source: libcfa/src/concurrency/kernel/startup.cfa

Last change on this file was 24e321c, checked in by Thierry Delisle <tdelisle@…>, 3 weeks ago

Unpark now takes a hint on locality.

  • Property mode set to 100644
File size: 24.6 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 *);
102
103//-----------------------------------------------------------------------------
104// Kernel storage
105KERNEL_STORAGE(cluster,              mainCluster);
106KERNEL_STORAGE(processor,            mainProcessor);
107KERNEL_STORAGE(thread$,              mainThread);
108KERNEL_STORAGE(__stack_t,            mainThreadCtx);
109KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock);
110#if !defined(__CFA_NO_STATISTICS__)
111KERNEL_STORAGE(__stats_t, mainProcStats);
112#endif
113
114cluster              * mainCluster;
115processor            * mainProcessor;
116thread$              * mainThread;
117__scheduler_RWLock_t * __scheduler_lock;
118
119extern "C" {
120        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
121}
122
123extern size_t __page_size;
124extern int __map_prot;
125
126//-----------------------------------------------------------------------------
127// Global state
128thread_local struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
129        NULL,                                                                                           // cannot use 0p
130        NULL,
131        false,
132        { 1, false, false },
133        0,
134        { 0, 0 },
135        NULL,
136        #ifdef __CFA_WITH_VERIFY__
137                false,
138                0,
139        #endif
140};
141
142#if   defined(CFA_HAVE_LINUX_LIBRSEQ)
143        // No data needed
144#elif defined(CFA_HAVE_LINUX_RSEQ_H)
145        extern "Cforall" {
146                __attribute__((aligned(128))) thread_local volatile struct rseq __cfaabi_rseq @= {
147                        .cpu_id : RSEQ_CPU_ID_UNINITIALIZED,
148                };
149        }
150#else
151        // No data needed
152#endif
153
154//-----------------------------------------------------------------------------
155// Struct to steal stack
156struct current_stack_info_t {
157        __stack_t * storage;  // pointer to stack object
158        void * base;          // base of stack
159        void * limit;         // stack grows towards stack limit
160        void * context;       // address of cfa_context_t
161};
162
163void ?{}( current_stack_info_t & this ) {
164        __stack_context_t ctx;
165        CtxGet( ctx );
166        this.base = ctx.FP;
167
168        rlimit r;
169        getrlimit( RLIMIT_STACK, &r);
170        size_t size = r.rlim_cur;
171
172        this.limit = (void *)(((intptr_t)this.base) - size);
173        this.context = &storage_mainThreadCtx;
174}
175
176
177
178//=============================================================================================
179// Kernel Setup logic
180//=============================================================================================
181//-----------------------------------------------------------------------------
182// Kernel boot procedures
183static void __kernel_startup(void) {
184        /* paranoid */ verify( ! __preemption_enabled() );
185        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
186
187        __cfa_dbg_global_clusters.list{ __get };
188        __cfa_dbg_global_clusters.lock{};
189
190        /* paranoid */ verify( verify_fwd_bck_rng() );
191
192        // Initialize the global scheduler lock
193        __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock;
194        (*__scheduler_lock){};
195
196        // Initialize the main cluster
197        mainCluster = (cluster *)&storage_mainCluster;
198        (*mainCluster){"Main Cluster", 0};
199
200        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
201
202        // Construct the processor context of the main processor
203        void ?{}(processorCtx_t & this, processor * proc) {
204                (this.__cor){ "Processor" };
205                this.__cor.starter = 0p;
206                this.proc = proc;
207        }
208
209        void ?{}(processor & this) with( this ) {
210                ( this.terminated ){};
211                ( this.runner ){};
212                init( this, "Main Processor", *mainCluster, 0p );
213                kernel_thread = pthread_self();
214
215                runner{ &this };
216                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
217        }
218
219        // Initialize the main processor and the main processor ctx
220        // (the coroutine that contains the processing control flow)
221        mainProcessor = (processor *)&storage_mainProcessor;
222        (*mainProcessor){};
223
224        register_tls( mainProcessor );
225
226        // Start by initializing the main thread
227        // SKULLDUGGERY: the mainThread steals the process main thread
228        // which will then be scheduled by the mainProcessor normally
229        mainThread = (thread$ *)&storage_mainThread;
230        current_stack_info_t info;
231        info.storage = (__stack_t*)&storage_mainThreadCtx;
232        (*mainThread){ &info };
233
234        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
235
236        //initialize the global state variables
237        __cfaabi_tls.this_processor = mainProcessor;
238        __cfaabi_tls.this_thread    = mainThread;
239
240        #if !defined( __CFA_NO_STATISTICS__ )
241                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
242                __init_stats( __cfaabi_tls.this_stats );
243        #endif
244        mainProcessor->local_data = &__cfaabi_tls;
245
246        // Enable preemption
247        __kernel_alarm_startup();
248
249        // Add the main thread to the ready queue
250        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
251        schedule_thread$(mainThread, UNPARK_LOCAL);
252
253        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
254        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
255        // mainThread is on the ready queue when this call is made.
