source: libcfa/src/concurrency/kernel.cfa @ 276ae57e

arm-ehjacob/cs343-translationnew-astnew-ast-unique-expr
Last change on this file since 276ae57e was 276ae57e, checked in by Thierry Delisle <tdelisle@…>, 19 months ago

Fix yesterday's checks where the stale flag was wrong

  • Property mode set to 100644
File size: 31.5 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($thread, mainThread);
120KERNEL_STORAGE(__stack_t,       mainThreadCtx);
121
122cluster     * mainCluster;
123processor   * mainProcessor;
124$thread * 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 & 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 & 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 * __next_thread(cluster * this);
267static void __run_thread(processor * this, $thread * 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 * 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 * thrd_dst) {
326        $coroutine * 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                __cfaabi_dbg_debug_do(
346                        thrd_dst->park_stale   = true;
347                        thrd_dst->unpark_stale = true;
348                )
349
350                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
351                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) || thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst ); // add escape condition if we are setting up the processor
352                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) || thrd_dst->curr_cor == proc_cor, "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst ); // add escape condition if we are setting up the processor
353
354                // set context switch to the thread that the processor is executing
355                verify( thrd_dst->context.SP );
356                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
357                // when __cfactx_switch returns we are back in the processor coroutine
358
359                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too large.\n", thrd_dst );
360                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ), "ERROR : Destination $thread %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
361                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
362
363
364                // We just finished running a thread, there are a few things that could have happened.
365                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
366                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
367                // 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
368                // 4 - Preempted
369                // In case 1, we may have won a race so we can't write to the state again.
370                // In case 2, we lost the race so we now own the thread.
371                // In case 3, we lost the race but can just reschedule the thread.
372
373                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
374                        // The thread was preempted, reschedule it and reset the flag
375                        __schedule_thread( thrd_dst );
376                        break RUNNING;
377                }
378
379                // set state of processor coroutine to active and the thread to inactive
380                static_assert(sizeof(thrd_dst->state) == sizeof(int));
381                enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Inactive, __ATOMIC_SEQ_CST);
382                switch(old_state) {
383                        case Halted:
384                                // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
385                                thrd_dst->state = Halted;
386
387                                // We may need to wake someone up here since
388                                unpark( this->destroyer __cfaabi_dbg_ctx2 );
389                                this->destroyer = 0p;
390                                break RUNNING;
391                        case Active:
392                                // This is case 1, the regular case, nothing more is needed
393                                break RUNNING;
394                        case Rerun:
395                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
396                                // In this case, just run it again.
397                                continue RUNNING;
398                        default:
399                                // This makes no sense, something is wrong abort
400                                abort("Finished running a thread that was Inactive/Start/Primed %d\n", old_state);
401                }
402        }
403
404        // Just before returning to the processor, set the processor coroutine to active
405        proc_cor->state = Active;
406}
407
408// KERNEL_ONLY
409void returnToKernel() {
410        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
411        $coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
412        $thread * thrd_src = kernelTLS.this_thread;
413
414        // Run the thread on this processor
415        {
416                int local_errno = *__volatile_errno();
417                #if defined( __i386 ) || defined( __x86_64 )
418                        __x87_store;
419                #endif
420                verify( proc_cor->context.SP );
421                __cfactx_switch( &thrd_src->context, &proc_cor->context );
422                #if defined( __i386 ) || defined( __x86_64 )
423                        __x87_load;
424                #endif
425                *__volatile_errno() = local_errno;
426        }
427
428        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
429        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too small.\n", thrd_src );
430        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit), "ERROR : Returning $thread %p has been corrupted.\n StackPointer too large.\n", thrd_src );
431}
432
433// KERNEL_ONLY
434// Context invoker for processors
435// This is the entry point for processors (kernel threads)
436// It effectively constructs a coroutine by stealing the pthread stack
437static void * __invoke_processor(void * arg) {
438        processor * proc = (processor *) arg;
439        kernelTLS.this_processor = proc;
440        kernelTLS.this_thread    = 0p;
441        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
442        // SKULLDUGGERY: We want to create a context for the processor coroutine
443        // which is needed for the 2-step context switch. However, there is no reason
444        // to waste the perfectly valid stack create by pthread.
