source: libcfa/src/concurrency/kernel.cfa @ 6c12fd2

arm-ehjacob/cs343-translationnew-astnew-ast-unique-expr
Last change on this file since 6c12fd2 was 6c12fd2, checked in by Thierry Delisle <tdelisle@…>, 19 months ago

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc

  • Property mode set to 100644
File size: 35.1 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        next = 0p;
200
201        node.next = 0p;
202        node.prev = 0p;
203        doregister(curr_cluster, this);
204
205        monitors{ &self_mon_p, 1, (fptr_t)0 };
206}
207
208//-----------------------------------------------------------------------------
209// Processor coroutine
210void ?{}(processorCtx_t & this) {
211
212}
213
214// Construct the processor context of non-main processors
215static void ?{}(processorCtx_t & this, processor * proc, current_stack_info_t * info) {
216        (this.__cor){ info };
217        this.proc = proc;
218}
219
220static void * __invoke_processor(void * arg);
221
222void ?{}(processor & this, const char name[], cluster & cltr) with( this ) {
223        this.name = name;
224        this.cltr = &cltr;
225        terminated{ 0 };
226        destroyer = 0p;
227        do_terminate = false;
228        preemption_alarm = 0p;
229        pending_preemption = false;
230        runner.proc = &this;
231
232        idle{};
233
234        __cfadbg_print_safe(runtime_core, "Kernel : Starting core %p\n", &this);
235
236        this.stack = __create_pthread( &this.kernel_thread, __invoke_processor, (void *)&this );
237
238        __cfadbg_print_safe(runtime_core, "Kernel : core %p created\n", &this);
239}
240
241void ^?{}(processor & this) with( this ){
242        if( ! __atomic_load_n(&do_terminate, __ATOMIC_ACQUIRE) ) {
243                __cfadbg_print_safe(runtime_core, "Kernel : core %p signaling termination\n", &this);
244
245                __atomic_store_n(&do_terminate, true, __ATOMIC_RELAXED);
246                __wake_proc( &this );
247
248                P( terminated );
249                verify( kernelTLS.this_processor != &this);
250        }
251
252        int err = pthread_join( kernel_thread, 0p );
253        if( err != 0 ) abort("KERNEL ERROR: joining processor %p caused error %s\n", &this, strerror(err));
254
255        free( this.stack );
256}
257
258void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned io_flags) with( this ) {
259        this.name = name;
260        this.preemption_rate = preemption_rate;
261        ready_queue{};
262        ready_queue_lock{};
263
264        #if !defined(__CFA_NO_STATISTICS__)
265                print_stats = false;
266        #endif
267
268        procs{ __get };
269        idles{ __get };
270        threads{ __get };
271
272        __kernel_io_startup( this, io_flags, &this == mainCluster );
273
274        doregister(this);
275}
276
277void ^?{}(cluster & this) {
278        __kernel_io_shutdown( this, &this == mainCluster );
279
280        unregister(this);
281}
282
283//=============================================================================================
284// Kernel Scheduling logic
285//=============================================================================================
286//Main of the processor contexts
287void main(processorCtx_t & runner) {
288        // Because of a bug, we couldn't initialized the seed on construction
289        // Do it here
290        kernelTLS.rand_seed ^= rdtscl();
291
292        processor * this = runner.proc;
293        verify(this);
294
295        __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
296
297        doregister(this->cltr, this);
298
299        {
300                // Setup preemption data
301                preemption_scope scope = { this };
302
303                __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
304
305                $thread * readyThread = 0p;
306                for( unsigned int spin_count = 0; ! __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST); spin_count++ ) {
307                        // Try to get the next thread
308                        readyThread = __next_thread( this->cltr );
309
310                        // If no ready thread
311                        if( readyThread == 0p ) {
312                                // Block until a thread is ready
313                                readyThread = __halt(this);
314                        }
315
316                        // Check if we actually found a thread
317                        if( readyThread ) {
318                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
319                                /* paranoid */ verifyf( readyThread->state == Ready || readyThread->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", readyThread->state, readyThread->preempted);
320                                /* paranoid */ verifyf( readyThread->next == 0p, "Expected null got %p", readyThread->next );
321
322                                // We found a thread run it
323                                __run_thread(this, readyThread);
324
325                                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
326                        }
327                }
328
329                __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
330        }
331
332        unregister(this->cltr, this);
333
334        V( this->terminated );
335
336        __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
337
338        // HACK : the coroutine context switch expects this_thread to be set
339        // and it make sense for it to be set in all other cases except here
340        // fake it
341        if( this == mainProcessor ) kernelTLS.this_thread = mainThread;
342}
343
344static int * __volatile_errno() __attribute__((noinline));
345static int * __volatile_errno() { asm(""); return &errno; }
346
347// KERNEL ONLY
348// runThread runs a thread by context switching
349// from the processor coroutine to the target thread
350static void __run_thread(processor * this, $thread * thrd_dst) {
