the necessary struct for internal/external scheduling operations */ \
-        monitor_desc ** monitors = mons;        /* Save the targeted monitors                                                          */ \
-        unsigned short count = cnt;             /* Save the count to a local variable                                                  */ \
-        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 BlockInternal                               */ \
+#define wait_ctx(thrd, user_info)                               /* Create the necessary information to use the signaller stack                         */ \
+        __condition_node_t waiter = { thrd, count, user_info };   /* Create the node specific to this wait operation                                     */ \
+        __condition_criterion_t criteria[count];                  /* Create the creteria this wait operation needs to wake up                            */ \
+        init( count, monitors, waiter, criteria );                /* Link everything together                                                            */ \
+
+#define wait_ctx_primed(thrd, user_info)                        /* Create the necessary information to use the signaller stack                         */ \
+        __condition_node_t waiter = { thrd, count, user_info };   /* Create the node specific to this wait operation                                     */ \
+        __condition_criterion_t criteria[count];                  /* Create the creteria this wait operation needs to wake up                            */ \
+        init_push( count, monitors, waiter, criteria );           /* Link everything together and push it to the AS-Stack                                */ \
+
+#define monitor_ctx( mons, cnt )                                /* Define that create the necessary struct for internal/external scheduling operations */ \
+        monitor_desc ** monitors = mons;                          /* Save the targeted monitors                                                          */ \
+        __lock_size_t count = cnt;                                /* Save the count to a local variable                                                  */ \
+        unsigned int recursions[ count ];                         /* Save the current recursion levels to restore them later                             */ \
+        __waitfor_mask_t masks [ count ];                         /* Save the current waitfor masks to restore them later                                */ \
+        spinlock *   locks     [ count ];                         /* We need to pass-in an array of locks to BlockInternal                               */ \
+
+#define monitor_save    save   ( monitors, count, locks, recursions, masks )
+#define monitor_restore restore( monitors, count, locks, recursions, masks )
+
 
 //-----------------------------------------------------------------------------
@@ -68 +83 @@
 extern "C" {
         // Enter single monitor
-        static void __enter_monitor_desc( monitor_desc * this, monitor_desc ** group, int group_cnt, void (*func)() ) {
+        static void __enter_monitor_desc( monitor_desc * this, const __monitor_group_t & group ) {
                 // Lock the monitor spinlock, lock_yield to reduce contention
                 lock_yield( &this->lock DEBUG_CTX2 );
@@ -75 +90 @@
                 LIB_DEBUG_PRINT_SAFE("Kernel : %10p Entering mon %p (%p)\n", thrd, this, this->owner);
 
-                this->accepted_index = -1;
                 if( !this->owner ) {
                         // No one has the monitor, just take it
@@ -83 +97 @@
                 }
                 else if( this->owner == thrd) {
-                        // We already have the monitor, just not how many times we took it
-                        verify( this->recursion > 0 );
+                        // We already have the monitor, just note how many times we took it
                         this->recursion += 1;
 
                         LIB_DEBUG_PRINT_SAFE("Kernel :  mon already owned \n");
                 }
-                else if( (this->accepted_index = is_accepted( thrd, this, group, group_cnt, func)) >= 0 ) {
+                else if( is_accepted( this, group) ) {
                         // Some one was waiting for us, enter
                         set_owner( this, thrd );
 
+                        // Reset mask
+                        reset_mask( this );
+
                         LIB_DEBUG_PRINT_SAFE("Kernel :  mon accepts \n");
                 }
@@ -99 +115 @@
 
                         // Some one else has the monitor, wait in line for it
-                        append( &this->entry_queue, thrd );
+                        append( this->entry_queue, thrd );
                         BlockInternal( &this->lock );
 
