source: libcfa/src/concurrency/kernel/startup.cfa @ 9f99799

ADTast-experimentalpthread-emulation
Last change on this file since 9f99799 was 1bcbf02, checked in by Thierry Delisle <tdelisle@…>, 2 years ago

Changed declarations using _Thread_local to use thread.
I'm fairly sure they do exactly the same, but
thread is:

  • more consistent with uC++.
  • Shorter.
  • Better documented.
  • Property mode set to 100644
File size: 26.4 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#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 "iofwd.hfa"
36#include "startup.hfa"                                  // STARTUP_PRIORITY_XXX
37#include "limits.hfa"
38#include "math.hfa"
39
40#define CFA_PROCESSOR_USE_MMAP 0
41
42//-----------------------------------------------------------------------------
43// Some assembly required
44#if defined( __i386 )
45        #define CtxGet( ctx ) __asm__ volatile ( \
46                "movl %%esp,%0\n" \
47                "movl %%ebp,%1\n" \
48                : "=rm" (ctx.SP), \
49                  "=rm" (ctx.FP) \
50        )
51#elif defined( __x86_64 )
52        #define CtxGet( ctx ) __asm__ volatile ( \
53                "movq %%rsp,%0\n" \
54                "movq %%rbp,%1\n" \
55                : "=rm" (ctx.SP), \
56                  "=rm" (ctx.FP) \
57        )
58#elif defined( __aarch64__ )
59        #define CtxGet( ctx ) __asm__ volatile ( \
60                "mov %0, sp\n" \
61                "mov %1, fp\n" \
62                : "=rm" (ctx.SP), \
63                  "=rm" (ctx.FP) \
64        )
65#else
66        #error unknown hardware architecture
67#endif
68
69//-----------------------------------------------------------------------------
70// Start and stop routine for the kernel, declared first to make sure they run first
71static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
72static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
73
74//-----------------------------------------------------------------------------
75// Static Forward Declarations
76struct current_stack_info_t;
77
78static void * __invoke_processor(void * arg);
79static void __kernel_first_resume( processor * this );
80static void __kernel_last_resume ( processor * this );
81static void init(processor & this, const char name[], cluster & _cltr, thread$ * initT);
82static void deinit(processor & this);
83static void doregister( struct cluster & cltr );
84static void unregister( struct cluster & cltr );
85static void register_tls( processor * this );
86static void unregister_tls( processor * this );
87static void ?{}( coroutine$ & this, current_stack_info_t * info);
88static void ?{}( thread$ & this, current_stack_info_t * info);
89static void ?{}(processorCtx_t & this) {}
90static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info);
91
92#if defined(__CFA_WITH_VERIFY__)
93        static bool verify_fwd_bck_rng(void);
94#endif
95
96//-----------------------------------------------------------------------------
97// Forward Declarations for other modules
98extern void __kernel_alarm_startup(void);
99extern void __kernel_alarm_shutdown(void);
100extern void __cfa_io_start( processor * );
101extern void __cfa_io_stop ( processor * );
102
103//-----------------------------------------------------------------------------
104// Other Forward Declarations
105extern void __wake_proc(processor *);
106extern int cfa_main_returned;                                                   // from interpose.cfa
107uint32_t __global_random_prime = 4_294_967_291u, __global_random_mask = false;
108
109//-----------------------------------------------------------------------------
110// Kernel storage
111KERNEL_STORAGE(cluster,              mainCluster);
112KERNEL_STORAGE(processor,            mainProcessor);
113KERNEL_STORAGE(thread$,              mainThread);
114KERNEL_STORAGE(__stack_t,            mainThreadCtx);
115// KERNEL_STORAGE(__scheduler_RWLock_t, __scheduler_lock);
116KERNEL_STORAGE(eventfd_t,            mainIdleEventFd);
117KERNEL_STORAGE(io_future_t,          mainIdleFuture);
118#if !defined(__CFA_NO_STATISTICS__)
119KERNEL_STORAGE(__stats_t, mainProcStats);
120#endif
121
122cluster              * mainCluster libcfa_public;
123processor            * mainProcessor;
124thread$              * mainThread;
125
126extern "C" {
127        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
128}
129
130extern size_t __page_size;
131extern int __map_prot;
132
133//-----------------------------------------------------------------------------
134// Global state
135__thread struct KernelThreadData __cfaabi_tls __attribute__ ((tls_model ( "initial-exec" ))) @= {
136        NULL,                                                                                           // cannot use 0p
137        NULL,
138        false,
139        { 1, false, false },
140        0,
141        { 0, 0 },
142        NULL,
143        #ifdef __CFA_WITH_VERIFY__
