source: libcfa/src/concurrency/kernel.cfa @ 6a490b2

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum
Last change on this file since 6a490b2 was 6a490b2, checked in by Thierry Delisle <tdelisle@…>, 4 years ago

Merge branch 'master' into relaxed_ready
  • Property mode set to 100644
File size: 35.6 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// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
18
19//C Includes
20#include <stddef.h>
21#include <errno.h>
22#include <string.h>
23extern "C" {
24#include <stdio.h>
25#include <fenv.h>
26#include <sys/resource.h>
27#include <signal.h>
28#include <unistd.h>
29#include <limits.h>                                                                             // PTHREAD_STACK_MIN
30#include <sys/mman.h>                                                                   // mprotect
31}
32
33//CFA Includes
34#include "time.hfa"
35#include "kernel_private.hfa"
36#include "preemption.hfa"
37#include "startup.hfa"
38
39//Private includes
40#define __CFA_INVOKE_PRIVATE__
41#include "invoke.h"
42
43
44//-----------------------------------------------------------------------------
45// Some assembly required
46#if defined( __i386 )
47        #define CtxGet( ctx )        \
48                __asm__ volatile (     \
49                        "movl %%esp,%0\n"\
50                        "movl %%ebp,%1\n"\
51                        : "=rm" (ctx.SP),\
52                                "=rm" (ctx.FP) \
53                )
54
55        // mxcr : SSE Status and Control bits (control bits are preserved across function calls)
56        // fcw  : X87 FPU control word (preserved across function calls)
57        #define __x87_store         \
58                uint32_t __mxcr;      \
59                uint16_t __fcw;       \
60                __asm__ volatile (    \
61                        "stmxcsr %0\n"  \
62                        "fnstcw  %1\n"  \
63                        : "=m" (__mxcr),\
64                                "=m" (__fcw)  \
65                )
66
67        #define __x87_load         \
68                __asm__ volatile (   \
69                        "fldcw  %1\n"  \
70                        "ldmxcsr %0\n" \
71                        ::"m" (__mxcr),\
72                                "m" (__fcw)  \
73                )
74
75#elif defined( __x86_64 )
76        #define CtxGet( ctx )        \
77                __asm__ volatile (     \
78                        "movq %%rsp,%0\n"\
79                        "movq %%rbp,%1\n"\
80                        : "=rm" (ctx.SP),\
81                                "=rm" (ctx.FP) \
82                )
83
84        #define __x87_store         \
85                uint32_t __mxcr;      \
86                uint16_t __fcw;       \
87                __asm__ volatile (    \
88                        "stmxcsr %0\n"  \
89                        "fnstcw  %1\n"  \
90                        : "=m" (__mxcr),\
91                                "=m" (__fcw)  \
92                )
93
94        #define __x87_load          \
95                __asm__ volatile (    \
96                        "fldcw  %1\n"   \
97                        "ldmxcsr %0\n"  \
98                        :: "m" (__mxcr),\
99                                "m" (__fcw)  \
100                )
101
102
103#elif defined( __ARM_ARCH )
104#define CtxGet( ctx ) __asm__ ( \
105                "mov %0,%%sp\n"   \
106                "mov %1,%%r11\n"   \
107        : "=rm" (ctx.SP), "=rm" (ctx.FP) )
108#else
109        #error unknown hardware architecture
110#endif
111
112//-----------------------------------------------------------------------------
113//Start and stop routine for the kernel, declared first to make sure they run first
114static void __kernel_startup (void) __attribute__(( constructor( STARTUP_PRIORITY_KERNEL ) ));
115static void __kernel_shutdown(void) __attribute__(( destructor ( STARTUP_PRIORITY_KERNEL ) ));
116
117//-----------------------------------------------------------------------------
118// Kernel Scheduling logic
119static $thread * __next_thread(cluster * this);
120static void __run_thread(processor * this, $thread * dst);
121static $thread * __halt(processor * this);
122static bool __wake_one(cluster * cltr, bool was_empty);
123static bool __wake_proc(processor *);
124
125//-----------------------------------------------------------------------------
126// Kernel storage
127KERNEL_STORAGE(cluster,         mainCluster);
128KERNEL_STORAGE(processor,       mainProcessor);
129KERNEL_STORAGE($thread, mainThread);
130KERNEL_STORAGE(__stack_t,       mainThreadCtx);
131
132cluster     * mainCluster;
133processor   * mainProcessor;
134$thread * mainThread;
135
136extern "C" {
137        struct { __dllist_t(cluster) list; __spinlock_t lock; } __cfa_dbg_global_clusters;
138}
139
140size_t __page_size = 0;
141
142//-----------------------------------------------------------------------------
143// Global state
144thread_local struct KernelThreadData kernelTLS __attribute__ ((tls_model ( "initial-exec" ))) = {
145        NULL,                                                                                           // cannot use 0p
146        NULL,
