source: libcfa/src/concurrency/kernel/startup.cfa @ 1757f98

enumforall-pointer-decaypthread-emulation
Last change on this file since 1757f98 was 1757f98, checked in by Thierry Delisle <tdelisle@…>, 9 months ago

Refactoring idle sleep to try and help the change from idle on read to idle on io_uring_enter.

  • Property mode set to 100644
File size: 24.7 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#define _GNU_SOURCE
18
19// C Includes
20#include <errno.h>              // errno
21#include <signal.h>
22#include <string.h>             // strerror
23#include <unistd.h>             // sysconf
24
25extern "C" {
26      #include <limits.h>       // PTHREAD_STACK_MIN
27        #include <unistd.h>       // syscall
28        #include <sys/eventfd.h>  // eventfd
29      #include <sys/mman.h>     // mprotect
30      #include <sys/resource.h> // getrlimit
31}
32
33// CFA Includes
34#include "kernel_private.hfa"
35#include "startup.hfa"          // STARTUP_PRIORITY_XXX
36#include "math.hfa"
37
38#define CFA_PROCESSOR_USE_MMAP 0
39
40//-----------------------------------------------------------------------------
41// Some assembly required
42#if defined( __i386 )
43        #define CtxGet( ctx ) __asm__ volatile ( \
44                "movl %%esp,%0\n" \
45                "movl %%ebp,%1\n" \
46                : "=rm" (ctx.SP), \
47                  "=rm" (ctx.FP) \
48        )
49#elif defined( __x86_64 )
50        #define CtxGet( ctx ) __asm__ volatile ( \
51                "movq %%rsp,%0\n" \
52                "movq %%rbp,%1\n" \
53                : "=rm" (ctx.SP), \
54                  "=rm" (ctx.FP) \
55        )
56#elif defined( __aarch64__ )
57        #define CtxGet( ctx ) __asm__ volatile ( \
58                "mov %0, sp\n" \
59                "mov %1, fp\n" \
60                : "=rm" (ctx.SP), \
61                  "=rm" (ctx.FP) \
62        )
63#else
64        #error unknown hardware architecture
65#endif
66
67//-----------------------------------------------------------------------------
68// Start and stop routine for the kernel, declared first to make sure they run first
69static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
70static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
71
72//-----------------------------------------------------------------------------
73// Static Forward Declarations
74struct current_stack_info_t;
75
76static void * __invoke_processor(void * arg);
77static void __kernel_first_resume( processor * this );
78static void __kernel_last_resume ( processor * this );
79static void init(processor & this, const char name[], cluster & _cltr, thread$ * initT);
80static void deinit(processor & this);
81static void doregister( struct cluster & cltr );
82static void unregister( struct cluster & cltr );
83static void register_tls( processor * this );
84static void unregister_tls( processor * this );
85static void ?{}( coroutine$ & this, current_stack_info_t * info);
86static void ?{}( thread$ & this, current_stack_info_t * info);
87static void ?{}(processorCtx_t & this) {}
88static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info);
89
90#if defined(__CFA_WITH_VERIFY__)
91        static bool verify_fwd_bck_rng(void);
92#endif
93
94//-----------------------------------------------------------------------------
95// Forward Declarations for other modules
96extern void __kernel_alarm_startup(void);
97extern void __kernel_alarm_shutdown(void);
98
99//-----------------------------------------------------------------------------
100// Other Forward Declarations
101extern void __wake_proc(processor *);
102extern int cfa_main_returned;                                                   // from interpose.cfa
103
104//-----------------------------------------------------------------------------
105// Kernel storage
106KERNEL_STORAGE(cluster,              mainCluster);
107KERNEL_STORAGE(processor,            mainProcessor);
108KERNEL_STORAGE(thread$,              mainThread);
109KERNEL_STORAGE(__stack_t,            mainThreadCtx);
110KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock);
111#if !defined(__CFA_NO_STATISTICS__)
112KERNEL_STORAGE(__stats_t, mainProcStats);
113#endif
114
115cluster              * mainCluster;
116processor            * mainProcessor;
117thread$              * mainThread;
118__scheduler_RWLock_t * __scheduler_lock;
119
120extern "C" {
121        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
122}
123
124extern size_t __page_size;
125extern int __map_prot;
126
127//-----------------------------------------------------------------------------
128// Global state
129thread_local struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
130        NULL,                                                                                           // cannot use 0p
