Changes in / [16ba4a6f:18f0b70]

Files:

• benchmark/readyQ/cycle.cfa (modified) (2 diffs)
• benchmark/readyQ/cycle.go (modified) (1 diff)
• benchmark/readyQ/rq_bench.hfa (modified) (1 diff)
• benchmark/readyQ/rq_bench.hpp (modified) (1 diff)
• benchmark/rmit.py (modified) (2 diffs)
• libcfa/src/concurrency/coroutine.cfa (modified) (2 diffs)
• libcfa/src/concurrency/coroutine.hfa (modified) (5 diffs)
• libcfa/src/concurrency/exception.cfa (modified) (1 diff)
• libcfa/src/concurrency/io.cfa (modified) (4 diffs)
• libcfa/src/concurrency/io/setup.cfa (modified) (2 diffs)
• libcfa/src/concurrency/iocall.cfa (deleted)
• libcfa/src/concurrency/kernel.cfa (modified) (29 diffs)
• libcfa/src/concurrency/kernel.hfa (modified) (1 diff)
• libcfa/src/concurrency/kernel/fwd.hfa (modified) (5 diffs)
• libcfa/src/concurrency/kernel/startup.cfa (modified) (14 diffs)
• libcfa/src/concurrency/kernel_private.hfa (modified) (4 diffs)
• libcfa/src/concurrency/locks.cfa (modified) (8 diffs)
• libcfa/src/concurrency/monitor.cfa (modified) (27 diffs)
• libcfa/src/concurrency/mutex.cfa (modified) (5 diffs)
• libcfa/src/concurrency/preemption.cfa (modified) (16 diffs)
• libcfa/src/concurrency/ready_queue.cfa (modified) (5 diffs)
• libcfa/src/stdlib.hfa (modified) (1 diff)
• tests/alloc.cfa (modified) (3 diffs)
• tests/malloc.cfa (modified) (1 diff)
• tools/stat.py (modified) (2 diffs)

benchmark/readyQ/cycle.cfa

@@ -84 +84 @@
                 }
 
-                printf("Duration (ms)        : %'ld\n", (end - start)`ms);
+                printf("Duration (ms)        : %'ld\n", (end - start)`dms);
                 printf("Number of processors : %'d\n", nprocs);
                 printf("Number of threads    : %'d\n", tthreads);
@@ -90 +90 @@
                 printf("Total Operations(ops): %'15llu\n", global_counter);
                 printf("Total blocks         : %'15llu\n", global_blocks);
-                printf("Ops per second       : %'18.2lf\n", ((double)global_counter) / (end - start)`s);
-                printf("ns per ops           : %'18.2lf\n", ((double)(end - start)`ns) / global_counter);
+                printf("Ops per second       : %'18.2lf\n", ((double)global_counter) / (end - start)`ds);
+                printf("ns per ops           : %'18.2lf\n", (end - start)`dns / global_counter);
                 printf("Ops per threads      : %'15llu\n", global_counter / tthreads);
                 printf("Ops per procs        : %'15llu\n", global_counter / nprocs);
-                printf("Ops/sec/procs        : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`s);
-                printf("ns per ops/procs     : %'18.2lf\n", ((double)(end - start)`ns) / (global_counter / nprocs));
+                printf("Ops/sec/procs        : %'18.2lf\n", (((double)global_counter) / nprocs) / (end - start)`ds);
+                printf("ns per ops/procs     : %'18.2lf\n", (end - start)`dns / (global_counter / nprocs));
                 fflush(stdout);
         }

benchmark/readyQ/cycle.go

@@ -72 +72 @@
         p.Printf("Cycle size (# thrds) : %d\n", ring_size);
         p.Printf("Total Operations(ops): %15d\n", global_counter)
-        p.Printf("Yields per second    : %18.2f\n", float64(global_counter) / delta.Seconds())
+        p.Printf("Ops per second       : %18.2f\n", float64(global_counter) / delta.Seconds())
         p.Printf("ns per ops           : %18.2f\n", float64(delta.Nanoseconds()) / float64(global_counter))
         p.Printf("Ops per threads      : %15d\n", global_counter / uint64(tthreads))

benchmark/readyQ/rq_bench.hfa

@@ -88 +88 @@
 }
 
-struct bench_sem {
+struct __attribute__((aligned(128))) bench_sem {
         struct $thread * volatile ptr;
 };

benchmark/readyQ/rq_bench.hpp

@@ -75 +75 @@
 }
 
-class bench_sem {
+class __attribute__((aligned(128))) bench_sem {
         Fibre * volatile ptr = nullptr;
 public:

benchmark/rmit.py

@@ -173 +173 @@
         # ================================================================================
         # Prepare to run
-        print(actions)
 
         # find expected time
@@ -226 +225 @@
                         d = [r[0], r[1]]
                         for k in headers[2:]:
-                                d.append(r[2][k])
+                                try:
+                                        d.append(r[2][k])
+                                except:
+                                        d.append(0.0)
 
                         data.append(d)

libcfa/src/concurrency/coroutine.cfa

@@ -134 +134 @@
 void ^?{}($coroutine& this) {
         if(this.state != Halted && this.state != Start && this.state != Primed) {
-                $coroutine * src = TL_GET( this_thread )->curr_cor;
+                $coroutine * src = active_coroutine();
                 $coroutine * dst = &this;
 
@@ -240 +240 @@
 
         struct $coroutine * __cfactx_cor_finish(void) {
-                struct $coroutine * cor = kernelTLS.this_thread->curr_cor;
+                struct $coroutine * cor = active_coroutine();
 
                 if(cor->state == Primed) {

libcfa/src/concurrency/coroutine.hfa

@@ -63 +63 @@
 void prime(T & cor);
 
-static inline struct $coroutine * active_coroutine() { return TL_GET( this_thread )->curr_cor; }
+static inline struct $coroutine * active_coroutine() { return active_thread()->curr_cor; }
 
 //-----------------------------------------------------------------------------
@@ -87 +87 @@
 
         // set new coroutine that task is executing
-        TL_GET( this_thread )->curr_cor = dst;
+        active_thread()->curr_cor = dst;
 
         // context switch to specified coroutine
@@ -112 +112 @@
                 // will also migrate which means this value will
                 // stay in syn with the TLS
-                $coroutine * src = TL_GET( this_thread )->curr_cor;
+                $coroutine * src = active_coroutine();
 
                 assertf( src->last != 0,
@@ -138 +138 @@
         // will also migrate which means this value will
         // stay in syn with the TLS
-        $coroutine * src = TL_GET( this_thread )->curr_cor;
+        $coroutine * src = active_coroutine();
         $coroutine * dst = get_coroutine(cor);
 
         if( unlikely(dst->context.SP == 0p) ) {
-                TL_GET( this_thread )->curr_cor = dst;
+                active_thread()->curr_cor = dst;
                 __stack_prepare(&dst->stack, 65000);
                 __cfactx_start(main, dst, cor, __cfactx_invoke_coroutine);
-                TL_GET( this_thread )->curr_cor = src;
+                active_thread()->curr_cor = src;
         }
 
@@ -175 +175 @@
         // will also migrate which means this value will
         // stay in syn with the TLS
-        $coroutine * src = TL_GET( this_thread )->curr_cor;
+        $coroutine * src = active_coroutine();
 
         // not resuming self ?

libcfa/src/concurrency/exception.cfa

@@ -72 +72 @@
         void * stop_param;
 
-        struct $thread * this_thread = TL_GET( this_thread );
+        struct $thread * this_thread = active_thread();
         if ( &this_thread->self_cor != this_thread->curr_cor ) {
                 struct $coroutine * cor = this_thread->curr_cor;

libcfa/src/concurrency/io.cfa

@@ -76 +76 @@
 
         static inline bool next( __leaderlock_t & this ) {
-                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( ! __preemption_enabled() );
                 struct $thread * nextt;
                 for() {
@@ -168 +168 @@
         // This is NOT thread-safe
         static [int, bool] __drain_io( & struct __io_data ring ) {
-                /* paranoid */ verify( !kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( ! __preemption_enabled() );
 
                 unsigned to_submit = 0;
@@ -404 +404 @@
                                         return;
                                 }
-                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                                /* paranoid */ verify( ! __preemption_enabled() );
                                 __STATS__( true,
                                         io.submit_q.leader += 1;
@@ -442 +442 @@
 