147        { 1, false, false },
148        6u //this should be seeded better but due to a bug calling rdtsc doesn't work
149};
150
151//-----------------------------------------------------------------------------
152// Struct to steal stack
153struct current_stack_info_t {
154        __stack_t * storage;                                                            // pointer to stack object
155        void * base;                                                                            // base of stack
156        void * limit;                                                                           // stack grows towards stack limit
157        void * context;                                                                         // address of cfa_context_t
158};
159
160void ?{}( current_stack_info_t & this ) {
161        __stack_context_t ctx;
162        CtxGet( ctx );
163        this.base = ctx.FP;
164
165        rlimit r;
166        getrlimit( RLIMIT_STACK, &r);
167        size_t size = r.rlim_cur;
168
169        this.limit = (void *)(((intptr_t)this.base) - size);
170        this.context = &storage_mainThreadCtx;
171}
172
173//-----------------------------------------------------------------------------
174// Main thread construction
175
176void ?{}( $coroutine & this, current_stack_info_t * info) with( this ) {
177        stack.storage = info->storage;
178        with(*stack.storage) {
179                limit     = info->limit;
180                base      = info->base;
181        }
182        __attribute__((may_alias)) intptr_t * istorage = (intptr_t*) &stack.storage;
183        *istorage |= 0x1;
184        name = "Main Thread";
185        state = Start;
186        starter = 0p;
187        last = 0p;
188        cancellation = 0p;
189}
190
191void ?{}( $thread & this, current_stack_info_t * info) with( this ) {
192        state = Start;
193        self_cor{ info };
194        curr_cor = &self_cor;
195        curr_cluster = mainCluster;
196        self_mon.owner = &this;
197        self_mon.recursion = 1;
198        self_mon_p = &self_mon;
199        link.next = 0p;
200        link.prev = 0p;
201
202        node.next = 0p;
203        node.prev = 0p;
204        doregister(curr_cluster, this);
205
206        monitors{ &self_mon_p, 1, (fptr_t)0 };
207}
208
209//-----------------------------------------------------------------------------
210// Processor coroutine
211void ?{}(processorCtx_t & this) {
212
213}
214
215// Construct the processor context of non-main processors
216static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
217        (this.__cor){ info };
218        this.proc = proc;
219}
220
221static void * __invoke_processor(void * arg);
222
223void ?{}(processor & this, const char name[], cluster & cltr) with( this ) {
224        this.name = name;
225        this.cltr = &cltr;
226        id = -1u;
227        terminated{ 0 };
228        destroyer = 0p;
229        do_terminate = false;
230        preemption_alarm = 0p;
231        pending_preemption = false;
232        runner.proc = &this;
233
234        idle{};
235
236        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
237
238        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
239
240        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
241}
242
243void ^?{}(processor & this) with( this ){
244        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
245                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
246
247                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
248                __wake_proc( &this );
249
250                P( terminated );
251                verify( kernelTLS.this_processor != &this);
252        }
253
254        int err = pthread_join( kernel_thread, 0p );
255        if( err != 0 ) abort("KERNEL ERROR: joining processor %p caused error %s\n", &this, strerror(err));
256
257        free( this.stack );
258}
259
260void ?{}(cluster & this, const char name[], Duration preemption_rate, int io_flags) with( this ) {
261        this.name = name;
262        this.preemption_rate = preemption_rate;
263        ready_queue{};
264        ready_lock{};
265
266        #if !defined(__CFA_NO_STATISTICS__)
267                print_stats = false;
268        #endif
269
270        procs{ __get };
271        idles{ __get };
272        threads{ __get };
273
274        __kernel_io_startup( this, io_flags, &this == mainCluster );
275
276        doregister(this);
277}
278
279void ^?{}(cluster & this) {
280        __kernel_io_shutdown( this, &this == mainCluster );
281
282        unregister(this);
283}
284
285//=============================================================================================
286// Kernel Scheduling logic
287//=============================================================================================
288//Main of the processor contexts
289void main(processorCtx_t & runner) {
290        // Because of a bug, we couldn't initialized the seed on construction
291        // Do it here
292        kernelTLS.rand_seed ^= rdtscl();
293
294        processor * this = runner.proc;
295        verify(this);
296
297        __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
298
