source: libcfa/src/concurrency/kernel.cfa @ c7a900a

arm-ehjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-expr
Last change on this file since c7a900a was c7a900a, checked in by Thierry Delisle <tdelisle@…>, 22 months ago

More renames and clean-ups

  • Property mode set to 100644
File size: 29.5 KB
Line 
1//
2// Cforall Version 1.0.0 Copyright (C) 2016 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.c --
8//
9// Author           : Thierry Delisle
10// Created On       : Tue Jan 17 12:27:26 2017
11// Last Modified By : Peter A. Buhr
12// Last Modified On : Tue Feb  4 13:03:15 2020
13// Update Count     : 58
14//
15
16#define __cforall_thread__
17
18//C Includes
19#include <stddef.h>
20#include <errno.h>
21#include <string.h>
22extern "C" {
23#include <stdio.h>
24#include <fenv.h>
25#include <sys/resource.h>
26#include <signal.h>
27#include <unistd.h>
28#include <limits.h>                                                                             // PTHREAD_STACK_MIN
29#include <sys/mman.h>                                                                   // mprotect
30}
31
32//CFA Includes
33#include "time.hfa"
34#include "kernel_private.hfa"
35#include "preemption.hfa"
36#include "startup.hfa"
37
38//Private includes
39#define __CFA_INVOKE_PRIVATE__
40#include "invoke.h"
41
42//-----------------------------------------------------------------------------
43// Some assembly required
44#if defined( __i386 )
45        #define CtxGet( ctx )        \
46                __asm__ volatile (     \
47                        "movl %%esp,%0\n"\
48                        "movl %%ebp,%1\n"\
49                        : "=rm" (ctx.SP),\
50                                "=rm" (ctx.FP) \
51                )
52
53        // mxcr : SSE Status and Control bits (control bits are preserved across function calls)
54        // fcw  : X87 FPU control word (preserved across function calls)
55        #define __x87_store         \
56                uint32_t __mxcr;      \
57                uint16_t __fcw;       \
58                __asm__ volatile (    \
59                        "stmxcsr %0\n"  \
60                        "fnstcw  %1\n"  \
61                        : "=m" (__mxcr),\
62                                "=m" (__fcw)  \
63                )
64
65        #define __x87_load         \
66                __asm__ volatile (   \
67                        "fldcw  %1\n"  \
68                        "ldmxcsr %0\n" \
69                        ::"m" (__mxcr),\
70                                "m" (__fcw)  \
71                )
72
73#elif defined( __x86_64 )
74        #define CtxGet( ctx )        \
75                __asm__ volatile (     \
76                        "movq %%rsp,%0\n"\
77                        "movq %%rbp,%1\n"\
78                        : "=rm" (ctx.SP),\
79                                "=rm" (ctx.FP) \
80                )
81
82        #define __x87_store         \
83                uint32_t __mxcr;      \
84                uint16_t __fcw;       \
85                __asm__ volatile (    \
86                        "stmxcsr %0\n"  \
87                        "fnstcw  %1\n"  \
88                        : "=m" (__mxcr),\
89                                "=m" (__fcw)  \
90                )
91
92        #define __x87_load          \
93                __asm__ volatile (    \
94                        "fldcw  %1\n"   \
95                        "ldmxcsr %0\n"  \
96                        :: "m" (__mxcr),\
97                                "m" (__fcw)  \
98                )
99
100
101#elif defined( __ARM_ARCH )
102#define CtxGet( ctx ) __asm__ ( \
103                "mov %0,%%sp\n"   \
104                "mov %1,%%r11\n"   \
105        : "=rm" (ctx.SP), "=rm" (ctx.FP) )
106#else
107        #error unknown hardware architecture
108#endif
109
110//-----------------------------------------------------------------------------
111//Start and stop routine for the kernel, declared first to make sure they run first
112static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
113static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
114
115//-----------------------------------------------------------------------------
116// Kernel storage
117KERNEL_STORAGE(cluster,         mainCluster);
118KERNEL_STORAGE(processor,       mainProcessor);
119KERNEL_STORAGE(thread_desc,     mainThread);
120KERNEL_STORAGE(__stack_t,       mainThreadCtx);
121
122cluster     * mainCluster;
123processor   * mainProcessor;
124thread_desc * mainThread;
125
126extern "C" {
127        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
128}
129
130size_t __page_size = 0;
131
132//-----------------------------------------------------------------------------
133// Global state
134thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = {
135        NULL,                                                                                           // cannot use 0p
136        NULL,
137        { 1, false, false },
138        6u //this should be seeded better but due to a bug calling rdtsc doesn't work