144                false,
145                0,
146        #endif
147};
148
149__scheduler_RWLock_t __scheduler_lock @= { 0 };
150
151#if   defined(CFA_HAVE_LINUX_LIBRSEQ)
152        // No data needed
153#elif defined(CFA_HAVE_LINUX_RSEQ_H)
154        extern "Cforall" {
155                __attribute__((aligned(64))) __thread volatile struct rseq __cfaabi_rseq @= {
156                        .cpu_id : RSEQ_CPU_ID_UNINITIALIZED,
157                };
158        }
159#else
160        // No data needed
161#endif
162
163//-----------------------------------------------------------------------------
164// Struct to steal stack
165struct current_stack_info_t {
166        __stack_t * storage;  // pointer to stack object
167        void * base;          // base of stack
168        void * limit;         // stack grows towards stack limit
169        void * context;       // address of cfa_context_t
170};
171
172static void ?{}( current_stack_info_t & this ) {
173        __stack_context_t ctx;
174        CtxGet( ctx );
175        this.base = ctx.FP;
176
177        rlimit r;
178        getrlimit( RLIMIT_STACK, &r);
179        size_t size = r.rlim_cur;
180
181        this.limit = (void *)(((intptr_t)this.base) - size);
182        this.context = &storage_mainThreadCtx;
183}
184
185
186//=============================================================================================
187// Kernel Setup logic
188//=============================================================================================
189//-----------------------------------------------------------------------------
190// Kernel boot procedures
191static void __kernel_startup(void) {
192        /* paranoid */ verify( ! __preemption_enabled() );
193        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
194
195        __cfa_dbg_global_clusters.list{ __get };
196        __cfa_dbg_global_clusters.lock{};
197
198        /* paranoid */ verify( verify_fwd_bck_rng() );
199
200        // Initialize the global scheduler lock
201        // __scheduler_lock = (__scheduler_RWLock_t*)&storage___scheduler_lock;
202        (__scheduler_lock){};
203
204        // Initialize the main cluster
205        mainCluster = (cluster *)&storage_mainCluster;
206        (*mainCluster){"Main Cluster", 0};
207
208        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
209
210        // Construct the processor context of the main processor
211        void ?{}(processorCtx_t & this, processor * proc) {
212                (this.self){ "Processor" };
213                this.self.starter = 0p;
214                this.proc = proc;
215        }
216
217        void ?{}(processor & this) with( this ) {
218                ( this.terminated ){};
219                ( this.runner ){};
220                init( this, "Main Processor", *mainCluster, 0p );
221                kernel_thread = pthread_self();
222
223                runner{ &this };
224                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
225        }
226
227        // Initialize the main processor and the main processor ctx
228        // (the coroutine that contains the processing control flow)
229        mainProcessor = (processor *)&storage_mainProcessor;
230        (*mainProcessor){};
231
232        mainProcessor->idle_wctx.rdbuf = &storage_mainIdleEventFd;
233        mainProcessor->idle_wctx.ftr   = (io_future_t*)&storage_mainIdleFuture;
234        /* paranoid */ verify( sizeof(storage_mainIdleEventFd) == sizeof(eventfd_t) );
235
236        __cfa_io_start( mainProcessor );
237        register_tls( mainProcessor );
238
239        // Start by initializing the main thread
240        // SKULLDUGGERY: the mainThread steals the process main thread
241        // which will then be scheduled by the mainProcessor normally
242        mainThread = (thread$ *)&storage_mainThread;
243        current_stack_info_t info;
244        info.storage = (__stack_t*)&storage_mainThreadCtx;
245        (*mainThread){ &info };
246
247        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
248
249        //initialize the global state variables
250        __cfaabi_tls.this_processor = mainProcessor;
251        __cfaabi_tls.this_thread    = mainThread;
252
253        #if !defined( __CFA_NO_STATISTICS__ )
254                __cfaabi_tls.this_stats = (__stats_t *)& storage_mainProcStats;
255                __init_stats( __cfaabi_tls.this_stats );
256        #endif
257        mainProcessor->local_data = &__cfaabi_tls;
258
259        // Enable preemption
260        __kernel_alarm_startup();
261
262        // Add the main thread to the ready queue
263        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
264        schedule_thread$(mainThread, UNPARK_LOCAL);
265
266        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
267        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
268        // mainThread is on the ready queue when this call is made.