299        // register the processor unless it's the main thread which is handled in the boot sequence
300        if(this != mainProcessor) {
301                this->id = doregister(this->cltr, this);
302                ready_queue_grow( this->cltr );
303        }
304
305
306        {
307                // Setup preemption data
308                preemption_scope scope = { this };
309
310                __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
311
312                $thread * readyThread = 0p;
313                for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
314                        // Try to get the next thread
315                        readyThread = __next_thread( this->cltr );
316
317                        // If no ready thread
318                        if( readyThread == 0p ) {
319                                // Block until a thread is ready
320                                readyThread = __halt(this);
321                        }
322
323                        // Check if we actually found a thread
324                        if( readyThread ) {
325                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
326                                /* paranoid */ verifyf( readyThread->state == Ready || readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
327                                /* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );
328
329                                // We found a thread run it
330                                __run_thread(this, readyThread);
331
332                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
333                        }
334                }
335
336                __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
337        }
338
339        V( this->terminated );
340
341        // unregister the processor unless it's the main thread which is handled in the boot sequence
342        if(this != mainProcessor) {
343                ready_queue_shrink( this->cltr );
344                unregister(this->cltr, this);
345        }
346        else {
347                // HACK : the coroutine context switch expects this_thread to be set
348                // and it make sense for it to be set in all other cases except here
349                // fake it
350                kernelTLS.this_thread = mainThread;
351        }
352
353        __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
354
355        stats_tls_tally(this->cltr);
356}
357
358static int * __volatile_errno() __attribute__((noinline));
359static int * __volatile_errno() { asm(""); return &errno; }
360
361// KERNEL ONLY
362// runThread runs a thread by context switching
363// from the processor coroutine to the target thread
364static void __run_thread(processor * this, $thread * thrd_dst) {
365        $coroutine * proc_cor = get_coroutine(this->runner);
366
367        // Update global state
368        kernelTLS.this_thread = thrd_dst;
369
370        // set state of processor coroutine to inactive
371        verify(proc_cor->state == Active);
372        proc_cor->state = Blocked;
373
374        // Actually run the thread
375        RUNNING:  while(true) {
376                if(unlikely(thrd_dst->preempted)) {
377                        thrd_dst->preempted = __NO_PREEMPTION;
378                        verify(thrd_dst->state == Active  || thrd_dst->state == Rerun);
379                } else {
380                        verify(thrd_dst->state == Blocked || thrd_dst->state == Ready); // Ready means scheduled normally, blocked means rerun
381                        thrd_dst->state = Active;
382                }
383
384                __cfaabi_dbg_debug_do(
385                        thrd_dst->park_stale   = true;
386                        thrd_dst->unpark_stale = true;
387                )
388
389                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
390                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
391                /* 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
392                /* 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
393
394                // set context switch to the thread that the processor is executing
395                verify( thrd_dst->context.SP );
396                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
397                // when __cfactx_switch returns we are back in the processor coroutine
398
399                /* 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 );
400                /* 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 );
401                /* paranoid */ verify( kernelTLS.this_thread == thrd_dst );
402                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
403
404
405                // We just finished running a thread, there are a few things that could have happened.
406                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
407                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
408                // 4 - Preempted
409                // In case 1, we may have won a race so we can't write to the state again.
410                // In case 2, we lost the race so we now own the thread.