139};
140
141//-----------------------------------------------------------------------------
142// Struct to steal stack
143struct current_stack_info_t {
144        __stack_t * storage;                                                            // pointer to stack object
145        void * base;                                                                            // base of stack
146        void * limit;                                                                           // stack grows towards stack limit
147        void * context;                                                                         // address of cfa_context_t
148};
149
150void ?{}( current_stack_info_t & this ) {
151        __stack_context_t ctx;
152        CtxGet( ctx );
153        this.base = ctx.FP;
154
155        rlimit r;
156        getrlimit( RLIMIT_STACK, &r);
157        size_t size = r.rlim_cur;
158
159        this.limit = (void *)(((intptr_t)this.base) - size);
160        this.context = &storage_mainThreadCtx;
161}
162
163//-----------------------------------------------------------------------------
164// Main thread construction
165
166void ?{}( coroutine_desc & this, current_stack_info_t * info) with( this ) {
167        stack.storage = info->storage;
168        with(*stack.storage) {
169                limit     = info->limit;
170                base      = info->base;
171        }
172        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
173        *istorage |= 0x1;
174        name = "Main Thread";
175        state = Start;
176        starter = 0p;
177        last = 0p;
178        cancellation = 0p;
179}
180
181void ?{}( thread_desc & this, current_stack_info_t * info) with( this ) {
182        state = Start;
183        self_cor{ info };
184        curr_cor = &self_cor;
185        curr_cluster = mainCluster;
186        self_mon.owner = &this;
187        self_mon.recursion = 1;
188        self_mon_p = &self_mon;
189        next = 0p;
190
191        node.next = 0p;
192        node.prev = 0p;
193        doregister(curr_cluster, this);
194
195        monitors{ &self_mon_p, 1, (fptr_t)0 };
196}
197
198//-----------------------------------------------------------------------------
199// Processor coroutine
200void ?{}(processorCtx_t & this) {
201
202}
203
204// Construct the processor context of non-main processors
205static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
206        (this.__cor){ info };
207        this.proc = proc;
208}
209
210static void * __invoke_processor(void * arg);
211
212void ?{}(processor & this, const char name[], cluster & cltr) with( this ) {
213        this.name = name;
214        this.cltr = &cltr;
215        terminated{ 0 };
216        destroyer = 0p;
217        do_terminate = false;
218        preemption_alarm = 0p;
219        pending_preemption = false;
220        runner.proc = &this;
221
222        idleLock{};
223
224        __cfaabi_dbg_print_safe("Kernel : Starting core %p\n", &this);
225
226        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
227
228        __cfaabi_dbg_print_safe("Kernel : core %p started\n", &this);
229}
230
231void ^?{}(processor & this) with( this ){
232        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
233                __cfaabi_dbg_print_safe("Kernel : core %p signaling termination\n", &this);
234
235                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
236                wake( &this );
237
238                P( terminated );
239                verify( kernelTLS.this_processor != &this);
240        }
241
242        pthread_join( kernel_thread, 0p );
243        free( this.stack );
244}
245
246void ?{}(cluster & this, const char name[], Duration preemption_rate) with( this ) {
247        this.name = name;
248        this.preemption_rate = preemption_rate;
249        ready_queue{};
250        ready_queue_lock{};
251
252        procs{ __get };
253        idles{ __get };
254        threads{ __get };
255
256        doregister(this);
257}
258
259void ^?{}(cluster & this) {
260        unregister(this);
261}
262
263//=============================================================================================
264// Kernel Scheduling logic
265//=============================================================================================
266static thread_desc * __next_thread(cluster * this);
267static void __run_thread(processor * this, thread_desc * dst);
268static void __halt(processor * this);
269
270//Main of the processor contexts
271void main(processorCtx_t & runner) {
272        // Because of a bug, we couldn't initialized the seed on construction
273        // Do it here
274        kernelTLS.rand_seed ^= rdtscl();
275
276        processor * this = runner.proc;
277        verify(this);
278
279        __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this);
280
281        doregister(this->cltr, this);
282
283        {
284                // Setup preemption data
285                preemption_scope scope = { this };
286
