source: libcfa/src/concurrency/kernel/startup.cfa @ 2b96031

arm-ehenumforall-pointer-decayjacob/cs343-translationnew-ast-unique-exprpthread-emulation
Last change on this file since 2b96031 was 2b96031, checked in by Thierry Delisle <tdelisle@…>, 16 months ago

Added new subqueue implementation.
Seems faster will test on another machine before full replacement.

  • Property mode set to 100644
File size: 24.0 KB
Line 
1//
2// Cforall Version 1.0.0 Copyright (C) 2020 University of Waterloo
3//
4// The contents of this file are covered under the licence agreement in the
5// file "LICENCE" distributed with Cforall.
6//
7// kernel/startup.cfa --
8//
9// Author           : Thierry Delisle
10// Created On       : Thu Jul 30 15:12:54 2020
11// Last Modified By :
12// Last Modified On :
13// Update Count     :
14//
15
16#define __cforall_thread__
17
18// C Includes
19#include <errno.h>              // errno
20#include <string.h>             // strerror
21#include <unistd.h>             // sysconf
22extern "C" {
23      #include <limits.h>       // PTHREAD_STACK_MIN
24        #include <sys/eventfd.h>  // eventfd
25      #include <sys/mman.h>     // mprotect
26      #include <sys/resource.h> // getrlimit
27}
28
29// CFA Includes
30#include "kernel_private.hfa"
31#include "startup.hfa"          // STARTUP_PRIORITY_XXX
32#include "math.hfa"
33
34#define CFA_PROCESSOR_USE_MMAP 0
35
36//-----------------------------------------------------------------------------
37// Some assembly required
38#if defined( __i386 )
39        #define CtxGet( ctx ) __asm__ volatile ( \
40                "movl %%esp,%0\n" \
41                "movl %%ebp,%1\n" \
42                : "=rm" (ctx.SP), \
43                  "=rm" (ctx.FP) \
44        )
45#elif defined( __x86_64 )
46        #define CtxGet( ctx ) __asm__ volatile ( \
47                "movq %%rsp,%0\n" \
48                "movq %%rbp,%1\n" \
49                : "=rm" (ctx.SP), \
50                  "=rm" (ctx.FP) \
51        )
52#elif defined( __aarch64__ )
53        #define CtxGet( ctx ) __asm__ volatile ( \
54                "mov %0, sp\n" \
55                "mov %1, fp\n" \
56                : "=rm" (ctx.SP), \
57                  "=rm" (ctx.FP) \
58        )
59#else
60        #error unknown hardware architecture
61#endif
62
63//-----------------------------------------------------------------------------
64// Start and stop routine for the kernel, declared first to make sure they run first
65static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
66static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
67
68//-----------------------------------------------------------------------------
69// Static Forward Declarations
70struct current_stack_info_t;
71
72static void * __invoke_processor(void * arg);
73static void __kernel_first_resume( processor * this );
74static void __kernel_last_resume ( processor * this );
75static void init(processor & this, const char name[], cluster & _cltr, $thread * initT);
76static void deinit(processor & this);
77static void doregister( struct cluster & cltr );
78static void unregister( struct cluster & cltr );
79static void register_tls( processor * this );
80static void unregister_tls( processor * this );
81static void ?{}( $coroutine & this, current_stack_info_t * info);
82static void ?{}( $thread & this, current_stack_info_t * info);
83static void ?{}(processorCtx_t & this) {}
84static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info);
85
86#if defined(__CFA_WITH_VERIFY__)
87        static bool verify_fwd_bck_rng(void);
88#endif
89
90//-----------------------------------------------------------------------------
91// Forward Declarations for other modules
92extern void __kernel_alarm_startup(void);
93extern void __kernel_alarm_shutdown(void);
94
95//-----------------------------------------------------------------------------
96// Other Forward Declarations
97extern void __wake_proc(processor *);
98
99//-----------------------------------------------------------------------------
100// Kernel storage
101KERNEL_STORAGE(cluster,              mainCluster);
102KERNEL_STORAGE(processor,            mainProcessor);
103KERNEL_STORAGE($thread,              mainThread);
104KERNEL_STORAGE(__stack_t,            mainThreadCtx);
105KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock);
106#if !defined(__CFA_NO_STATISTICS__)
107KERNEL_STORAGE(__stats_t, mainProcStats);
108#endif
109
110cluster              * mainCluster;
111processor            * mainProcessor;
112$thread              * mainThread;
113__scheduler_RWLock_t * __scheduler_lock;
114
115extern "C" {
116        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
