source: libcfa/src/concurrency/kernel.cfa @ e307e12

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationjenkins-sandboxnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since e307e12 was 09d4b22, checked in by Peter A. Buhr <pabuhr@…>, 5 years ago

move stack for preemptive pthread from TLS to static variable

  • Property mode set to 100644
File size: 27.9 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 : Thu Dec  5 16:25:52 2019
13// Update Count     : 52
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 start(processor * this);
211void ?{}(processor & this, const char * name, cluster & cltr) with( this ) {
212        this.name = name;
213        this.cltr = &cltr;
214        terminated{ 0 };
215        do_terminate = false;
216        preemption_alarm = 0p;
217        pending_preemption = false;
218        runner.proc = &this;
219
220        idleLock{};
221
222        start( &this );
223}
224
225void ^?{}(processor & this) with( this ){
226        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
227                __cfaabi_dbg_print_safe("Kernel : core %p signaling termination\n", &this);
228
229                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
230                wake( &this );
231
232                P( terminated );
233                verify( kernelTLS.this_processor != &this);
234        }
235
236        pthread_join( kernel_thread, 0p );
237        free( this.stack );
238}
239
240void ?{}(cluster & this, const char * name, Duration preemption_rate) with( this ) {
241        this.name = name;
242        this.preemption_rate = preemption_rate;
243        ready_queue{};
244        ready_queue_lock{};
245
246        procs{ __get };
247        idles{ __get };
248        threads{ __get };
249
250        doregister(this);
251}
252
253void ^?{}(cluster & this) {
254        unregister(this);
255}
256
257//=============================================================================================
258// Kernel Scheduling logic
259//=============================================================================================
260static void runThread(processor * this, thread_desc * dst);
261static void finishRunning(processor * this);
262static void halt(processor * this);
263
264//Main of the processor contexts
265void main(processorCtx_t & runner) {
266        // Because of a bug, we couldn't initialized the seed on construction
267        // Do it here
268        kernelTLS.rand_seed ^= rdtscl();
269
270        processor * this = runner.proc;
271        verify(this);
272
273        __cfaabi_dbg_print_safe("Kernel : core %p starting\n", this);
274
275        doregister(this->cltr, this);
276
277        {
278                // Setup preemption data
279                preemption_scope scope = { this };
280
281                __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);
282
283                thread_desc * readyThread = 0p;
284                for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
285                        readyThread = nextThread( this->cltr );
286
287                        if(readyThread) {
288                                verify( ! kernelTLS.preemption_state.enabled );
289
290                                runThread(this, readyThread);
291
292                                verify( ! kernelTLS.preemption_state.enabled );
293
294                                //Some actions need to be taken from the kernel
295                                finishRunning(this);
296
297                                spin_count = 0;
298                        } else {
299                                // spin(this, &spin_count);
300                                halt(this);
301                        }
302                }
303
304                __cfaabi_dbg_print_safe("Kernel : core %p stopping\n", this);
305        }
306
307        unregister(this->cltr, this);
308
309        V( this->terminated );
310
311        __cfaabi_dbg_print_safe("Kernel : core %p terminated\n", this);
312}
313
314static int * __volatile_errno() __attribute__((noinline));
315static int * __volatile_errno() { asm(""); return &errno; }
316
317// KERNEL ONLY
318// runThread runs a thread by context switching
319// from the processor coroutine to the target thread
320static void runThread(processor * this, thread_desc * thrd_dst) {
321        coroutine_desc * proc_cor = get_coroutine(this->runner);
322
323        // Reset the terminating actions here
324        this->finish.action_code = No_Action;
325
326        // Update global state
327        kernelTLS.this_thread = thrd_dst;
328
329        // set state of processor coroutine to inactive and the thread to active
330        proc_cor->state = proc_cor->state == Halted ? Halted : Inactive;
331        thrd_dst->state = Active;
332
333        // set context switch to the thread that the processor is executing
334        verify( thrd_dst->context.SP );
335        CtxSwitch( &proc_cor->context, &thrd_dst->context );
336        // when CtxSwitch returns we are back in the processor coroutine
337