                         #if defined(LEADER_LOCK)
-                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                                /* paranoid */ verify( ! __preemption_enabled() );
                                 next(ring.submit_q.submit_lock);
                         #else

libcfa/src/concurrency/io/setup.cfa

@@ -149 +149 @@
                 id.full_proc = false;
                 id.id = doregister(&id);
-                kernelTLS.this_proc_id = &id;
+                __cfaabi_tls.this_proc_id = &id;
                 __cfaabi_dbg_print_safe( "Kernel : IO poller thread starting\n" );
 
@@ -179 +179 @@
                                 __cfadbg_print_safe(io_core, "Kernel I/O : Unparking io poller %p\n", io_ctx);
                                 #if !defined( __CFA_NO_STATISTICS__ )
-                                        kernelTLS.this_stats = io_ctx->self.curr_cluster->stats;
+                                        __cfaabi_tls.this_stats = io_ctx->self.curr_cluster->stats;
                                 #endif
                                 post( io_ctx->sem );

libcfa/src/concurrency/kernel.cfa

@@ -122 +122 @@
         // Because of a bug, we couldn't initialized the seed on construction
         // Do it here
-        kernelTLS.rand_seed ^= rdtscl();
-        kernelTLS.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
+        __cfaabi_tls.rand_seed ^= rdtscl();
+        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
         __tls_rand_advance_bck();
 
@@ -217 +217 @@
                 // and it make sense for it to be set in all other cases except here
                 // fake it
-                kernelTLS.this_thread = mainThread;
+                __cfaabi_tls.this_thread = mainThread;
         }
 
@@ -230 +230 @@
 // from the processor coroutine to the target thread
 static void __run_thread(processor * this, $thread * thrd_dst) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verifyf( thrd_dst->state == Ready || thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
         /* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
@@ -247 +247 @@
 
                 // Update global state
-                kernelTLS.this_thread = thrd_dst;
-
-                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
+                kernelTLS().this_thread = thrd_dst;
+
+                /* paranoid */ verify( ! __preemption_enabled() );
+                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
                 /* paranoid */ verify( thrd_dst->context.SP );
                 /* paranoid */ verify( thrd_dst->state != Halted );
@@ -267 +267 @@
                 /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
                 /* paranoid */ verify( thrd_dst->context.SP );
-                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
-                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
+                /* paranoid */ verify( ! __preemption_enabled() );
 
                 // Reset global state
-                kernelTLS.this_thread = 0p;
+                kernelTLS().this_thread = 0p;
 
                 // We just finished running a thread, there are a few things that could have happened.
@@ -315 +315 @@
         proc_cor->state = Active;
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
 // KERNEL_ONLY
 void returnToKernel() {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-        $coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
-        $thread * thrd_src = kernelTLS.this_thread;
+        /* paranoid */ verify( ! __preemption_enabled() );
+        $coroutine * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
+        $thread * thrd_src = kernelTLS().this_thread;
 
         #if !defined(__CFA_NO_STATISTICS__)
-                struct processor * last_proc = kernelTLS.this_processor;
+                struct processor * last_proc = kernelTLS().this_processor;
         #endif
 
@@ -345 +345 @@
 
         #if !defined(__CFA_NO_STATISTICS__)
-                if(last_proc != kernelTLS.this_processor) {
+                if(last_proc != kernelTLS().this_processor) {
                         __tls_stats()->ready.threads.migration++;
                 }
         #endif
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too small.\n", thrd_src );
         /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too large.\n", thrd_src );
@@ -359 +359 @@
 // KERNEL ONLY
 void __schedule_thread( $thread * thrd ) {
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verify( thrd );
         /* paranoid */ verify( thrd->state != Halted );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-        /* paranoid */ verify( kernelTLS.this_proc_id );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
         /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
         /* paranoid */  if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
@@ -380 +380 @@
         ready_schedule_unlock();
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
 // KERNEL ONLY
 static inline $thread * __next_thread(cluster * this) with( *this ) {
-        /* paranoid */ verify( kernelTLS.this_proc_id );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
 
         ready_schedule_lock();
@@ -392 +392 @@
         ready_schedule_unlock();
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-        /* paranoid */ verify( kernelTLS.this_proc_id );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
+        /* paranoid */ verify( ! __preemption_enabled() );
         return thrd;
 }
@@ -399 +399 @@
 // KERNEL ONLY
 static inline $thread * __next_thread_slow(cluster * this) with( *this ) {
-        /* paranoid */ verify( kernelTLS.this_proc_id );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
 
         ready_schedule_lock();
@@ -406 +406 @@
         ready_schedule_unlock();
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-        /* paranoid */ verify( kernelTLS.this_proc_id );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
+        /* paranoid */ verify( ! __preemption_enabled() );
         return thrd;
 }
@@ -414 +414 @@
         if( !thrd ) return;
 
-        /* paranoid */ verify( kernelTLS.this_proc_id );
-        bool full = kernelTLS.this_proc_id->full_proc;
-        if(full) disable_interrupts();
-
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
         int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
         switch(old_ticket) {
@@ -428 +423 @@
                         /* paranoid */ verify( thrd->state == Blocked );
 
-                        // Wake lost the race,
-                        __schedule_thread( thrd );
+                        {
+                                /* paranoid */ verify( publicTLS_get(this_proc_id) );
+                                bool full = publicTLS_get(this_proc_id)->full_proc;
+                                if(full) disable_interrupts();
+
+                                /* paranoid */ verify( ! __preemption_enabled() );
+
+                                // Wake lost the race,
+                                __schedule_thread( thrd );
+
+                                /* paranoid */ verify( ! __preemption_enabled() );
+
+                                if(full) enable_interrupts( __cfaabi_dbg_ctx );
+                                /* paranoid */ verify( publicTLS_get(this_proc_id) );
+                        }
+
                         break;
                 default:
@@ -435 +444 @@
                         abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
         }
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-
-        if(full) enable_interrupts( __cfaabi_dbg_ctx );
-        /* paranoid */ verify( kernelTLS.this_proc_id );
 }
 
 void park( void ) {
-        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( __preemption_enabled() );
         disable_interrupts();
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
 
         returnToKernel();
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         enable_interrupts( __cfaabi_dbg_ctx );
-        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( __preemption_enabled() );
 
 }
@@ -460 +465 @@
         // Should never return
         void __cfactx_thrd_leave() {
-                $thread * thrd = TL_GET( this_thread );
+                $thread * thrd = active_thread();
                 $monitor * this = &thrd->self_mon;
 
@@ -473 +478 @@
 
                 // Leave the thread
-                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( ! __preemption_enabled() );
                 returnToKernel();
                 abort();
@@ -483 +488 @@
 // KERNEL ONLY
 bool force_yield( __Preemption_Reason reason ) {
-        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( __preemption_enabled() );
         disable_interrupts();
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-
-        $thread * thrd = kernelTLS.this_thread;
+        /* paranoid */ verify( ! __preemption_enabled() );
+
+        $thread * thrd = kernelTLS().this_thread;
         /* paranoid */ verify(thrd->state == Active);
 
@@ -501 +506 @@
         }
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         enable_interrupts_noPoll();
-        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( __preemption_enabled() );
 
         return preempted;
@@ -513 +518 @@
 // Wake a thread from the front if there are any
 static void __wake_one(cluster * this) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verify( ready_schedule_islocked() );
 
@@ -533 +538 @@
 
         /* paranoid */ verify( ready_schedule_islocked() );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 
         return;
@@ -543 +548 @@
 
         disable_interrupts();
-                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+                /* paranoid */ verify( ! __preemption_enabled() );
                 post( this->idle );
         enable_interrupts( __cfaabi_dbg_ctx );
@@ -549 +554 @@
 
 static void push  (__cluster_idles & this, processor & proc) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         lock( this );
                 this.idle++;
@@ -556 +561 @@
                 insert_first(this.list, proc);
         unlock( this );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
 static void remove(__cluster_idles & this, processor & proc) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         lock( this );
                 this.idle--;
@@ -567 +572 @@
                 remove(proc);
         unlock( this );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
@@ -611 +616 @@
         }
 