131        NULL,
132        false,
133        { 1, false, false },
134        0,
135        { 0, 0 },
136        NULL,
137        #ifdef __CFA_WITH_VERIFY__
138                false,
139                0,
140        #endif
141};
142
143#if   defined(CFA_HAVE_LINUX_LIBRSEQ)
144        // No data needed
145#elif defined(CFA_HAVE_LINUX_RSEQ_H)
146        extern "Cforall" {
147                __attribute__((aligned(128))) thread_local volatile struct rseq __cfaabi_rseq @= {
148                        .cpu_id : RSEQ_CPU_ID_UNINITIALIZED,
149                };
150        }
151#else
152        // No data needed
153#endif
154
155//-----------------------------------------------------------------------------
156// Struct to steal stack
157struct current_stack_info_t {
158        __stack_t * storage;  // pointer to stack object
159        void * base;          // base of stack
160        void * limit;         // stack grows towards stack limit
161        void * context;       // address of cfa_context_t
162};
163
164void ?{}( current_stack_info_t & this ) {
165        __stack_context_t ctx;
166        CtxGet( ctx );
167        this.base = ctx.FP;
168
169        rlimit r;
170        getrlimit( RLIMIT_STACK, &r);
171        size_t size = r.rlim_cur;
172
173        this.limit = (void *)(((intptr_t)this.base) - size);
174        this.context = &storage_mainThreadCtx;
175}
176
177
178
179//=============================================================================================
180// Kernel Setup logic
181//=============================================================================================
182//-----------------------------------------------------------------------------
183// Kernel boot procedures
184static void __kernel_startup(void) {
185        /* paranoid */ verify( ! __preemption_enabled() );
186        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
187
188        __cfa_dbg_global_clusters.list{ __get };
189        __cfa_dbg_global_clusters.lock{};
190
191        /* paranoid */ verify( verify_fwd_bck_rng() );
192
193        // Initialize the global scheduler lock
194        __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock;
195        (*__scheduler_lock){};
196
197        // Initialize the main cluster
198        mainCluster = (cluster *)&storage_mainCluster;
199        (*mainCluster){"Main Cluster", 0};
200
201        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
202
203        // Construct the processor context of the main processor
204        void ?{}(processorCtx_t & this, processor * proc) {
205                (this.__cor){ "Processor" };
206                this.__cor.starter = 0p;
207                this.proc = proc;
208        }
209
210        void ?{}(processor & this) with( this ) {
211                ( this.terminated ){};
212                ( this.runner ){};
213                init( this, "Main Processor", *mainCluster, 0p );
214                kernel_thread = pthread_self();
215
216                runner{ &this };
217                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
218        }
219
220        // Initialize the main processor and the main processor ctx
221        // (the coroutine that contains the processing control flow)
222        mainProcessor = (processor *)&storage_mainProcessor;
223        (*mainProcessor){};
224
225        register_tls( mainProcessor );
226
227        // Start by initializing the main thread
228        // SKULLDUGGERY: the mainThread steals the process main thread
229        // which will then be scheduled by the mainProcessor normally
230        mainThread = (thread$ *)&storage_mainThread;
231        current_stack_info_t info;
232        info.storage = (__stack_t*)&storage_mainThreadCtx;
233        (*mainThread){ &info };
234
235        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
236
237        //initialize the global state variables
238        __cfaabi_tls.this_processor = mainProcessor;
239        __cfaabi_tls.this_thread    = mainThread;
240
241        #if !defined( __CFA_NO_STATISTICS__ )
242                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
243                __init_stats( __cfaabi_tls.this_stats );
244        #endif
245        mainProcessor->local_data = &__cfaabi_tls;
246
247        // Enable preemption
248        __kernel_alarm_startup();
249
250        // Add the main thread to the ready queue
251        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
252        schedule_thread$(mainThread, UNPARK_LOCAL);
253
254        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
255        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
256        // mainThread is on the ready queue when this call is made.