338        // set state of processor coroutine to active and the thread to inactive
339        thrd_dst->state = thrd_dst->state == Halted ? Halted : Inactive;
340        proc_cor->state = Active;
341}
342
343// KERNEL_ONLY
344static void returnToKernel() {
345        coroutine_desc * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
346        thread_desc * thrd_src = kernelTLS.this_thread;
347
348        // set state of current coroutine to inactive
349        thrd_src->state = thrd_src->state == Halted ? Halted : Inactive;
350        proc_cor->state = Active;
351        int local_errno = *__volatile_errno();
352        #if defined( __i386 ) || defined( __x86_64 )
353                __x87_store;
354        #endif
355
356        // set new coroutine that the processor is executing
357        // and context switch to it
358        verify( proc_cor->context.SP );
359        CtxSwitch( &thrd_src->context, &proc_cor->context );
360
361        // set state of new coroutine to active
362        proc_cor->state = proc_cor->state == Halted ? Halted : Inactive;
363        thrd_src->state = Active;
364
365        #if defined( __i386 ) || defined( __x86_64 )
366                __x87_load;
367        #endif
368        *__volatile_errno() = local_errno;
369}
370
371// KERNEL_ONLY
372// Once a thread has finished running, some of
373// its final actions must be executed from the kernel
374static void finishRunning(processor * this) with( this->finish ) {
375        verify( ! kernelTLS.preemption_state.enabled );
376        choose( action_code ) {
377        case No_Action:
378                break;
379        case Release:
380                unlock( *lock );
381        case Schedule:
382                ScheduleThread( thrd );
383        case Release_Schedule:
384                unlock( *lock );
385                ScheduleThread( thrd );
386        case Release_Multi:
387                for(int i = 0; i < lock_count; i++) {
388                        unlock( *locks[i] );
389                }
390        case Release_Multi_Schedule:
391                for(int i = 0; i < lock_count; i++) {
392                        unlock( *locks[i] );
393                }
394                for(int i = 0; i < thrd_count; i++) {
395                        ScheduleThread( thrds[i] );
396                }
397        case Callback:
398                callback();
399        default:
400                abort("KERNEL ERROR: Unexpected action to run after thread");
401        }
402}
403
404// KERNEL_ONLY
405// Context invoker for processors
406// This is the entry point for processors (kernel threads)
407// It effectively constructs a coroutine by stealing the pthread stack
408static void * CtxInvokeProcessor(void * arg) {
409        processor * proc = (processor *) arg;
410        kernelTLS.this_processor = proc;
411        kernelTLS.this_thread    = 0p;
412        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
413        // SKULLDUGGERY: We want to create a context for the processor coroutine
414        // which is needed for the 2-step context switch. However, there is no reason
415        // to waste the perfectly valid stack create by pthread.
416        current_stack_info_t info;
417        __stack_t ctx;
418        info.storage = &ctx;
419        (proc->runner){ proc, &info };
420
421        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
422
423        //Set global state
424        kernelTLS.this_thread = 0p;
425
426        //We now have a proper context from which to schedule threads
427        __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
428
429        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
430        // resume it to start it like it normally would, it will just context switch
431        // back to here. Instead directly call the main since we already are on the
432        // appropriate stack.
433        get_coroutine(proc->runner)->state = Active;
434        main( proc->runner );
435        get_coroutine(proc->runner)->state = Halted;
436
437        // Main routine of the core returned, the core is now fully terminated
438        __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner);
439
440        return 0p;
441}
442
443static void Abort( int ret, const char * func ) {
444        if ( ret ) {                                                                            // pthread routines return errno values
445                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
446        } // if
447} // Abort
448
449void * create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
450        pthread_attr_t attr;
451
452        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
453
454        size_t stacksize;
455        // default stack size, normally defined by shell limit
456        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
457        assert( stacksize >= PTHREAD_STACK_MIN );
458
459        void * stack;
460        __cfaabi_dbg_debug_do(
461                stack = memalign( __page_size, stacksize + __page_size );
462                // pthread has no mechanism to create the guard page in user supplied stack.