351        $coroutine * proc_cor = get_coroutine(this->runner);
352
353        // Update global state
354        kernelTLS.this_thread = thrd_dst;
355
356        // set state of processor coroutine to inactive
357        verify(proc_cor->state == Active);
358        proc_cor->state = Blocked;
359
360        // Actually run the thread
361        RUNNING:  while(true) {
362                if(unlikely(thrd_dst->preempted)) {
363                        thrd_dst->preempted = __NO_PREEMPTION;
364                        verify(thrd_dst->state == Active  || thrd_dst->state == Rerun);
365                } else {
366                        verify(thrd_dst->state == Blocked || thrd_dst->state == Ready); // Ready means scheduled normally, blocked means rerun
367                        thrd_dst->state = Active;
368                }
369
370                __cfaabi_dbg_debug_do(
371                        thrd_dst->park_stale   = true;
372                        thrd_dst->unpark_stale = true;
373                )
374
375                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
376                /* 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
377                /* 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
378
379                // set context switch to the thread that the processor is executing
380                verify( thrd_dst->context.SP );
381                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
382                // when __cfactx_switch returns we are back in the processor coroutine
383
384                /* 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 );
385                /* 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 );
386                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
387
388
389                // We just finished running a thread, there are a few things that could have happened.
390                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
391                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
392                // 4 - Preempted
393                // In case 1, we may have won a race so we can't write to the state again.
394                // In case 2, we lost the race so we now own the thread.
395
396                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
397                        // The thread was preempted, reschedule it and reset the flag
398                        __schedule_thread( thrd_dst );
399                        break RUNNING;
400                }
401
402                // set state of processor coroutine to active and the thread to inactive
403                static_assert(sizeof(thrd_dst->state) == sizeof(int));
404                enum coroutine_state old_state = __atomic_exchange_n(&thrd_dst->state, Blocked, __ATOMIC_SEQ_CST);
405                __cfaabi_dbg_debug_do( thrd_dst->park_result = old_state; )
406                switch(old_state) {
407                        case Halted:
408                                // The thread has halted, it should never be scheduled/run again, leave it back to Halted and move on
409                                thrd_dst->state = Halted;
410
411                                // We may need to wake someone up here since
412                                unpark( this->destroyer __cfaabi_dbg_ctx2 );
413                                this->destroyer = 0p;
414                                break RUNNING;
415                        case Active:
416                                // This is case 1, the regular case, nothing more is needed
417                                break RUNNING;
418                        case Rerun:
419                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
420                                // In this case, just run it again.
421                                continue RUNNING;
422                        default:
423                                // This makes no sense, something is wrong abort
424                                abort("Finished running a thread that was Blocked/Start/Primed %d\n", old_state);
425                }
426        }
427
428        // Just before returning to the processor, set the processor coroutine to active
429        proc_cor->state = Active;
430        kernelTLS.this_thread = 0p;
431}
432
433// KERNEL_ONLY
434void returnToKernel() {
435        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
436        $coroutine * proc_cor = get_coroutine(kernelTLS.this_processor->runner);
437        $thread * thrd_src = kernelTLS.this_thread;
438
439        // Run the thread on this processor
440        {
441                int local_errno = *__volatile_errno();
442                #if defined( __i386 ) || defined( __x86_64 )
443                        __x87_store;
444                #endif
445                verify( proc_cor->context.SP );
446                __cfactx_switch( &thrd_src->context, &proc_cor->context );
447                #if defined( __i386 ) || defined( __x86_64 )
448                        __x87_load;
449                #endif
450                *__volatile_errno() = local_errno;
451        }
452
453        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
454        /* 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 );
455        /* 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 );
456}
457
458// KERNEL_ONLY
459// Context invoker for processors
460// This is the entry point for processors (kernel threads)
461// It effectively constructs a coroutine by stealing the pthread stack
462static void * __invoke_processor(void * arg) {
463        processor * proc = (processor *) arg;
464        kernelTLS.this_processor = proc;
465        kernelTLS.this_thread    = 0p;
466        kernelTLS.preemption_state.[enabled, disable_count] = [false, 1];
467        // SKULLDUGGERY: We want to create a context for the processor coroutine
468        // which is needed for the 2-step context switch. However, there is no reason
469        // to waste the perfectly valid stack create by pthread.