-        return kernelTLS.this_thread;
+        return __cfaabi_tls.this_thread;
 }
 
@@ -636 +641 @@
 
 int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
-        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
+        return get_coroutine(kernelTLS().this_thread) == get_coroutine(mainThread) ? 4 : 2;
 }
 
@@ -668 +673 @@
         if ( count < 0 ) {
                 // queue current task
-                append( waiting, kernelTLS.this_thread );
+                append( waiting, active_thread() );
 
                 // atomically release spin lock and block
@@ -718 +723 @@
                 void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
                         this.prev_name = prev_name;
-                        this.prev_thrd = kernelTLS.this_thread;
+                        this.prev_thrd = kernelTLS().this_thread;
                 }
         }

libcfa/src/concurrency/kernel.hfa

@@ -275 +275 @@
 static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; }
 
-static inline struct processor * active_processor() { return TL_GET( this_processor ); } // UNSAFE
-static inline struct cluster   * active_cluster  () { return TL_GET( this_processor )->cltr; }
+static inline struct processor * active_processor() { return publicTLS_get( this_processor ); } // UNSAFE
+static inline struct cluster   * active_cluster  () { return publicTLS_get( this_processor )->cltr; }
 
 #if !defined(__CFA_NO_STATISTICS__)
+        void print_stats_now( cluster & this, int flags );
+
         static inline void print_stats_at_exit( cluster & this, int flags ) {
                 this.print_stats |= flags;

libcfa/src/concurrency/kernel/fwd.hfa

@@ -55 +55 @@
                                 uint64_t bck_seed;
                         } ready_rng;
-                } kernelTLS __attribute__ ((tls_model ( "initial-exec" )));
+                } __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" )));
 
+                extern bool __preemption_enabled();
 
+                static inline KernelThreadData & kernelTLS( void ) {
+                        /* paranoid */ verify( ! __preemption_enabled() );
+                        return __cfaabi_tls;
+                }
+
+                extern uintptr_t __cfatls_get( unsigned long int member );
+                // #define publicTLS_get( member ) ((typeof(__cfaabi_tls.member))__cfatls_get( __builtin_offsetof(KernelThreadData, member) ))
+                #define publicTLS_get( member ) (__cfaabi_tls.member)
+                // extern forall(otype T) T __cfatls_get( T * member, T value );
+                // #define publicTLS_set( member, value ) __cfatls_set( (typeof(member)*)__builtin_offsetof(KernelThreadData, member), value );
 
                 static inline uint64_t __tls_rand() {
                         #if defined(__SIZEOF_INT128__)
-                                return __lehmer64( kernelTLS.rand_seed );
+                                return __lehmer64( kernelTLS().rand_seed );
                         #else
-                                return __xorshift64( kernelTLS.rand_seed );
+                                return __xorshift64( kernelTLS().rand_seed );
                         #endif
                 }
@@ -75 +86 @@
                 static inline unsigned __tls_rand_fwd() {
 
-                        kernelTLS.ready_rng.fwd_seed = (A * kernelTLS.ready_rng.fwd_seed + C) & (M - 1);
-                        return kernelTLS.ready_rng.fwd_seed >> D;
+                        kernelTLS().ready_rng.fwd_seed = (A * kernelTLS().ready_rng.fwd_seed + C) & (M - 1);
+                        return kernelTLS().ready_rng.fwd_seed >> D;
                 }
 
                 static inline unsigned __tls_rand_bck() {
-                        unsigned int r = kernelTLS.ready_rng.bck_seed >> D;
-                        kernelTLS.ready_rng.bck_seed = AI * (kernelTLS.ready_rng.bck_seed - C) & (M - 1);
+                        unsigned int r = kernelTLS().ready_rng.bck_seed >> D;
+                        kernelTLS().ready_rng.bck_seed = AI * (kernelTLS().ready_rng.bck_seed - C) & (M - 1);
                         return r;
                 }
@@ -92 +103 @@
 
                 static inline void __tls_rand_advance_bck(void) {
-                        kernelTLS.ready_rng.bck_seed = kernelTLS.ready_rng.fwd_seed;
+                        kernelTLS().ready_rng.bck_seed = kernelTLS().ready_rng.fwd_seed;
                 }
         }
 
-        #if 0 // def __ARM_ARCH
-                // function prototypes are only really used by these macros on ARM
-                void disable_global_interrupts();
-                void enable_global_interrupts();
 
-                #define TL_GET( member ) ( { __typeof__( kernelTLS.member ) target; \
-                        disable_global_interrupts(); \
-                        target = kernelTLS.member; \
-                        enable_global_interrupts(); \
-                        target; } )
-                #define TL_SET( member, value ) disable_global_interrupts(); \
-                        kernelTLS.member = value; \
-                        enable_global_interrupts();
-        #else
-                #define TL_GET( member ) kernelTLS.member
-                #define TL_SET( member, value ) kernelTLS.member = value;
-        #endif
 
         extern void disable_interrupts();
@@ -121 +116 @@
                 extern void park( void );
                 extern void unpark( struct $thread * this );
-                static inline struct $thread * active_thread () { return TL_GET( this_thread ); }
+                static inline struct $thread * active_thread () {
+                        struct $thread * t = publicTLS_get( this_thread );
+                        /* paranoid */ verify( t );
+                        return t;
+                }
 
                 extern bool force_yield( enum __Preemption_Reason );
@@ -140 +139 @@
                 #if !defined(__CFA_NO_STATISTICS__)
                         static inline struct __stats_t * __tls_stats() {
-                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
-                                /* paranoid */ verify( kernelTLS.this_stats );
-                                return kernelTLS.this_stats;
+                                /* paranoid */ verify( ! __preemption_enabled() );
+                                /* paranoid */ verify( kernelTLS().this_stats );
+                                return kernelTLS().this_stats;
                         }
 

libcfa/src/concurrency/kernel/startup.cfa

@@ -118 +118 @@
 //-----------------------------------------------------------------------------
 // Global state
-thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) @= {
+thread_local struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
         NULL,                                                                                           // cannot use 0p
         NULL,
@@ -156 +156 @@
 // Kernel boot procedures
 static void __kernel_startup(void) {
-        verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
 
@@ -212 +212 @@
 
         //initialize the global state variables
-        kernelTLS.this_processor = mainProcessor;
-        kernelTLS.this_proc_id   = (__processor_id_t*)mainProcessor;
-        kernelTLS.this_thread    = mainThread;
+        __cfaabi_tls.this_processor = mainProcessor;
+        __cfaabi_tls.this_proc_id   = (__processor_id_t*)mainProcessor;
+        __cfaabi_tls.this_thread    = mainThread;
 
         #if !defined( __CFA_NO_STATISTICS__ )
-                kernelTLS.this_stats = (__stats_t *)& storage_mainProcStats;
-                __init_stats( kernelTLS.this_stats );
+                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
+                __init_stats( __cfaabi_tls.this_stats );
         #endif
 
@@ -234 +234 @@
         // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
         // mainThread is on the ready queue when this call is made.
-        __kernel_first_resume( kernelTLS.this_processor );
+        __kernel_first_resume( __cfaabi_tls.this_processor );
 
 
@@ -251 +251 @@
         __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
 
-        verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         enable_interrupts( __cfaabi_dbg_ctx );
-        verify( TL_GET( preemption_state.enabled ) );
+        /* paranoid */ verify( __preemption_enabled() );
+
 }
 
@@ -262 +263 @@
         mainCluster->io.ctxs = 0p;
 
-        /* paranoid */ verify( TL_GET( preemption_state.enabled ) );
+        /* paranoid */ verify( __preemption_enabled() );
         disable_interrupts();
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 
         __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
@@ -272 +273 @@
         // which is currently here
         __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
-        __kernel_last_resume( kernelTLS.this_processor );
+        __kernel_last_resume( __cfaabi_tls.this_processor );
         mainThread->self_cor.state = Halted;
 
