source: libcfa/src/concurrency/kernel/startup.cfa @ 7039ab9

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since 7039ab9 was a5a01faa, checked in by Thierry Delisle <tdelisle@…>, 4 years ago

Fix a bug where stats from the main processor where not tallied properly.

  • Property mode set to 100644
File size: 23.4 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 ?{}( $coroutine & this, current_stack_info_t * info);
80static void ?{}( $thread & this, current_stack_info_t * info);
81static void ?{}(processorCtx_t & this) {}
82static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info);
83
84#if defined(__CFA_WITH_VERIFY__)
85        static bool verify_fwd_bck_rng(void);
86#endif
87
88//-----------------------------------------------------------------------------
89// Forward Declarations for other modules
90extern void __kernel_alarm_startup(void);
91extern void __kernel_alarm_shutdown(void);
92
93//-----------------------------------------------------------------------------
94// Other Forward Declarations
95extern void __wake_proc(processor *);
96
97//-----------------------------------------------------------------------------
98// Kernel storage
99KERNEL_STORAGE(cluster,              mainCluster);
100KERNEL_STORAGE(processor,            mainProcessor);
101KERNEL_STORAGE($thread,              mainThread);
102KERNEL_STORAGE(__stack_t,            mainThreadCtx);
103KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock);
104#if !defined(__CFA_NO_STATISTICS__)
105KERNEL_STORAGE(__stats_t, mainProcStats);
106#endif
107
108cluster              * mainCluster;
109processor            * mainProcessor;
110$thread              * mainThread;
111__scheduler_RWLock_t * __scheduler_lock;
112
113extern "C" {
114        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
115}
116
117extern size_t __page_size;
118extern int __map_prot;
119
120//-----------------------------------------------------------------------------
121// Global state
122thread_local struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
123        NULL,                                                                                           // cannot use 0p
124        NULL,
125        NULL,
126        NULL,
127        { 1, false, false },
128};
129
130//-----------------------------------------------------------------------------
131// Struct to steal stack
132struct current_stack_info_t {
133        __stack_t * storage;  // pointer to stack object
134        void * base;          // base of stack
135        void * limit;         // stack grows towards stack limit
136        void * context;       // address of cfa_context_t
137};
138
139void ?{}( current_stack_info_t & this ) {
140        __stack_context_t ctx;
141        CtxGet( ctx );
142        this.base = ctx.FP;
143
144        rlimit r;
145        getrlimit( RLIMIT_STACK, &r);
146        size_t size = r.rlim_cur;
147
148        this.limit = (void *)(((intptr_t)this.base) - size);
149        this.context = &storage_mainThreadCtx;
150}
151
152
153
154//=============================================================================================
155// Kernel Setup logic
156//=============================================================================================
157//-----------------------------------------------------------------------------
158// Kernel boot procedures
159static void __kernel_startup(void) {
160        /* paranoid */ verify( ! __preemption_enabled() );
161        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
162
163        __cfa_dbg_global_clusters.list{ __get };
164        __cfa_dbg_global_clusters.lock{};
165
166        /* paranoid */ verify( verify_fwd_bck_rng() );
167
168        // Initialize the global scheduler lock
169        __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock;
170        (*__scheduler_lock){};
171
172        // Initialize the main cluster
173        mainCluster = (cluster *)&storage_mainCluster;
174        (*mainCluster){"Main Cluster", 0};
175
176        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
177
178        // Start by initializing the main thread
179        // SKULLDUGGERY: the mainThread steals the process main thread
180        // which will then be scheduled by the mainProcessor normally
181        mainThread = ($thread *)&storage_mainThread;
182        current_stack_info_t info;
183        info.storage = (__stack_t*)&storage_mainThreadCtx;
184        (*mainThread){ &info };
185
186        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
187
188
189
190        // Construct the processor context of the main processor
191        void ?{}(processorCtx_t & this, processor * proc) {
192                (this.__cor){ "Processor" };
193                this.__cor.starter = 0p;
194                this.proc = proc;
195        }
196
197        void ?{}(processor & this) with( this ) {
198                ( this.terminated ){};
199                ( this.runner ){};
200                init( this, "Main Processor", *mainCluster, 0p );
201                kernel_thread = pthread_self();
202
203                runner{ &this };
204                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
205        }
206
207        // Initialize the main processor and the main processor ctx
208        // (the coroutine that contains the processing control flow)
209        mainProcessor = (processor *)&storage_mainProcessor;
210        (*mainProcessor){};
211
212        //initialize the global state variables
213        __cfaabi_tls.this_processor = mainProcessor;
214        __cfaabi_tls.this_proc_id   = (__processor_id_t*)mainProcessor;
215        __cfaabi_tls.this_thread    = mainThread;
216
217        #if !defined( __CFA_NO_STATISTICS__ )
218                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
219                __init_stats( __cfaabi_tls.this_stats );
220        #endif
221
222        // Enable preemption
223        __kernel_alarm_startup();
224
225        // Add the main thread to the ready queue
226        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
227        __schedule_thread(mainThread);
228
229        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
230        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
231        // mainThread is on the ready queue when this call is made.