269        __kernel_first_resume( __cfaabi_tls.this_processor );
270
271
272        // THE SYSTEM IS NOW COMPLETELY RUNNING
273
274        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
275
276        /* paranoid */ verify( ! __preemption_enabled() );
277        enable_interrupts();
278        /* paranoid */ verify( __preemption_enabled() );
279
280}
281
282static void __kernel_shutdown(void) {
283        if(!cfa_main_returned) return;
284        /* paranoid */ verify( __preemption_enabled() );
285        disable_interrupts();
286        /* paranoid */ verify( ! __preemption_enabled() );
287
288        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
289
290        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
291        // When its coroutine terminates, it return control to the mainThread
292        // which is currently here
293        /* paranoid */ verify( !__atomic_load_n(&mainProcessor->do_terminate, __ATOMIC_ACQUIRE) );
294        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
295        __wake_proc( mainProcessor );
296        __kernel_last_resume( __cfaabi_tls.this_processor );
297        mainThread->self_cor.state = Halted;
298
299        // THE SYSTEM IS NOW COMPLETELY STOPPED
300
301        // Disable preemption
302        __kernel_alarm_shutdown();
303
304        #if !defined( __CFA_NO_STATISTICS__ )
305                __stats_t * st = (__stats_t *)& storage_mainProcStats;
306                __tally_stats(mainCluster->stats, st);
307                if( 0 != mainProcessor->print_stats ) {
308                        __print_stats( st, mainProcessor->print_stats, "Processor ", mainProcessor->name, (void*)mainProcessor );
309                }
310                #if defined(CFA_STATS_ARRAY)
311                        __flush_stat( st, "Processor", mainProcessor );
312                #endif
313        #endif
314
315        mainProcessor->local_data = 0p;
316
317        unregister_tls( mainProcessor );
318        __cfa_io_stop( mainProcessor );
319
320        // Destroy the main processor and its context in reverse order of construction
321        // These were manually constructed so we need manually destroy them
322        void ^?{}(processor & this) with( this ){
323                deinit( this );
324
325                /* paranoid */ verify( this.do_terminate == true );
326                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
327        }
328
329        ^(*mainProcessor){};
330
331        // Final step, destroy the main thread since it is no longer needed
332
333        // Since we provided a stack to this taxk it will not destroy anything
334        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
335        ^(*mainThread){};
336
337        ^(*mainCluster){};
338
339        ^(__scheduler_lock){};
340
341        ^(__cfa_dbg_global_clusters.list){};
342        ^(__cfa_dbg_global_clusters.lock){};
343
344        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
345}
346
347//=============================================================================================
348// Kernel Initial Scheduling logic
349//=============================================================================================
350
351// Context invoker for processors
352// This is the entry point for processors (kernel threads) *except* for the main processor
353// It effectively constructs a coroutine by stealing the pthread stack
354static void * __invoke_processor(void * arg) {
355        #if !defined( __CFA_NO_STATISTICS__ )
356                __stats_t local_stats;
357                __init_stats( &local_stats );
358                __cfaabi_tls.this_stats = &local_stats;
359        #endif
360
361        processor * proc = (processor *) arg;
362        __cfaabi_tls.this_processor = proc;
363        __cfaabi_tls.this_thread    = 0p;
364        __cfaabi_tls.preemption_state.[enabled, disable_count] = [false, 1];
365        proc->local_data = &__cfaabi_tls;
366
367        __cfa_io_start( proc );
368        register_tls( proc );
369
370        // used for idle sleep when io_uring is present
371        io_future_t future;
372        eventfd_t idle_buf;
373        proc->idle_wctx.ftr = &future;
374        proc->idle_wctx.rdbuf = &idle_buf;
375
376
377        // SKULLDUGGERY: We want to create a context for the processor coroutine
378        // which is needed for the 2-step context switch. However, there is no reason
379        // to waste the perfectly valid stack create by pthread.