117}
118
119extern size_t __page_size;
120extern int __map_prot;
121
122//-----------------------------------------------------------------------------
123// Global state
124thread_local struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
125        NULL,                                                                                           // cannot use 0p
126        NULL,
127        false,
128        { 1, false, false },
129        0,
130        { 0, 0 },
131        NULL,
132        #ifdef __CFA_WITH_VERIFY__
133                false,
134                0,
135        #endif
136};
137
138//-----------------------------------------------------------------------------
139// Struct to steal stack
140struct current_stack_info_t {
141        __stack_t * storage;  // pointer to stack object
142        void * base;          // base of stack
143        void * limit;         // stack grows towards stack limit
144        void * context;       // address of cfa_context_t
145};
146
147void ?{}( current_stack_info_t & this ) {
148        __stack_context_t ctx;
149        CtxGet( ctx );
150        this.base = ctx.FP;
151
152        rlimit r;
153        getrlimit( RLIMIT_STACK, &r);
154        size_t size = r.rlim_cur;
155
156        this.limit = (void *)(((intptr_t)this.base) - size);
157        this.context = &storage_mainThreadCtx;
158}
159
160
161
162//=============================================================================================
163// Kernel Setup logic
164//=============================================================================================
165//-----------------------------------------------------------------------------
166// Kernel boot procedures
167static void __kernel_startup(void) {
168        /* paranoid */ verify( ! __preemption_enabled() );
169        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
170
171        __cfa_dbg_global_clusters.list{ __get };
172        __cfa_dbg_global_clusters.lock{};
173
174        /* paranoid */ verify( verify_fwd_bck_rng() );
175
176        // Initialize the global scheduler lock
177        __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock;
178        (*__scheduler_lock){};
179
180        // Initialize the main cluster
181        mainCluster = (cluster *)&storage_mainCluster;
182        (*mainCluster){"Main Cluster", 0};
183
184        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
185
186        // Start by initializing the main thread
187        // SKULLDUGGERY: the mainThread steals the process main thread
188        // which will then be scheduled by the mainProcessor normally
189        mainThread = ($thread *)&storage_mainThread;
190        current_stack_info_t info;
191        info.storage = (__stack_t*)&storage_mainThreadCtx;
192        (*mainThread){ &info };
193
194        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
195
196
197
198        // Construct the processor context of the main processor
199        void ?{}(processorCtx_t & this, processor * proc) {
200                (this.__cor){ "Processor" };
201                this.__cor.starter = 0p;
202                this.proc = proc;
203        }
204
205        void ?{}(processor & this) with( this ) {
206                ( this.terminated ){};
207                ( this.runner ){};
208                init( this, "Main Processor", *mainCluster, 0p );
209                kernel_thread = pthread_self();
210
211                runner{ &this };
212                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
213        }
214
215        // Initialize the main processor and the main processor ctx
216        // (the coroutine that contains the processing control flow)
217        mainProcessor = (processor *)&storage_mainProcessor;
218        (*mainProcessor){};
219
220        register_tls( mainProcessor );
221
222        //initialize the global state variables
223        __cfaabi_tls.this_processor = mainProcessor;
224        __cfaabi_tls.this_thread    = mainThread;
225
226        #if !defined( __CFA_NO_STATISTICS__ )
227                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
228                __init_stats( __cfaabi_tls.this_stats );
229        #endif
230
231        // Enable preemption
232        __kernel_alarm_startup();
233
234        // Add the main thread to the ready queue
235        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
236        schedule_thread$(mainThread);
237
238        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
239        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
240        // mainThread is on the ready queue when this call is made.