232        __kernel_first_resume( __cfaabi_tls.this_processor );
233
234
235        // THE SYSTEM IS NOW COMPLETELY RUNNING
236
237        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
238
239        /* paranoid */ verify( ! __preemption_enabled() );
240        enable_interrupts( __cfaabi_dbg_ctx );
241        /* paranoid */ verify( __preemption_enabled() );
242
243}
244
245static void __kernel_shutdown(void) {
246        /* paranoid */ verify( __preemption_enabled() );
247        disable_interrupts();
248        /* paranoid */ verify( ! __preemption_enabled() );
249
250        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
251
252        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
253        // When its coroutine terminates, it return control to the mainThread
254        // which is currently here
255        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
256        __kernel_last_resume( __cfaabi_tls.this_processor );
257        mainThread->self_cor.state = Halted;
258
259        // THE SYSTEM IS NOW COMPLETELY STOPPED
260
261        // Disable preemption
262        __kernel_alarm_shutdown();
263
264        #if !defined( __CFA_NO_STATISTICS__ )
265                __stats_t * st = (__stats_t *)& storage_mainProcStats;
266                __tally_stats(mainCluster->stats, st);
267                if( 0 != mainProcessor->print_stats ) {
268                        __print_stats( st, mainProcessor->print_stats, "Processor ", mainProcessor->name, (void*)mainProcessor );
269                }
270        #endif
271
272        // Destroy the main processor and its context in reverse order of construction
273        // These were manually constructed so we need manually destroy them
274        void ^?{}(processor & this) with( this ){
275                deinit( this );
276
277                /* paranoid */ verify( this.do_terminate == true );
278                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
279        }
280
281        ^(*mainProcessor){};
282
283        // Final step, destroy the main thread since it is no longer needed
284
285        // Since we provided a stack to this taxk it will not destroy anything
286        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
287        ^(*mainThread){};
288
289        ^(*mainCluster){};
290
291        ^(*__scheduler_lock){};
292
293        ^(__cfa_dbg_global_clusters.list){};
294        ^(__cfa_dbg_global_clusters.lock){};
295
296        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
297}
298
299//=============================================================================================
300// Kernel Initial Scheduling logic
301//=============================================================================================
302
303// Context invoker for processors
304// This is the entry point for processors (kernel threads) *except* for the main processor
305// It effectively constructs a coroutine by stealing the pthread stack
306static void * __invoke_processor(void * arg) {
307        #if !defined( __CFA_NO_STATISTICS__ )
308                __stats_t local_stats;
309                __init_stats( &local_stats );
310                __cfaabi_tls.this_stats = &local_stats;
311        #endif
312
313        processor * proc = (processor *) arg;
314        __cfaabi_tls.this_processor = proc;
315        __cfaabi_tls.this_proc_id   = (__processor_id_t*)proc;
316        __cfaabi_tls.this_thread    = 0p;
317        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
318        // SKULLDUGGERY: We want to create a context for the processor coroutine
319        // which is needed for the 2-step context switch. However, there is no reason
320        // to waste the perfectly valid stack create by pthread.