380        current_stack_info_t info;
381        __stack_t ctx;
382        info.storage = &ctx;
383        (proc->runner){ proc, &info };
384
385        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
386
387        //Set global state
388        __cfaabi_tls.this_thread = 0p;
389
390        //We now have a proper context from which to schedule threads
391        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
392
393        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
394        // resume it to start it like it normally would, it will just context switch
395        // back to here. Instead directly call the main since we already are on the
396        // appropriate stack.
397        get_coroutine(proc->runner)->state = Active;
398        main( proc->runner );
399        get_coroutine(proc->runner)->state = Halted;
400
401        // Main routine of the core returned, the core is now fully terminated
402        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
403
404        #if !defined(__CFA_NO_STATISTICS__)
405                __tally_stats(proc->cltr->stats, &local_stats);
406                if( 0 != proc->print_stats ) {
407                        __print_stats( &local_stats, proc->print_stats, "Processor ", proc->name, (void*)proc );
408                }
409                #if defined(CFA_STATS_ARRAY)
410                        __flush_stat( &local_stats, "Processor", proc );
411                #endif
412        #endif
413
414        proc->local_data = 0p;
415
416        unregister_tls( proc );
417        __cfa_io_stop( proc );
418
419        return 0p;
420}
421
422static void __kernel_first_resume( processor * this ) {
423        thread$ * src = mainThread;
424        coroutine$ * dst = get_coroutine(this->runner);
425
426        /* paranoid */ verify( ! __preemption_enabled() );
427
428        __cfaabi_tls.this_thread->curr_cor = dst;
429        __stack_prepare( &dst->stack, DEFAULT_STACK_SIZE );
430        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
431
432        /* paranoid */ verify( ! __preemption_enabled() );
433
434        dst->last = &src->self_cor;
435        dst->starter = dst->starter ? dst->starter : &src->self_cor;
436
437        // make sure the current state is still correct
438        /* paranoid */ verify(src->state == Ready);
439        src->corctx_flag = true;
440
441        // context switch to specified coroutine
442        verify( dst->context.SP );
443        __cfactx_switch( &src->context, &dst->context );
444        // when __cfactx_switch returns we are back in the src coroutine
445
446        mainThread->curr_cor = &mainThread->self_cor;
447
448        // make sure the current state has been update
449        /* paranoid */ verify(src->state == Active);
450
451        /* paranoid */ verify( ! __preemption_enabled() );
452}
453
454// KERNEL_ONLY
455static void __kernel_last_resume( processor * this ) {
456        coroutine$ * src = &mainThread->self_cor;
457        coroutine$ * dst = get_coroutine(this->runner);
458
459        /* paranoid */ verify( ! __preemption_enabled() );
460        /* paranoid */ verify( dst->starter == src );
461        /* paranoid */ verify( dst->context.SP );
462
463        // SKULLDUGGERY in debug the processors check that the
464        // stack is still within the limit of the stack limits after running a thread.
465        // that check doesn't make sense if we context switch to the processor using the
466        // coroutine semantics. Since this is a special case, use the current context
467        // info to populate these fields.