@@ -321 +322 @@
                 __stats_t local_stats;
                 __init_stats( &local_stats );
-                kernelTLS.this_stats = &local_stats;
+                __cfaabi_tls.this_stats = &local_stats;
         #endif
 
         processor * proc = (processor *) arg;
-        kernelTLS.this_processor = proc;
-        kernelTLS.this_proc_id   = (__processor_id_t*)proc;
-        kernelTLS.this_thread    = 0p;
-        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
+        __cfaabi_tls.this_processor = proc;
+        __cfaabi_tls.this_proc_id   = (__processor_id_t*)proc;
+        __cfaabi_tls.this_thread    = 0p;
+        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
         // SKULLDUGGERY: We want to create a context for the processor coroutine
         // which is needed for the 2-step context switch. However, there is no reason
@@ -340 +341 @@
 
         //Set global state
-        kernelTLS.this_thread = 0p;
+        __cfaabi_tls.this_thread = 0p;
 
         //We now have a proper context from which to schedule threads
@@ -370 +371 @@
         $coroutine * dst = get_coroutine(this->runner);
 
-        verify( ! kernelTLS.preemption_state.enabled );
-
-        kernelTLS.this_thread->curr_cor = dst;
+        /* paranoid */ verify( ! __preemption_enabled() );
+
+        __cfaabi_tls.this_thread->curr_cor = dst;
         __stack_prepare( &dst->stack, 65000 );
         __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
 
-        verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 
         dst->last = &src->self_cor;
@@ -394 +395 @@
         /* paranoid */ verify(src->state == Active);
 
-        verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
@@ -402 +403 @@
         $coroutine * dst = get_coroutine(this->runner);
 
-        verify( ! kernelTLS.preemption_state.enabled );
-        verify( dst->starter == src );
-        verify( dst->context.SP );
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( dst->starter == src );
+        /* paranoid */ verify( dst->context.SP );
 
         // SKULLDUGGERY in debug the processors check that the
@@ -546 +547 @@
 
                 P( terminated );
-                verify( kernelTLS.this_processor != &this);
+                /* paranoid */ verify( active_processor() != &this);
         }
 
@@ -696 +697 @@
 #if defined(__CFA_WITH_VERIFY__)
 static bool verify_fwd_bck_rng(void) {
-        kernelTLS.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
+        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
 
         unsigned values[10];

libcfa/src/concurrency/kernel_private.hfa

@@ -38 +38 @@
 #endif
 ;
+
+extern bool __preemption_enabled();
 
 //release/wake-up the following resources
@@ -181 +183 @@
 //  creating/destroying queues
 static inline void ready_schedule_lock(void) with(*__scheduler_lock) {
-        /*paranoid*/ verify( kernelTLS.this_proc_id );
-
-        unsigned iproc = kernelTLS.this_proc_id->id;
-        /*paranoid*/ verify(data[iproc].handle == kernelTLS.this_proc_id);
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
+
+        unsigned iproc = kernelTLS().this_proc_id->id;
+        /*paranoid*/ verify(data[iproc].handle == kernelTLS().this_proc_id);
         /*paranoid*/ verify(iproc < ready);
 
@@ -207 +210 @@
 
 static inline void ready_schedule_unlock(void) with(*__scheduler_lock) {
-        /*paranoid*/ verify( kernelTLS.this_proc_id );
-
-        unsigned iproc = kernelTLS.this_proc_id->id;
-        /*paranoid*/ verify(data[iproc].handle == kernelTLS.this_proc_id);
+        /* paranoid */ verify( ! __preemption_enabled() );
+        /* paranoid */ verify( kernelTLS().this_proc_id );
+
+        unsigned iproc = kernelTLS().this_proc_id->id;
+        /*paranoid*/ verify(data[iproc].handle == kernelTLS().this_proc_id);
         /*paranoid*/ verify(iproc < ready);
         /*paranoid*/ verify(data[iproc].lock);
@@ -223 +227 @@
 #ifdef __CFA_WITH_VERIFY__
         static inline bool ready_schedule_islocked(void) {
-                /*paranoid*/ verify( kernelTLS.this_proc_id );
-                __processor_id_t * proc = kernelTLS.this_proc_id;
+                /* paranoid */ verify( ! __preemption_enabled() );
+                /*paranoid*/ verify( kernelTLS().this_proc_id );
+                __processor_id_t * proc = kernelTLS().this_proc_id;
                 return __scheduler_lock->data[proc->id].owned;
         }

libcfa/src/concurrency/locks.cfa

@@ -76 +76 @@
 
 void lock( blocking_lock & this ) with( this ) {
+        $thread * thrd = active_thread();
         lock( lock __cfaabi_dbg_ctx2 );
-        if ( owner == kernelTLS.this_thread && !multi_acquisition) {
+        if ( owner == thrd && !multi_acquisition) {
                 fprintf(stderr, "A single acquisition lock holder attempted to reacquire the lock resulting in a deadlock."); // Possibly throw instead
         exit(EXIT_FAILURE);
-        } else if ( owner != 0p && owner != kernelTLS.this_thread ) {
-                append( blocked_threads, kernelTLS.this_thread );
+        } else if ( owner != 0p && owner != thrd ) {
+                append( blocked_threads, thrd );
                 wait_count++;
                 unlock( lock );
                 park( );
-        } else if ( owner == kernelTLS.this_thread && multi_acquisition ) {
+        } else if ( owner == thrd && multi_acquisition ) {
                 recursion_count++;
                 unlock( lock );
         } else {
-                owner = kernelTLS.this_thread;
+                owner = thrd;
                 recursion_count = 1;
                 unlock( lock );
@@ -96 +97 @@
 
 bool try_lock( blocking_lock & this ) with( this ) {
+        $thread * thrd = active_thread();
         bool ret = false;
         lock( lock __cfaabi_dbg_ctx2 );
         if ( owner == 0p ) {
-                owner = kernelTLS.this_thread;
+                owner = thrd;
                 if ( multi_acquisition ) recursion_count = 1;
                 ret = true;
-        } else if ( owner == kernelTLS.this_thread && multi_acquisition ) {
+        } else if ( owner == thrd && multi_acquisition ) {
                 recursion_count++;
                 ret = true;
@@ -113 +115 @@
         lock( lock __cfaabi_dbg_ctx2 );
         if ( owner == 0p ){ // no owner implies lock isn't held
-                fprintf( stderr, "There was an attempt to release a lock that isn't held" ); 
+                fprintf( stderr, "There was an attempt to release a lock that isn't held" );
                 return;
-        } else if ( strict_owner && owner != kernelTLS.this_thread ) {
-                fprintf( stderr, "A thread other than the owner attempted to release an owner lock" ); 
+        } else if ( strict_owner && active_thread() ) {
+                fprintf( stderr, "A thread other than the owner attempted to release an owner lock" );
                 return;
         }
@@ -163 +165 @@
     lock( lock __cfaabi_dbg_ctx2 );
         if ( owner == 0p ){ // no owner implies lock isn't held
-                fprintf( stderr, "A lock that is not held was passed to a synchronization lock" ); 
-        } else if ( strict_owner && owner != kernelTLS.this_thread ) {
-                fprintf( stderr, "A thread other than the owner of a lock passed it to a synchronization lock" ); 
+                fprintf( stderr, "A lock that is not held was passed to a synchronization lock" );
+        } else if ( strict_owner && active_thread() ) {
+                fprintf( stderr, "A thread other than the owner of a lock passed it to a synchronization lock" );
         } else {
                 $thread * thrd = pop_head( blocked_threads );
@@ -246 +248 @@
                                 unlock( cond->lock );
                                 #if !defined( __CFA_NO_STATISTICS__ )
+                                        #warning unprotected access to tls TODO discuss this
                                         kernelTLS.this_stats = copy->t->curr_cluster->stats;
                                 #endif
@@ -338 +341 @@
                         remove_( *i.lock );
                 }
-                
+
                 unlock( lock );
                 park( ); // blocks here
@@ -374 +377 @@
 
         void wait( condition_variable(L) & this ) with(this) {
-                info_thread( L ) i = { kernelTLS.this_thread };
+                info_thread( L ) i = { active_thread() };
                 queue_info_thread( this, i );
         }
 
         void wait( condition_variable(L) & this, uintptr_t info ) with(this) {
-                info_thread( L ) i = { kernelTLS.this_thread, info };
+                info_thread( L ) i = { active_thread(), info };
                 queue_info_thread( this, i );
         }
-        
+
         void wait( condition_variable(L) & this, Duration duration ) with(this) {
-                info_thread( L ) i = { kernelTLS.this_thread };
+                info_thread( L ) i = { active_thread() };
                 queue_info_thread_timeout(this, i, __kernel_get_time() + duration );
         }
 