@@ -113 +129 @@
                 unlock( &this->lock );
                 return;
+        }
+
+        static void __enter_monitor_dtor( monitor_desc * this, fptr_t func ) {
+                // Lock the monitor spinlock, lock_yield to reduce contention
+                lock_yield( &this->lock DEBUG_CTX2 );
+                thread_desc * thrd = this_thread;
+
+                LIB_DEBUG_PRINT_SAFE("Kernel : %10p Entering dtor for mon %p (%p)\n", thrd, this, this->owner);
+
+
+                if( !this->owner ) {
+                        LIB_DEBUG_PRINT_SAFE("Kernel : Destroying free mon %p\n", this);
+
+                        // No one has the monitor, just take it
+                        set_owner( this, thrd );
+
+                        unlock( &this->lock );
+                        return;
+                }
+                else if( this->owner == thrd) {
+                        // We already have the monitor... but where about to destroy it so the nesting will fail
+                        // Abort!
+                        abortf("Attempt to destroy monitor %p by thread \"%.256s\" (%p) in nested mutex.");
+                }
+
+                __lock_size_t count = 1;
+                monitor_desc ** monitors = &this;
+                __monitor_group_t group = { &this, 1, func };
+                if( is_accepted( this, group) ) {
+                        LIB_DEBUG_PRINT_SAFE("Kernel :  mon accepts dtor, block and signal it \n");
+
+                        // Wake the thread that is waiting for this
+                        __condition_criterion_t * urgent = pop( this->signal_stack );
+                        verify( urgent );
+
+                        // Reset mask
+                        reset_mask( this );
+
+                        // Create the node specific to this wait operation
+                        wait_ctx_primed( this_thread, 0 )
+
+                        // Some one else has the monitor, wait for him to finish and then run
+                        BlockInternal( &this->lock, urgent->owner->waiting_thread );
+
+                        // Some one was waiting for us, enter
+                        set_owner( this, thrd );
+                }
+                else {
+                        LIB_DEBUG_PRINT_SAFE("Kernel :  blocking \n");
+
+                        wait_ctx( this_thread, 0 )
+                        this->dtor_node = &waiter;
+
+                        // Some one else has the monitor, wait in line for it
+                        append( this->entry_queue, thrd );
+                        BlockInternal( &this->lock );
+
+                        // BlockInternal will unlock spinlock, no need to unlock ourselves
+                        return;
+                }
+
+                LIB_DEBUG_PRINT_SAFE("Kernel : Destroying %p\n", this);
+
         }
 
@@ -120 +199 @@
                 lock_yield( &this->lock DEBUG_CTX2 );
 
-                verifyf( this_thread == this->owner, "Expected owner to be %p, got %p (r: %i)", this_thread, this->owner, this->recursion );
+                LIB_DEBUG_PRINT_SAFE("Kernel : %10p Leaving mon %p (%p)\n", this_thread, this, this->owner);
+
+                verifyf( this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", this_thread, this->owner, this->recursion, this );
 
                 // Leaving a recursion level, decrement the counter
@@ -128 +209 @@
                 // it means we don't need to do anything
                 if( this->recursion != 0) {
+                        LIB_DEBUG_PRINT_SAFE("Kernel :  recursion still %d\n", this->recursion);
                         unlock( &this->lock );
                         return;
@@ -140 +222 @@
                 //We need to wake-up the thread
                 WakeThread( new_owner );
+        }
+
+        // Leave single monitor for the last time
+        void __leave_dtor_monitor_desc( monitor_desc * this ) {
+                LIB_DEBUG_DO(
+                        if( this_thread != this->owner ) {
+                                abortf("Destroyed monitor %p has inconsistent owner, expected %p got %p.\n", this, this_thread, this->owner);
+                        }
+                        if( this->recursion != 1 ) {
+                                abortf("Destroyed monitor %p has %d outstanding nested calls.\n", this, this->recursion - 1);
+                        }
+                )
         }
 
@@ -146 +240 @@
         // Should never return
         void __leave_thread_monitor( thread_desc * thrd ) {
-                monitor_desc * this = &thrd->mon;
+                monitor_desc * this = &thrd->self_mon;
 
                 // Lock the monitor now
@@ -153 +247 @@
                 disable_interrupts();
 
-                thrd->cor.state = Halted;
-
-                verifyf( thrd == this->owner, "Expected owner to be %p, got %p (r: %i)", thrd, this->owner, this->recursion );
+                thrd->self_cor.state = Halted;
+
+                verifyf( thrd == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", thrd, this->owner, this->recursion, this );
 
                 // Leaving a recursion level, decrement the counter
@@ -178 +272 @@
 // Enter multiple monitor
 // relies on the monitor array being sorted
-static inline void enter(monitor_desc ** monitors, int count, void (*func)() ) {
-        for(int i = 0; i < count; i++) {
-                __enter_monitor_desc( monitors[i], monitors, count, func );
+static inline void enter( __monitor_group_t monitors ) {
+        for( __lock_size_t i = 0; i < monitors.size; i++) {
+                __enter_monitor_desc( monitors.list[i], monitors );
         }
 }
@@ -186 +280 @@
 // Leave multiple monitor
 // relies on the monitor array being sorted
-static inline void leave(monitor_desc ** monitors, int count) {
-        for(int i = count - 1; i >= 0; i--) {
+static inline void leave(monitor_desc * monitors [], __lock_size_t count) {
+        for( __lock_size_t i = count - 1; i >= 0; i--) {
                 __leave_monitor_desc( monitors[i] );
         }
@@ -194 +288 @@
 // Ctor for monitor guard
 // Sorts monitors before entering
-void ?{}( monitor_guard_t & this, monitor_desc ** m, int count, void (*func)() ) {
+void ?{}( monitor_guard_t & this, monitor_desc * m [], __lock_size_t count, fptr_t func ) {
         // Store current array
         this.m = m;
@@ -200 +294 @@
 
         // Sort monitors based on address -> TODO use a sort specialized for small numbers
-        qsort(this.m, count);
+        __libcfa_small_sort(this.m, count);
 