468        __cfaabi_dbg_debug_do(
469                __stack_context_t ctx;
470                CtxGet( ctx );
471                mainThread->context.SP = ctx.SP;
472                mainThread->context.FP = ctx.FP;
473        )
474
475        // context switch to the processor
476        __cfactx_switch( &src->context, &dst->context );
477}
478
479
480//=============================================================================================
481// Kernel Object Constructors logic
482//=============================================================================================
483//-----------------------------------------------------------------------------
484// Main thread construction
485static void ?{}( coroutine$ & this, current_stack_info_t * info) with( this ) {
486        stack.storage = info->storage;
487        with(*stack.storage) {
488                limit     = info->limit;
489                base      = info->base;
490        }
491        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
492        *istorage |= 0x1;
493        name = "Main Thread";
494        state = Start;
495        starter = 0p;
496        last = 0p;
497        cancellation = 0p;
498}
499
500static void ?{}( thread$ & this, current_stack_info_t * info) with( this ) {
501        ticket = TICKET_RUNNING;
502        state = Start;
503        self_cor{ info };
504        curr_cor = &self_cor;
505        curr_cluster = mainCluster;
506        self_mon.owner = &this;
507        self_mon.recursion = 1;
508        self_mon_p = &self_mon;
509        link.next = 0p;
510        link.ts   = MAX;
511        preferred = ready_queue_new_preferred();
512        last_proc = 0p;
513        random_state = __global_random_mask ? __global_random_prime : __global_random_prime ^ rdtscl();
514        #if defined( __CFA_WITH_VERIFY__ )
515                canary = 0x0D15EA5E0D15EA5Ep;
516        #endif
517
518        node.next = 0p;
519        node.prev = 0p;
520        doregister(curr_cluster, this);
521
522        monitors{ &self_mon_p, 1, (fptr_t)0 };
523}
524
525//-----------------------------------------------------------------------------
526// Processor
527// Construct the processor context of non-main processors
528static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
529        (this.self){ info };
530        this.proc = proc;
531}
532
533static void init(processor & this, const char name[], cluster & _cltr, thread$ * initT) with( this ) {
534        this.name = name;
535        this.cltr = &_cltr;
536        this.rdq.its = 0;
537        this.rdq.itr = 0;
538        this.rdq.id  = 0;
539        this.rdq.target = MAX;
540        this.rdq.last = MAX;
541        this.rdq.cpu = 0;
542        // this.rdq.cutoff = 0ull;
543        do_terminate = false;
544        preemption_alarm = 0p;
545        pending_preemption = false;
546
547        this.io.ctx = 0p;
548        this.io.pending = false;
549        this.io.dirty   = false;
550
551        this.init.thrd = initT;
552
553        this.local_data = 0p;
554
555        idle_wctx.evfd = eventfd(0, 0);
556        if (idle_wctx.evfd < 0) {
557                abort("KERNEL ERROR: PROCESSOR EVENTFD - %s\n", strerror(errno));
558        }
559
560        idle_wctx.sem = 0;
561        idle_wctx.wake__time = 0;
562
563        // I'm assuming these two are reserved for standard input and output
564        // so I'm using them as sentinels with idle_wctx.