-        void wait( condition_variable(L) & this, uintptr_t info, Duration duration ) with(this) { 
-                info_thread( L ) i = { kernelTLS.this_thread, info };
+        void wait( condition_variable(L) & this, uintptr_t info, Duration duration ) with(this) {
+                info_thread( L ) i = { active_thread(), info };
                 queue_info_thread_timeout(this, i, __kernel_get_time() + duration );
         }
 
         void wait( condition_variable(L) & this, Time time ) with(this) {
-                info_thread( L ) i = { kernelTLS.this_thread };
+                info_thread( L ) i = { active_thread() };
                 queue_info_thread_timeout(this, i, time);
         }
 
         void wait( condition_variable(L) & this, uintptr_t info, Time time ) with(this) {
-                info_thread( L ) i = { kernelTLS.this_thread, info };
+                info_thread( L ) i = { active_thread(), info };
                 queue_info_thread_timeout(this, i, time);
         }
 
         void wait( condition_variable(L) & this, L & l ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread };
+                info_thread(L) i = { active_thread() };
                 i.lock = &l;
                 queue_info_thread( this, i );
@@ -410 +413 @@
 
         void wait( condition_variable(L) & this, L & l, uintptr_t info ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread, info };
+                info_thread(L) i = { active_thread(), info };
                 i.lock = &l;
                 queue_info_thread( this, i );
         }
-        
+
         void wait( condition_variable(L) & this, L & l, Duration duration ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread };
+                info_thread(L) i = { active_thread() };
                 i.lock = &l;
                 queue_info_thread_timeout(this, i, __kernel_get_time() + duration );
         }
-        
+
         void wait( condition_variable(L) & this, L & l, uintptr_t info, Duration duration ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread, info };
+                info_thread(L) i = { active_thread(), info };
                 i.lock = &l;
                 queue_info_thread_timeout(this, i, __kernel_get_time() + duration );
         }
-        
+
         void wait( condition_variable(L) & this, L & l, Time time ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread };
+                info_thread(L) i = { active_thread() };
                 i.lock = &l;
                 queue_info_thread_timeout(this, i, time );
         }
-        
+
         void wait( condition_variable(L) & this, L & l, uintptr_t info, Time time ) with(this) {
-                info_thread(L) i = { kernelTLS.this_thread, info };
+                info_thread(L) i = { active_thread(), info };
                 i.lock = &l;
                 queue_info_thread_timeout(this, i, time );

libcfa/src/concurrency/monitor.cfa

@@ -82 +82 @@
 // Enter single monitor
 static void __enter( $monitor * this, const __monitor_group_t & group ) {
+        $thread * thrd = active_thread();
+
         // Lock the monitor spinlock
         lock( this->lock __cfaabi_dbg_ctx2 );
-        // Interrupts disable inside critical section
-        $thread * thrd = kernelTLS.this_thread;
 
         __cfaabi_dbg_print_safe( "Kernel : %10p Entering mon %p (%p)\n", thrd, this, this->owner);
@@ -126 +126 @@
                 __cfaabi_dbg_print_safe( "Kernel : %10p Entered  mon %p\n", thrd, this);
 
-                /* paranoid */ verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                /* paranoid */ verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
                 return;
         }
@@ -132 +132 @@
         __cfaabi_dbg_print_safe( "Kernel : %10p Entered  mon %p\n", thrd, this);
 
-        /* paranoid */ verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+        /* paranoid */ verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
         /* paranoid */ verify( this->lock.lock );
 
@@ -141 +141 @@
 
 static void __dtor_enter( $monitor * this, fptr_t func, bool join ) {
+        $thread * thrd = active_thread();
+
         // Lock the monitor spinlock
         lock( this->lock __cfaabi_dbg_ctx2 );
-        // Interrupts disable inside critical section
-        $thread * thrd = kernelTLS.this_thread;
 
         __cfaabi_dbg_print_safe( "Kernel : %10p Entering dtor for mon %p (%p)\n", thrd, this, this->owner);
@@ -155 +155 @@
                 __set_owner( this, thrd );
 
-                verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
 
                 unlock( this->lock );
@@ -174 +174 @@
                 this->owner = thrd;
 
-                verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
 
                 unlock( this->lock );
@@ -200 +200 @@
 
                 // Release the next thread
-                /* paranoid */ verifyf( urgent->owner->waiting_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                /* paranoid */ verifyf( urgent->owner->waiting_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
                 unpark( urgent->owner->waiting_thread );
 
@@ -207 +207 @@
 
                 // Some one was waiting for us, enter
-                /* paranoid */ verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                /* paranoid */ verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
         }
         else {
@@ -224 +224 @@
                 park();
 
-                /* paranoid */ verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                /* paranoid */ verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
                 return;
         }
@@ -237 +237 @@
         lock( this->lock __cfaabi_dbg_ctx2 );
 
-        __cfaabi_dbg_print_safe( "Kernel : %10p Leaving mon %p (%p)\n", kernelTLS.this_thread, this, this->owner);
-
-        /* paranoid */ verifyf( kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+        __cfaabi_dbg_print_safe( "Kernel : %10p Leaving mon %p (%p)\n", active_thread(), this, this->owner);
+
+        /* paranoid */ verifyf( active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
 
         // Leaving a recursion level, decrement the counter
@@ -270 +270 @@
 void __dtor_leave( $monitor * this, bool join ) {
         __cfaabi_dbg_debug_do(
-                if( TL_GET( this_thread ) != this->owner ) {
-                        abort( "Destroyed monitor %p has inconsistent owner, expected %p got %p.\n", this, TL_GET( this_thread ), this->owner);
+                if( active_thread() != this->owner ) {
+                        abort( "Destroyed monitor %p has inconsistent owner, expected %p got %p.\n", this, active_thread(), this->owner);
                 }
                 if( this->recursion != 1  && !join ) {
@@ -287 +287 @@
         /* paranoid */ verify( this->lock.lock );
         /* paranoid */ verifyf( thrd == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", thrd, this->owner, this->recursion, this );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verify( thrd->state == Halted );
         /* paranoid */ verify( this->recursion == 1 );
@@ -303 +303 @@
         // Unpark the next owner if needed
         /* paranoid */ verifyf( !new_owner || new_owner == this->owner, "Expected owner to be %p, got %p (m: %p)", new_owner, this->owner, this );
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         /* paranoid */ verify( thrd->state == Halted );
         unpark( new_owner );
@@ -327 +327 @@
 // Sorts monitors before entering
 void ?{}( monitor_guard_t & this, $monitor * m [], __lock_size_t count, fptr_t func ) {
-        $thread * thrd = TL_GET( this_thread );
+        $thread * thrd = active_thread();
 
         // Store current array
@@ -362 +362 @@
 
         // Restore thread context
-        TL_GET( this_thread )->monitors = this.prev;
+        active_thread()->monitors = this.prev;
 }
 
@@ -369 +369 @@
 void ?{}( monitor_dtor_guard_t & this, $monitor * m [], fptr_t func, bool join ) {
         // optimization
-        $thread * thrd = TL_GET( this_thread );
+        $thread * thrd = active_thread();
 
         // Store current array
@@ -392 +392 @@
 
         // Restore thread context
-        TL_GET( this_thread )->monitors = this.prev;
+        active_thread()->monitors = this.prev;
 }
 
@@ -432 +432 @@
 
         // Create the node specific to this wait operation
-        wait_ctx( TL_GET( this_thread ), user_info );
+        wait_ctx( active_thread(), user_info );
 