256        __kernel_first_resume( __cfaabi_tls.this_processor );
257
258
259        // THE SYSTEM IS NOW COMPLETELY RUNNING
260
261        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
262
263        /* paranoid */ verify( ! __preemption_enabled() );
264        enable_interrupts();
265        /* paranoid */ verify( __preemption_enabled() );
266
267}
268
269static void __kernel_shutdown(void) {
270        /* paranoid */ verify( __preemption_enabled() );
271        disable_interrupts();
272        /* paranoid */ verify( ! __preemption_enabled() );
273
274        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
275
276        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
277        // When its coroutine terminates, it return control to the mainThread
278        // which is currently here
279        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
280        __kernel_last_resume( __cfaabi_tls.this_processor );
281        mainThread->self_cor.state = Halted;
282
283        // THE SYSTEM IS NOW COMPLETELY STOPPED
284
285        // Disable preemption
286        __kernel_alarm_shutdown();
287
288        #if !defined( __CFA_NO_STATISTICS__ )
289                __stats_t * st = (__stats_t *)& storage_mainProcStats;
290                __tally_stats(mainCluster->stats, st);
291                if( 0 != mainProcessor->print_stats ) {
292                        __print_stats( st, mainProcessor->print_stats, "Processor ", mainProcessor->name, (void*)mainProcessor );
293                }
294                #if defined(CFA_STATS_ARRAY)
295                        __flush_stat( st, "Processor", mainProcessor );
296                #endif
297        #endif
298
299        mainProcessor->local_data = 0p;
300
301        unregister_tls( mainProcessor );
302
303        // Destroy the main processor and its context in reverse order of construction
304        // These were manually constructed so we need manually destroy them
305        void ^?{}(processor & this) with( this ){
306                deinit( this );
307
308                /* paranoid */ verify( this.do_terminate == true );
309                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
310        }
311
312        ^(*mainProcessor){};
313
314        // Final step, destroy the main thread since it is no longer needed
315
316        // Since we provided a stack to this taxk it will not destroy anything
317        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
318        ^(*mainThread){};
319
320        ^(*mainCluster){};
321
322        ^(*__scheduler_lock){};
323
324        ^(__cfa_dbg_global_clusters.list){};
325        ^(__cfa_dbg_global_clusters.lock){};
326
327        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
328}
329
330//=============================================================================================
331// Kernel Initial Scheduling logic
332//=============================================================================================
333
334// Context invoker for processors
335// This is the entry point for processors (kernel threads) *except* for the main processor
336// It effectively constructs a coroutine by stealing the pthread stack
337static void * __invoke_processor(void * arg) {
338        #if !defined( __CFA_NO_STATISTICS__ )
339                __stats_t local_stats;
340                __init_stats( &local_stats );
341                __cfaabi_tls.this_stats = &local_stats;
342        #endif
343
344        processor * proc = (processor *) arg;
345        __cfaabi_tls.this_processor = proc;
346        __cfaabi_tls.this_thread    = 0p;
347        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
348        proc->local_data = &__cfaabi_tls;
349
350        register_tls( proc );
351
352        // SKULLDUGGERY: We want to create a context for the processor coroutine
353        // which is needed for the 2-step context switch. However, there is no reason
354        // to waste the perfectly valid stack create by pthread.
355        current_stack_info_t info;
356        __stack_t ctx;
357        info.storage = &ctx;
358        (proc->runner){ proc, &info };
359
360        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
361
362        //Set global state
363        __cfaabi_tls.this_thread = 0p;
364
365        //We now have a proper context from which to schedule threads
366        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
367
368        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
369        // resume it to start it like it normally would, it will just context switch
370        // back to here. Instead directly call the main since we already are on the
371        // appropriate stack.