321        current_stack_info_t info;
322        __stack_t ctx;
323        info.storage = &ctx;
324        (proc->runner){ proc, &info };
325
326        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
327
328        //Set global state
329        __cfaabi_tls.this_thread = 0p;
330
331        //We now have a proper context from which to schedule threads
332        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
333
334        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
335        // resume it to start it like it normally would, it will just context switch
336        // back to here. Instead directly call the main since we already are on the
337        // appropriate stack.
338        get_coroutine(proc->runner)->state = Active;
339        main( proc->runner );
340        get_coroutine(proc->runner)->state = Halted;
341
342        // Main routine of the core returned, the core is now fully terminated
343        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
344
345        #if !defined(__CFA_NO_STATISTICS__)
346                __tally_stats(proc->cltr->stats, &local_stats);
347                if( 0 != proc->print_stats ) {
348                        __print_stats( &local_stats, proc->print_stats, "Processor ", proc->name, (void*)proc );
349                }
350        #endif
351
352        return 0p;
353}
354
355static void __kernel_first_resume( processor * this ) {
356        $thread * src = mainThread;
357        $coroutine * dst = get_coroutine(this->runner);
358
359        /* paranoid */ verify( ! __preemption_enabled() );
360
361        __cfaabi_tls.this_thread->curr_cor = dst;
362        __stack_prepare( &dst->stack, 65000 );
363        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
364
365        /* paranoid */ verify( ! __preemption_enabled() );
366
367        dst->last = &src->self_cor;
368        dst->starter = dst->starter ? dst->starter : &src->self_cor;
369
370        // make sure the current state is still correct
371        /* paranoid */ verify(src->state == Ready);
372
373        // context switch to specified coroutine
374        verify( dst->context.SP );
375        __cfactx_switch( &src->context, &dst->context );
376        // when __cfactx_switch returns we are back in the src coroutine
377
378        mainThread->curr_cor = &mainThread->self_cor;
379
380        // make sure the current state has been update
381        /* paranoid */ verify(src->state == Active);
382
383        /* paranoid */ verify( ! __preemption_enabled() );
384}
385
386// KERNEL_ONLY
387static void __kernel_last_resume( processor * this ) {
388        $coroutine * src = &mainThread->self_cor;
389        $coroutine * dst = get_coroutine(this->runner);
390
391        /* paranoid */ verify( ! __preemption_enabled() );
392        /* paranoid */ verify( dst->starter == src );
393        /* paranoid */ verify( dst->context.SP );
394
395        // SKULLDUGGERY in debug the processors check that the
396        // stack is still within the limit of the stack limits after running a thread.
397        // that check doesn't make sense if we context switch to the processor using the
398        // coroutine semantics. Since this is a special case, use the current context
399        // info to populate these fields.