470        current_stack_info_t info;
471        __stack_t ctx;
472        info.storage = &ctx;
473        (proc->runner){ proc, &info };
474
475        __cfaabi_dbg_print_safe("Coroutine : created stack %p\n", get_coroutine(proc->runner)->stack.storage);
476
477        //Set global state
478        kernelTLS.this_thread = 0p;
479
480        //We now have a proper context from which to schedule threads
481        __cfadbg_print_safe(runtime_core, "Kernel : core %p created (%p, %p)\n", proc, &proc->runner, &ctx);
482
483        // SKULLDUGGERY: Since the coroutine doesn't have its own stack, we can't
484        // resume it to start it like it normally would, it will just context switch
485        // back to here. Instead directly call the main since we already are on the
486        // appropriate stack.
487        get_coroutine(proc->runner)->state = Active;
488        main( proc->runner );
489        get_coroutine(proc->runner)->state = Halted;
490
491        // Main routine of the core returned, the core is now fully terminated
492        __cfadbg_print_safe(runtime_core, "Kernel : core %p main ended (%p)\n", proc, &proc->runner);
493
494        return 0p;
495}
496
497static void Abort( int ret, const char func[] ) {
498        if ( ret ) {                                                                            // pthread routines return errno values
499                abort( "%s : internal error, error(%d) %s.", func, ret, strerror( ret ) );
500        } // if
501} // Abort
502
503void * __create_pthread( pthread_t * pthread, void * (*start)(void *), void * arg ) {
504        pthread_attr_t attr;
505
506        Abort( pthread_attr_init( &attr ), "pthread_attr_init" ); // initialize attribute
507
508        size_t stacksize;
509        // default stack size, normally defined by shell limit
510        Abort( pthread_attr_getstacksize( &attr, &stacksize ), "pthread_attr_getstacksize" );
511        assert( stacksize >= PTHREAD_STACK_MIN );
512
513        void * stack;
514        __cfaabi_dbg_debug_do(
515                stack = memalign( __page_size, stacksize + __page_size );
516                // pthread has no mechanism to create the guard page in user supplied stack.
517                if ( mprotect( stack, __page_size, PROT_NONE ) == -1 ) {
518                        abort( "mprotect : internal error, mprotect failure, error(%d) %s.", errno, strerror( errno ) );
519                } // if
520        );
521        __cfaabi_dbg_no_debug_do(
522                stack = malloc( stacksize );
523        );
524
525        Abort( pthread_attr_setstack( &attr, stack, stacksize ), "pthread_attr_setstack" );
526
527        Abort( pthread_create( pthread, &attr, start, arg ), "pthread_create" );
528        return stack;
529}
530
531// KERNEL_ONLY
532static void __kernel_first_resume( processor * this ) {
533        $thread * src = mainThread;
534        $coroutine * dst = get_coroutine(this->runner);
535
536        verify( ! kernelTLS.preemption_state.enabled );
537
538        kernelTLS.this_thread->curr_cor = dst;
539        __stack_prepare( &dst->stack, 65000 );
540        __cfactx_start(main, dst, this->runner, __cfactx_invoke_coroutine);
541
542        verify( ! kernelTLS.preemption_state.enabled );
543
544        dst->last = &src->self_cor;
545        dst->starter = dst->starter ? dst->starter : &src->self_cor;
546
547        // make sure the current state is still correct
548        /* paranoid */ verify(src->state == Ready);
549
550        // context switch to specified coroutine
551        verify( dst->context.SP );
552        __cfactx_switch( &src->context, &dst->context );
553        // when __cfactx_switch returns we are back in the src coroutine
554
555        mainThread->curr_cor = &mainThread->self_cor;
556
557        // make sure the current state has been update
558        /* paranoid */ verify(src->state == Active);
559
560        verify( ! kernelTLS.preemption_state.enabled );
561}
562
563// KERNEL_ONLY
564static void __kernel_last_resume( processor * this ) {
565        $coroutine * src = &mainThread->self_cor;
566        $coroutine * dst = get_coroutine(this->runner);
567
568        verify( ! kernelTLS.preemption_state.enabled );
569        verify( dst->starter == src );
570        verify( dst->context.SP );
571
572        // SKULLDUGGERY in debug the processors check that the
573        // stack is still within the limit of the stack limits after running a thread.