257        __kernel_first_resume( __cfaabi_tls.this_processor );
258
259
260        // THE SYSTEM IS NOW COMPLETELY RUNNING
261
262        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
263
264        /* paranoid */ verify( ! __preemption_enabled() );
265        enable_interrupts();
266        /* paranoid */ verify( __preemption_enabled() );
267
268}
269
270static void __kernel_shutdown(void) {
271        if(!cfa_main_returned) return;
272        /* paranoid */ verify( __preemption_enabled() );
273        disable_interrupts();
274        /* paranoid */ verify( ! __preemption_enabled() );
275
276        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
277
278        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
279        // When its coroutine terminates, it return control to the mainThread
280        // which is currently here
281        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
282        __kernel_last_resume( __cfaabi_tls.this_processor );
283        mainThread->self_cor.state = Halted;
284
285        // THE SYSTEM IS NOW COMPLETELY STOPPED
286
287        // Disable preemption
288        __kernel_alarm_shutdown();
289
290        #if !defined( __CFA_NO_STATISTICS__ )
291                __stats_t * st = (__stats_t *)& storage_mainProcStats;
292                __tally_stats(mainCluster->stats, st);
293                if( 0 != mainProcessor->print_stats ) {
294                        __print_stats( st, mainProcessor->print_stats, "Processor ", mainProcessor->name, (void*)mainProcessor );
295                }
296                #if defined(CFA_STATS_ARRAY)
297                        __flush_stat( st, "Processor", mainProcessor );
298                #endif
299        #endif
300
301        mainProcessor->local_data = 0p;
302
303        unregister_tls( mainProcessor );
304
305        // Destroy the main processor and its context in reverse order of construction
306        // These were manually constructed so we need manually destroy them
307        void ^?{}(processor & this) with( this ){
308                deinit( this );
309
310                /* paranoid */ verify( this.do_terminate == true );
311                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
312        }
313
314        ^(*mainProcessor){};
315
316        // Final step, destroy the main thread since it is no longer needed
317
318        // Since we provided a stack to this taxk it will not destroy anything
319        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
320        ^(*mainThread){};
321
322        ^(*mainCluster){};
323
324        ^(*__scheduler_lock){};
325
326        ^(__cfa_dbg_global_clusters.list){};
327        ^(__cfa_dbg_global_clusters.lock){};
328
329        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
330}
331
332//=============================================================================================
333// Kernel Initial Scheduling logic
334//=============================================================================================
335
336// Context invoker for processors
337// This is the entry point for processors (kernel threads) *except* for the main processor
338// It effectively constructs a coroutine by stealing the pthread stack
339static void * __invoke_processor(void * arg) {
340        #if !defined( __CFA_NO_STATISTICS__ )
341                __stats_t local_stats;
342                __init_stats( &local_stats );
343                __cfaabi_tls.this_stats = &local_stats;
344        #endif
345
346        processor * proc = (processor *) arg;
347        __cfaabi_tls.this_processor = proc;
348        __cfaabi_tls.this_thread    = 0p;
349        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
350        proc->local_data = &__cfaabi_tls;
351
352        register_tls( proc );
353
354        // SKULLDUGGERY: We want to create a context for the processor coroutine
355        // which is needed for the 2-step context switch. However, there is no reason
356        // to waste the perfectly valid stack create by pthread.
357        current_stack_info_t info;
358        __stack_t ctx;
359        info.storage = &ctx;
360        (proc->runner){ proc, &info };
361
362        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
363
364        //Set global state
365        __cfaabi_tls.this_thread = 0p;
366
367        //We now have a proper context from which to schedule threads
368        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
369
370        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
371        // resume it to start it like it normally would, it will just context switch
372        // back to here. Instead directly call the main since we already are on the
373        // appropriate stack.