372        get_coroutine(proc->runner)->state = Active;
373        main( proc->runner );
374        get_coroutine(proc->runner)->state = Halted;
375
376        // Main routine of the core returned, the core is now fully terminated
377        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
378
379        #if !defined(__CFA_NO_STATISTICS__)
380                __tally_stats(proc->cltr->stats, &local_stats);
381                if( 0 != proc->print_stats ) {
382                        __print_stats( &local_stats, proc->print_stats, "Processor ", proc->name, (void*)proc );
383                }
384                #if defined(CFA_STATS_ARRAY)
385                        __flush_stat( &local_stats, "Processor", proc );
386                #endif
387        #endif
388
389        proc->local_data = 0p;
390
391        unregister_tls( proc );
392
393        return 0p;
394}
395
396static void __kernel_first_resume( processor * this ) {
397        thread$ * src = mainThread;
398        coroutine$ * dst = get_coroutine(this->runner);
399
400        /* paranoid */ verify( ! __preemption_enabled() );
401
402        __cfaabi_tls.this_thread->curr_cor = dst;
403        __stack_prepare( &dst->stack, 65000 );
404        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
405
406        /* paranoid */ verify( ! __preemption_enabled() );
407
408        dst->last = &src->self_cor;
409        dst->starter = dst->starter ? dst->starter : &src->self_cor;
410
411        // make sure the current state is still correct
412        /* paranoid */ verify(src->state == Ready);
413        src->corctx_flag = true;
414
415        // context switch to specified coroutine
416        verify( dst->context.SP );
417        __cfactx_switch( &src->context, &dst->context );
418        // when __cfactx_switch returns we are back in the src coroutine
419
420        mainThread->curr_cor = &mainThread->self_cor;
421
422        // make sure the current state has been update
423        /* paranoid */ verify(src->state == Active);
424
425        /* paranoid */ verify( ! __preemption_enabled() );
426}
427
428// KERNEL_ONLY
429static void __kernel_last_resume( processor * this ) {
430        coroutine$ * src = &mainThread->self_cor;
431        coroutine$ * dst = get_coroutine(this->runner);
432
433        /* paranoid */ verify( ! __preemption_enabled() );
434        /* paranoid */ verify( dst->starter == src );
435        /* paranoid */ verify( dst->context.SP );
436
437        // SKULLDUGGERY in debug the processors check that the
438        // stack is still within the limit of the stack limits after running a thread.
439        // that check doesn't make sense if we context switch to the processor using the
440        // coroutine semantics. Since this is a special case, use the current context
441        // info to populate these fields.
442        __cfaabi_dbg_debug_do(
443                __stack_context_t ctx;
444                CtxGet( ctx );
445                mainThread->context.SP = ctx.SP;
446                mainThread->context.FP = ctx.FP;
447        )
448
449        // context switch to the processor
450        __cfactx_switch( &src->context, &dst->context );
451}
452
453
454//=============================================================================================
455// Kernel Object Constructors logic
456//=============================================================================================
457//-----------------------------------------------------------------------------
458// Main thread construction
459static void ?{}( coroutine$ & this, current_stack_info_t * info) with( this ) {
460        stack.storage = info->storage;
461        with(*stack.storage) {
462                limit     = info->limit;
463                base      = info->base;
464        }
465        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
466        *istorage |= 0x1;
467        name = "Main Thread";
468        state = Start;
469        starter = 0p;
470        last = 0p;
471        cancellation = 0p;
472}
473
474static void ?{}( thread$ & this, current_stack_info_t * info) with( this ) {
475        ticket = TICKET_RUNNING;
476        state = Start;
477        self_cor{ info };
478        curr_cor = &self_cor;
479        curr_cluster = mainCluster;
480        self_mon.owner = &this;
481        self_mon.recursion = 1;
482        self_mon_p = &self_mon;
483        link.next = 0p;
484        link.ts   = -1llu;
485        preferred = ready_queue_new_preferred();
486        last_proc = 0p;
487        #if defined( __CFA_WITH_VERIFY__ )
488                canary = 0x0D15EA5E0D15EA5Ep;
489        #endif
490
491        node.next = 0p;
492        node.prev = 0p;
493        doregister(curr_cluster, this);
494
495        monitors{ &self_mon_p, 1, (fptr_t)0 };
496}
497
498//-----------------------------------------------------------------------------
499// Processor
500// Construct the processor context of non-main processors
501static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
502        (this.__cor){ info };
503        this.proc = proc;
504}
505
506static void init(processor & this, const char name[], cluster & _cltr, thread$ * initT) with( this ) {
507        this.name = name;
508        this.cltr = &_cltr;
509        this.rdq.its = 0;
510        this.rdq.itr = 0;
511        this.rdq.id  = -1u;
512        this.rdq.target = -1u;
513        this.rdq.last = -1u;