574        // that check doesn't make sense if we context switch to the processor using the
575        // coroutine semantics. Since this is a special case, use the current context
576        // info to populate these fields.
577        __cfaabi_dbg_debug_do(
578                __stack_context_t ctx;
579                CtxGet( ctx );
580                mainThread->context.SP = ctx.SP;
581                mainThread->context.FP = ctx.FP;
582        )
583
584        // context switch to the processor
585        __cfactx_switch( &src->context, &dst->context );
586}
587
588//-----------------------------------------------------------------------------
589// Scheduler routines
590// KERNEL ONLY
591void __schedule_thread( $thread * thrd ) with( *thrd->curr_cluster ) {
592        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
593        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
594        /* paranoid */ if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
595                          "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
596        /* paranoid */ if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active || thrd->state == Rerun,
597                          "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
598        /* paranoid */ #endif
599        /* paranoid */ verifyf( thrd->next == 0p, "Expected null got %p", thrd->next );
600
601        if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
602
603        lock  ( ready_queue_lock __cfaabi_dbg_ctx2 );
604        bool was_empty = !(ready_queue != 0);
605        append( ready_queue, thrd );
606        unlock( ready_queue_lock );
607
608        __wake_one(thrd->curr_cluster, was_empty);
609
610        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
611}
612
613// KERNEL ONLY
614static $thread * __next_thread(cluster * this) with( *this ) {
615        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
616
617        lock( ready_queue_lock __cfaabi_dbg_ctx2 );
618        $thread * head = pop_head( ready_queue );
619        unlock( ready_queue_lock );
620
621        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
622        return head;
623}
624
625// KERNEL ONLY unpark with out disabling interrupts
626void __unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
627        static_assert(sizeof(thrd->state) == sizeof(int));
628
629        // record activity
630        __cfaabi_dbg_record_thrd( *thrd, false, caller );
631
632        enum coroutine_state old_state = __atomic_exchange_n(&thrd->state, Rerun, __ATOMIC_SEQ_CST);
633        __cfaabi_dbg_debug_do( thrd->unpark_result = old_state; )
634        switch(old_state) {
635                case Active:
636                        // Wake won the race, the thread will reschedule/rerun itself
637                        break;
638                case Blocked:
639                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
640
641                        // Wake lost the race,
642                        thrd->state = Blocked;
643                        __schedule_thread( thrd );
644                        break;
645                case Rerun:
646                        abort("More than one thread attempted to schedule thread %p\n", thrd);
647                        break;
648                case Halted:
649                case Start:
650                case Primed:
651                default:
652                        // This makes no sense, something is wrong abort
653                        abort();
654        }
655}
656
657void unpark( $thread * thrd __cfaabi_dbg_ctx_param2 ) {
658        if( !thrd ) return;
659
660        disable_interrupts();
661        __unpark( thrd __cfaabi_dbg_ctx_fwd2 );
662        enable_interrupts( __cfaabi_dbg_ctx );
663}
664
665void park( __cfaabi_dbg_ctx_param ) {
666        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
667        disable_interrupts();
668        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
669        /* paranoid */ verify( kernelTLS.this_thread->preempted == __NO_PREEMPTION );
670
671        // record activity
672        __cfaabi_dbg_record_thrd( *kernelTLS.this_thread, true, caller );
673
674        returnToKernel();
675
676        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
677        enable_interrupts( __cfaabi_dbg_ctx );
678        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
679
680}
681
682// KERNEL ONLY
683void __leave_thread() {
684        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
685        returnToKernel();
686        abort();
687}
688
689// KERNEL ONLY
690bool force_yield( __Preemption_Reason reason ) {
691        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
692        disable_interrupts();
693        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
694
695        $thread * thrd = kernelTLS.this_thread;
696        /* paranoid */ verify(thrd->state == Active || thrd->state == Rerun);
697
698        // SKULLDUGGERY: It is possible that we are preempting this thread just before
699        // it was going to park itself. If that is the case and it is already using the
700        // intrusive fields then we can't use them to preempt the thread
701        // If that is the case, abandon the preemption.