287                __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);
288
289                thread_desc * readyThread = 0p;
290                for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
291                        readyThread = __next_thread( this->cltr );
292
293                        if(readyThread) {
294                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
295                                /* paranoid */ verifyf( readyThread->state == Inactive || readyThread->state == Start || readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
296                                /* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );
297
298                                __run_thread(this, readyThread);
299
300                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
301
302                                spin_count = 0;
303                        } else {
304                                // spin(this, &spin_count);
305                                __halt(this);
306                        }
307                }
308
309                __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this);
310        }
311
312        unregister(this->cltr, this);
313
314        V( this->terminated );
315
316        __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this);
317}
318
319static int * __volatile_errno() __attribute__((noinline));
320static int * __volatile_errno() { asm(""); return &errno; }
321
322// KERNEL ONLY
323// runThread runs a thread by context switching
324// from the processor coroutine to the target thread
325static void __run_thread(processor * this, thread_desc * thrd_dst) {
326        coroutine_desc * proc_cor = get_coroutine(this->runner);
327
328        // Update global state
329        kernelTLS.this_thread = thrd_dst;
330
331        // set state of processor coroutine to inactive
332        verify(proc_cor->state == Active);
333        proc_cor->state = Inactive;
334
335        // Actually run the thread
336        RUNNING:  while(true) {
337                if(unlikely(thrd_dst->preempted)) {
338                        thrd_dst->preempted = __NO_PREEMPTION;
339                        verify(thrd_dst->state == Active || thrd_dst->state == Rerun);
340                } else {
341                        verify(thrd_dst->state == Start || thrd_dst->state == Primed || thrd_dst->state == Inactive);
342                        thrd_dst->state = Active;
343                }
344
345                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
346
347                // set context switch to the thread that the processor is executing
348                verify( thrd_dst->context.SP );
349                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
350                // when __cfactx_switch returns we are back in the processor coroutine
351
352                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
353
354
355                // We just finished running a thread, there are a few things that could have happened.
356                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
357                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
358                // 3 - Polite Racy case : the thread has blocked, someone has already tried to schedule it, but the thread is nice and wants to go through the ready-queue any way
359                // 4 - Preempted
360                // In case 1, we may have won a race so we can't write to the state again.
361                // In case 2, we lost the race so we now own the thread.
362                // In case 3, we lost the race but can just reschedule the thread.
363
364                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
365                        // The thread was preempted, reschedule it and reset the flag
366                        __schedule_thread( thrd_dst );
367                        break RUNNING;
368                }
369
370                // set state of processor coroutine to active and the thread to inactive
371                static_assert(sizeof(thrd_dst->state) == sizeof(int));
372                enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Inactive, __ATOMIC_SEQ_CST);
373                switch(old_state) {
374                        case Halted:
375                                // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
376                                thrd_dst->state = Halted;
377
378                                // We may need to wake someone up here since
379                                unpark( this->destroyer );
380                                this->destroyer = 0p;
381                                break RUNNING;
382                        case Active:
383                                // This is case 1, the regular case, nothing more is needed
384                                break RUNNING;
385                        case Rerun:
386                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
387                                // In this case, just run it again.