411
412                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
413                        // The thread was preempted, reschedule it and reset the flag
414                        __schedule_thread( thrd_dst );
415                        break RUNNING;
416                }
417
418                // set state of processor coroutine to active and the thread to inactive
419                static_assert(sizeof(thrd_dst->state) == sizeof(int));
420                enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Blocked, __ATOMIC_SEQ_CST);
421                __cfaabi_dbg_debug_do( thrd_dst->park_result = old_state; )
422                switch(old_state) {
423                        case Halted:
424                                // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
425                                thrd_dst->state = Halted;
426
427                                // We may need to wake someone up here since
428                                unpark( this->destroyer __cfaabi_dbg_ctx2 );
429                                this->destroyer = 0p;
430                                break RUNNING;
431                        case Active:
432                                // This is case 1, the regular case, nothing more is needed
433                                break RUNNING;
434                        case Rerun:
435                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
436                                // In this case, just run it again.
437                                continue RUNNING;
438                        default:
439                                // This makes no sense, something is wrong abort
440                                abort("Finished running a thread that was Blocked/Start/Primed %d\n", old_state);
441                }
442        }
443
444        // Just before returning to the processor, set the processor coroutine to active
445        proc_cor->state = Active;
446        kernelTLS.this_thread = 0p;
447}
448
449// KERNEL_ONLY
450void returnToKernel() {
451        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
452        $coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
453        $thread * thrd_src = kernelTLS.this_thread;
454
455        // Run the thread on this processor
456        {
457                int local_errno = *__volatile_errno();
458                #if defined( __i386 ) || defined( __x86_64 )
459                        __x87_store;
460                #endif
461                verify( proc_cor->context.SP );
462                __cfactx_switch( &thrd_src->context, &proc_cor->context );
463                #if defined( __i386 ) || defined( __x86_64 )
464                        __x87_load;
465                #endif
466                *__volatile_errno() = local_errno;
467        }
468
469        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
470        /* 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 );
471        /* 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 );
472}
473
474// KERNEL_ONLY
475// Context invoker for processors
476// This is the entry point for processors (kernel threads)
477// It effectively constructs a coroutine by stealing the pthread stack
478static void * __invoke_processor(void * arg) {
479        processor * proc = (processor *) arg;
480        kernelTLS.this_processor = proc;
481        kernelTLS.this_thread    = 0p;
482        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
483        // SKULLDUGGERY: We want to create a context for the processor coroutine
484        // which is needed for the 2-step context switch. However, there is no reason
485        // to waste the perfectly valid stack create by pthread.
486        current_stack_info_t info;
487        __stack_t ctx;
488        info.storage = &ctx;
489        (proc->runner){ proc, &info };
490
491        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
492
493        //Set global state
494        kernelTLS.this_thread = 0p;
495
496        //We now have a proper context from which to schedule threads
497        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
498
499        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
500        // resume it to start it like it normally would, it will just context switch
501        // back to here. Instead directly call the main since we already are on the
502        // appropriate stack.
503        get_coroutine(proc->runner)->state = Active;
504        main( proc->runner );
505        get_coroutine(proc->runner)->state = Halted;
506
507        // Main routine of the core returned, the core is now fully terminated
508        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
509
510        return 0p;
511}
512
513static void Abort( int ret, const char func[] ) {
514        if ( ret ) {                                                                            // pthread routines return errno values
515                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
516        } // if
517} // Abort
518
519void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
520        pthread_attr_t attr;
521
522        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
523
524        size_t stacksize;
525        // default stack size, normally defined by shell limit
526        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
527        assert( stacksize >= PTHREAD_STACK_MIN );
528
529        void * stack;
530        __cfaabi_dbg_debug_do(
531                stack = memalign( __page_size, stacksize + __page_size );
532                // pthread has no mechanism to create the guard page in user supplied stack.