445        current_stack_info_t info;
446        __stack_t ctx;
447        info.storage = &ctx;
448        (proc->runner){ proc, &info };
449
450        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
451
452        //Set global state
453        kernelTLS.this_thread = 0p;
454
455        //We now have a proper context from which to schedule threads
456        __cfaabi_dbg_print_safe("Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
457
458        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
459        // resume it to start it like it normally would, it will just context switch
460        // back to here. Instead directly call the main since we already are on the
461        // appropriate stack.
462        get_coroutine(proc->runner)->state = Active;
463        main( proc->runner );
464        get_coroutine(proc->runner)->state = Halted;
465
466        // Main routine of the core returned, the core is now fully terminated
467        __cfaabi_dbg_print_safe("Kernel : core %p main ended (%p)\n", proc, &proc->runner);
468
469        return 0p;
470}
471
472static void Abort( int ret, const char func[] ) {
473        if ( ret ) {                                                                            // pthread routines return errno values
474                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
475        } // if
476} // Abort
477
478void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
479        pthread_attr_t attr;
480
481        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
482
483        size_t stacksize;
484        // default stack size, normally defined by shell limit
485        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
486        assert( stacksize >= PTHREAD_STACK_MIN );
487
488        void * stack;
489        __cfaabi_dbg_debug_do(
490                stack = memalign( __page_size, stacksize + __page_size );
491                // pthread has no mechanism to create the guard page in user supplied stack.
492                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
493                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
494                } // if
495        );
496        __cfaabi_dbg_no_debug_do(
497                stack = malloc( stacksize );
498        );
499
500        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
501
502        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
503        return stack;
504}
505
506// KERNEL_ONLY
507static void __kernel_first_resume( processor * this ) {
508        $thread * src = mainThread;
509        $coroutine * dst = get_coroutine(this->runner);
510
511        verify( ! kernelTLS.preemption_state.enabled );
512
513        kernelTLS.this_thread->curr_cor = dst;
514        __stack_prepare( &dst->stack, 65000 );
515        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
516
517        verify( ! kernelTLS.preemption_state.enabled );
518
519        dst->last = &src->self_cor;
520        dst->starter = dst->starter ? dst->starter : &src->self_cor;
521
522        // set state of current coroutine to inactive
523        src->state = src->state == Halted ? Halted : Inactive;
524
525        // context switch to specified coroutine
526        verify( dst->context.SP );
527        __cfactx_switch( &src->context, &dst->context );
528        // when __cfactx_switch returns we are back in the src coroutine
529
530        mainThread->curr_cor = &mainThread->self_cor;
531
532        // set state of new coroutine to active
533        src->state = Active;
534
535        verify( ! kernelTLS.preemption_state.enabled );
536}
537
538// KERNEL_ONLY
539static void __kernel_last_resume( processor * this ) {
540        $coroutine * src = &mainThread->self_cor;
541        $coroutine * dst = get_coroutine(this->runner);
542
543        verify( ! kernelTLS.preemption_state.enabled );
544        verify( dst->starter == src );
545        verify( dst->context.SP );
546
547        // context switch to the processor
548        __cfactx_switch( &src->context, &dst->context );
549}
550
551//-----------------------------------------------------------------------------
552// Scheduler routines
553// KERNEL ONLY
554void __schedule_thread( $thread * thrd ) with( *thrd->curr_cluster ) {
555        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
556        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
557        /* paranoid */ if( thrd->state == Inactive || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
558                          "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
559        /* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active || thrd->state == Rerun,
560                          "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
561        /* paranoid */ #endif
562        /* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
563
564        lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
565        bool was_empty = !(ready_queue != 0);
566        append( ready_queue, thrd );
567        unlock( ready_queue_lock );
568
569        if(was_empty) {
570                lock      (proc_list_lock __cfaabi_dbg_ctx2);