514        this.rdq.cutoff = 0ull;
515        do_terminate = false;
516        preemption_alarm = 0p;
517        pending_preemption = false;
518
519        this.io.ctx = 0p;
520        this.io.pending = false;
521        this.io.dirty   = false;
522
523        this.init.thrd = initT;
524
525        this.local_data = 0p;
526
527        this.idle = eventfd(0, 0);
528        if (idle < 0) {
529                abort("KERNEL ERROR: PROCESSOR EVENTFD - %s\n", strerror(errno));
530        }
531
532        #if !defined(__CFA_NO_STATISTICS__)
533                print_stats = 0;
534                print_halts = false;
535        #endif
536
537        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
538}
539
540// Not a ctor, it just preps the destruction but should not destroy members
541static void deinit(processor & this) {
542        close(this.idle);
543}
544
545void ?{}(processor & this, const char name[], cluster & _cltr, thread$ * initT) {
546        ( this.terminated ){};
547        ( this.runner ){};
548
549        disable_interrupts();
550                init( this, name, _cltr, initT );
551        enable_interrupts();
552
553        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
554
555        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
556}
557
558void ?{}(processor & this, const char name[], cluster & _cltr) {
559        (this){name, _cltr, 0p};
560}
561
562extern size_t __page_size;
563void ^?{}(processor & this) with( this ){
564        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
565                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
566
567                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
568                __wake_proc( &this );
569
570                wait( terminated );
571                /* paranoid */ verify( active_processor() != &this);
572        }
573
574        __destroy_pthread( kernel_thread, this.stack, 0p );
575
576        disable_interrupts();
577                deinit( this );
578        enable_interrupts();
579}
580
581//-----------------------------------------------------------------------------
582// Cluster
583static void ?{}(__cluster_proc_list & this) {
584        this.lock  = 0;
585        this.idle  = 0;
586        this.total = 0;
587}
588
589void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) with( this ) {
590        this.name = name;
591        this.preemption_rate = preemption_rate;
592        ready_queue{};
593
594        #if !defined(__CFA_NO_STATISTICS__)
595                print_stats = 0;
596                stats = alloc();
597                __init_stats( stats );
598        #endif
599
600        threads{ __get };
601
602        io.arbiter = create();
603        io.params = io_params;
604
605        doregister(this);
606
607        // Lock the RWlock so no-one pushes/pops while we are changing the queue
608        disable_interrupts();
609        uint_fast32_t last_size = ready_mutate_lock();
610
611                // Adjust the ready queue size
612                ready_queue_grow( &this );
613
614        // Unlock the RWlock
615        ready_mutate_unlock( last_size );
616        enable_interrupts( false ); // Don't poll, could be in main cluster
617}
618
619void ^?{}(cluster & this) {
620        destroy(this.io.arbiter);
621
622        // Lock the RWlock so no-one pushes/pops while we are changing the queue
623        disable_interrupts();
624        uint_fast32_t last_size = ready_mutate_lock();
625
626                // Adjust the ready queue size
627                ready_queue_shrink( &this );
628
629        // Unlock the RWlock
630        ready_mutate_unlock( last_size );
631        enable_interrupts( false ); // Don't poll, could be in main cluster
632
633        #if !defined(__CFA_NO_STATISTICS__)
634                if( 0 != this.print_stats ) {
635                        __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
636                }
637                #if defined(CFA_STATS_ARRAY)
638                        __flush_stat( this.stats, "Cluster", &this );
639                #endif
640                free( this.stats );
641        #endif
642
643        unregister(this);
644}
645
646//=============================================================================================
647// Miscellaneous Initialization
648//=============================================================================================
649//-----------------------------------------------------------------------------
650// Global Queues
651static void doregister( cluster     & cltr ) {
652        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
653        push_front( __cfa_dbg_global_clusters.list, cltr );
654        unlock    ( __cfa_dbg_global_clusters.lock );
655}
656
657static void unregister( cluster     & cltr ) {
658        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
659        remove( __cfa_dbg_global_clusters.list, cltr );
660        unlock( __cfa_dbg_global_clusters.lock );
661}
662
663void doregister( cluster * cltr, thread$ & thrd ) {
664        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
665        cltr->nthreads += 1;
666        push_front(cltr->threads, thrd);
667        unlock    (cltr->thread_list_lock);
668}
669
670void unregister( cluster * cltr, thread$ & thrd ) {
671        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
672        remove(cltr->threads, thrd );
673        cltr->nthreads -= 1;