702        bool preempted = false;
703        if(thrd->next == 0p) {
704                preempted = true;
705                thrd->preempted = reason;
706                returnToKernel();
707        }
708
709        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
710        enable_interrupts_noPoll();
711        /* paranoid */ verify( kernelTLS.preemption_state.enabled );
712
713        return preempted;
714}
715
716//=============================================================================================
717// Kernel Setup logic
718//=============================================================================================
719//-----------------------------------------------------------------------------
720// Kernel boot procedures
721static void __kernel_startup(void) {
722        verify( ! kernelTLS.preemption_state.enabled );
723        __cfadbg_print_safe(runtime_core, "Kernel : Starting\n");
724
725        __page_size = sysconf( _SC_PAGESIZE );
726
727        __cfa_dbg_global_clusters.list{ __get };
728        __cfa_dbg_global_clusters.lock{};
729
730        // Initialize the main cluster
731        mainCluster = (cluster *)&storage_mainCluster;
732        (*mainCluster){"Main Cluster"};
733
734        __cfadbg_print_safe(runtime_core, "Kernel : Main cluster ready\n");
735
736        // Start by initializing the main thread
737        // SKULLDUGGERY: the mainThread steals the process main thread
738        // which will then be scheduled by the mainProcessor normally
739        mainThread = ($thread *)&storage_mainThread;
740        current_stack_info_t info;
741        info.storage = (__stack_t*)&storage_mainThreadCtx;
742        (*mainThread){ &info };
743
744        __cfadbg_print_safe(runtime_core, "Kernel : Main thread ready\n");
745
746
747
748        // Construct the processor context of the main processor
749        void ?{}(processorCtx_t & this, processor * proc) {
750                (this.__cor){ "Processor" };
751                this.__cor.starter = 0p;
752                this.proc = proc;
753        }
754
755        void ?{}(processor & this) with( this ) {
756                name = "Main Processor";
757                cltr = mainCluster;
758                terminated{ 0 };
759                do_terminate = false;
760                preemption_alarm = 0p;
761                pending_preemption = false;
762                kernel_thread = pthread_self();
763
764                runner{ &this };
765                __cfadbg_print_safe(runtime_core, "Kernel : constructed main processor context %p\n", &runner);
766        }
767
768        // Initialize the main processor and the main processor ctx
769        // (the coroutine that contains the processing control flow)
770        mainProcessor = (processor *)&storage_mainProcessor;
771        (*mainProcessor){};
772
773        //initialize the global state variables
774        kernelTLS.this_processor = mainProcessor;
775        kernelTLS.this_thread    = mainThread;
776
777        // Enable preemption
778        kernel_start_preemption();
779
780        // Add the main thread to the ready queue
781        // once resume is called on mainProcessor->runner the mainThread needs to be scheduled like any normal thread
782        __schedule_thread(mainThread);
783
784        // SKULLDUGGERY: Force a context switch to the main processor to set the main thread's context to the current UNIX
785        // context. Hence, the main thread does not begin through __cfactx_invoke_thread, like all other threads. The trick here is that
786        // mainThread is on the ready queue when this call is made.
787        __kernel_first_resume( kernelTLS.this_processor );
788
789
790        // THE SYSTEM IS NOW COMPLETELY RUNNING
791
792
793        // Now that the system is up, finish creating systems that need threading
794        __kernel_io_finish_start( *mainCluster );
795
796
797        __cfadbg_print_safe(runtime_core, "Kernel : Started\n--------------------------------------------------\n\n");
798
799        verify( ! kernelTLS.preemption_state.enabled );
800        enable_interrupts( __cfaabi_dbg_ctx );
801        verify( TL_GET( preemption_state.enabled ) );
802}
803
804static void __kernel_shutdown(void) {
805        //Before we start shutting things down, wait for systems that need threading to shutdown
806        __kernel_io_prepare_stop( *mainCluster );
807
808        /* paranoid */ verify( TL_GET( preemption_state.enabled ) );
809        disable_interrupts();
810        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
811
812        __cfadbg_print_safe(runtime_core, "\n--------------------------------------------------\nKernel : Shutting down\n");
813
814        // SKULLDUGGERY: Notify the mainProcessor it needs to terminates.