571                if(idles) {
572                        wake_fast(idles.head);
573                }
574                unlock    (proc_list_lock);
575        }
576        else if( struct processor * idle = idles.head ) {
577                wake_fast(idle);
578        }
579
580        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
581}
582
583// KERNEL ONLY
584static $thread * __next_thread(cluster * this) with( *this ) {
585        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
586
587        lock( ready_queue_lock __cfaabi_dbg_ctx2 );
588        $thread * head = pop_head( ready_queue );
589        unlock( ready_queue_lock );
590
591        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
592        return head;
593}
594
595void unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
596        if( !thrd ) return;
597
598        disable_interrupts();
599        static_assert(sizeof(thrd->state) == sizeof(int));
600       
601        // record activity
602        __cfaabi_dbg_record_thrd( *thrd, false, caller );
603
604        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
605        switch(old_state) {
606                case Active:
607                        // Wake won the race, the thread will reschedule/rerun itself
608                        break;
609                case Inactive:
610                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
611
612                        // Wake lost the race,
613                        thrd->state = Inactive;
614                        __schedule_thread( thrd );
615                        break;
616                case Rerun:
617                        abort("More than one thread attempted to schedule thread %p\n", thrd);
618                        break;
619                case Halted:
620                case Start:
621                case Primed:
622                default:
623                        // This makes no sense, something is wrong abort
624                        abort();
625        }
626        enable_interrupts( __cfaabi_dbg_ctx );
627}
628
629void park( __cfaabi_dbg_ctx_param ) {
630        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
631        disable_interrupts();
632        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
633        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
634
635        // record activity
636        __cfaabi_dbg_record_thrd( *kernelTLS.this_thread, true, caller );
637
638        returnToKernel();
639
640        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
641        enable_interrupts( __cfaabi_dbg_ctx );
642        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
643
644}
645
646// KERNEL ONLY
647void __leave_thread() {
648        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
649        returnToKernel();
650        abort();
651}
652
653// KERNEL ONLY
654bool force_yield( __Preemption_Reason reason ) {
655        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
656        disable_interrupts();
657        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
658
659        $thread * thrd = kernelTLS.this_thread;
660        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun);
661
662        // SKULLDUGGERY: It is possible that we are preempting this thread just before
663        // it was going to park itself. If that is the case and it is already using the
664        // intrusive fields then we can't use them to preempt the thread
665        // If that is the case, abandon the preemption.
666        bool preempted = false;
667        if(thrd->next == 0p) {
668                preempted = true;
669                thrd->preempted = reason;
670                returnToKernel();
671        }
672
673        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
674        enable_interrupts_noPoll();
675        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
676
677        return preempted;
678}
679
680//=============================================================================================
681// Kernel Setup logic
682//=============================================================================================
683//-----------------------------------------------------------------------------
684// Kernel boot procedures
685static void __kernel_startup(void) {
686        verify( ! kernelTLS.preemption_state.enabled );
687        __cfaabi_dbg_print_safe("Kernel : Starting\n");
688
689        __page_size = sysconf( _SC_PAGESIZE );
690
691        __cfa_dbg_global_clusters.list{ __get };
692        __cfa_dbg_global_clusters.lock{};
693
694        // Initialize the main cluster
695        mainCluster = (cluster *)&storage_mainCluster;
696        (*mainCluster){"Main Cluster"};
697
698        __cfaabi_dbg_print_safe("Kernel : Main cluster ready\n");
699
700        // Start by initializing the main thread
701        // SKULLDUGGERY: the mainThread steals the process main thread
702        // which will then be scheduled by the mainProcessor normally
703        mainThread = ($thread *)&storage_mainThread;
704        current_stack_info_t info;
705        info.storage = (__stack_t*)&storage_mainThreadCtx;