463                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
464                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
465                } // if
466        );
467        __cfaabi_dbg_no_debug_do(
468                stack = malloc( stacksize );
469        );
470
471        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
472
473        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
474        return stack;
475}
476
477static void start(processor * this) {
478        __cfaabi_dbg_print_safe("Kernel : Starting core %p\n", this);
479
480        this->stack = create_pthread( &this->kernel_thread, CtxInvokeProcessor, (void *)this );
481
482        __cfaabi_dbg_print_safe("Kernel : core %p started\n", this);
483}
484
485// KERNEL_ONLY
486void kernel_first_resume( processor * this ) {
487        thread_desc * src = mainThread;
488        coroutine_desc * dst = get_coroutine(this->runner);
489
490        verify( ! kernelTLS.preemption_state.enabled );
491
492        __stack_prepare( &dst->stack, 65000 );
493        CtxStart(&this->runner, CtxInvokeCoroutine);
494
495        verify( ! kernelTLS.preemption_state.enabled );
496
497        dst->last = &src->self_cor;
498        dst->starter = dst->starter ? dst->starter : &src->self_cor;
499
500        // set state of current coroutine to inactive
501        src->state = src->state == Halted ? Halted : Inactive;
502
503        // context switch to specified coroutine
504        verify( dst->context.SP );
505        CtxSwitch( &src->context, &dst->context );
506        // when CtxSwitch returns we are back in the src coroutine
507
508        // set state of new coroutine to active
509        src->state = Active;
510
511        verify( ! kernelTLS.preemption_state.enabled );
512}
513
514// KERNEL_ONLY
515void kernel_last_resume( processor * this ) {
516        coroutine_desc * src = &mainThread->self_cor;
517        coroutine_desc * dst = get_coroutine(this->runner);
518
519        verify( ! kernelTLS.preemption_state.enabled );
520        verify( dst->starter == src );
521        verify( dst->context.SP );
522
523        // context switch to the processor
524        CtxSwitch( &src->context, &dst->context );
525}
526
527//-----------------------------------------------------------------------------
528// Scheduler routines
529
530// KERNEL ONLY
531void ScheduleThread( thread_desc * thrd ) {
532        verify( thrd );
533        verify( thrd->state != Halted );
534
535        verify( ! kernelTLS.preemption_state.enabled );
536
537        verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
538
539        with( *thrd->curr_cluster ) {
540                lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
541                bool was_empty = !(ready_queue != 0);
542                append( ready_queue, thrd );
543                unlock( ready_queue_lock );
544
545                if(was_empty) {
546                        lock      (proc_list_lock __cfaabi_dbg_ctx2);
547                        if(idles) {
548                                wake_fast(idles.head);
549                        }
550                        unlock    (proc_list_lock);
551                }
552                else if( struct processor * idle = idles.head ) {
553                        wake_fast(idle);
554                }
555
556        }
557
558        verify( ! kernelTLS.preemption_state.enabled );
559}
560
561// KERNEL ONLY
562thread_desc * nextThread(cluster * this) with( *this ) {
563        verify( ! kernelTLS.preemption_state.enabled );
564        lock( ready_queue_lock __cfaabi_dbg_ctx2 );
565        thread_desc * head = pop_head( ready_queue );
566        unlock( ready_queue_lock );
567        verify( ! kernelTLS.preemption_state.enabled );
568        return head;
569}
570
571void BlockInternal() {
572        disable_interrupts();
573        verify( ! kernelTLS.preemption_state.enabled );
574        returnToKernel();
575        verify( ! kernelTLS.preemption_state.enabled );
576        enable_interrupts( __cfaabi_dbg_ctx );
577}
578
579void BlockInternal( __spinlock_t * lock ) {
580        disable_interrupts();
581        with( *kernelTLS.this_processor ) {
582                finish.action_code = Release;
583                finish.lock        = lock;
584        }
585
586        verify( ! kernelTLS.preemption_state.enabled );
587        returnToKernel();
588        verify( ! kernelTLS.preemption_state.enabled );
589
590        enable_interrupts( __cfaabi_dbg_ctx );
591}
592
593void BlockInternal( thread_desc * thrd ) {
594        disable_interrupts();
595        with( * kernelTLS.this_processor ) {
596                finish.action_code = Schedule;