815        // When its coroutine terminates, it return control to the mainThread
816        // which is currently here
817        __atomic_store_n(&mainProcessor->do_terminate, true, __ATOMIC_RELEASE);
818        __kernel_last_resume( kernelTLS.this_processor );
819        mainThread->self_cor.state = Halted;
820
821        // THE SYSTEM IS NOW COMPLETELY STOPPED
822
823        // Disable preemption
824        kernel_stop_preemption();
825
826        // Destroy the main processor and its context in reverse order of construction
827        // These were manually constructed so we need manually destroy them
828        void ^?{}(processor & this) with( this ){
829                /* paranoid */ verify( this.do_terminate == true );
830        }
831
832        ^(*mainProcessor){};
833
834        // Final step, destroy the main thread since it is no longer needed
835        // Since we provided a stack to this taxk it will not destroy anything
836        /* paranoid */ verify(mainThread->self_cor.stack.storage == (__stack_t*)(((uintptr_t)&storage_mainThreadCtx)| 0x1));
837        ^(*mainThread){};
838
839        ^(*mainCluster){};
840
841        ^(__cfa_dbg_global_clusters.list){};
842        ^(__cfa_dbg_global_clusters.lock){};
843
844        __cfadbg_print_safe(runtime_core, "Kernel : Shutdown complete\n");
845}
846
847//=============================================================================================
848// Kernel Idle Sleep
849//=============================================================================================
850static $thread * __halt(processor * this) with( *this ) {
851        if( do_terminate ) return 0p;
852
853        // First, lock the cluster idle
854        lock( cltr->idle_lock __cfaabi_dbg_ctx2 );
855
856        // Check if we can find a thread
857        if( $thread * found = __next_thread( cltr ) ) {
858                unlock( cltr->idle_lock );
859                return found;
860        }
861
862        // Move this processor from the active list to the idle list
863        move_to_front(cltr->procs, cltr->idles, *this);
864
865        // Unlock the idle lock so we don't go to sleep with a lock
866        unlock    (cltr->idle_lock);
867
868        // We are ready to sleep
869        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p ready to sleep\n", this);
870        wait( idle );
871
872        // We have woken up
873        __cfadbg_print_safe(runtime_core, "Kernel : Processor %p woke up and ready to run\n", this);
874
875        // Get ourself off the idle list
876        with( *cltr ) {
877                lock  (idle_lock __cfaabi_dbg_ctx2);
878                move_to_front(idles, procs, *this);
879                unlock(idle_lock);
880        }
881
882        // Don't check the ready queue again, we may not be in a position to run a thread
883        return 0p;
884}
885
886// Wake a thread from the front if there are any
887static bool __wake_one(cluster * this, __attribute__((unused)) bool force) {
888        // if we don't want to force check if we know it's false
889        // if( !this->idles.head && !force ) return false;
890
891        // First, lock the cluster idle
892        lock( this->idle_lock __cfaabi_dbg_ctx2 );
893
894        // Check if there is someone to wake up
895        if( !this->idles.head ) {
896                // Nope unlock and return false
897                unlock( this->idle_lock );
898                return false;
899        }
900
901        // Wake them up
902        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this->idles.head);
903        /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
904        post( this->idles.head->idle );
905
906        // Unlock and return true
907        unlock( this->idle_lock );
908        return true;
909}
910
911// Unconditionnaly wake a thread
912static bool __wake_proc(processor * this) {
913        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
914
915        disable_interrupts();
916                /* paranoid */ verify( ! kernelTLS.preemption_state.enabled );
917                bool ret = post( this->idle );
918        enable_interrupts( __cfaabi_dbg_ctx );
919
920        return ret;
921}
922
923//=============================================================================================
924// Unexpected Terminating logic
925//=============================================================================================
926static __spinlock_t kernel_abort_lock;
927static bool kernel_abort_called = false;
928
929void * kernel_abort(void) __attribute__ ((__nothrow__)) {
930        // abort cannot be recursively entered by the same or different processors because all signal handlers return when
931        // the globalAbort flag is true.