388                                continue RUNNING;
389                        default:
390                                // This makes no sense, something is wrong abort
391                                abort("Finished running a thread that was Inactive/Start/Primed %d\n", old_state);
392                }
393        }
394
395        // Just before returning to the processor, set the processor coroutine to active
396        proc_cor->state = Active;
397}
398
399// KERNEL_ONLY
400void returnToKernel() {
401        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
402        coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
403        thread_desc * thrd_src = kernelTLS.this_thread;
404
405        // Run the thread on this processor
406        {
407                int local_errno = *__volatile_errno();
408                #if defined( __i386 ) || defined( __x86_64 )
409                        __x87_store;
410                #endif
411                verify( proc_cor->context.SP );
412                __cfactx_switch( &thrd_src->context, &proc_cor->context );
413                #if defined( __i386 ) || defined( __x86_64 )
414                        __x87_load;
415                #endif
416                *__volatile_errno() = local_errno;
417        }
418
419        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
420}
421
422// KERNEL_ONLY
423// Context invoker for processors
424// This is the entry point for processors (kernel threads)
425// It effectively constructs a coroutine by stealing the pthread stack
426static void * __invoke_processor(void * arg) {
427        processor * proc = (processor *) arg;
428        kernelTLS.this_processor = proc;
429        kernelTLS.this_thread    = 0p;
430        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
431        // SKULLDUGGERY: We want to create a context for the processor coroutine
432        // which is needed for the 2-step context switch. However, there is no reason
433        // to waste the perfectly valid stack create by pthread.
434        current_stack_info_t info;
435        __stack_t ctx;
436        info.storage = &ctx;
437        (proc->runner){ proc, &info };
438
439        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
440
441        //Set global state
442        kernelTLS.this_thread = 0p;
443
444        //We now have a proper context from which to schedule threads
445        __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
446
447        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
448        // resume it to start it like it normally would, it will just context switch
449        // back to here. Instead directly call the main since we already are on the
450        // appropriate stack.
451        get_coroutine(proc->runner)->state = Active;
452        main( proc->runner );
453        get_coroutine(proc->runner)->state = Halted;
454
455        // Main routine of the core returned, the core is now fully terminated
456        __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner);
457
458        return 0p;
459}
460
461static void Abort( int ret, const char func[] ) {
462        if ( ret ) {                                                                            // pthread routines return errno values
463                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
464        } // if
465} // Abort
466
467void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
468        pthread_attr_t attr;
469
470        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
471
472        size_t stacksize;
473        // default stack size, normally defined by shell limit
474        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
475        assert( stacksize >= PTHREAD_STACK_MIN );
476
477        void * stack;
478        __cfaabi_dbg_debug_do(
479                stack = memalign( __page_size, stacksize + __page_size );
480                // pthread has no mechanism to create the guard page in user supplied stack.
481                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
482                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
483                } // if
484        );
485        __cfaabi_dbg_no_debug_do(
486                stack = malloc( stacksize );
487        );
488
489        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
490
491        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
492        return stack;
493}
494
495// KERNEL_ONLY
496static void __kernel_first_resume( processor * this ) {
497        thread_desc * src = mainThread;
498        coroutine_desc * dst = get_coroutine(this->runner);
499
500        verify( ! kernelTLS.preemption_state.enabled );
501
502        kernelTLS.this_thread->curr_cor = dst;
503        __stack_prepare( &dst->stack, 65000 );
504        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
505
506        verify( ! kernelTLS.preemption_state.enabled );
507
508        dst->last = &src->self_cor;
509        dst->starter = dst->starter ? dst->starter : &src->self_cor;
510
511        // set state of current coroutine to inactive
512        src->state = src->state == Halted ? Halted : Inactive;
513
514        // context switch to specified coroutine