706        (*mainThread){ &info };
707
708        __cfaabi_dbg_print_safe("Kernel : Main thread ready\n");
709
710
711
712        // Construct the processor context of the main processor
713        void ?{}(processorCtx_t & this, processor * proc) {
714                (this.__cor){ "Processor" };
715                this.__cor.starter = 0p;
716                this.proc = proc;
717        }
718
719        void ?{}(processor & this) with( this ) {
720                name = "Main Processor";
721                cltr = mainCluster;
722                terminated{ 0 };
723                do_terminate = false;
724                preemption_alarm = 0p;
725                pending_preemption = false;
726                kernel_thread = pthread_self();
727
728                runner{ &this };
729                __cfaabi_dbg_print_safe("Kernel : constructed main processor context %p\n", &runner);
730        }
731
732        // Initialize the main processor and the main processor ctx
733        // (the coroutine that contains the processing control flow)
734        mainProcessor = (processor *)&storage_mainProcessor;
735        (*mainProcessor){};
736
737        //initialize the global state variables
738        kernelTLS.this_processor = mainProcessor;
739        kernelTLS.this_thread    = mainThread;
740
741        // Enable preemption
742        kernel_start_preemption();
743
744        // Add the main thread to the ready queue
745        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
746        __schedule_thread(mainThread);
747
748        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
749        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
750        // mainThread is on the ready queue when this call is made.
751        __kernel_first_resume( kernelTLS.this_processor );
752
753
754
755        // THE SYSTEM IS NOW COMPLETELY RUNNING
756        __cfaabi_dbg_print_safe("Kernel : Started\n--------------------------------------------------\n\n");
757
758        verify( ! kernelTLS.preemption_state.enabled );
759        enable_interrupts( __cfaabi_dbg_ctx );
760        verify( TL_GET( preemption_state.enabled ) );
761}
762
763static void __kernel_shutdown(void) {
764        __cfaabi_dbg_print_safe("\n--------------------------------------------------\nKernel : Shutting down\n");
765
766        verify( TL_GET( preemption_state.enabled ) );
767        disable_interrupts();
768        verify( ! kernelTLS.preemption_state.enabled );
769
770        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
771        // When its coroutine terminates, it return control to the mainThread
772        // which is currently here
773        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
774        __kernel_last_resume( kernelTLS.this_processor );
775        mainThread->self_cor.state = Halted;
776
777        // THE SYSTEM IS NOW COMPLETELY STOPPED
778
779        // Disable preemption
780        kernel_stop_preemption();
781
782        // Destroy the main processor and its context in reverse order of construction
783        // These were manually constructed so we need manually destroy them
784        ^(*mainProcessor){};
785
786        // Final step, destroy the main thread since it is no longer needed
787        // Since we provided a stack to this taxk it will not destroy anything
788        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
789        ^(*mainThread){};
790
791        ^(__cfa_dbg_global_clusters.list){};
792        ^(__cfa_dbg_global_clusters.lock){};
793
794        __cfaabi_dbg_print_safe("Kernel : Shutdown complete\n");
795}
796
797//=============================================================================================
798// Kernel Quiescing
799//=============================================================================================
800static void __halt(processor * this) with( *this ) {
801        // verify( ! __atomic_load_n(&do_terminate, __ATOMIC_SEQ_CST) );
802
803        with( *cltr ) {
804                lock      (proc_list_lock __cfaabi_dbg_ctx2);
805                remove    (procs, *this);
806                push_front(idles, *this);
807                unlock    (proc_list_lock);
808        }
809
810        __cfaabi_dbg_print_safe("Kernel : Processor %p ready to sleep\n", this);
811
812        wait( idleLock );
813
814        __cfaabi_dbg_print_safe("Kernel : Processor %p woke up and ready to run\n", this);
815
816        with( *cltr ) {
817                lock      (proc_list_lock __cfaabi_dbg_ctx2);
818                remove    (idles, *this);
819                push_front(procs, *this);
820                unlock    (proc_list_lock);
821        }
822}
823
824//=============================================================================================
825// Unexpected Terminating logic
826//=============================================================================================
827static __spinlock_t kernel_abort_lock;
828static bool kernel_abort_called = false;
829
830void * kernel_abort(void) __attribute__ ((__nothrow__)) {
831        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
832        // the globalAbort flag is true.