241        __kernel_first_resume( __cfaabi_tls.this_processor );
242
243
244        // THE SYSTEM IS NOW COMPLETELY RUNNING
245
246        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
247
248        /* paranoid */ verify( ! __preemption_enabled() );
249        enable_interrupts();
250        /* paranoid */ verify( __preemption_enabled() );
251
252}
253
254static void __kernel_shutdown(void) {
255        /* paranoid */ verify( __preemption_enabled() );
256        disable_interrupts();
257        /* paranoid */ verify( ! __preemption_enabled() );
258
259        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
260
261        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
262        // When its coroutine terminates, it return control to the mainThread
263        // which is currently here
264        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
265        __kernel_last_resume( __cfaabi_tls.this_processor );
266        mainThread->self_cor.state = Halted;
267
268        // THE SYSTEM IS NOW COMPLETELY STOPPED
269
270        // Disable preemption
271        __kernel_alarm_shutdown();
272
273        #if !defined( __CFA_NO_STATISTICS__ )
274                __stats_t * st = (__stats_t *)& storage_mainProcStats;
275                __tally_stats(mainCluster->stats, st);
276                if( 0 != mainProcessor->print_stats ) {
277                        __print_stats( st, mainProcessor->print_stats, "Processor ", mainProcessor->name, (void*)mainProcessor );
278                }
279                #if defined(CFA_STATS_ARRAY)
280                        __flush_stat( st, "Processor", mainProcessor );
281                #endif
282        #endif
283
284        unregister_tls( mainProcessor );
285
286        // Destroy the main processor and its context in reverse order of construction
287        // These were manually constructed so we need manually destroy them
288        void ^?{}(processor & this) with( this ){
289                deinit( this );
290
291                /* paranoid */ verify( this.do_terminate == true );
292                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
293        }
294
295        ^(*mainProcessor){};
296
297        // Final step, destroy the main thread since it is no longer needed
298
299        // Since we provided a stack to this taxk it will not destroy anything
300        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
301        ^(*mainThread){};
302
303        ^(*mainCluster){};
304
305        ^(*__scheduler_lock){};
306
307        ^(__cfa_dbg_global_clusters.list){};
308        ^(__cfa_dbg_global_clusters.lock){};
309
310        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
311}
312
313//=============================================================================================
314// Kernel Initial Scheduling logic
315//=============================================================================================
316
317// Context invoker for processors
318// This is the entry point for processors (kernel threads) *except* for the main processor
319// It effectively constructs a coroutine by stealing the pthread stack
320static void * __invoke_processor(void * arg) {
321        #if !defined( __CFA_NO_STATISTICS__ )
322                __stats_t local_stats;
323                __init_stats( &local_stats );
324                __cfaabi_tls.this_stats = &local_stats;
325        #endif
326
327        processor * proc = (processor *) arg;
328        __cfaabi_tls.this_processor = proc;
329        __cfaabi_tls.this_thread    = 0p;
330        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
331
332        register_tls( proc );
333
334        // SKULLDUGGERY: We want to create a context for the processor coroutine
335        // which is needed for the 2-step context switch. However, there is no reason
336        // to waste the perfectly valid stack create by pthread.
337        current_stack_info_t info;
338        __stack_t ctx;
339        info.storage = &ctx;
340        (proc->runner){ proc, &info };
341
342        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
343
344        //Set global state
345        __cfaabi_tls.this_thread = 0p;
346
347        //We now have a proper context from which to schedule threads
348        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
349
350        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
351        // resume it to start it like it normally would, it will just context switch
352        // back to here. Instead directly call the main since we already are on the
353        // appropriate stack.