533                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
534                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
535                } // if
536        );
537        __cfaabi_dbg_no_debug_do(
538                stack = malloc( stacksize );
539        );
540
541        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
542
543        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
544        return stack;
545}
546
547// KERNEL_ONLY
548static void __kernel_first_resume( processor * this ) {
549        $thread * src = mainThread;
550        $coroutine * dst = get_coroutine(this->runner);
551
552        verify( ! kernelTLS.preemption_state.enabled );
553
554        kernelTLS.this_thread->curr_cor = dst;
555        __stack_prepare( &dst->stack, 65000 );
556        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
557
558        verify( ! kernelTLS.preemption_state.enabled );
559
560        dst->last = &src->self_cor;
561        dst->starter = dst->starter ? dst->starter : &src->self_cor;
562
563        // make sure the current state is still correct
564        /* paranoid */ verify(src->state == Ready);
565
566        // context switch to specified coroutine
567        verify( dst->context.SP );
568        __cfactx_switch( &src->context, &dst->context );
569        // when __cfactx_switch returns we are back in the src coroutine
570
571        mainThread->curr_cor = &mainThread->self_cor;
572
573        // make sure the current state has been update
574        /* paranoid */ verify(src->state == Active);
575
576        verify( ! kernelTLS.preemption_state.enabled );
577}
578
579// KERNEL_ONLY
580static void __kernel_last_resume( processor * this ) {
581        $coroutine * src = &mainThread->self_cor;
582        $coroutine * dst = get_coroutine(this->runner);
583
584        verify( ! kernelTLS.preemption_state.enabled );
585        verify( dst->starter == src );
586        verify( dst->context.SP );
587
588        // SKULLDUGGERY in debug the processors check that the
589        // stack is still within the limit of the stack limits after running a thread.
590        // that check doesn't make sense if we context switch to the processor using the
591        // coroutine semantics. Since this is a special case, use the current context
592        // info to populate these fields.
593        __cfaabi_dbg_debug_do(
594                __stack_context_t ctx;
595                CtxGet( ctx );
596                mainThread->context.SP = ctx.SP;
597                mainThread->context.FP = ctx.FP;
598        )
599
600        // context switch to the processor
601        __cfactx_switch( &src->context, &dst->context );
602}
603
604//-----------------------------------------------------------------------------
605// Scheduler routines
606// KERNEL ONLY
607void __schedule_thread( $thread * thrd ) {
608        /* paranoid */ verify( thrd );
609        /* paranoid */ verify( thrd->state != Halted );
610        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
611        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
612        /* paranoid */ if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
613                          "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
614        /* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active || thrd->state == Rerun,
615                          "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
616        /* paranoid */ #endif
617        /* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
618
619        if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
620
621        ready_schedule_lock(thrd->curr_cluster, kernelTLS.this_processor);
622                bool was_empty = push( thrd->curr_cluster, thrd );
623        ready_schedule_unlock(thrd->curr_cluster, kernelTLS.this_processor);
624
625        __wake_one(thrd->curr_cluster, was_empty);
626
627        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
628}
629
630// KERNEL ONLY
631static $thread * __next_thread(cluster * this) with( *this ) {
632        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
633
634        ready_schedule_lock(this, kernelTLS.this_processor);
635                $thread * head = pop( this );
636        ready_schedule_unlock(this, kernelTLS.this_processor);
637
638        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
639        return head;
640}
641
642// KERNEL ONLY unpark with out disabling interrupts
643void __unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
644        static_assert(sizeof(thrd->state) == sizeof(int));
645
646        // record activity
647        __cfaabi_dbg_record_thrd( *thrd, false, caller );
648
649        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
650        __cfaabi_dbg_debug_do( thrd->unpark_result = old_state; )
651        switch(old_state) {
652                case Active:
653                        // Wake won the race, the thread will reschedule/rerun itself
654                        break;
655                case Blocked:
656                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
657
658                        // Wake lost the race,
659                        thrd->state = Blocked;
660                        __schedule_thread( thrd );
661                        break;
662                case Rerun:
663                        abort("More than one thread attempted to schedule thread %p\n", thrd);
664                        break;
665                case Halted:
666                case Start:
667                case Primed:
668                default:
669                        // This makes no sense, something is wrong abort
670                        abort();
671        }
672}
673
674void unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
675        if( !thrd ) return;
676
677        disable_interrupts();
678        __unpark( thrd __cfaabi_dbg_ctx_fwd2 );
679        enable_interrupts( __cfaabi_dbg_ctx );
680}
681
682void park( __cfaabi_dbg_ctx_param ) {
683        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
684        disable_interrupts();
685        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
686        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
687
688        // record activity
689        __cfaabi_dbg_record_thrd( *kernelTLS.this_thread, true, caller );
690
691        returnToKernel();
692
693        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
694        enable_interrupts( __cfaabi_dbg_ctx );
695        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
696
697}
698
699// KERNEL ONLY
700void __leave_thread() {
701        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
702        returnToKernel();
703        abort();
704}
705
706// KERNEL ONLY
707bool force_yield( __Preemption_Reason reason ) {
708        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
709        disable_interrupts();
710        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
711
712        $thread * thrd = kernelTLS.this_thread;
713        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun);
714
715        // SKULLDUGGERY: It is possible that we are preempting this thread just before
716        // it was going to park itself. If that is the case and it is already using the
717        // intrusive fields then we can't use them to preempt the thread
718        // If that is the case, abandon the preemption.