         // Save previous thread context
-        this.prev_mntrs = this_thread->current_monitors;
-        this.prev_count = this_thread->current_monitor_count;
-        this.prev_func  = this_thread->current_monitor_func;
+        this.[prev_mntrs, prev_count, prev_func] = this_thread->monitors.[list, size, func];
 
         // Update thread context (needed for conditions)
-        this_thread->current_monitors      = m;
-        this_thread->current_monitor_count = count;
-        this_thread->current_monitor_func  = func;
+        this_thread->monitors.[list, size, func] = [m, count, func];
+
+        // LIB_DEBUG_PRINT_SAFE("MGUARD : enter %d\n", count);
 
         // Enter the monitors in order
-        enter( this.m, this.count, func );
+        __monitor_group_t group = {this.m, this.count, func};
+        enter( group );
+
+        // LIB_DEBUG_PRINT_SAFE("MGUARD : entered\n");
 }
 
@@ -219 +314 @@
 // Dtor for monitor guard
 void ^?{}( monitor_guard_t & this ) {
+        // LIB_DEBUG_PRINT_SAFE("MGUARD : leaving %d\n", this.count);
+
         // Leave the monitors in order
         leave( this.m, this.count );
 
+        // LIB_DEBUG_PRINT_SAFE("MGUARD : left\n");
+
         // Restore thread context
-        this_thread->current_monitors      = this.prev_mntrs;
-        this_thread->current_monitor_count = this.prev_count;
-        this_thread->current_monitor_func  = this.prev_func;
+        this_thread->monitors.[list, size, func] = this.[prev_mntrs, prev_count, prev_func];
+}
+
+// Ctor for monitor guard
+// Sorts monitors before entering
+void ?{}( monitor_dtor_guard_t & this, monitor_desc * m [], fptr_t func ) {
+        // Store current array
+        this.m = *m;
+
+        // Save previous thread context
+        this.[prev_mntrs, prev_count, prev_func] = this_thread->monitors.[list, size, func];
+
+        // Update thread context (needed for conditions)
+        this_thread->monitors.[list, size, func] = [m, 1, func];
+
+        __enter_monitor_dtor( this.m, func );
+}
+
+// Dtor for monitor guard
+void ^?{}( monitor_dtor_guard_t & this ) {
+        // Leave the monitors in order
+        __leave_dtor_monitor_desc( this.m );
+
+        // Restore thread context
+        this_thread->monitors.[list, size, func] = this.[prev_mntrs, prev_count, prev_func];
 }
 
 //-----------------------------------------------------------------------------
 // Internal scheduling types
-void ?{}(__condition_node_t & this, thread_desc * waiting_thread, unsigned short count, uintptr_t user_info ) {
+void ?{}(__condition_node_t & this, thread_desc * waiting_thread, __lock_size_t count, uintptr_t user_info ) {
         this.waiting_thread = waiting_thread;
         this.count = count;
@@ -244 +365 @@
 }
 
-void ?{}(__condition_criterion_t & this, monitor_desc * target, __condition_node_t * owner ) {
+void ?{}(__condition_criterion_t & this, monitor_desc * target, __condition_node_t & owner ) {
         this.ready  = false;
         this.target = target;
-        this.owner  = owner;
+        this.owner  = &owner;
         this.next   = NULL;
 }
@@ -253 +374 @@
 //-----------------------------------------------------------------------------
 // Internal scheduling
-void wait( condition * this, uintptr_t user_info = 0 ) {
+void wait( condition & this, uintptr_t user_info = 0 ) {
         brand_condition( this );
 
         // Check that everything is as expected
-        assertf( this->monitors != NULL, "Waiting with no monitors (%p)", this->monitors );
-        verifyf( this->monitor_count != 0, "Waiting with 0 monitors (%i)", this->monitor_count );
-        verifyf( this->monitor_count < 32u, "Excessive monitor count (%i)", this->monitor_count );
+        assertf( this.monitors != NULL, "Waiting with no monitors (%p)", this.monitors );
+        verifyf( this.monitor_count != 0, "Waiting with 0 monitors (%"PRIiFAST16")", this.monitor_count );
+        verifyf( this.monitor_count < 32u, "Excessive monitor count (%"PRIiFAST16")", this.monitor_count );
 
         // Create storage for monitor context
-        monitor_ctx( this->monitors, this->monitor_count );
+        monitor_ctx( this.monitors, this.monitor_count );
 
         // Create the node specific to this wait operation
@@ -269 +390 @@
         // Append the current wait operation to the ones already queued on the condition
         // We don't need locks for that since conditions must always be waited on inside monitor mutual exclusion
-        append( &this->blocked, &waiter );
-
-        // Lock all monitors (aggregates the lock them as well)
+        append( this.blocked, &waiter );
+
+        // Lock all monitors (aggregates the locks as well)
         lock_all( monitors, locks, count );
 