         // Append the current wait operation to the ones already queued on the condition
@@ -483 +483 @@
         //Some more checking in debug
         __cfaabi_dbg_debug_do(
-                $thread * this_thrd = TL_GET( this_thread );
+                $thread * this_thrd = active_thread();
                 if ( this.monitor_count != this_thrd->monitors.size ) {
                         abort( "Signal on condition %p made with different number of monitor(s), expected %zi got %zi", &this, this.monitor_count, this_thrd->monitors.size );
@@ -531 +531 @@
 
         // Create the node specific to this wait operation
-        wait_ctx_primed( kernelTLS.this_thread, 0 )
+        wait_ctx_primed( active_thread(), 0 )
 
         //save contexts
@@ -630 +630 @@
 
                                 // Create the node specific to this wait operation
-                                wait_ctx_primed( kernelTLS.this_thread, 0 );
+                                wait_ctx_primed( active_thread(), 0 );
 
                                 // Save monitor states
@@ -682 +682 @@
 
         // Create the node specific to this wait operation
-        wait_ctx_primed( kernelTLS.this_thread, 0 );
+        wait_ctx_primed( active_thread(), 0 );
 
         monitor_save;
@@ -688 +688 @@
 
         for( __lock_size_t i = 0; i < count; i++) {
-                verify( monitors[i]->owner == kernelTLS.this_thread );
+                verify( monitors[i]->owner == active_thread() );
         }
 
@@ -724 +724 @@
 static inline void __set_owner( $monitor * monitors [], __lock_size_t count, $thread * owner ) {
         /* paranoid */ verify ( monitors[0]->lock.lock );
-        /* paranoid */ verifyf( monitors[0]->owner == kernelTLS.this_thread, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, monitors[0]->owner, monitors[0]->recursion, monitors[0] );
+        /* paranoid */ verifyf( monitors[0]->owner == active_thread(), "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), monitors[0]->owner, monitors[0]->recursion, monitors[0] );
         monitors[0]->owner        = owner;
         monitors[0]->recursion    = 1;
         for( __lock_size_t i = 1; i < count; i++ ) {
                 /* paranoid */ verify ( monitors[i]->lock.lock );
-                /* paranoid */ verifyf( monitors[i]->owner == kernelTLS.this_thread, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, monitors[i]->owner, monitors[i]->recursion, monitors[i] );
+                /* paranoid */ verifyf( monitors[i]->owner == active_thread(), "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), monitors[i]->owner, monitors[i]->recursion, monitors[i] );
                 monitors[i]->owner        = owner;
                 monitors[i]->recursion    = 0;
@@ -755 +755 @@
                 //regardless of if we are ready to baton pass,
                 //we need to set the monitor as in use
-                /* paranoid */ verifyf( !this->owner || kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+                /* paranoid */ verifyf( !this->owner || active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
                 __set_owner( this,  urgent->owner->waiting_thread );
 
@@ -764 +764 @@
         // Get the next thread in the entry_queue
         $thread * new_owner = pop_head( this->entry_queue );
-        /* paranoid */ verifyf( !this->owner || kernelTLS.this_thread == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", kernelTLS.this_thread, this->owner, this->recursion, this );
+        /* paranoid */ verifyf( !this->owner || active_thread() == this->owner, "Expected owner to be %p, got %p (r: %i, m: %p)", active_thread(), this->owner, this->recursion, this );
         /* paranoid */ verify( !new_owner || new_owner->link.next == 0p );
         __set_owner( this, new_owner );
@@ -892 +892 @@
 
 static inline void brand_condition( condition & this ) {
-        $thread * thrd = TL_GET( this_thread );
+        $thread * thrd = active_thread();
         if( !this.monitors ) {
                 // __cfaabi_dbg_print_safe( "Branding\n" );

libcfa/src/concurrency/mutex.cfa

@@ -40 +40 @@
         lock( lock __cfaabi_dbg_ctx2 );
         if( is_locked ) {
-                append( blocked_threads, kernelTLS.this_thread );
+                append( blocked_threads, active_thread() );
                 unlock( lock );
                 park();
@@ -86 +86 @@
         lock( lock __cfaabi_dbg_ctx2 );
         if( owner == 0p ) {
-                owner = kernelTLS.this_thread;
+                owner = active_thread();
                 recursion_count = 1;
                 unlock( lock );
         }
-        else if( owner == kernelTLS.this_thread ) {
+        else if( owner == active_thread() ) {
                 recursion_count++;
                 unlock( lock );
         }
         else {
-                append( blocked_threads, kernelTLS.this_thread );
+                append( blocked_threads, active_thread() );
                 unlock( lock );
                 park();
@@ -105 +105 @@
         lock( lock __cfaabi_dbg_ctx2 );
         if( owner == 0p ) {
-                owner = kernelTLS.this_thread;
+                owner = active_thread();
                 recursion_count = 1;
                 ret = true;
         }
-        else if( owner == kernelTLS.this_thread ) {
+        else if( owner == active_thread() ) {
                 recursion_count++;
                 ret = true;
@@ -159 +159 @@
 void wait(condition_variable & this) {
         lock( this.lock __cfaabi_dbg_ctx2 );
-        append( this.blocked_threads, kernelTLS.this_thread );
+        append( this.blocked_threads, active_thread() );
         unlock( this.lock );
         park();
@@ -167 +167 @@
 void wait(condition_variable & this, L & l) {
         lock( this.lock __cfaabi_dbg_ctx2 );
-        append( this.blocked_threads, kernelTLS.this_thread );
+        append( this.blocked_threads, active_thread() );
         unlock(l);
         unlock(this.lock);

libcfa/src/concurrency/preemption.cfa

@@ -10 +10 @@
 // Created On       : Mon Jun 5 14:20:42 2017
 // Last Modified By : Peter A. Buhr
-// Last Modified On : Wed Aug 26 16:46:03 2020
-// Update Count     : 53
+// Last Modified On : Fri Nov  6 07:42:13 2020
+// Update Count     : 54
 //
 
@@ -163 +163 @@
 // Kernel Signal Tools
 //=============================================================================================
-
-__cfaabi_dbg_debug_do( static thread_local void * last_interrupt = 0; )
+// In a user-level threading system, there are handful of thread-local variables where this problem occurs on the ARM.
+//
+// For each kernel thread running user-level threads, there is a flag variable to indicate if interrupts are
+// enabled/disabled for that kernel thread. Therefore, this variable is made thread local.
+//
+// For example, this code fragment sets the state of the "interrupt" variable in thread-local memory.
+//
+// _Thread_local volatile int interrupts;
+// int main() {
+//     interrupts = 0; // disable interrupts }
+//
+// which generates the following code on the ARM
+//
+// (gdb) disassemble main
+// Dump of assembler code for function main:
+//    0x0000000000000610 <+0>:  mrs     x1, tpidr_el0
+//    0x0000000000000614 <+4>:  mov     w0, #0x0                        // #0
+//    0x0000000000000618 <+8>:  add     x1, x1, #0x0, lsl #12
+//    0x000000000000061c <+12>: add     x1, x1, #0x10
+//    0x0000000000000620 <+16>: str     wzr, [x1]
+//    0x0000000000000624 <+20>: ret
+//
+// The mrs moves a pointer from coprocessor register tpidr_el0 into register x1.  Register w0 is set to 0. The two adds
+// increase the TLS pointer with the displacement (offset) 0x10, which is the location in the TSL of variable
+// "interrupts".  Finally, 0 is stored into "interrupts" through the pointer in register x1 that points into the
+// TSL. Now once x1 has the pointer to the location of the TSL for kernel thread N, it can be be preempted at a
+// user-level and the user thread is put on the user-level ready-queue. When the preempted thread gets to the front of
+// the user-level ready-queue it is run on kernel thread M. It now stores 0 into "interrupts" back on kernel thread N,
+// turning off interrupt on the wrong kernel thread.
+//
+// On the x86, the following code is generated for the same code fragment.
+//
+// (gdb) disassemble main
+// Dump of assembler code for function main:
+//    0x0000000000400420 <+0>:  movl   $0x0,%fs:0xfffffffffffffffc
+//    0x000000000040042c <+12>: xor    %eax,%eax
+//    0x000000000040042e <+14>: retq
+//
+// and there is base-displacement addressing used to atomically reset variable "interrupts" off of the TSL pointer in
+// register "fs".
+//
+// Hence, the ARM has base-displacement address for the general purpose registers, BUT not to the coprocessor
+// registers. As a result, generating the address for the write into variable "interrupts" is no longer atomic.
+//
+// Note this problem does NOT occur when just using multiple kernel threads because the preemption ALWAYS restarts the
+// thread on the same kernel thread.
+//
+// The obvious question is why does ARM use a coprocessor register to store the TSL pointer given that coprocessor
+// registers are second-class registers with respect to the instruction set. One possible answer is that they did not
+// want to dedicate one of the general registers to hold the TLS pointer and there was a free coprocessor register
+// available.
+
+//----------
+// special case for preemption since used often
+bool __preemption_enabled() {
+        // create a assembler label before
+        // marked as clobber all to avoid movement
+        asm volatile("__cfaasm_check_before:":::"memory");
+
+        // access tls as normal
+        bool enabled = __cfaabi_tls.preemption_state.enabled;
+
+        // create a assembler label after
+        // marked as clobber all to avoid movement
+        asm volatile("__cfaasm_check_after:":::"memory");
+        return enabled;
+}
+
+//----------
+// Get data from the TLS block
+uintptr_t __cfatls_get( unsigned long int offset ) __attribute__((__noinline__)); //no inline to avoid problems
+uintptr_t __cfatls_get( unsigned long int offset ) {
+        // create a assembler label before
+        // marked as clobber all to avoid movement
+        asm volatile("__cfaasm_get_before:":::"memory");
+
+        // access tls as normal (except for pointer arithmetic)
+        uintptr_t val = *(uintptr_t*)((uintptr_t)&__cfaabi_tls + offset);
+
+        // create a assembler label after
+        // marked as clobber all to avoid movement
+        asm volatile("__cfaasm_get_after:":::"memory");
+        return val;
+}
 