719        bool preempted = false;
720        if(thrd->next == 0p) {
721                preempted = true;
722                thrd->preempted = reason;
723                returnToKernel();
724        }
725
726        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
727        enable_interrupts_noPoll();
728        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
729
730        return preempted;
731}
732
733//=============================================================================================
734// Kernel Setup logic
735//=============================================================================================
736//-----------------------------------------------------------------------------
737// Kernel boot procedures
738static void __kernel_startup(void) {
739        verify( ! kernelTLS.preemption_state.enabled );
740        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
741
742        __page_size = sysconf( _SC_PAGESIZE );
743
744        __cfa_dbg_global_clusters.list{ __get };
745        __cfa_dbg_global_clusters.lock{};
746
747        // Initialize the main cluster
748        mainCluster = (cluster *)&storage_mainCluster;
749        (*mainCluster){"Main Cluster"};
750
751        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
752
753        // Start by initializing the main thread
754        // SKULLDUGGERY: the mainThread steals the process main thread
755        // which will then be scheduled by the mainProcessor normally
756        mainThread = ($thread *)&storage_mainThread;
757        current_stack_info_t info;
758        info.storage = (__stack_t*)&storage_mainThreadCtx;
759        (*mainThread){ &info };
760
761        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
762
763
764
765        // Construct the processor context of the main processor
766        void ?{}(processorCtx_t & this, processor * proc) {
767                (this.__cor){ "Processor" };
768                this.__cor.starter = 0p;
769                this.proc = proc;
770        }
771
772        void ?{}(processor & this) with( this ) {
773                name = "Main Processor";
774                cltr = mainCluster;
775                terminated{ 0 };
776                do_terminate = false;
777                preemption_alarm = 0p;
778                pending_preemption = false;
779                kernel_thread = pthread_self();
780                id = -1u;
781
782                runner{ &this };
783                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
784        }
785
786        // Initialize the main processor and the main processor ctx
787        // (the coroutine that contains the processing control flow)
788        mainProcessor = (processor *)&storage_mainProcessor;
789        (*mainProcessor){};
790
791        mainProcessor->id = doregister(mainCluster, mainProcessor);
792
793        //initialize the global state variables
794        kernelTLS.this_processor = mainProcessor;
795        kernelTLS.this_thread    = mainThread;
796
797        // Enable preemption
798        kernel_start_preemption();
799
800        // Add the main thread to the ready queue
801        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
802        __schedule_thread(mainThread);
803
804        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
805        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
806        // mainThread is on the ready queue when this call is made.