-        // DON'T unlock, ask the kernel to do it
-
-        // Save monitor state
-        save_recursion( monitors, recursions, count );
-
         // Find the next thread(s) to run
-        unsigned short thread_count = 0;
+        __lock_size_t thread_count = 0;
         thread_desc * threads[ count ];
-        for(int i = 0; i < count; i++) {
-                threads[i] = 0;
-        }
+        __builtin_memset( threads, 0, sizeof( threads ) );
+
+        // Save monitor states
+        monitor_save;
 
         // Remove any duplicate threads
-        for( int i = 0; i < count; i++) {
+        for( __lock_size_t i = 0; i < count; i++) {
                 thread_desc * new_owner = next_thread( monitors[i] );
-                thread_count = insert_unique( threads, thread_count, new_owner );
+                insert_unique( threads, thread_count, new_owner );
         }
 
@@ -295 +412 @@
         BlockInternal( locks, count, threads, thread_count );
 
-
-        // WE WOKE UP
-
-
         // We are back, restore the owners and recursions
-        lock_all( locks, count );
-        restore_recursion( monitors, recursions, count );
-        unlock_all( locks, count );
-}
-
-bool signal( condition * this ) {
+        monitor_restore;
+}
+
+bool signal( condition & this ) {
         if( is_empty( this ) ) { return false; }
 
         //Check that everything is as expected
-        verify( this->monitors );
-        verify( this->monitor_count != 0 );
+        verify( this.monitors );
+        verify( this.monitor_count != 0 );
 
         //Some more checking in debug
         LIB_DEBUG_DO(
                 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 );
-                }
-
-                for(int i = 0; i < this->monitor_count; i++) {
-                        if ( this->monitors[i] != this_thrd->current_monitors[i] ) {
-                                abortf( "Signal on condition %p made with different monitor, expected %p got %i", this, this->monitors[i], this_thrd->current_monitors[i] );
+                if ( this.monitor_count != this_thrd->monitors.size ) {
+                        abortf( "Signal on condition %p made with different number of monitor(s), expected %i got %i", &this, this.monitor_count, this_thrd->monitors.size );
+                }
+
+                for(int i = 0; i < this.monitor_count; i++) {
+                        if ( this.monitors[i] != this_thrd->monitors.list[i] ) {
+                                abortf( "Signal on condition %p made with different monitor, expected %p got %i", &this, this.monitors[i], this_thrd->monitors.list[i] );
                         }
                 }
         );
 
-        unsigned short count = this->monitor_count;
+        __lock_size_t count = this.monitor_count;
 
         // Lock all monitors
-        lock_all( this->monitors, NULL, count );
+        lock_all( this.monitors, NULL, count );
 
         //Pop the head of the waiting queue
-        __condition_node_t * node = pop_head( &this->blocked );
+        __condition_node_t * node = pop_head( this.blocked );
 
         //Add the thread to the proper AS stack
@@ -338 +449 @@
                 __condition_criterion_t * crit = &node->criteria[i];
                 assert( !crit->ready );
-                push( &crit->target->signal_stack, crit );
+                push( crit->target->signal_stack, crit );
         }
 
         //Release
-        unlock_all( this->monitors, count );
+        unlock_all( this.monitors, count );
 
         return true;
 }
 
-bool signal_block( condition * this ) {
-        if( !this->blocked.head ) { return false; }
+bool signal_block( condition & this ) {
+        if( !this.blocked.head ) { return false; }
 
         //Check that everything is as expected
-        verifyf( this->monitors != NULL, "Waiting with no monitors (%p)", this->monitors );
-        verifyf( this->monitor_count != 0, "Waiting with 0 monitors (%i)", this->monitor_count );
+        verifyf( this.monitors != NULL, "Waiting with no monitors (%p)", this.monitors );
+        verifyf( this.monitor_count != 0, "Waiting with 0 monitors (%"PRIiFAST16")", this.monitor_count );
 
         // Create storage for monitor context
-        monitor_ctx( this->monitors, this->monitor_count );
+        monitor_ctx( this.monitors, this.monitor_count );
 
         // Lock all monitors (aggregates the locks them as well)
@@ -364 +475 @@
 
         //save contexts
-        save_recursion( monitors, recursions, count );
+        monitor_save;
 
         //Find the thread to run
-        thread_desc * signallee = pop_head( &this->blocked )->waiting_thread;
-        for(int i = 0; i < count; i++) {
-                set_owner( monitors[i], signallee );
-        }
+        thread_desc * signallee = pop_head( this.blocked )->waiting_thread;
+        set_owner( monitors, count, signallee );
+
+        LIB_DEBUG_PRINT_BUFFER_DECL( "Kernel : signal_block condition %p (s: %p)\n", &this, signallee );
 