597                finish.thrd        = thrd;
598        }
599
600        verify( ! kernelTLS.preemption_state.enabled );
601        returnToKernel();
602        verify( ! kernelTLS.preemption_state.enabled );
603
604        enable_interrupts( __cfaabi_dbg_ctx );
605}
606
607void BlockInternal( __spinlock_t * lock, thread_desc * thrd ) {
608        assert(thrd);
609        disable_interrupts();
610        with( * kernelTLS.this_processor ) {
611                finish.action_code = Release_Schedule;
612                finish.lock        = lock;
613                finish.thrd        = thrd;
614        }
615
616        verify( ! kernelTLS.preemption_state.enabled );
617        returnToKernel();
618        verify( ! kernelTLS.preemption_state.enabled );
619
620        enable_interrupts( __cfaabi_dbg_ctx );
621}
622
623void BlockInternal(__spinlock_t * locks [], unsigned short count) {
624        disable_interrupts();
625        with( * kernelTLS.this_processor ) {
626                finish.action_code = Release_Multi;
627                finish.locks       = locks;
628                finish.lock_count  = count;
629        }
630
631        verify( ! kernelTLS.preemption_state.enabled );
632        returnToKernel();
633        verify( ! kernelTLS.preemption_state.enabled );
634
635        enable_interrupts( __cfaabi_dbg_ctx );
636}
637
638void BlockInternal(__spinlock_t * locks [], unsigned short lock_count, thread_desc * thrds [], unsigned short thrd_count) {
639        disable_interrupts();
640        with( *kernelTLS.this_processor ) {
641                finish.action_code = Release_Multi_Schedule;
642                finish.locks       = locks;
643                finish.lock_count  = lock_count;
644                finish.thrds       = thrds;
645                finish.thrd_count  = thrd_count;
646        }
647
648        verify( ! kernelTLS.preemption_state.enabled );
649        returnToKernel();
650        verify( ! kernelTLS.preemption_state.enabled );
651
652        enable_interrupts( __cfaabi_dbg_ctx );
653}
654
655void BlockInternal(__finish_callback_fptr_t callback) {
656        disable_interrupts();
657        with( *kernelTLS.this_processor ) {
658                finish.action_code = Callback;
659                finish.callback    = callback;
660        }
661
662        verify( ! kernelTLS.preemption_state.enabled );
663        returnToKernel();
664        verify( ! kernelTLS.preemption_state.enabled );
665
666        enable_interrupts( __cfaabi_dbg_ctx );
667}
668
669// KERNEL ONLY
670void LeaveThread(__spinlock_t * lock, thread_desc * thrd) {
671        verify( ! kernelTLS.preemption_state.enabled );
672        with( * kernelTLS.this_processor ) {
673                finish.action_code = thrd ? Release_Schedule : Release;
674                finish.lock        = lock;
675                finish.thrd        = thrd;
676        }
677
678        returnToKernel();
679}
680
681//=============================================================================================
682// Kernel Setup logic
683//=============================================================================================
684//-----------------------------------------------------------------------------
685// Kernel boot procedures
686static void kernel_startup(void) {
687        verify( ! kernelTLS.preemption_state.enabled );
688        __cfaabi_dbg_print_safe("Kernel : Starting\n");
689
690        __page_size = sysconf( _SC_PAGESIZE );
691
692        __cfa_dbg_global_clusters.list{ __get };
693        __cfa_dbg_global_clusters.lock{};
694
695        // Initialize the main cluster
696        mainCluster = (cluster *)&storage_mainCluster;
697        (*mainCluster){"Main Cluster"};
698
699        __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n");
700
701        // Start by initializing the main thread
702        // SKULLDUGGERY: the mainThread steals the process main thread
703        // which will then be scheduled by the mainProcessor normally
704        mainThread = (thread_desc *)&storage_mainThread;
705        current_stack_info_t info;
706        info.storage = (__stack_t*)&storage_mainThreadCtx;
707        (*mainThread){ &info };
708
709        __cfaabi_dbg_print_safe("Kernel : Main thread ready\n");
710
711
712
713        // Construct the processor context of the main processor
714        void ?{}(processorCtx_t & this, processor * proc) {
715                (this.__cor){ "Processor" };
716                this.__cor.starter = 0p;
717                this.proc = proc;
718        }
719
720        void ?{}(processor & this) with( this ) {
721                name = "Main Processor";
722                cltr = mainCluster;
723                terminated{ 0 };