833        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
834
835        // first task to abort ?
836        if ( kernel_abort_called ) {                    // not first task to abort ?
837                unlock( kernel_abort_lock );
838
839                sigset_t mask;
840                sigemptyset( &mask );
841                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
842                sigaddset( &mask, SIGUSR1 );            // 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 * 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                unlock( lock );
912                park( __cfaabi_dbg_ctx );
913        }
914        else {
915            unlock( lock );
916        }
917}
918
919void V(semaphore & this) with( this ) {
920        $thread * thrd = 0p;
921        lock( lock __cfaabi_dbg_ctx2 );
922        count += 1;
923        if ( count <= 0 ) {
924                // remove task at head of waiting list
925                thrd = pop_head( waiting );
926        }
927
928        unlock( lock );
929
930        // make new owner
931        unpark( thrd __cfaabi_dbg_ctx2 );
932}
933
934//-----------------------------------------------------------------------------
935// Global Queues
936void doregister( cluster     & cltr ) {
937        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
938        push_front( __cfa_dbg_global_clusters.list, cltr );
939        unlock    ( __cfa_dbg_global_clusters.lock );
940}
941
942void unregister( cluster     & cltr ) {
943        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
944        remove( __cfa_dbg_global_clusters.list, cltr );
945        unlock( __cfa_dbg_global_clusters.lock );
946}
947
948void doregister( cluster * cltr, $thread & thrd ) {
949        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
950        cltr->nthreads += 1;
951        push_front(cltr->threads, thrd);
952        unlock    (cltr->thread_list_lock);
953}
954
955void unregister( cluster * cltr, $thread & thrd ) {
956        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
957        remove(cltr->threads, thrd );
958        cltr->nthreads -= 1;
959        unlock(cltr->thread_list_lock);
960}
961
962void doregister( cluster * cltr, processor * proc ) {
963        lock      (cltr->proc_list_lock __cfaabi_dbg_ctx2);
964        cltr->nprocessors += 1;
965        push_front(cltr->procs, *proc);
966        unlock    (cltr->proc_list_lock);
967}
968
969void unregister( cluster * cltr, processor * proc ) {
970        lock  (cltr->proc_list_lock __cfaabi_dbg_ctx2);
971        remove(cltr->procs, *proc );
972        cltr->nprocessors -= 1;
973        unlock(cltr->proc_list_lock);
974}
975
976//-----------------------------------------------------------------------------
977// Debug
978__cfaabi_dbg_debug_do(
979        extern "C" {
980                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
981                        this.prev_name = prev_name;
982                        this.prev_thrd = kernelTLS.this_thread;
983                }
984
985                void __cfaabi_dbg_record_thrd($thread & this, bool park, const char prev_name[]) {
986                        if(park) {
987                                this.prev_park    = prev_name;
988                                this.park_stale   = false;
989                        }
990                        else {
991                                this.prev_unpark  = prev_name;
992                                this.unpark_stale = false;
993                        }
994                }
995        }
996)
997
998//-----------------------------------------------------------------------------
999// Debug
1000bool threading_enabled(void) __attribute__((const)) {
1001        return true;
1002}
1003// Local Variables: //
1004// mode: c //
1005// tab-width: 4 //
1006// End: //
Note: See TracBrowser for help on using the repository browser.