374        get_coroutine(proc->runner)->state = Active;
375        main( proc->runner );
376        get_coroutine(proc->runner)->state = Halted;
377
378        // Main routine of the core returned, the core is now fully terminated
379        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
380
381        #if !defined(__CFA_NO_STATISTICS__)
382                __tally_stats(proc->cltr->stats, &local_stats);
383                if( 0 != proc->print_stats ) {
384                        __print_stats( &local_stats, proc->print_stats, "Processor ", proc->name, (void*)proc );
385                }
386                #if defined(CFA_STATS_ARRAY)
387                        __flush_stat( &local_stats, "Processor", proc );
388                #endif
389        #endif
390
391        proc->local_data = 0p;
392
393        unregister_tls( proc );
394
395        return 0p;
396}
397
398static void __kernel_first_resume( processor * this ) {
399        thread$ * src = mainThread;
400        coroutine$ * dst = get_coroutine(this->runner);
401
402        /* paranoid */ verify( ! __preemption_enabled() );
403
404        __cfaabi_tls.this_thread->curr_cor = dst;
405        __stack_prepare( &dst->stack, 65000 );
406        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
407
408        /* paranoid */ verify( ! __preemption_enabled() );
409
410        dst->last = &src->self_cor;
411        dst->starter = dst->starter ? dst->starter : &src->self_cor;
412
413        // make sure the current state is still correct
414        /* paranoid */ verify(src->state == Ready);
415        src->corctx_flag = true;
416
417        // context switch to specified coroutine
418        verify( dst->context.SP );
419        __cfactx_switch( &src->context, &dst->context );
420        // when __cfactx_switch returns we are back in the src coroutine
421
422        mainThread->curr_cor = &mainThread->self_cor;
423
424        // make sure the current state has been update
425        /* paranoid */ verify(src->state == Active);
426
427        /* paranoid */ verify( ! __preemption_enabled() );
428}
429
430// KERNEL_ONLY
431static void __kernel_last_resume( processor * this ) {
432        coroutine$ * src = &mainThread->self_cor;
433        coroutine$ * dst = get_coroutine(this->runner);
434
435        /* paranoid */ verify( ! __preemption_enabled() );
436        /* paranoid */ verify( dst->starter == src );
437        /* paranoid */ verify( dst->context.SP );
438
439        // SKULLDUGGERY in debug the processors check that the
440        // stack is still within the limit of the stack limits after running a thread.
441        // that check doesn't make sense if we context switch to the processor using the
442        // coroutine semantics. Since this is a special case, use the current context
443        // info to populate these fields.
444        __cfaabi_dbg_debug_do(
445                __stack_context_t ctx;
446                CtxGet( ctx );
447                mainThread->context.SP = ctx.SP;
448                mainThread->context.FP = ctx.FP;
449        )
450
451        // context switch to the processor
452        __cfactx_switch( &src->context, &dst->context );
453}
454
455
456//=============================================================================================
457// Kernel Object Constructors logic
458//=============================================================================================
459//-----------------------------------------------------------------------------
460// Main thread construction
461static void ?{}( coroutine$ & this, current_stack_info_t * info) with( this ) {
462        stack.storage = info->storage;
463        with(*stack.storage) {
464                limit     = info->limit;
465                base      = info->base;
466        }
467        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
468        *istorage |= 0x1;
469        name = "Main Thread";
470        state = Start;
471        starter = 0p;
472        last = 0p;
473        cancellation = 0p;
474}
475
476static void ?{}( thread$ & this, current_stack_info_t * info) with( this ) {
477        ticket = TICKET_RUNNING;
478        state = Start;
479        self_cor{ info };
480        curr_cor = &self_cor;
481        curr_cluster = mainCluster;
482        self_mon.owner = &this;
483        self_mon.recursion = 1;
484        self_mon_p = &self_mon;
485        link.next = 0p;
486        link.ts   = -1llu;
487        preferred = ready_queue_new_preferred();
488        last_proc = 0p;
489        #if defined( __CFA_WITH_VERIFY__ )
490                canary = 0x0D15EA5E0D15EA5Ep;
491        #endif
492
493        node.next = 0p;
494        node.prev = 0p;
495        doregister(curr_cluster, this);
496
497        monitors{ &self_mon_p, 1, (fptr_t)0 };
498}
499
500//-----------------------------------------------------------------------------
501// Processor
502// Construct the processor context of non-main processors
503static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
504        (this.__cor){ info };
505        this.proc = proc;
506}
507
508static void init(processor & this, const char name[], cluster & _cltr, thread$ * initT) with( this ) {
509        this.name = name;
510        this.cltr = &_cltr;
511        this.rdq.its = 0;
512        this.rdq.itr = 0;
513        this.rdq.id  = -1u;
514        this.rdq.target = -1u;
515        this.rdq.last = -1u;