807        __kernel_first_resume( kernelTLS.this_processor );
808
809
810        // THE SYSTEM IS NOW COMPLETELY RUNNING
811
812
813        // Now that the system is up, finish creating systems that need threading
814        __kernel_io_finish_start( *mainCluster );
815
816
817        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
818
819        verify( ! kernelTLS.preemption_state.enabled );
820        enable_interrupts( __cfaabi_dbg_ctx );
821        verify( TL_GET( preemption_state.enabled ) );
822}
823
824static void __kernel_shutdown(void) {
825        //Before we start shutting things down, wait for systems that need threading to shutdown
826        __kernel_io_prepare_stop( *mainCluster );
827
828        /* paranoid */ verify( TL_GET( preemption_state.enabled ) );
829        disable_interrupts();
830        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
831
832        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
833
834        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
835        // When its coroutine terminates, it return control to the mainThread
836        // which is currently here
837        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
838        __kernel_last_resume( kernelTLS.this_processor );
839        mainThread->self_cor.state = Halted;
840
841        // THE SYSTEM IS NOW COMPLETELY STOPPED
842
843        // Disable preemption
844        kernel_stop_preemption();
845
846        unregister(mainCluster, mainProcessor);
847
848        // Destroy the main processor and its context in reverse order of construction
849        // These were manually constructed so we need manually destroy them
850        void ^?{}(processor & this) with( this ){
851                /* paranoid */ verify( this.do_terminate == true );
852                __cfaabi_dbg_print_safe("Kernel : destroyed main processor context %p\n", &runner);
853        }
854
855        ^(*mainProcessor){};
856
857        // Final step, destroy the main thread since it is no longer needed
858
859        // Since we provided a stack to this taxk it will not destroy anything
860        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
861        ^(*mainThread){};
862
863        ^(*mainCluster){};
864
865        ^(__cfa_dbg_global_clusters.list){};
866        ^(__cfa_dbg_global_clusters.lock){};
867
868        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
869}
870
871//=============================================================================================
872// Kernel Idle Sleep
873//=============================================================================================
874static $thread * __halt(processor * this) with( *this ) {
875        if( do_terminate ) return 0p;
876
877        // First, lock the cluster idle
878        lock( cltr->idle_lock __cfaabi_dbg_ctx2 );
879
880        // Check if we can find a thread
881        if( $thread * found = __next_thread( cltr ) ) {
882                unlock( cltr->idle_lock );
883                return found;
884        }
885
886        // Move this processor from the active list to the idle list
887        move_to_front(cltr->procs, cltr->idles, *this);
888
889        // Unlock the idle lock so we don't go to sleep with a lock
890        unlock    (cltr->idle_lock);
891
892        // We are ready to sleep
893        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p ready to sleep\n", this);
894        wait( idle );
895
896        // We have woken up
897        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p woke up and ready to run\n", this);
898
899        // Get ourself off the idle list
900        with( *cltr ) {
901                lock  (idle_lock __cfaabi_dbg_ctx2);
902                move_to_front(idles, procs, *this);
903                unlock(idle_lock);
904        }
905
906        // Don't check the ready queue again, we may not be in a position to run a thread
907        return 0p;
908}
909
910// Wake a thread from the front if there are any
911static bool __wake_one(cluster * this, __attribute__((unused)) bool force) {
912        // if we don't want to force check if we know it's false
913        // if( !this->idles.head && !force ) return false;
914
915        // First, lock the cluster idle
916        lock( this->idle_lock __cfaabi_dbg_ctx2 );
917
918        // Check if there is someone to wake up
919        if( !this->idles.head ) {
920                // Nope unlock and return false
921                unlock( this->idle_lock );
922                return false;
923        }
924
925        // Wake them up
926        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this->idles.head);
927        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
928        post( this->idles.head->idle );
929
930        // Unlock and return true
931        unlock( this->idle_lock );
932        return true;
933}
934
935// Unconditionnaly wake a thread
936static bool __wake_proc(processor * this) {
937        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
938
939        disable_interrupts();
940                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
941                bool ret = post( this->idle );
942        enable_interrupts( __cfaabi_dbg_ctx );
943
944        return ret;
945}
946
947//=============================================================================================
948// Unexpected Terminating logic
949//=============================================================================================
950static __spinlock_t kernel_abort_lock;
951static bool kernel_abort_called = false;
952
953void * kernel_abort(void) __attribute__ ((__nothrow__)) {
954        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
955        // the globalAbort flag is true.
956        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
957
958        // first task to abort ?
959        if ( kernel_abort_called ) {                    // not first task to abort ?