 extern "C" {
         // Disable interrupts by incrementing the counter
         void disable_interrupts() {
-                with( kernelTLS.preemption_state ) {
+                // create a assembler label before
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_disable_before:":::"memory");
+
+                with( __cfaabi_tls.preemption_state ) {
                         #if GCC_VERSION > 50000
                         static_assert(__atomic_always_lock_free(sizeof(enabled), &enabled), "Must be lock-free");
@@ -186 +272 @@
                         verify( new_val < 65_000u );              // If this triggers someone is disabling interrupts without enabling them
                 }
+
+                // create a assembler label after
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_disable_after:":::"memory");
         }
 
@@ -191 +281 @@
         // If counter reaches 0, execute any pending __cfactx_switch
         void enable_interrupts( __cfaabi_dbg_ctx_param ) {
-                processor   * proc = kernelTLS.this_processor; // Cache the processor now since interrupts can start happening after the atomic store
+                // create a assembler label before
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_enable_before:":::"memory");
+
+                processor   * proc = __cfaabi_tls.this_processor; // Cache the processor now since interrupts can start happening after the atomic store
                 /* paranoid */ verify( proc );
 
-                with( kernelTLS.preemption_state ){
+                with( __cfaabi_tls.preemption_state ){
                         unsigned short prev = disable_count;
                         disable_count -= 1;
@@ -221 +315 @@
                 // For debugging purposes : keep track of the last person to enable the interrupts
                 __cfaabi_dbg_debug_do( proc->last_enable = caller; )
+
+                // create a assembler label after
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_enable_after:":::"memory");
         }
 
@@ -226 +324 @@
         // Don't execute any pending __cfactx_switch even if counter reaches 0
         void enable_interrupts_noPoll() {
-                unsigned short prev = kernelTLS.preemption_state.disable_count;
-                kernelTLS.preemption_state.disable_count -= 1;
+                // create a assembler label before
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_nopoll_before:":::"memory");
+
+                unsigned short prev = __cfaabi_tls.preemption_state.disable_count;
+                __cfaabi_tls.preemption_state.disable_count -= 1;
                 verifyf( prev != 0u, "Incremented from %u\n", prev );                     // If this triggers someone is enabled already enabled interrupts
                 if( prev == 1 ) {
                         #if GCC_VERSION > 50000
-                        static_assert(__atomic_always_lock_free(sizeof(kernelTLS.preemption_state.enabled), &kernelTLS.preemption_state.enabled), "Must be lock-free");
+                        static_assert(__atomic_always_lock_free(sizeof(__cfaabi_tls.preemption_state.enabled), &__cfaabi_tls.preemption_state.enabled), "Must be lock-free");
                         #endif
                         // Set enabled flag to true
                         // should be atomic to avoid preemption in the middle of the operation.
                         // use memory order RELAXED since there is no inter-thread on this variable requirements
-                        __atomic_store_n(&kernelTLS.preemption_state.enabled, true, __ATOMIC_RELAXED);
+                        __atomic_store_n(&__cfaabi_tls.preemption_state.enabled, true, __ATOMIC_RELAXED);
 
                         // Signal the compiler that a fence is needed but only for signal handlers
                         __atomic_signal_fence(__ATOMIC_RELEASE);
                 }
+
+                // create a assembler label after
+                // marked as clobber all to avoid movement
+                asm volatile("__cfaasm_nopoll_after:":::"memory");
         }
 }
@@ -275 +381 @@
 static void timeout( $thread * this ) {
         #if !defined( __CFA_NO_STATISTICS__ )
-                kernelTLS.this_stats = this->curr_cluster->stats;
+                kernelTLS().this_stats = this->curr_cluster->stats;
         #endif
         unpark( this );
@@ -286 +392 @@
 static inline bool preemption_ready() {
         // Check if preemption is safe
-        bool ready = kernelTLS.preemption_state.enabled && ! kernelTLS.preemption_state.in_progress;
+        bool ready = __cfaabi_tls.preemption_state.enabled && ! __cfaabi_tls.preemption_state.in_progress;
 
         // Adjust the pending flag accordingly
-        kernelTLS.this_processor->pending_preemption = !ready;
+        __cfaabi_tls.this_processor->pending_preemption = !ready;
         return ready;
 }
@@ -303 +409 @@
 
         // Start with preemption disabled until ready
-        kernelTLS.preemption_state.enabled = false;
-        kernelTLS.preemption_state.disable_count = 1;
+        __cfaabi_tls.preemption_state.enabled = false;
+        __cfaabi_tls.preemption_state.disable_count = 1;
 
         // Initialize the event kernel
@@ -362 +468 @@
 // Kernel Signal Handlers
 //=============================================================================================
+struct asm_region {
+        void * before;
+        void * after;
+};
+
+//-----------------------------------------------------------------------------
+// Some assembly required
+#if defined( __i386 )
+        #define __cfaasm_label( label ) \
+                ({ \
+                        struct asm_region region; \
+                        asm( \
+                                "movl $__cfaasm_" #label "_before, %[vb]\n\t" \
+                                "movl $__cfaasm_" #label "_after , %[va]\n\t" \
+                                 : [vb]"=r"(region.before), [vb]"=r"(region.before) \
+                        ); \
+                        region; \
+                });
+#elif defined( __x86_64 )
+        #ifdef __PIC__
+                #define PLT "@PLT"
+        #else
+                #define PLT ""
+        #endif
+        #define __cfaasm_label( label ) \
+                ({ \
+                        struct asm_region region; \
+                        asm( \
+                                "movq $__cfaasm_" #label "_before" PLT ", %[vb]\n\t" \
+                                "movq $__cfaasm_" #label "_after"  PLT ", %[va]\n\t" \
+                                 : [vb]"=r"(region.before), [va]"=r"(region.after) \
+                        ); \
+                        region; \
+                });
+#elif defined( __aarch64__ )
+        #error __cfaasm_label undefined for arm
+#else
+        #error unknown hardware architecture
+#endif
+
+__cfaabi_dbg_debug_do( static thread_local void * last_interrupt = 0; )
 
 // Context switch signal handler
 // Receives SIGUSR1 signal and causes the current thread to yield
 static void sigHandler_ctxSwitch( __CFA_SIGPARMS__ ) {
-        __cfaabi_dbg_debug_do( last_interrupt = (void *)(cxt->uc_mcontext.CFA_REG_IP); )
+        void * ip = (void *)(cxt->uc_mcontext.CFA_REG_IP);
+        __cfaabi_dbg_debug_do( last_interrupt = ip; )
 