565        /* paranoid */ verify( idle_wctx.evfd != 0 );
566        /* paranoid */ verify( idle_wctx.evfd != 1 );
567
568        #if !defined(__CFA_NO_STATISTICS__)
569                print_stats = 0;
570                print_halts = false;
571        #endif
572
573        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
574}
575
576// Not a ctor, it just preps the destruction but should not destroy members
577static void deinit(processor & this) {
578        close(this.idle_wctx.evfd);
579}
580
581void ?{}(processor & this, const char name[], cluster & _cltr, thread$ * initT) libcfa_public {
582        ( this.terminated ){};
583        ( this.runner ){};
584
585        disable_interrupts();
586                init( this, name, _cltr, initT );
587        enable_interrupts();
588
589        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
590
591        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
592}
593
594void ?{}(processor & this, const char name[], cluster & _cltr) libcfa_public {
595        (this){name, _cltr, 0p};
596}
597
598extern size_t __page_size;
599void ^?{}(processor & this) libcfa_public with( this ) {
600        /* paranoid */ verify( !__atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) );
601        __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
602
603        __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
604        __disable_interrupts_checked();
605                __wake_proc( &this );
606        __enable_interrupts_checked();
607
608        wait( terminated );
609        /* paranoid */ verify( active_processor() != &this);
610
611        __destroy_pthread( kernel_thread, this.stack, 0p );
612
613        disable_interrupts();
614                deinit( this );
615        enable_interrupts();
616}
617
618//-----------------------------------------------------------------------------
619// Cluster
620static void ?{}(__cluster_proc_list & this) {
621        this.fdw   = 0p;
622        this.idle  = 0;
623        this.total = 0;
624}
625
626void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned num_io, const io_context_params & io_params) libcfa_public with( this ) {
627        this.name = name;
628        this.preemption_rate = preemption_rate;
629        this.sched.readyQ.data = 0p;
630        this.sched.readyQ.tscs = 0p;
631        this.sched.readyQ.count = 0;
632        this.sched.io.tscs = 0p;
633        this.sched.io.data = 0p;
634        this.sched.caches = 0p;
635
636        #if !defined(__CFA_NO_STATISTICS__)
637                print_stats = 0;
638                stats = alloc();
639                __init_stats( stats );
640        #endif
641
642        threads{ __get };
643
644        io.arbiter = create();
645        io.params = io_params;
646
647        doregister(this);
648
649        // Lock the RWlock so no-one pushes/pops while we are changing the queue
650        disable_interrupts();
651        uint_fast32_t last_size = ready_mutate_lock();
652
653                // Adjust the ready queue size
654                ready_queue_grow( &this );
655
656        // Unlock the RWlock
657        ready_mutate_unlock( last_size );
658        enable_interrupts( false ); // Don't poll, could be in main cluster
659}
660
661void ^?{}(cluster & this) libcfa_public {
662        destroy(this.io.arbiter);
663
664        // Lock the RWlock so no-one pushes/pops while we are changing the queue
665        disable_interrupts();
666        uint_fast32_t last_size = ready_mutate_lock();
667
668                // Adjust the ready queue size
669                ready_queue_shrink( &this );
670
671        // Unlock the RWlock
672        ready_mutate_unlock( last_size );
673
674        ready_queue_close( &this );
675        /* paranoid */ verify( this.sched.readyQ.data == 0p );
676        /* paranoid */ verify( this.sched.readyQ.tscs == 0p );
677        /* paranoid */ verify( this.sched.readyQ.count == 0 );
678        /* paranoid */ verify( this.sched.io.tscs == 0p );
679        /* paranoid */ verify( this.sched.caches == 0p );
680
681        enable_interrupts( false ); // Don't poll, could be in main cluster
682
683
684        #if !defined(__CFA_NO_STATISTICS__)
685                if( 0 != this.print_stats ) {
686                        __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
687                }
688                #if defined(CFA_STATS_ARRAY)
689                        __flush_stat( this.stats, "Cluster", &this );
690                #endif
691                free( this.stats );
692        #endif
693
694        unregister(this);
695}
696
697//=============================================================================================
698// Miscellaneous Initialization
699//=============================================================================================
700//-----------------------------------------------------------------------------
701// Global Queues
702static void doregister( cluster     & cltr ) {
703        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
704        push_front( __cfa_dbg_global_clusters.list, cltr );
705        unlock    ( __cfa_dbg_global_clusters.lock );
706}
707
708static void unregister( cluster     & cltr ) {
709        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
710        remove( __cfa_dbg_global_clusters.list, cltr );
711        unlock( __cfa_dbg_global_clusters.lock );
712}
713
714void doregister( cluster * cltr, thread$ & thrd ) {
715        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
716        cltr->nthreads += 1;
717        push_front(cltr->threads, thrd);