724                do_terminate = false;
725                preemption_alarm = 0p;
726                pending_preemption = false;
727                kernel_thread = pthread_self();
728
729                runner{ &this };
730                __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner);
731        }
732
733        // Initialize the main processor and the main processor ctx
734        // (the coroutine that contains the processing control flow)
735        mainProcessor = (processor *)&storage_mainProcessor;
736        (*mainProcessor){};
737
738        //initialize the global state variables
739        kernelTLS.this_processor = mainProcessor;
740        kernelTLS.this_thread    = mainThread;
741
742        // Enable preemption
743        kernel_start_preemption();
744
745        // Add the main thread to the ready queue
746        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
747        ScheduleThread(mainThread);
748
749        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
750        // context. Hence, the main thread does not begin through CtxInvokeThread, like all other threads. The trick here is that
751        // mainThread is on the ready queue when this call is made.
752        kernel_first_resume( kernelTLS.this_processor );
753
754
755
756        // THE SYSTEM IS NOW COMPLETELY RUNNING
757        __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n");
758
759        verify( ! kernelTLS.preemption_state.enabled );
760        enable_interrupts( __cfaabi_dbg_ctx );
761        verify( TL_GET( preemption_state.enabled ) );
762}
763
764static void kernel_shutdown(void) {
765        __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n");
766
767        verify( TL_GET( preemption_state.enabled ) );
768        disable_interrupts();
769        verify( ! kernelTLS.preemption_state.enabled );
770
771        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
772        // When its coroutine terminates, it return control to the mainThread
773        // which is currently here
774        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
775        kernel_last_resume( kernelTLS.this_processor );
776        mainThread->self_cor.state = Halted;
777
778        // THE SYSTEM IS NOW COMPLETELY STOPPED
779
780        // Disable preemption
781        kernel_stop_preemption();
782
783        // Destroy the main processor and its context in reverse order of construction
784        // These were manually constructed so we need manually destroy them
785        ^(mainProcessor->runner){};
786        ^(mainProcessor){};
787
788        // Final step, destroy the main thread since it is no longer needed
789        // Since we provided a stack to this taxk it will not destroy anything
790        ^(mainThread){};
791
792        ^(__cfa_dbg_global_clusters.list){};
793        ^(__cfa_dbg_global_clusters.lock){};
794
795        __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n");
796}
797
798//=============================================================================================
799// Kernel Quiescing
800//=============================================================================================
801static void halt(processor * this) with( *this ) {
802        // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) );
803
804        with( *cltr ) {
805                lock      (proc_list_lock __cfaabi_dbg_ctx2);
806                remove    (procs, *this);
807                push_front(idles, *this);
808                unlock    (proc_list_lock);
809        }
810
811        __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this);
812
813        wait( idleLock );
814
815        __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this);
816
817        with( *cltr ) {
818                lock      (proc_list_lock __cfaabi_dbg_ctx2);
819                remove    (idles, *this);
820                push_front(procs, *this);
821                unlock    (proc_list_lock);
822        }
823}
824
825//=============================================================================================
826// Unexpected Terminating logic
827//=============================================================================================
828static __spinlock_t kernel_abort_lock;
829static bool kernel_abort_called = false;
830
831void * kernel_abort(void) __attribute__ ((__nothrow__)) {
832        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
833        // the globalAbort flag is true.
834        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
835
836        // first task to abort ?
837        if ( kernel_abort_called ) {                    // not first task to abort ?