         // SKULLDUGGERY: if a thread creates a processor and the immediately deletes it,
@@ -372 +520 @@
         // before the kernel thread has even started running. When that happens, an interrupt
         // with a null 'this_processor' will be caught, just ignore it.
-        if(! kernelTLS.this_processor ) return;
+        if(! __cfaabi_tls.this_processor ) return;
 
         choose(sfp->si_value.sival_int) {
                 case PREEMPT_NORMAL   : ;// Normal case, nothing to do here
-                case PREEMPT_TERMINATE: verify( __atomic_load_n( &kernelTLS.this_processor->do_terminate, __ATOMIC_SEQ_CST ) );
+                case PREEMPT_TERMINATE: verify( __atomic_load_n( &__cfaabi_tls.this_processor->do_terminate, __ATOMIC_SEQ_CST ) );
                 default:
                         abort( "internal error, signal value is %d", sfp->si_value.sival_int );
@@ -384 +532 @@
         if( !preemption_ready() ) { return; }
 
-        __cfaabi_dbg_print_buffer_decl( " KERNEL: preempting core %p (%p @ %p).\n", kernelTLS.this_processor, kernelTLS.this_thread, (void *)(cxt->uc_mcontext.CFA_REG_IP) );
+        struct asm_region region;
+        region = __cfaasm_label( get     ); if( ip >= region.before && ip <= region.after ) return;
+        region = __cfaasm_label( check   ); if( ip >= region.before && ip <= region.after ) return;
+        region = __cfaasm_label( disable ); if( ip >= region.before && ip <= region.after ) return;
+        region = __cfaasm_label( enable  ); if( ip >= region.before && ip <= region.after ) return;
+        region = __cfaasm_label( nopoll  ); if( ip >= region.before && ip <= region.after ) return;
+
+        __cfaabi_dbg_print_buffer_decl( " KERNEL: preempting core %p (%p @ %p).\n", __cfaabi_tls.this_processor, __cfaabi_tls.this_thread, (void *)(cxt->uc_mcontext.CFA_REG_IP) );
 
         // Sync flag : prevent recursive calls to the signal handler
-        kernelTLS.preemption_state.in_progress = true;
+        __cfaabi_tls.preemption_state.in_progress = true;
 
         // Clear sighandler mask before context switching.
@@ -397 +552 @@
         }
 
-        // TODO: this should go in finish action
         // Clear the in progress flag
-        kernelTLS.preemption_state.in_progress = false;
+        __cfaabi_tls.preemption_state.in_progress = false;
 
         // Preemption can occur here
@@ -416 +570 @@
         id.full_proc = false;
         id.id = doregister(&id);
-        kernelTLS.this_proc_id = &id;
+        __cfaabi_tls.this_proc_id = &id;
 
         // Block sigalrms to control when they arrive
@@ -484 +638 @@
 
 void __cfaabi_check_preemption() {
-        bool ready = kernelTLS.preemption_state.enabled;
+        bool ready = __preemption_enabled();
         if(!ready) { abort("Preemption should be ready"); }
 
@@ -507 +661 @@
 #ifdef __CFA_WITH_VERIFY__
 bool __cfaabi_dbg_in_kernel() {
-        return !kernelTLS.preemption_state.enabled;
+        return !__preemption_enabled();
 }
 #endif

libcfa/src/concurrency/ready_queue.cfa

@@ -150 +150 @@
 //  queues or removing them.
 uint_fast32_t ready_mutate_lock( void ) with(*__scheduler_lock) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 
         // Step 1 : lock global lock
@@ -166 +166 @@
         }
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
         return s;
 }
 
 void ready_mutate_unlock( uint_fast32_t last_s ) with(*__scheduler_lock) {
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 
         // Step 1 : release local locks
@@ -188 +188 @@
         __atomic_store_n(&lock, (bool)false, __ATOMIC_RELEASE);
 
-        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
+        /* paranoid */ verify( ! __preemption_enabled() );
 }
 
@@ -252 +252 @@
                 preferred =
                         //*
-                        kernelTLS.this_processor ? kernelTLS.this_processor->id * 4 : -1;
+                        kernelTLS().this_processor ? kernelTLS().this_processor->id * 4 : -1;
                         /*/
                         thrd->link.preferred * 4;
@@ -331 +331 @@
                 // Don't bother trying locally too much
                 int local_tries = 8;
-                preferred = kernelTLS.this_processor->id * 4;
+                preferred = kernelTLS().this_processor->id * 4;
         #endif
 

libcfa/src/stdlib.hfa

@@ -101 +101 @@
                 return (T *)pvalloc( sizeof(T) );                               // C pvalloc
         } // pvalloc
+
+        void free( T * addr ) {
+                free( (void *) addr );                                                  // C free
+        } // free
+} // distribution
+
+static inline forall( ttype TT | { void free( TT ); } ) {
+        // T* does not take void* and vice-versa
+
+        void free( void * addr, TT rest ) {
+                free( addr );
+                free( rest );
+        } // free
+
+        forall( dtype T | sized(T) )
+        void free( T * addr, TT rest ) {
+                free( addr );
+                free( rest );
+        } // free
 } // distribution
 

tests/alloc.cfa

@@ -208 +208 @@
 
 
+        int const_count, dest_count;
         struct Struct { int x; double y; };
+        void  ?{}( Struct & a ) {                                       // construct
+                a.[ x, y ] = [ -1, -1.0 ];
+        }
+        void  ?{}( Struct & a, int x, double y ) {      // initialize
+                a.[ x, y ] = [ x, y ];
+                const_count++;
+        }
+        void ^?{}( Struct & a ) {  dest_count++; }      // destruct
         Struct st, st1, sta[dim], sta1[dim], * stp, * stp1;
 
@@ -329 +338 @@
         printf( "\n" );
 
+        const_count = dest_count = 0;
         stp = new( 42, 42.5 );
+        assert( const_count == 1 && dest_count == 0 );                                          // assertion for testing
         stp1 = new( 42, 42.5 );
+        assert( const_count == 2 && dest_count == 0 );                                          // assertion for testing
+
         printf( "CFA new initialize\n%d %g %d %g\n", stp->x, stp->y, stp1->x, stp1->y );
         delete( stp, stp1 );
+        assert( const_count == 2 && dest_count == 2 );                                          // assertion for testing
 
         // new, array types
         stp = anew( dim, 42, 42.5 );
+        assert( const_count == 2 + dim && dest_count == 2 );                            // assertion for testing
         printf( "CFA array new initialize\n" );
         for ( i; dim ) { printf( "%d %g, ", stp[i].x, stp[i].y ); }
         printf( "\n" );
+
         stp1 = anew( dim, 42, 42.5 );
+        assert( const_count == 2 + 2 * dim && dest_count == 2 );                        // assertion for testing
         for ( i; dim ) { printf( "%d %g, ", stp1[i].x, stp1[i].y ); }
         printf( "\n" );
         adelete( stp, stp1 );
+        assert( const_count == 2 + 2 * dim && dest_count == 2 + 2 * dim);       // assertion for testing
 
         // extras
@@ -354 +372 @@
         *ip = 0xdeadbeef;
         printf( "CFA deep malloc %#x\n", *ip );
-        free( ip );
+
+        dp = alloc(5.0`fill); // just for testing multiple free
+        assert(*dp == 5.0);
+        free( ip, dp );
 
 #ifdef ERR1

tests/malloc.cfa

@@ -319 +319 @@
         free(ip);
 
-        free( 0p ); // sanity check
+        free( (void*) 0p ); // sanity check
         free( NULL ); // sanity check
 

tools/stat.py

@@ -1 @@
-#!/usr/bin/python
+#!/usr/bin/python3
 
 import sys
@@ -17 +17 @@
                 avg = numpy.mean  (content)
                 std = numpy.std   (content)
-                print "median {0:.1f} avg {1:.1f} stddev {2:.1f}".format( med, avg, std )
+                print("median {0:.1f} avg {1:.1f} stddev {2:.1f}".format( med, avg, std ))