         //Everything is ready to go to sleep
@@ -379 +490 @@
 
 
-        //We are back, restore the owners and recursions
-        lock_all( locks, count );
-        restore_recursion( monitors, recursions, count );
-        unlock_all( locks, count );
+        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel :   signal_block returned\n" );
+
+        //We are back, restore the masks and recursions
+        monitor_restore;
 
         return true;
@@ -388 +499 @@
 
 // Access the user_info of the thread waiting at the front of the queue
-uintptr_t front( condition * this ) {
+uintptr_t front( condition & this ) {
         verifyf( !is_empty(this),
                 "Attempt to access user data on an empty condition.\n"
                 "Possible cause is not checking if the condition is empty before reading stored data."
         );
-        return this->blocked.head->user_info;
+        return this.blocked.head->user_info;
 }
 
 //-----------------------------------------------------------------------------
-// Internal scheduling
-int __accept_internal( unsigned short acc_count, __acceptable_t * acceptables ) {
-        thread_desc * thrd = this_thread;
+// External scheduling
+// cases to handle :
+//      - target already there :
+//              block and wake
+//      - dtor already there
+//              put thread on signaller stack
+//      - non-blocking
+//              return else
+//      - timeout
+//              return timeout
+//      - block
+//              setup mask
+//              block
+void __waitfor_internal( const __waitfor_mask_t & mask, int duration ) {
+        // This statment doesn't have a contiguous list of monitors...
+        // Create one!
+        __lock_size_t max = count_max( mask );
+        monitor_desc * mon_storage[max];
+        __builtin_memset( mon_storage, 0, sizeof( mon_storage ) );
+        __lock_size_t actual_count = aggregate( mon_storage, mask );
+
+        LIB_DEBUG_PRINT_BUFFER_DECL( "Kernel : waitfor %d (s: %d, m: %d)\n", actual_count, mask.size, (__lock_size_t)max);
+
+        if(actual_count == 0) return;
+
+        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : waitfor internal proceeding\n");
 
         // Create storage for monitor context
-        monitor_ctx( acceptables->monitors, acceptables->count );
-
-        // Lock all monitors (aggregates the lock them as well)
+        monitor_ctx( mon_storage, actual_count );
+
+        // Lock all monitors (aggregates the locks as well)
         lock_all( monitors, locks, count );
 
+        {
+                // Check if the entry queue
+                thread_desc * next; int index;
+                [next, index] = search_entry_queue( mask, monitors, count );
+
+                if( next ) {
+                        *mask.accepted = index;
+                        if( mask.clauses[index].is_dtor ) {
+                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : dtor already there\n");
+                                verifyf( mask.clauses[index].size == 1        , "ERROR: Accepted dtor has more than 1 mutex parameter." );
+
+                                monitor_desc * mon2dtor = mask.clauses[index].list[0];
+                                verifyf( mon2dtor->dtor_node, "ERROR: Accepted monitor has no dtor_node." );
+
+                                __condition_criterion_t * dtor_crit = mon2dtor->dtor_node->criteria;
+                                push( mon2dtor->signal_stack, dtor_crit );
+
+                                unlock_all( locks, count );
+                        }
+                        else {
+                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : thread present, baton-passing\n");
+
+                                // Create the node specific to this wait operation
+                                wait_ctx_primed( this_thread, 0 );
+
+                                // Save monitor states
+                                monitor_save;
+
+                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel :  baton of %d monitors : ", count );
+                                #ifdef __CFA_DEBUG_PRINT__
+                                        for( int i = 0; i < count; i++) {
+                                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "%p %p ", monitors[i], monitors[i]->signal_stack.top );
+                                        }
+                                #endif
+                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "\n");
+
+                                // Set the owners to be the next thread
+                                set_owner( monitors, count, next );
+
+                                // Everything is ready to go to sleep
+                                BlockInternal( locks, count, &next, 1 );
+
+                                // We are back, restore the owners and recursions
+                                monitor_restore;
+
+                                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : thread present, returned\n");
+                        }
+
+                        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : accepted %d\n", *mask.accepted);
+
+                        return;
+                }
+        }
+
+
+        if( duration == 0 ) {
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : non-blocking, exiting\n");
+
+                unlock_all( locks, count );
+
+                LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : accepted %d\n", *mask.accepted);
+                return;
+        }
+
+
+        verifyf( duration < 0, "Timeout on waitfor statments not supported yet.");
+
+        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : blocking waitfor\n");
+
         // Create the node specific to this wait operation
-        wait_ctx_primed( thrd, 0 );
-
-        // Check if the entry queue
-        thread_desc * next = search_entry_queue( acceptables, acc_count, monitors, count );
-
-        LIB_DEBUG_PRINT_SAFE("Owner(s) :");
-        for(int i = 0; i < count; i++) {
-                LIB_DEBUG_PRINT_SAFE(" %p", monitors[i]->owner );
-        }
-        LIB_DEBUG_PRINT_SAFE("\n");
-
-        LIB_DEBUG_PRINT_SAFE("Passing mon to %p\n", next);
-
-        if( !next ) {
-                // Update acceptables on the current monitors
-                for(int i = 0; i < count; i++) {
-                        monitors[i]->acceptables = acceptables;
-                        monitors[i]->acceptable_count = acc_count;
-                }
-        }
-        else {
-                for(int i = 0; i < count; i++) {
-                        set_owner( monitors[i], next );
-                }
-        }
-
-
-        save_recursion( monitors, recursions, count );
-
-
-        // Everything is ready to go to sleep
-        BlockInternal( locks, count, &next, next ? 1 : 0 );
-
-
-        //WE WOKE UP
-
-
-        //We are back, restore the owners and recursions
-        lock_all( locks, count );
-        restore_recursion( monitors, recursions, count );
-        int acc_idx = monitors[0]->accepted_index;
-        unlock_all( locks, count );
-
-        return acc_idx;
+        wait_ctx_primed( this_thread, 0 );
+
+        monitor_save;
+        set_mask( monitors, count, mask );
+
+        for( __lock_size_t i = 0; i < count; i++) {
+                verify( monitors[i]->owner == this_thread );
+        }
+
+        //Everything is ready to go to sleep
+        BlockInternal( locks, count );
+
+
+        // WE WOKE UP
+
+
+        //We are back, restore the masks and recursions
+        monitor_restore;
+
+        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : exiting\n");
+
+        LIB_DEBUG_PRINT_BUFFER_LOCAL( "Kernel : accepted %d\n", *mask.accepted);
 }
 