515        verify( dst->context.SP );
516        __cfactx_switch( &src->context, &dst->context );
517        // when __cfactx_switch returns we are back in the src coroutine
518
519        mainThread->curr_cor = &mainThread->self_cor;
520
521        // set state of new coroutine to active
522        src->state = Active;
523
524        verify( ! kernelTLS.preemption_state.enabled );
525}
526
527// KERNEL_ONLY
528static void __kernel_last_resume( processor * this ) {
529        coroutine_desc * src = &mainThread->self_cor;
530        coroutine_desc * dst = get_coroutine(this->runner);
531
532        verify( ! kernelTLS.preemption_state.enabled );
533        verify( dst->starter == src );
534        verify( dst->context.SP );
535
536        // context switch to the processor
537        __cfactx_switch( &src->context, &dst->context );
538}
539
540//-----------------------------------------------------------------------------
541// Scheduler routines
542// KERNEL ONLY
543void __schedule_thread( thread_desc * thrd ) with( *thrd->curr_cluster ) {
544        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
545        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
546        /* paranoid */ if( thrd->state == Inactive || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
547                          "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
548        /* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active || thrd->state == Rerun,
549                          "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
550        /* paranoid */ #endif
551        /* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
552
553        lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
554        bool was_empty = !(ready_queue != 0);
555        append( ready_queue, thrd );
556        unlock( ready_queue_lock );
557
558        if(was_empty) {
559                lock      (proc_list_lock __cfaabi_dbg_ctx2);
560                if(idles) {
561                        wake_fast(idles.head);
562                }
563                unlock    (proc_list_lock);
564        }
565        else if( struct processor * idle = idles.head ) {
566                wake_fast(idle);
567        }
568
569        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
570}
571
572// KERNEL ONLY
573static thread_desc * __next_thread(cluster * this) with( *this ) {
574        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
575
576        lock( ready_queue_lock __cfaabi_dbg_ctx2 );
577        thread_desc * head = pop_head( ready_queue );
578        unlock( ready_queue_lock );
579
580        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
581        return head;
582}
583
584void unpark( thread_desc * thrd ) {
585        if( !thrd ) return;
586
587        disable_interrupts();
588        static_assert(sizeof(thrd->state) == sizeof(int));
589        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
590        switch(old_state) {
591                case Active:
592                        // Wake won the race, the thread will reschedule/rerun itself
593                        break;
594                case Inactive:
595                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
596
597                        // Wake lost the race,
598                        thrd->state = Inactive;
599                        __schedule_thread( thrd );
600                        break;
601                case Rerun:
602                        abort("More than one thread attempted to schedule thread %p\n", thrd);
603                        break;
604                case Halted:
605                case Start:
606                case Primed:
607                default:
608                        // This makes no sense, something is wrong abort
609                        abort();
610        }
611        enable_interrupts( __cfaabi_dbg_ctx );
612}
613
614void park( void ) {
615        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
616        disable_interrupts();
617        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
618        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
619
620        returnToKernel();
621
622        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
623        enable_interrupts( __cfaabi_dbg_ctx );
624        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
625
626}
627
628// KERNEL ONLY
629void __leave_thread() {
630        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
631        returnToKernel();
632        abort();
633}
634
635// KERNEL ONLY
636bool force_yield( __Preemption_Reason reason ) {
637        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
638        disable_interrupts();
639        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
640
641        thread_desc * thrd = kernelTLS.this_thread;
642        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun);
643
644        // SKULLDUGGERY: It is possible that we are preempting this thread just before
645        // it was going to park itself. If that is the case and it is already using the
646        // intrusive fields then we can't use them to preempt the thread
647        // If that is the case, abandon the preemption.