516        this.rdq.cutoff = 0ull;
517        do_terminate = false;
518        preemption_alarm = 0p;
519        pending_preemption = false;
520
521        this.io.ctx = 0p;
522        this.io.pending = false;
523        this.io.dirty   = false;
524
525        this.init.thrd = initT;
526
527        this.local_data = 0p;
528
529        this.idle_fd = eventfd(0, 0);
530        if (idle_fd < 0) {
531                abort("KERNEL ERROR: PROCESSOR EVENTFD - %s\n", strerror(errno));
532        }
533
534        #if !defined(__CFA_NO_STATISTICS__)
535                print_stats = 0;
536                print_halts = false;
537        #endif
538
539        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
540}
541
542// Not a ctor, it just preps the destruction but should not destroy members
543static void deinit(processor & this) {
544        close(this.idle_fd);
545}
546
547void ?{}(processor & this, const char name[], cluster & _cltr, thread$ * initT) {
548        ( this.terminated ){};
549        ( this.runner ){};
550
551        disable_interrupts();
552                init( this, name, _cltr, initT );
553        enable_interrupts();
554
555        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
556
557        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
558}
559
560void ?{}(processor & this, const char name[], cluster & _cltr) {
561        (this){name, _cltr, 0p};
562}
563
564extern size_t __page_size;
565void ^?{}(processor & this) with( this ){
566        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
567                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
568
569                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
570                __wake_proc( &this );
571
572                wait( terminated );
573                /* paranoid */ verify( active_processor() != &this);
574        }
575
576        __destroy_pthread( kernel_thread, this.stack, 0p );
577
578        disable_interrupts();
579                deinit( this );
580        enable_interrupts();
581}
582
583//-----------------------------------------------------------------------------
584// Cluster
585static void ?{}(__cluster_proc_list & this) {
586        this.fd    = 0;
587        this.idle  = 0;
588        this.total = 0;
589}
590
591void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) with( this ) {
592        this.name = name;
593        this.preemption_rate = preemption_rate;
594        ready_queue{};
595
596        #if !defined(__CFA_NO_STATISTICS__)
597                print_stats = 0;
598                stats = alloc();
599                __init_stats( stats );
600        #endif
601
602        threads{ __get };
603
604        io.arbiter = create();
605        io.params = io_params;
606
607        doregister(this);
608
609        // Lock the RWlock so no-one pushes/pops while we are changing the queue
610        disable_interrupts();
611        uint_fast32_t last_size = ready_mutate_lock();
612
613                // Adjust the ready queue size
614                ready_queue_grow( &this );
615
616        // Unlock the RWlock
617        ready_mutate_unlock( last_size );
618        enable_interrupts( false ); // Don't poll, could be in main cluster
619}
620
621void ^?{}(cluster & this) {
622        destroy(this.io.arbiter);
623
624        // Lock the RWlock so no-one pushes/pops while we are changing the queue
625        disable_interrupts();
626        uint_fast32_t last_size = ready_mutate_lock();
627
628                // Adjust the ready queue size
629                ready_queue_shrink( &this );
630
631        // Unlock the RWlock
632        ready_mutate_unlock( last_size );
633        enable_interrupts( false ); // Don't poll, could be in main cluster
634
635        #if !defined(__CFA_NO_STATISTICS__)
636                if( 0 != this.print_stats ) {
637                        __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
638                }
639                #if defined(CFA_STATS_ARRAY)
640                        __flush_stat( this.stats, "Cluster", &this );
641                #endif
642                free( this.stats );
643        #endif
644
645        unregister(this);
646}
647
648//=============================================================================================
649// Miscellaneous Initialization
650//=============================================================================================
651//-----------------------------------------------------------------------------
652// Global Queues
653static void doregister( cluster     & cltr ) {
654        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
655        push_front( __cfa_dbg_global_clusters.list, cltr );
656        unlock    ( __cfa_dbg_global_clusters.lock );
657}
658
659static void unregister( cluster     & cltr ) {
660        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
661        remove( __cfa_dbg_global_clusters.list, cltr );
662        unlock( __cfa_dbg_global_clusters.lock );
663}
664
665void doregister( cluster * cltr, thread$ & thrd ) {
666        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
667        cltr->nthreads += 1;
668        push_front(cltr->threads, thrd);
669        unlock    (cltr->thread_list_lock);
670}
671
672void unregister( cluster * cltr, thread$ & thrd ) {
673        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
674        remove(cltr->threads, thrd );
675        cltr->nthreads -= 1;