@@ -458 +639 @@
 
 static inline void set_owner( monitor_desc * this, thread_desc * owner ) {
+        // LIB_DEBUG_PRINT_SAFE("Kernal :   Setting owner of %p to %p ( was %p)\n", this, owner, this->owner );
+
         //Pass the monitor appropriately
         this->owner = owner;
@@ -465 +648 @@
 }
 
+static inline void set_owner( monitor_desc * monitors [], __lock_size_t count, thread_desc * owner ) {
+        monitors[0]->owner     = owner;
+        monitors[0]->recursion = 1;
+        for( __lock_size_t i = 1; i < count; i++ ) {
+                monitors[i]->owner     = owner;
+                monitors[i]->recursion = 0;
+        }
+}
+
+static inline void set_mask( monitor_desc * storage [], __lock_size_t count, const __waitfor_mask_t & mask ) {
+        for( __lock_size_t i = 0; i < count; i++) {
+                storage[i]->mask = mask;
+        }
+}
+
+static inline void reset_mask( monitor_desc * this ) {
+        this->mask.accepted = NULL;
+        this->mask.clauses = NULL;
+        this->mask.size = 0;
+}
+
 static inline thread_desc * next_thread( monitor_desc * this ) {
         //Check the signaller stack
-        __condition_criterion_t * urgent = pop( &this->signal_stack );
+        LIB_DEBUG_PRINT_SAFE("Kernel :  mon %p AS-stack top %p\n", this, this->signal_stack.top);
+        __condition_criterion_t * urgent = pop( this->signal_stack );
         if( urgent ) {
                 //The signaller stack is not empty,
@@ -479 +684 @@
         // No signaller thread
         // Get the next thread in the entry_queue
-        thread_desc * new_owner = pop_head( &this->entry_queue );
+        thread_desc * new_owner = pop_head( this->entry_queue );
         set_owner( this, new_owner );
 
@@ -485 +690 @@
 }
 
-static inline int is_accepted( thread_desc * owner, monitor_desc * this, monitor_desc ** group, int group_cnt, void (*func)() ) {
-        __acceptable_t* accs = this->acceptables; // Optim
-        int acc_cnt = this->acceptable_count;
+static inline bool is_accepted( monitor_desc * this, const __monitor_group_t & group ) {
+        __acceptable_t * it = this->mask.clauses; // Optim
+        __lock_size_t count = this->mask.size;
 
         // Check if there are any acceptable functions
-        if( !accs ) return -1;
+        if( !it ) return false;
 
         // If this isn't the first monitor to test this, there is no reason to repeat the test.
-        if( this != group[0] ) return group[0]->accepted_index;
+        if( this != group[0] ) return group[0]->mask.accepted >= 0;
 
         // For all acceptable functions check if this is the current function.
-        OUT_LOOP:
-        for( int i = 0; i < acc_cnt; i++ ) {
-                __acceptable_t * acc = &accs[i];
-
-                // if function matches, check the monitors
-                if( acc->func == func ) {
-
-                        // If the group count is different then it can't be a match
-                        if( acc->count != group_cnt ) return -1;
-
-                        // Check that all the monitors match
-                        for( int j = 0; j < group_cnt; j++ ) {
-                                // If not a match, check next function
-                                if( acc->monitors[j] != group[j] ) continue OUT_LOOP;
-                        }
-
-                        // It's a complete match, accept the call
-                        return i;
+        for( __lock_size_t i = 0; i < count; i++, it++ ) {
+                if( *it == group ) {
+                        *this->mask.accepted = i;
+                        return true;
                 }
         }
 