960                unlock( kernel_abort_lock );
961
962                sigset_t mask;
963                sigemptyset( &mask );
964                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
965                sigaddset( &mask, SIGUSR1 );            // block SIGALRM signals
966                sigsuspend( &mask );                            // block the processor to prevent further damage during abort
967                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it
968        }
969        else {
970                kernel_abort_called = true;
971                unlock( kernel_abort_lock );
972        }
973
974        return kernelTLS.this_thread;
975}
976
977void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
978        $thread * thrd = kernel_data;
979
980        if(thrd) {
981                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
982                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
983
984                if ( &thrd->self_cor != thrd->curr_cor ) {
985                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
986                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
987                }
988                else {
989                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
990                }
991        }
992        else {
993                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
994                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
995        }
996}
997
998int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
999        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
1000}
1001
1002static __spinlock_t kernel_debug_lock;
1003
1004extern "C" {
1005        void __cfaabi_bits_acquire() {
1006                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
1007        }
1008
1009        void __cfaabi_bits_release() {
1010                unlock( kernel_debug_lock );
1011        }
1012}
1013
1014//=============================================================================================
1015// Kernel Utilities
1016//=============================================================================================
1017//-----------------------------------------------------------------------------
1018// Locks
1019void  ?{}( semaphore & this, int count = 1 ) {
1020        (this.lock){};
1021        this.count = count;
1022        (this.waiting){};
1023}
1024void ^?{}(semaphore & this) {}
1025
1026bool P(semaphore & this) with( this ){
1027        lock( lock __cfaabi_dbg_ctx2 );
1028        count -= 1;
1029        if ( count < 0 ) {
1030                // queue current task
1031                append( waiting, kernelTLS.this_thread );
1032
1033                // atomically release spin lock and block
1034                unlock( lock );
1035                park( __cfaabi_dbg_ctx );
1036                return true;
1037        }
1038        else {
1039            unlock( lock );
1040            return false;
1041        }
1042}
1043
1044bool V(semaphore & this) with( this ) {
1045        $thread * thrd = 0p;
1046        lock( lock __cfaabi_dbg_ctx2 );
1047        count += 1;
1048        if ( count <= 0 ) {
1049                // remove task at head of waiting list
1050                thrd = pop_head( waiting );
1051        }
1052
1053        unlock( lock );
1054
1055        // make new owner
1056        unpark( thrd __cfaabi_dbg_ctx2 );
1057
1058        return thrd != 0p;
1059}
1060
1061bool V(semaphore & this, unsigned diff) with( this ) {
1062        $thread * thrd = 0p;
1063        lock( lock __cfaabi_dbg_ctx2 );
1064        int release = max(-count, (int)diff);
1065        count += diff;
1066        for(release) {
1067                unpark( pop_head( waiting ) __cfaabi_dbg_ctx2 );
1068        }
1069
1070        unlock( lock );
1071
1072        return thrd != 0p;
1073}
1074
1075//-----------------------------------------------------------------------------
1076// Global Queues
1077void doregister( cluster     & cltr ) {
1078        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1079        push_front( __cfa_dbg_global_clusters.list, cltr );
1080        unlock    ( __cfa_dbg_global_clusters.lock );
1081}
1082
1083void unregister( cluster     & cltr ) {
1084        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1085        remove( __cfa_dbg_global_clusters.list, cltr );
1086        unlock( __cfa_dbg_global_clusters.lock );
1087}
1088
1089void doregister( cluster * cltr, $thread & thrd ) {
1090        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1091        cltr->nthreads += 1;
1092        push_front(cltr->threads, thrd);
1093        unlock    (cltr->thread_list_lock);
1094}
1095
1096void unregister( cluster * cltr, $thread & thrd ) {
1097        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1098        remove(cltr->threads, thrd );
1099        cltr->nthreads -= 1;
1100        unlock(cltr->thread_list_lock);
1101}
1102
1103//-----------------------------------------------------------------------------
1104// Debug
1105__cfaabi_dbg_debug_do(
1106        extern "C" {
1107                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
1108                        this.prev_name = prev_name;
1109                        this.prev_thrd = kernelTLS.this_thread;
1110                }
1111
1112                void __cfaabi_dbg_record_thrd($thread & this, bool park, const char prev_name[]) {
1113                        if(park) {
1114                                this.park_caller   = prev_name;
1115                                this.park_stale    = false;
1116                        }
1117                        else {
1118                                this.unpark_caller = prev_name;
1119                                this.unpark_stale  = false;
1120                        }
1121                }
1122        }
1123)
1124
1125//-----------------------------------------------------------------------------
1126// Debug
1127bool threading_enabled(void) __attribute__((const)) {
1128        return true;
1129}
1130// Local Variables: //
1131// mode: c //
1132// tab-width: 4 //
1133// End: //
Note: See TracBrowser for help on using the repository browser.