354        get_coroutine(proc->runner)->state = Active;
355        main( proc->runner );
356        get_coroutine(proc->runner)->state = Halted;
357
358        // Main routine of the core returned, the core is now fully terminated
359        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
360
361        #if !defined(__CFA_NO_STATISTICS__)
362                __tally_stats(proc->cltr->stats, &local_stats);
363                if( 0 != proc->print_stats ) {
364                        __print_stats( &local_stats, proc->print_stats, "Processor ", proc->name, (void*)proc );
365                }
366                #if defined(CFA_STATS_ARRAY)
367                        __flush_stat( &local_stats, "Processor", proc );
368                #endif
369        #endif
370
371        unregister_tls( proc );
372
373        return 0p;
374}
375
376static void __kernel_first_resume( processor * this ) {
377        $thread * src = mainThread;
378        $coroutine * dst = get_coroutine(this->runner);
379
380        /* paranoid */ verify( ! __preemption_enabled() );
381
382        __cfaabi_tls.this_thread->curr_cor = dst;
383        __stack_prepare( &dst->stack, 65000 );
384        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
385
386        /* paranoid */ verify( ! __preemption_enabled() );
387
388        dst->last = &src->self_cor;
389        dst->starter = dst->starter ? dst->starter : &src->self_cor;
390
391        // make sure the current state is still correct
392        /* paranoid */ verify(src->state == Ready);
393
394        // context switch to specified coroutine
395        verify( dst->context.SP );
396        __cfactx_switch( &src->context, &dst->context );
397        // when __cfactx_switch returns we are back in the src coroutine
398
399        mainThread->curr_cor = &mainThread->self_cor;
400
401        // make sure the current state has been update
402        /* paranoid */ verify(src->state == Active);
403
404        /* paranoid */ verify( ! __preemption_enabled() );
405}
406
407// KERNEL_ONLY
408static void __kernel_last_resume( processor * this ) {
409        $coroutine * src = &mainThread->self_cor;
410        $coroutine * dst = get_coroutine(this->runner);
411
412        /* paranoid */ verify( ! __preemption_enabled() );
413        /* paranoid */ verify( dst->starter == src );
414        /* paranoid */ verify( dst->context.SP );
415
416        // SKULLDUGGERY in debug the processors check that the
417        // stack is still within the limit of the stack limits after running a thread.
418        // that check doesn't make sense if we context switch to the processor using the
419        // coroutine semantics. Since this is a special case, use the current context
420        // info to populate these fields.
421        __cfaabi_dbg_debug_do(
422                __stack_context_t ctx;
423                CtxGet( ctx );
424                mainThread->context.SP = ctx.SP;
425                mainThread->context.FP = ctx.FP;
426        )
427
428        // context switch to the processor
429        __cfactx_switch( &src->context, &dst->context );
430}
431
432
433//=============================================================================================
434// Kernel Object Constructors logic
435//=============================================================================================
436//-----------------------------------------------------------------------------
437// Main thread construction
438static void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) {
439        stack.storage = info->storage;
440        with(*stack.storage) {
441                limit     = info->limit;
442                base      = info->base;
443        }
444        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
445        *istorage |= 0x1;
446        name = "Main Thread";
447        state = Start;
448        starter = 0p;
449        last = 0p;
450        cancellation = 0p;
451}
452
453static void ?{}( $thread & this, current_stack_info_t * info) with( this ) {
454        ticket = TICKET_RUNNING;
455        state = Start;
456        self_cor{ info };
457        curr_cor = &self_cor;
458        curr_cluster = mainCluster;
459        self_mon.owner = &this;
460        self_mon.recursion = 1;
461        self_mon_p = &self_mon;
462        link.next = 0p;
463        link.prev = 0p;
464        link.ts   = 0;
465        link.preferred = -1u;
466        last_proc = 0p;
467        #if defined( __CFA_WITH_VERIFY__ )
468                canary = 0x0D15EA5E0D15EA5Ep;
469        #endif
470
471        node.next = 0p;
472        node.prev = 0p;
473        doregister(curr_cluster, this);
474
475        monitors{ &self_mon_p, 1, (fptr_t)0 };
476}
477
478//-----------------------------------------------------------------------------
479// Processor
480// Construct the processor context of non-main processors
481static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
482        (this.__cor){ info };
483        this.proc = proc;
484}
485
486static void init(processor & this, const char name[], cluster & _cltr, $thread * initT) with( this ) {
487        this.name = name;
488        this.cltr = &_cltr;
489        this.rdq.its = 0;
490        this.rdq.itr = 0;
491        this.rdq.id  = -1u;
492        this.rdq.target = -1u;
493        this.rdq.cutoff = -1ull;
494        do_terminate = false;