674        unlock(cltr->thread_list_lock);
675}
676
677static void register_tls( processor * this ) {
678        // Register and Lock the RWlock so no-one pushes/pops while we are changing the queue
679        uint_fast32_t last_size;
680        [this->unique_id, last_size] = ready_mutate_register();
681
682                this->cltr->procs.total += 1u;
683                insert_last(this->cltr->procs.actives, *this);
684
685                // Adjust the ready queue size
686                ready_queue_grow( this->cltr );
687
688        // Unlock the RWlock
689        ready_mutate_unlock( last_size );
690}
691
692
693static void unregister_tls( processor * this ) {
694        // Lock the RWlock so no-one pushes/pops while we are changing the queue
695        uint_fast32_t last_size = ready_mutate_lock();
696                this->cltr->procs.total -= 1u;
697                remove(*this);
698
699                // clear the cluster so nothing gets pushed to local queues
700                cluster * cltr = this->cltr;
701                this->cltr = 0p;
702
703                // Adjust the ready queue size
704                ready_queue_shrink( cltr );
705
706        // Unlock the RWlock and unregister: we don't need the read_lock any more
707        ready_mutate_unregister( this->unique_id, last_size );
708}
709
710static void check( int ret, const char func[] ) {
711        if ( ret ) {                                                                            // pthread routines return errno values
712                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
713        } // if
714} // Abort
715
716void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
717        pthread_attr_t attr;
718
719        check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
720
721        size_t stacksize;
722        // default stack size, normally defined by shell limit
723        check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
724        assert( stacksize >= PTHREAD_STACK_MIN );
725
726        void * stack;
727        #if CFA_PROCESSOR_USE_MMAP
728                stacksize = ceiling( stacksize, __page_size ) + __page_size;
729                stack = mmap(0p, stacksize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
730                if(stack == ((void*)-1)) {
731                        abort( "pthread stack creation : internal error, mmap failure, error(%d) %s.", errno, strerror( errno ) );
732                }
733                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
734                        abort( "pthread stack creation : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
735                } // if
736        #else
737                __cfaabi_dbg_debug_do(
738                        stack = memalign( __page_size, stacksize + __page_size );
739                        // pthread has no mechanism to create the guard page in user supplied stack.
740                        if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
741                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
742                        } // if
743                );
744                __cfaabi_dbg_no_debug_do(
745                        stack = malloc( stacksize );
746                );
747        #endif
748
749
750        check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
751
752        check( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
753        return stack;
754}
755
756void __destroy_pthread( pthread_t pthread, void * stack, void ** retval ) {
757        int err = pthread_join( pthread, retval );
758        if( err != 0 ) abort("KERNEL ERROR: joining pthread %p caused error %s\n", (void*)pthread, strerror(err));
759
760        #if CFA_PROCESSOR_USE_MMAP
761                pthread_attr_t attr;
762
763                check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
764
765                size_t stacksize;
766                // default stack size, normally defined by shell limit
767                check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
768                assert( stacksize >= PTHREAD_STACK_MIN );
769                stacksize += __page_size;
770
771                if(munmap(stack, stacksize) == -1) {
772                        abort( "pthread stack destruction : internal error, munmap failure, error(%d) %s.", errno, strerror( errno ) );
773                }
774        #else
775                __cfaabi_dbg_debug_do(
776                        // pthread has no mechanism to create the guard page in user supplied stack.
777                        if ( mprotect( stack, __page_size, __map_prot ) == -1 ) {
778                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
779                        } // if
780                );
781                free( stack );
782        #endif
783}
784
785#if defined(__CFA_WITH_VERIFY__)
786static bool verify_fwd_bck_rng(void) {
787        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
788
789        unsigned values[10];
790        for(i; 10) {
791                values[i] = __tls_rand_fwd();
792        }
793
794        __tls_rand_advance_bck();
795
796        for ( i; 9 -~= 0 ) {
797                if(values[i] != __tls_rand_bck()) {
798                        return false;
799                }
800        }
801
802        return true;
803}
804#endif
Note: See TracBrowser for help on using the repository browser.