648        bool preempted = false;
649        if(thrd->next == 0p) {
650                preempted = true;
651                thrd->preempted = reason;
652                returnToKernel();
653        }
654
655        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
656        enable_interrupts_noPoll();
657        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
658
659        return preempted;
660}
661
662//=============================================================================================
663// Kernel Setup logic
664//=============================================================================================
665//-----------------------------------------------------------------------------
666// Kernel boot procedures
667static void __kernel_startup(void) {
668        verify( ! kernelTLS.preemption_state.enabled );
669        __cfaabi_dbg_print_safe("Kernel : Starting\n");
670
671        __page_size = sysconf( _SC_PAGESIZE );
672
673        __cfa_dbg_global_clusters.list{ __get };
674        __cfa_dbg_global_clusters.lock{};
675
676        // Initialize the main cluster
677        mainCluster = (cluster *)&storage_mainCluster;
678        (*mainCluster){"Main Cluster"};
679
680        __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n");
681
682        // Start by initializing the main thread
683        // SKULLDUGGERY: the mainThread steals the process main thread
684        // which will then be scheduled by the mainProcessor normally
685        mainThread = (thread_desc *)&storage_mainThread;
686        current_stack_info_t info;
687        info.storage = (__stack_t*)&storage_mainThreadCtx;
688        (*mainThread){ &info };
689
690        __cfaabi_dbg_print_safe("Kernel : Main thread ready\n");
691
692
693
694        // Construct the processor context of the main processor
695        void ?{}(processorCtx_t & this, processor * proc) {
696                (this.__cor){ "Processor" };
697                this.__cor.starter = 0p;
698                this.proc = proc;
699        }
700
701        void ?{}(processor & this) with( this ) {
702                name = "Main Processor";
703                cltr = mainCluster;
704                terminated{ 0 };
705                do_terminate = false;
706                preemption_alarm = 0p;
707                pending_preemption = false;
708                kernel_thread = pthread_self();
709
710                runner{ &this };
711                __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner);
712        }
713
714        // Initialize the main processor and the main processor ctx
715        // (the coroutine that contains the processing control flow)
716        mainProcessor = (processor *)&storage_mainProcessor;
717        (*mainProcessor){};
718
719        //initialize the global state variables
720        kernelTLS.this_processor = mainProcessor;
721        kernelTLS.this_thread    = mainThread;
722
723        // Enable preemption
724        kernel_start_preemption();
725
726        // Add the main thread to the ready queue
727        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
728        __schedule_thread(mainThread);
729
730        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
731        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
732        // mainThread is on the ready queue when this call is made.
733        __kernel_first_resume( kernelTLS.this_processor );
734
735
736
737        // THE SYSTEM IS NOW COMPLETELY RUNNING
738        __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n");
739
740        verify( ! kernelTLS.preemption_state.enabled );
741        enable_interrupts( __cfaabi_dbg_ctx );
742        verify( TL_GET( preemption_state.enabled ) );
743}
744
745static void __kernel_shutdown(void) {
746        __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n");
747
748        verify( TL_GET( preemption_state.enabled ) );
749        disable_interrupts();
750        verify( ! kernelTLS.preemption_state.enabled );
751
752        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
753        // When its coroutine terminates, it return control to the mainThread
754        // which is currently here
755        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
756        __kernel_last_resume( kernelTLS.this_processor );
757        mainThread->self_cor.state = Halted;
758
759        // THE SYSTEM IS NOW COMPLETELY STOPPED
760
761        // Disable preemption
762        kernel_stop_preemption();
763
764        // Destroy the main processor and its context in reverse order of construction
765        // These were manually constructed so we need manually destroy them
766        ^(mainProcessor->runner){};
767        ^(mainProcessor){};
768
769        // Final step, destroy the main thread since it is no longer needed
770        // Since we provided a stack to this taxk it will not destroy anything
771        ^(mainThread){};
772
773        ^(__cfa_dbg_global_clusters.list){};
774        ^(__cfa_dbg_global_clusters.lock){};
775
776        __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n");
777}
778
779//=============================================================================================
780// Kernel Quiescing
781//=============================================================================================
782static void __halt(processor * this) with( *this ) {
783        // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) );
784
785        with( *cltr ) {
786                lock      (proc_list_lock __cfaabi_dbg_ctx2);
787                remove    (procs, *this);
788                push_front(idles, *this);
789                unlock    (proc_list_lock);
790        }
791
792        __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this);
793
794        wait( idleLock );
795
796        __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this);
797
798        with( *cltr ) {
799                lock      (proc_list_lock __cfaabi_dbg_ctx2);
800                remove    (idles, *this);
801                push_front(procs, *this);
802                unlock    (proc_list_lock);
803        }
804}
805
806//=============================================================================================
807// Unexpected Terminating logic
808//=============================================================================================
809static __spinlock_t kernel_abort_lock;
810static bool kernel_abort_called = false;
811
812void * kernel_abort(void) __attribute__ ((__nothrow__)) {
813        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
814        // the globalAbort flag is true.
815        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
816
817        // first task to abort ?
818        if ( kernel_abort_called ) {                    // not first task to abort ?