495        preemption_alarm = 0p;
496        pending_preemption = false;
497
498        this.io.ctx = 0p;
499        this.io.pending = false;
500        this.io.dirty   = false;
501
502        this.init.thrd = initT;
503
504        this.idle = eventfd(0, 0);
505        if (idle < 0) {
506                abort("KERNEL ERROR: PROCESSOR EVENTFD - %s\n", strerror(errno));
507        }
508
509        #if !defined(__CFA_NO_STATISTICS__)
510                print_stats = 0;
511                print_halts = false;
512        #endif
513
514        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
515}
516
517// Not a ctor, it just preps the destruction but should not destroy members
518static void deinit(processor & this) {
519        close(this.idle);
520}
521
522void ?{}(processor & this, const char name[], cluster & _cltr, $thread * initT) {
523        ( this.terminated ){};
524        ( this.runner ){};
525
526        disable_interrupts();
527                init( this, name, _cltr, initT );
528        enable_interrupts();
529
530        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
531
532        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
533}
534
535void ?{}(processor & this, const char name[], cluster & _cltr) {
536        (this){name, _cltr, 0p};
537}
538
539extern size_t __page_size;
540void ^?{}(processor & this) with( this ){
541        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
542                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
543
544                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
545                __wake_proc( &this );
546
547                wait( terminated );
548                /* paranoid */ verify( active_processor() != &this);
549        }
550
551        __destroy_pthread( kernel_thread, this.stack, 0p );
552
553        disable_interrupts();
554                deinit( this );
555        enable_interrupts();
556}
557
558//-----------------------------------------------------------------------------
559// Cluster
560static void ?{}(__cluster_proc_list & this) {
561        this.lock  = 0;
562        this.idle  = 0;
563        this.total = 0;
564}
565
566void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) with( this ) {
567        this.name = name;
568        this.preemption_rate = preemption_rate;
569        ready_queue{};
570
571        #if !defined(__CFA_NO_STATISTICS__)
572                print_stats = 0;
573                stats = alloc();
574                __init_stats( stats );
575        #endif
576
577        threads{ __get };
578
579        io.arbiter = create();
580        io.params = io_params;
581
582        doregister(this);
583
584        // Lock the RWlock so no-one pushes/pops while we are changing the queue
585        disable_interrupts();
586        uint_fast32_t last_size = ready_mutate_lock();
587
588                // Adjust the ready queue size
589                ready_queue_grow( &this );
590
591        // Unlock the RWlock
592        ready_mutate_unlock( last_size );
593        enable_interrupts( false ); // Don't poll, could be in main cluster
594}
595
596void ^?{}(cluster & this) {
597        destroy(this.io.arbiter);
598
599        // Lock the RWlock so no-one pushes/pops while we are changing the queue
600        disable_interrupts();
601        uint_fast32_t last_size = ready_mutate_lock();
602
603                // Adjust the ready queue size
604                ready_queue_shrink( &this );
605
606        // Unlock the RWlock
607        ready_mutate_unlock( last_size );
608        enable_interrupts( false ); // Don't poll, could be in main cluster
609
610        #if !defined(__CFA_NO_STATISTICS__)
611                if( 0 != this.print_stats ) {
612                        __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
613                }
614                #if defined(CFA_STATS_ARRAY)
615                        __flush_stat( this.stats, "Cluster", &this );
616                #endif
617                free( this.stats );
618        #endif
619
620        unregister(this);
621}
622
623//=============================================================================================
624// Miscellaneous Initialization
625//=============================================================================================
626//-----------------------------------------------------------------------------
627// Global Queues
628static void doregister( cluster     & cltr ) {
629        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
630        push_front( __cfa_dbg_global_clusters.list, cltr );
631        unlock    ( __cfa_dbg_global_clusters.lock );
632}
633
634static void unregister( cluster     & cltr ) {
635        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
636        remove( __cfa_dbg_global_clusters.list, cltr );
637        unlock( __cfa_dbg_global_clusters.lock );
638}
639
640void doregister( cluster * cltr, $thread & thrd ) {
641        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
642        cltr->nthreads += 1;
643        push_front(cltr->threads, thrd);
644        unlock    (cltr->thread_list_lock);
645}
646
647void unregister( cluster * cltr, $thread & thrd ) {
648        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
649        remove(cltr->threads, thrd );
650        cltr->nthreads -= 1;
651        unlock(cltr->thread_list_lock);