838                unlock( kernel_abort_lock );
839
840                sigset_t mask;
841                sigemptyset( &mask );
842                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
843                sigsuspend( &mask );                    // block the processor to prevent further damage during abort
844                _exit( EXIT_FAILURE );                  // if processor unblocks before it is killed, terminate it
845        }
846        else {
847                kernel_abort_called = true;
848                unlock( kernel_abort_lock );
849        }
850
851        return kernelTLS.this_thread;
852}
853
854void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
855        thread_desc * thrd = kernel_data;
856
857        if(thrd) {
858                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
859                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
860
861                if ( &thrd->self_cor != thrd->curr_cor ) {
862                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
863                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
864                }
865                else {
866                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
867                }
868        }
869        else {
870                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
871                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
872        }
873}
874
875int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
876        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
877}
878
879static __spinlock_t kernel_debug_lock;
880
881extern "C" {
882        void __cfaabi_bits_acquire() {
883                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
884        }
885
886        void __cfaabi_bits_release() {
887                unlock( kernel_debug_lock );
888        }
889}
890
891//=============================================================================================
892// Kernel Utilities
893//=============================================================================================
894//-----------------------------------------------------------------------------
895// Locks
896void  ?{}( semaphore & this, int count = 1 ) {
897        (this.lock){};
898        this.count = count;
899        (this.waiting){};
900}
901void ^?{}(semaphore & this) {}
902
903void P(semaphore & this) with( this ){
904        lock( lock __cfaabi_dbg_ctx2 );
905        count -= 1;
906        if ( count < 0 ) {
907                // queue current task
908                append( waiting, kernelTLS.this_thread );
909
910                // atomically release spin lock and block
911                BlockInternal( &lock );
912        }
913        else {
914            unlock( lock );
915        }
916}
917
918void V(semaphore & this) with( this ) {
919        thread_desc * thrd = 0p;
920        lock( lock __cfaabi_dbg_ctx2 );
921        count += 1;
922        if ( count <= 0 ) {
923                // remove task at head of waiting list
924                thrd = pop_head( waiting );
925        }
926
927        unlock( lock );
928
929        // make new owner
930        WakeThread( thrd );
931}
932
933//-----------------------------------------------------------------------------
934// Global Queues
935void doregister( cluster     & cltr ) {
936        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
937        push_front( __cfa_dbg_global_clusters.list, cltr );
938        unlock    ( __cfa_dbg_global_clusters.lock );
939}
940
941void unregister( cluster     & cltr ) {
942        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
943        remove( __cfa_dbg_global_clusters.list, cltr );
944        unlock( __cfa_dbg_global_clusters.lock );
945}
946
947void doregister( cluster * cltr, thread_desc & thrd ) {
948        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
949        cltr->nthreads += 1;
950        push_front(cltr->threads, thrd);
951        unlock    (cltr->thread_list_lock);
952}
953
954void unregister( cluster * cltr, thread_desc & thrd ) {
955        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
956        remove(cltr->threads, thrd );
957        cltr->nthreads -= 1;
958        unlock(cltr->thread_list_lock);
959}
960
961void doregister( cluster * cltr, processor * proc ) {
962        lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2);
963        cltr->nprocessors += 1;
964        push_front(cltr->procs, *proc);
965        unlock    (cltr->proc_list_lock);
966}
967
968void unregister( cluster * cltr, processor * proc ) {
969        lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2);
970        remove(cltr->procs, *proc );
971        cltr->nprocessors -= 1;
972        unlock(cltr->proc_list_lock);
973}
974
975//-----------------------------------------------------------------------------
976// Debug
977__cfaabi_dbg_debug_do(
978        extern "C" {
979                void __cfaabi_dbg_record(__spinlock_t & this, const char * prev_name) {
980                        this.prev_name = prev_name;
981                        this.prev_thrd = kernelTLS.this_thread;
982                }
983        }
984)
985
986//-----------------------------------------------------------------------------
987// Debug
988bool threading_enabled(void) {
989        return true;
990}
991// Local Variables: //
992// mode: c //
993// tab-width: 4 //
994// End: //
Note: See TracBrowser for help on using the repository browser.