         // No function matched
-        return -1;
-}
-
-static inline void init( int count, monitor_desc ** monitors, __condition_node_t * waiter, __condition_criterion_t * criteria ) {
-        for(int i = 0; i < count; i++) {
+        return false;
+}
+
+static inline void init( __lock_size_t count, monitor_desc * monitors [], __condition_node_t & waiter, __condition_criterion_t criteria [] ) {
+        for( __lock_size_t i = 0; i < count; i++) {
                 (criteria[i]){ monitors[i], waiter };
         }
 
-        waiter->criteria = criteria;
-}
-
-static inline void init_push( int count, monitor_desc ** monitors, __condition_node_t * waiter, __condition_criterion_t * criteria ) {
-        for(int i = 0; i < count; i++) {
+        waiter.criteria = criteria;
+}
+
+static inline void init_push( __lock_size_t count, monitor_desc * monitors [], __condition_node_t & waiter, __condition_criterion_t criteria [] ) {
+        for( __lock_size_t i = 0; i < count; i++) {
                 (criteria[i]){ monitors[i], waiter };
-                push( &criteria[i].target->signal_stack, &criteria[i] );
-        }
-
-        waiter->criteria = criteria;
-}
-
-static inline void lock_all( spinlock ** locks, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+                LIB_DEBUG_PRINT_SAFE( "Kernel :  target %p = %p\n", criteria[i].target, &criteria[i] );
+                push( criteria[i].target->signal_stack, &criteria[i] );
+        }
+
+        waiter.criteria = criteria;
+}
+
+static inline void lock_all( spinlock * locks [], __lock_size_t count ) {
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 lock_yield( locks[i] DEBUG_CTX2 );
         }
 }
 
-static inline void lock_all( monitor_desc ** source, spinlock ** /*out*/ locks, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+static inline void lock_all( monitor_desc * source [], spinlock * /*out*/ locks [], __lock_size_t count ) {
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 spinlock * l = &source[i]->lock;
                 lock_yield( l DEBUG_CTX2 );
@@ -552 +744 @@
 }
 
-static inline void unlock_all( spinlock ** locks, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+static inline void unlock_all( spinlock * locks [], __lock_size_t count ) {
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 unlock( locks[i] );
         }
 }
 
-static inline void unlock_all( monitor_desc ** locks, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+static inline void unlock_all( monitor_desc * locks [], __lock_size_t count ) {
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 unlock( &locks[i]->lock );
         }
 }
 
-
-static inline void save_recursion   ( monitor_desc ** ctx, unsigned int * /*out*/ recursions, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+static inline void save(
+        monitor_desc * ctx [],
+        __lock_size_t count,
+        __attribute((unused)) spinlock * locks [],
+        unsigned int /*out*/ recursions [],
+        __waitfor_mask_t /*out*/ masks []
+) {
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 recursions[i] = ctx[i]->recursion;
-        }
-}
-
-static inline void restore_recursion( monitor_desc ** ctx, unsigned int * /*in */ recursions, unsigned short count ) {
-        for( int i = 0; i < count; i++ ) {
+                masks[i]      = ctx[i]->mask;
+        }
+}
+
+static inline void restore(
+        monitor_desc * ctx [],
+        __lock_size_t count,
+        spinlock * locks [],
+        unsigned int /*out*/ recursions [],
+        __waitfor_mask_t /*out*/ masks []
+) {
+        lock_all( locks, count );
+        for( __lock_size_t i = 0; i < count; i++ ) {
                 ctx[i]->recursion = recursions[i];
-        }
+                ctx[i]->mask      = masks[i];
+        }
+        unlock_all( locks, count );
 }
 
@@ -599 +806 @@
         }
 
-        // LIB_DEBUG_PRINT_SAFE( "Runing %i\n", ready2run );
+        LIB_DEBUG_PRINT_SAFE( "Kernel :  Runing %i (%p)\n", ready2run, ready2run ? node->waiting_thread : NULL );
         return ready2run ? node->waiting_thread : NULL;
 }
 