652}
653
654static void register_tls( processor * this ) {
655        // Register and Lock the RWlock so no-one pushes/pops while we are changing the queue
656        uint_fast32_t last_size;
657        [this->unique_id, last_size] = ready_mutate_register();
658
659                this->cltr->procs.total += 1u;
660                insert_last(this->cltr->procs.actives, *this);
661
662                // Adjust the ready queue size
663                ready_queue_grow( this->cltr );
664
665        // Unlock the RWlock
666        ready_mutate_unlock( last_size );
667}
668
669
670static void unregister_tls( processor * this ) {
671        // Lock the RWlock so no-one pushes/pops while we are changing the queue
672        uint_fast32_t last_size = ready_mutate_lock();
673                this->cltr->procs.total -= 1u;
674                remove(*this);
675
676                // clear the cluster so nothing gets pushed to local queues
677                cluster * cltr = this->cltr;
678                this->cltr = 0p;
679
680                // Adjust the ready queue size
681                ready_queue_shrink( cltr );
682
683        // Unlock the RWlock and unregister: we don't need the read_lock any more
684        ready_mutate_unregister( this->unique_id, last_size );
685}
686
687static void check( int ret, const char func[] ) {
688        if ( ret ) {                                                                            // pthread routines return errno values
689                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
690        } // if
691} // Abort
692
693void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
694        pthread_attr_t attr;
695
696        check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
697
698        size_t stacksize;
699        // default stack size, normally defined by shell limit
700        check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
701        assert( stacksize >= PTHREAD_STACK_MIN );
702
703        void * stack;
704        #if CFA_PROCESSOR_USE_MMAP
705                stacksize = ceiling( stacksize, __page_size ) + __page_size;
706                stack = mmap(0p, stacksize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
707                if(stack == ((void*)-1)) {
708                        abort( "pthread stack creation : internal error, mmap failure, error(%d) %s.", errno, strerror( errno ) );
709                }
710                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
711                        abort( "pthread stack creation : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
712                } // if
713        #else
714                __cfaabi_dbg_debug_do(
715                        stack = memalign( __page_size, stacksize + __page_size );
716                        // pthread has no mechanism to create the guard page in user supplied stack.
717                        if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
718                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
719                        } // if
720                );
721                __cfaabi_dbg_no_debug_do(
722                        stack = malloc( stacksize );
723                );
724        #endif
725
726
727        check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
728
729        check( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
730        return stack;
731}
732
733void __destroy_pthread( pthread_t pthread, void * stack, void ** retval ) {
734        int err = pthread_join( pthread, retval );
735        if( err != 0 ) abort("KERNEL ERROR: joining pthread %p caused error %s\n", (void*)pthread, strerror(err));
736
737        #if CFA_PROCESSOR_USE_MMAP
738                pthread_attr_t attr;
739
740                check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
741
742                size_t stacksize;
743                // default stack size, normally defined by shell limit
744                check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
745                assert( stacksize >= PTHREAD_STACK_MIN );
746                stacksize += __page_size;
747
748                if(munmap(stack, stacksize) == -1) {
749                        abort( "pthread stack destruction : internal error, munmap failure, error(%d) %s.", errno, strerror( errno ) );
750                }
751        #else
752                __cfaabi_dbg_debug_do(
753                        // pthread has no mechanism to create the guard page in user supplied stack.
754                        if ( mprotect( stack, __page_size, __map_prot ) == -1 ) {
755                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
756                        } // if
757                );
758                free( stack );
759        #endif
760}
761
762#if defined(__CFA_WITH_VERIFY__)
763static bool verify_fwd_bck_rng(void) {
764        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
765
766        unsigned values[10];
767        for(i; 10) {
768                values[i] = __tls_rand_fwd();
769        }
770
771        __tls_rand_advance_bck();
772
773        for ( i; 9 -~= 0 ) {
774                if(values[i] != __tls_rand_bck()) {
775                        return false;
776                }
777        }
778
779        return true;
780}
781#endif
Note: See TracBrowser for help on using the repository browser.