400        __cfaabi_dbg_debug_do(
401                __stack_context_t ctx;
402                CtxGet( ctx );
403                mainThread->context.SP = ctx.SP;
404                mainThread->context.FP = ctx.FP;
405        )
406
407        // context switch to the processor
408        __cfactx_switch( &src->context, &dst->context );
409}
410
411
412//=============================================================================================
413// Kernel Object Constructors logic
414//=============================================================================================
415//-----------------------------------------------------------------------------
416// Main thread construction
417static void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) {
418        stack.storage = info->storage;
419        with(*stack.storage) {
420                limit     = info->limit;
421                base      = info->base;
422        }
423        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
424        *istorage |= 0x1;
425        name = "Main Thread";
426        state = Start;
427        starter = 0p;
428        last = 0p;
429        cancellation = 0p;
430}
431
432static void ?{}( $thread & this, current_stack_info_t * info) with( this ) {
433        ticket = TICKET_RUNNING;
434        state = Start;
435        self_cor{ info };
436        curr_cor = &self_cor;
437        curr_cluster = mainCluster;
438        self_mon.owner = &this;
439        self_mon.recursion = 1;
440        self_mon_p = &self_mon;
441        link.next = 0p;
442        link.prev = 0p;
443        #if defined( __CFA_WITH_VERIFY__ )
444                canary = 0x0D15EA5E0D15EA5Ep;
445        #endif
446
447        node.next = 0p;
448        node.prev = 0p;
449        doregister(curr_cluster, this);
450
451        monitors{ &self_mon_p, 1, (fptr_t)0 };
452}
453
454//-----------------------------------------------------------------------------
455// Processor
456// Construct the processor context of non-main processors
457static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
458        (this.__cor){ info };
459        this.proc = proc;
460}
461
462static void init(processor & this, const char name[], cluster & _cltr, $thread * initT) with( this ) {
463        this.name = name;
464        this.cltr = &_cltr;
465        do_terminate = false;
466        preemption_alarm = 0p;
467        pending_preemption = false;
468
469        this.io.ctx = 0p;
470        this.io.pending = false;
471        this.io.dirty   = false;
472
473        this.init.thrd = initT;
474
475        this.idle = eventfd(0, 0);
476        if (idle < 0) {
477                abort("KERNEL ERROR: PROCESSOR EVENTFD - %s\n", strerror(errno));
478        }
479
480        #if !defined(__CFA_NO_STATISTICS__)
481                print_stats = 0;
482                print_halts = false;
483        #endif
484
485        lock( this.cltr->idles );
486                int target = this.cltr->idles.total += 1u;
487        unlock( this.cltr->idles );
488
489        id = doregister((__processor_id_t*)&this);
490
491        // Lock the RWlock so no-one pushes/pops while we are changing the queue
492        uint_fast32_t last_size = ready_mutate_lock();
493
494                // Adjust the ready queue size
495                this.cltr_id = ready_queue_grow( cltr, target );
496
497        // Unlock the RWlock
498        ready_mutate_unlock( last_size );
499
500        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
501}
502
503// Not a ctor, it just preps the destruction but should not destroy members
504static void deinit(processor & this) {
505        lock( this.cltr->idles );
506                int target = this.cltr->idles.total -= 1u;
507        unlock( this.cltr->idles );
508
509        // Lock the RWlock so no-one pushes/pops while we are changing the queue
510        uint_fast32_t last_size = ready_mutate_lock();
511
512                // Adjust the ready queue size
513                ready_queue_shrink( this.cltr, target );
514
515        // Unlock the RWlock
516        ready_mutate_unlock( last_size );
517
518        // Finally we don't need the read_lock any more
519        unregister((__processor_id_t*)&this);
520
521        close(this.idle);
522}
523
524void ?{}(processor & this, const char name[], cluster & _cltr, $thread * initT) {
525        ( this.terminated ){};
526        ( this.runner ){};
527
528        disable_interrupts();
529                init( this, name, _cltr, initT );
530        enable_interrupts( __cfaabi_dbg_ctx );
531
532        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
533
534        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
535}
536
537void ?{}(processor & this, const char name[], cluster & _cltr) {
538        (this){name, _cltr, 0p};
539}
540
541extern size_t __page_size;
542void ^?{}(processor & this) with( this ){
543        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
544                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
545
546                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
547                __wake_proc( &this );
548
549                wait( terminated );
550                /* paranoid */ verify( active_processor() != &this);
551        }
552
553        __destroy_pthread( kernel_thread, this.stack, 0p );
554
555        disable_interrupts();
556                deinit( this );
557        enable_interrupts( __cfaabi_dbg_ctx );
558}
559
560//-----------------------------------------------------------------------------
561// Cluster
562static void ?{}(__cluster_idles & this) {
563        this.lock  = 0;
564        this.idle  = 0;
565        this.total = 0;
566        (this.list){};
567}
568