-static inline void brand_condition( condition * this ) {
+static inline void brand_condition( condition & this ) {
         thread_desc * thrd = this_thread;
-        if( !this->monitors ) {
+        if( !this.monitors ) {
                 // LIB_DEBUG_PRINT_SAFE("Branding\n");
-                assertf( thrd->current_monitors != NULL, "No current monitor to brand condition %p", thrd->current_monitors );
-                this->monitor_count = thrd->current_monitor_count;
-
-                this->monitors = malloc( this->monitor_count * sizeof( *this->monitors ) );
-                for( int i = 0; i < this->monitor_count; i++ ) {
-                        this->monitors[i] = thrd->current_monitors[i];
-                }
-        }
-}
-
-static inline unsigned short insert_unique( thread_desc ** thrds, unsigned short end, thread_desc * val ) {
-        if( !val ) return end;
-
-        for(int i = 0; i <= end; i++) {
-                if( thrds[i] == val ) return end;
-        }
-
-        thrds[end] = val;
-        return end + 1;
-}
-
-
-static inline bool match( __acceptable_t * acc, thread_desc * thrd ) {
-        verify( thrd );
-        verify( acc );
-        if( acc->func != thrd->current_monitor_func ) return false;
-
-        return true;
-}
-
-static inline thread_desc * search_entry_queue( __acceptable_t * acceptables, int acc_count, monitor_desc ** monitors, int count ) {
-
-        __thread_queue_t * entry_queue = &monitors[0]->entry_queue;
+                assertf( thrd->monitors.list != NULL, "No current monitor to brand condition %p", thrd->monitors.list );
+                this.monitor_count = thrd->monitors.size;
+
+                this.monitors = malloc( this.monitor_count * sizeof( *this.monitors ) );
+                for( int i = 0; i < this.monitor_count; i++ ) {
+                        this.monitors[i] = thrd->monitors.list[i];
+                }
+        }
+}
+
+static inline [thread_desc *, int] search_entry_queue( const __waitfor_mask_t & mask, monitor_desc * monitors [], __lock_size_t count ) {
+
+        __thread_queue_t & entry_queue = monitors[0]->entry_queue;
 
         // For each thread in the entry-queue
-        for(    thread_desc ** thrd_it = &entry_queue->head;
+        for(    thread_desc ** thrd_it = &entry_queue.head;
                 *thrd_it;
-                thrd_it = &(*thrd_it)->next)
-        {
+                thrd_it = &(*thrd_it)->next
+        ) {
                 // For each acceptable check if it matches
-                __acceptable_t * acc_end = acceptables + acc_count;
-                for( __acceptable_t * acc_it = acceptables; acc_it != acc_end; acc_it++ ) {
+                int i = 0;
+                __acceptable_t * end = mask.clauses + mask.size;
+                for( __acceptable_t * it = mask.clauses; it != end; it++, i++ ) {
                         // Check if we have a match
-                        if( match( acc_it, *thrd_it ) ) {
+                        if( *it == (*thrd_it)->monitors ) {
 
                                 // If we have a match return it
                                 // after removeing it from the entry queue
-                                return remove( entry_queue, thrd_it );
+                                return [remove( entry_queue, thrd_it ), i];
                         }
                 }
         }
 
-        return NULL;
-}
+        return [0, -1];
+}
+
+forall(dtype T | sized( T ))
+static inline __lock_size_t insert_unique( T * array [], __lock_size_t & size, T * val ) {
+        if( !val ) return size;
+
+        for( __lock_size_t i = 0; i <= size; i++) {
+                if( array[i] == val ) return size;
+        }
+
+        array[size] = val;
+        size = size + 1;
+        return size;
+}
+
+static inline __lock_size_t count_max( const __waitfor_mask_t & mask ) {
+        __lock_size_t max = 0;
+        for( __lock_size_t i = 0; i < mask.size; i++ ) {
+                max += mask.clauses[i].size;
+        }
+        return max;
+}
+
+static inline __lock_size_t aggregate( monitor_desc * storage [], const __waitfor_mask_t & mask ) {
+        __lock_size_t size = 0;
+        for( __lock_size_t i = 0; i < mask.size; i++ ) {
+                __libcfa_small_sort( mask.clauses[i].list, mask.clauses[i].size );
+                for( __lock_size_t j = 0; j < mask.clauses[i].size; j++) {
+                        insert_unique( storage, size, mask.clauses[i].list[j] );
+                }
+        }
+        // TODO insertion sort instead of this
+        __libcfa_small_sort( storage, size );
+        return size;
+}
+
 void ?{}( __condition_blocked_queue_t & this ) {
         this.head = NULL;
@@ -666 +889 @@
 }
 
-void append( __condition_blocked_queue_t * this, __condition_node_t * c ) {
-        verify(this->tail != NULL);
-        *this->tail = c;
-        this->tail = &c->next;
-}
-
-__condition_node_t * pop_head( __condition_blocked_queue_t * this ) {
-        __condition_node_t * head = this->head;
+void append( __condition_blocked_queue_t & this, __condition_node_t * c ) {
+        verify(this.tail != NULL);
+        *this.tail = c;
+        this.tail = &c->next;
+}
+
+__condition_node_t * pop_head( __condition_blocked_queue_t & this ) {
+        __condition_node_t * head = this.head;
         if( head ) {
-                this->head = head->next;
+                this.head = head->next;
                 if( !head->next ) {
-                        this->tail = &this->head;
+                        this.tail = &this.head;
                 }
                 head->next = NULL;