819                unlock( kernel_abort_lock );
820
821                sigset_t mask;
822                sigemptyset( &mask );
823                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
824                sigaddset( &mask, SIGUSR1 );            // block SIGALRM signals
825                sigsuspend( &mask );                            // block the processor to prevent further damage during abort
826                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it
827        }
828        else {
829                kernel_abort_called = true;
830                unlock( kernel_abort_lock );
831        }
832
833        return kernelTLS.this_thread;
834}
835
836void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
837        thread_desc * thrd = kernel_data;
838
839        if(thrd) {
840                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
841                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
842
843                if ( &thrd->self_cor != thrd->curr_cor ) {
844                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
845                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
846                }
847                else {
848                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
849                }
850        }
851        else {
852                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
853                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
854        }
855}
856
857int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
858        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
859}
860
861static __spinlock_t kernel_debug_lock;
862
863extern "C" {
864        void __cfaabi_bits_acquire() {
865                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
866        }
867
868        void __cfaabi_bits_release() {
869                unlock( kernel_debug_lock );
870        }
871}
872
873//=============================================================================================
874// Kernel Utilities
875//=============================================================================================
876//-----------------------------------------------------------------------------
877// Locks
878void  ?{}( semaphore & this, int count = 1 ) {
879        (this.lock){};
880        this.count = count;
881        (this.waiting){};
882}
883void ^?{}(semaphore & this) {}
884
885void P(semaphore & this) with( this ){
886        lock( lock __cfaabi_dbg_ctx2 );
887        count -= 1;
888        if ( count < 0 ) {
889                // queue current task
890                append( waiting, kernelTLS.this_thread );
891
892                // atomically release spin lock and block
893                unlock( lock );
894                park();
895        }
896        else {
897            unlock( lock );
898        }
899}
900
901void V(semaphore & this) with( this ) {
902        thread_desc * thrd = 0p;
903        lock( lock __cfaabi_dbg_ctx2 );
904        count += 1;
905        if ( count <= 0 ) {
906                // remove task at head of waiting list
907                thrd = pop_head( waiting );
908        }
909
910        unlock( lock );
911
912        // make new owner
913        unpark( thrd );
914}
915
916//-----------------------------------------------------------------------------
917// Global Queues
918void doregister( cluster     & cltr ) {
919        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
920        push_front( __cfa_dbg_global_clusters.list, cltr );
921        unlock    ( __cfa_dbg_global_clusters.lock );
922}
923
924void unregister( cluster     & cltr ) {
925        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
926        remove( __cfa_dbg_global_clusters.list, cltr );
927        unlock( __cfa_dbg_global_clusters.lock );
928}
929
930void doregister( cluster * cltr, thread_desc & thrd ) {
931        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
932        cltr->nthreads += 1;
933        push_front(cltr->threads, thrd);
934        unlock    (cltr->thread_list_lock);
935}
936
937void unregister( cluster * cltr, thread_desc & thrd ) {
938        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
939        remove(cltr->threads, thrd );
940        cltr->nthreads -= 1;
941        unlock(cltr->thread_list_lock);
942}
943
944void doregister( cluster * cltr, processor * proc ) {
945        lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2);
946        cltr->nprocessors += 1;
947        push_front(cltr->procs, *proc);
948        unlock    (cltr->proc_list_lock);
949}
950
951void unregister( cluster * cltr, processor * proc ) {
952        lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2);
953        remove(cltr->procs, *proc );
954        cltr->nprocessors -= 1;
955        unlock(cltr->proc_list_lock);
956}
957
958//-----------------------------------------------------------------------------
959// Debug
960__cfaabi_dbg_debug_do(
961        extern "C" {
962                void __cfaabi_dbg_record(__spinlock_t & this, const char prev_name[]) {
963                        this.prev_name = prev_name;
964                        this.prev_thrd = kernelTLS.this_thread;
965                }
966        }
967)
968
969//-----------------------------------------------------------------------------
970// Debug
971bool threading_enabled(void) __attribute__((const)) {
972        return true;
973}
974// Local Variables: //
975// mode: c //
976// tab-width: 4 //
977// End: //
Note: See TracBrowser for help on using the repository browser.