718        unlock    (cltr->thread_list_lock);
719}
720
721void unregister( cluster * cltr, thread$ & thrd ) {
722        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
723        remove(cltr->threads, thrd );
724        cltr->nthreads -= 1;
725        unlock(cltr->thread_list_lock);
726}
727
728static void register_tls( processor * this ) {
729        // Register and Lock the RWlock so no-one pushes/pops while we are changing the queue
730        uint_fast32_t last_size;
731        [this->unique_id, last_size] = ready_mutate_register();
732
733                this->rdq.cpu = __kernel_getcpu();
734
735                this->cltr->procs.total += 1u;
736                insert_last(this->cltr->procs.actives, *this);
737
738                // Adjust the ready queue size
739                ready_queue_grow( this->cltr );
740
741        // Unlock the RWlock
742        ready_mutate_unlock( last_size );
743}
744
745
746static void unregister_tls( processor * this ) {
747        // Lock the RWlock so no-one pushes/pops while we are changing the queue
748        uint_fast32_t last_size = ready_mutate_lock();
749                this->cltr->procs.total -= 1u;
750                remove(*this);
751
752                // clear the cluster so nothing gets pushed to local queues
753                cluster * cltr = this->cltr;
754                this->cltr = 0p;
755
756                // Adjust the ready queue size
757                ready_queue_shrink( cltr );
758
759        // Unlock the RWlock and unregister: we don't need the read_lock any more
760        ready_mutate_unregister( this->unique_id, last_size );
761}
762
763static void check( int ret, const char func[] ) {
764        if ( ret ) {                                                                            // pthread routines return errno values
765                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
766        } // if
767} // Abort
768
769void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
770        pthread_attr_t attr;
771
772        check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
773
774        size_t stacksize = max( PTHREAD_STACK_MIN, DEFAULT_STACK_SIZE );
775
776        void * stack;
777        #if CFA_PROCESSOR_USE_MMAP
778                stacksize = ceiling( stacksize, __page_size ) + __page_size;
779                stack = mmap(0p, stacksize, __map_prot, MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
780                if(stack == ((void*)-1)) {
781                        abort( "pthread stack creation : internal error, mmap failure, error(%d) %s.", errno, strerror( errno ) );
782                }
783                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
784                        abort( "pthread stack creation : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
785                } // if
786        #else
787                __cfaabi_dbg_debug_do(
788                        stack = memalign( __page_size, stacksize + __page_size );
789                        // pthread has no mechanism to create the guard page in user supplied stack.
790                        if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
791                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
792                        } // if
793                );
794                __cfaabi_dbg_no_debug_do(
795                        stack = malloc( stacksize );
796                );
797        #endif
798
799        check( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
800        check( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
801        return stack;
802}
803
804void __destroy_pthread( pthread_t pthread, void * stack, void ** retval ) {
805        int err = pthread_join( pthread, retval );
806        if( err != 0 ) abort("KERNEL ERROR: joining pthread %p caused error %s\n", (void*)pthread, strerror(err));
807
808        #if CFA_PROCESSOR_USE_MMAP
809                pthread_attr_t attr;
810
811                check( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
812
813                size_t stacksize;
814                // default stack size, normally defined by shell limit
815                check( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
816                assert( stacksize >= PTHREAD_STACK_MIN );
817                stacksize += __page_size;
818
819                if(munmap(stack, stacksize) == -1) {
820                        abort( "pthread stack destruction : internal error, munmap failure, error(%d) %s.", errno, strerror( errno ) );
821                }
822        #else
823                __cfaabi_dbg_debug_do(
824                        // pthread has no mechanism to create the guard page in user supplied stack.
825                        if ( mprotect( stack, __page_size, __map_prot ) == -1 ) {
826                                abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
827                        } // if
828                );
829                free( stack );
830        #endif
831}
832
833#if defined(__CFA_WITH_VERIFY__)
834static bool verify_fwd_bck_rng(void) {
835        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&verify_fwd_bck_rng);
836
837        unsigned values[10];
838        for(i; 10) {
839                values[i] = __tls_rand_fwd();
840        }
841
842        __tls_rand_advance_bck();
843
844        for ( i; 9 -~= 0 ) {
845                if(values[i] != __tls_rand_bck()) {
846                        return false;
847                }
848        }
849
850        return true;
851}
852#endif
Note: See TracBrowser for help on using the repository browser.