676        unlock(cltr->thread_list_lock);
677}
678
679static void register_tls( processor * this ) {
680        // Register and Lock the RWlock so no-one pushes/pops while we are changing the queue
681        uint_fast32_t last_size;
682        [this->unique_id, last_size] = ready_mutate_register();
683
684                this->cltr->procs.total += 1u;
685                insert_last(this->cltr->procs.actives, *this);
686
687                // Adjust the ready queue size
688                ready_queue_grow( this->cltr );
689
690        // Unlock the RWlock
691        ready_mutate_unlock( last_size );
692}
693
694
695static void unregister_tls( processor * this ) {
696        // Lock the RWlock so no-one pushes/pops while we are changing the queue
697        uint_fast32_t last_size = ready_mutate_lock();
698                this->cltr->procs.total -= 1u;
699                remove(*this);
700
701                // clear the cluster so nothing gets pushed to local queues
702                cluster * cltr = this->cltr;
703                this->cltr = 0p;
704
705                // Adjust the ready queue size
706                ready_queue_shrink( cltr );
707
708        // Unlock the RWlock and unregister: we don't need the read_lock any more
709        ready_mutate_unregister( this->unique_id, last_size );
710}
711
712static void check( int ret, const char func[] ) {
713        if ( ret ) {                                                                            // pthread routines return errno values
714                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
715        } // if
716} // Abort
717
718void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
719        pthread_attr_t attr;
720
721        check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
722
723        size_t stacksize;
724        // default stack size, normally defined by shell limit
725        check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
726        assert( stacksize >= PTHREAD_STACK_MIN );
727
728        void * stack;
729        #if CFA_PROCESSOR_USE_MMAP
730                stacksize = ceiling( stacksize, __page_size ) + __page_size;
731                stack = mmap(0p, stacksize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
732                if(stack == ((void*)-1)) {
733                        abort( "pthread stack creation : internal error, mmap failure, error(%d) %s.", errno, strerror( errno ) );
734                }
735                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
736                        abort( "pthread stack creation : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
737                } // if
738        #else
739                __cfaabi_dbg_debug_do(
740                        stack = memalign( __page_size, stacksize + __page_size );
741                        // pthread has no mechanism to create the guard page in user supplied stack.
742                        if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
743                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
744                        } // if
745                );
746                __cfaabi_dbg_no_debug_do(
747                        stack = malloc( stacksize );
748                );
749        #endif
750
751
752        check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
753
754        check( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
755        return stack;
756}
757
758void __destroy_pthread( pthread_t pthread, void * stack, void ** retval ) {
759        int err = pthread_join( pthread, retval );
760        if( err != 0 ) abort("KERNEL ERROR: joining pthread %p caused error %s\n", (void*)pthread, strerror(err));
761
762        #if CFA_PROCESSOR_USE_MMAP
763                pthread_attr_t attr;
764
765                check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
766
767                size_t stacksize;
768                // default stack size, normally defined by shell limit
769                check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
770                assert( stacksize >= PTHREAD_STACK_MIN );
771                stacksize += __page_size;
772
773                if(munmap(stack, stacksize) == -1) {
774                        abort( "pthread stack destruction : internal error, munmap failure, error(%d) %s.", errno, strerror( errno ) );
775                }
776        #else
777                __cfaabi_dbg_debug_do(
778                        // pthread has no mechanism to create the guard page in user supplied stack.
779                        if ( mprotect( stack, __page_size, __map_prot ) == -1 ) {
780                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
781                        } // if
782                );
783                free( stack );
784        #endif
785}
786
787#if defined(__CFA_WITH_VERIFY__)
788static bool verify_fwd_bck_rng(void) {
789        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
790
791        unsigned values[10];
792        for(i; 10) {
793                values[i] = __tls_rand_fwd();
794        }
795
796        __tls_rand_advance_bck();
797
798        for ( i; 9 -~= 0 ) {
799                if(values[i] != __tls_rand_bck()) {
800                        return false;
801                }
802        }
803
804        return true;
805}
806#endif
Note: See TracBrowser for help on using the repository browser.