569void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) with( this ) {
570        this.name = name;
571        this.preemption_rate = preemption_rate;
572        ready_queue{};
573
574        #if !defined(__CFA_NO_STATISTICS__)
575                print_stats = 0;
576                stats = alloc();
577                __init_stats( stats );
578        #endif
579
580        threads{ __get };
581
582        io.arbiter = create();
583        io.params = io_params;
584
585        doregister(this);
586
587        // Lock the RWlock so no-one pushes/pops while we are changing the queue
588        disable_interrupts();
589        uint_fast32_t last_size = ready_mutate_lock();
590
591                // Adjust the ready queue size
592                ready_queue_grow( &this, 0 );
593
594        // Unlock the RWlock
595        ready_mutate_unlock( last_size );
596        enable_interrupts_noPoll(); // Don't poll, could be in main cluster
597}
598
599void ^?{}(cluster & this) {
600        destroy(this.io.arbiter);
601
602        // Lock the RWlock so no-one pushes/pops while we are changing the queue
603        disable_interrupts();
604        uint_fast32_t last_size = ready_mutate_lock();
605
606                // Adjust the ready queue size
607                ready_queue_shrink( &this, 0 );
608
609        // Unlock the RWlock
610        ready_mutate_unlock( last_size );
611        enable_interrupts_noPoll(); // Don't poll, could be in main cluster
612
613        #if !defined(__CFA_NO_STATISTICS__)
614                if( 0 != this.print_stats ) {
615                        __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
616                }
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 check( int ret, const char func[] ) {
655        if ( ret ) {                                                                            // pthread routines return errno values
656                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
657        } // if
658} // Abort
659
660void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
661        pthread_attr_t attr;
662
663        check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
664
665        size_t stacksize;
666        // default stack size, normally defined by shell limit
667        check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
668        assert( stacksize >= PTHREAD_STACK_MIN );
669
670        void * stack;
671        #if CFA_PROCESSOR_USE_MMAP
672                stacksize = ceiling( stacksize, __page_size ) + __page_size;
673                stack = mmap(0p, stacksize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
674                if(stack == ((void*)-1)) {
675                        abort( "pthread stack creation : internal error, mmap failure, error(%d) %s.", errno, strerror( errno ) );
676                }
677                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
678                        abort( "pthread stack creation : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
679                } // if
680        #else
681                __cfaabi_dbg_debug_do(
682                        stack = memalign( __page_size, stacksize + __page_size );
683                        // pthread has no mechanism to create the guard page in user supplied stack.
684                        if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
685                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
686                        } // if
687                );
688                __cfaabi_dbg_no_debug_do(
689                        stack = malloc( stacksize );
690                );
691        #endif
692
693
694        check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
695
696        check( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
697        return stack;
698}
699
700void __destroy_pthread( pthread_t pthread, void * stack, void ** retval ) {
701        int err = pthread_join( pthread, retval );
702        if( err != 0 ) abort("KERNEL ERROR: joining pthread %p caused error %s\n", (void*)pthread, strerror(err));
703
704        #if CFA_PROCESSOR_USE_MMAP
705                pthread_attr_t attr;
706
707                check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
708
709                size_t stacksize;
710                // default stack size, normally defined by shell limit
711                check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
712                assert( stacksize >= PTHREAD_STACK_MIN );
713                stacksize += __page_size;
714
715                if(munmap(stack, stacksize) == -1) {
716                        abort( "pthread stack destruction : internal error, munmap failure, error(%d) %s.", errno, strerror( errno ) );
717                }
718        #else
719                __cfaabi_dbg_debug_do(
720                        // pthread has no mechanism to create the guard page in user supplied stack.
721                        if ( mprotect( stack, __page_size, __map_prot ) == -1 ) {
722                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
723                        } // if
724                );
725                free( stack );
726        #endif
727}
728
729#if defined(__CFA_WITH_VERIFY__)
730static bool verify_fwd_bck_rng(void) {
731        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
732
733        unsigned values[10];
734        for(i; 10) {
735                values[i] = __tls_rand_fwd();
736        }
737
738        __tls_rand_advance_bck();
739
740        for ( i; 9 -~= 0 ) {
741                if(values[i] != __tls_rand_bck()) {
742                        return false;
743                }
744        }
745
746        return true;
747}
748#endif
Note: See TracBrowser for help on using the repository browser.