932        lock( kernel_abort_lock __cfaabi_dbg_ctx2 );
933
934        // first task to abort ?
935        if ( kernel_abort_called ) {                    // not first task to abort ?
936                unlock( kernel_abort_lock );
937
938                sigset_t mask;
939                sigemptyset( &mask );
940                sigaddset( &mask, SIGALRM );            // block SIGALRM signals
941                sigaddset( &mask, SIGUSR1 );            // block SIGALRM signals
942                sigsuspend( &mask );                            // block the processor to prevent further damage during abort
943                _exit( EXIT_FAILURE );                          // if processor unblocks before it is killed, terminate it
944        }
945        else {
946                kernel_abort_called = true;
947                unlock( kernel_abort_lock );
948        }
949
950        return kernelTLS.this_thread;
951}
952
953void kernel_abort_msg( void * kernel_data, char * abort_text, int abort_text_size ) {
954        $thread * thrd = kernel_data;
955
956        if(thrd) {
957                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
958                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
959
960                if ( &thrd->self_cor != thrd->curr_cor ) {
961                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
962                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
963                }
964                else {
965                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
966                }
967        }
968        else {
969                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
970                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
971        }
972}
973
974int kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
975        return get_coroutine(kernelTLS.this_thread) == get_coroutine(mainThread) ? 4 : 2;
976}
977
978static __spinlock_t kernel_debug_lock;
979
980extern "C" {
981        void __cfaabi_bits_acquire() {
982                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
983        }
984
985        void __cfaabi_bits_release() {
986                unlock( kernel_debug_lock );
987        }
988}
989
990//=============================================================================================
991// Kernel Utilities
992//=============================================================================================
993//-----------------------------------------------------------------------------
994// Locks
995void  ?{}( semaphore & this, int count = 1 ) {
996        (this.lock){};
997        this.count = count;
998        (this.waiting){};
999}
1000void ^?{}(semaphore & this) {}
1001
1002bool P(semaphore & this) with( this ){
1003        lock( lock __cfaabi_dbg_ctx2 );
1004        count -= 1;
1005        if ( count < 0 ) {
1006                // queue current task
1007                append( waiting, kernelTLS.this_thread );
1008
1009                // atomically release spin lock and block
1010                unlock( lock );
1011                park( __cfaabi_dbg_ctx );
1012                return true;
1013        }
1014        else {
1015            unlock( lock );
1016            return false;
1017        }
1018}
1019
1020bool V(semaphore & this) with( this ) {
1021        $thread * thrd = 0p;
1022        lock( lock __cfaabi_dbg_ctx2 );
1023        count += 1;
1024        if ( count <= 0 ) {
1025                // remove task at head of waiting list
1026                thrd = pop_head( waiting );
1027        }
1028
1029        unlock( lock );
1030
1031        // make new owner
1032        unpark( thrd __cfaabi_dbg_ctx2 );
1033
1034        return thrd != 0p;
1035}
1036
1037bool V(semaphore & this, unsigned diff) with( this ) {
1038        $thread * thrd = 0p;
1039        lock( lock __cfaabi_dbg_ctx2 );
1040        int release = max(-count, (int)diff);
1041        count += diff;
1042        for(release) {
1043                unpark( pop_head( waiting ) __cfaabi_dbg_ctx2 );
1044        }
1045
1046        unlock( lock );
1047
1048        return thrd != 0p;
1049}
1050
1051//-----------------------------------------------------------------------------
1052// Global Queues
1053void doregister( cluster     & cltr ) {
1054        lock      ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1055        push_front( __cfa_dbg_global_clusters.list, cltr );
1056        unlock    ( __cfa_dbg_global_clusters.lock );
1057}
1058
1059void unregister( cluster     & cltr ) {
1060        lock  ( __cfa_dbg_global_clusters.lock __cfaabi_dbg_ctx2);
1061        remove( __cfa_dbg_global_clusters.list, cltr );
1062        unlock( __cfa_dbg_global_clusters.lock );
1063}
1064
1065void doregister( cluster * cltr, $thread & thrd ) {
1066        lock      (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1067        cltr->nthreads += 1;
1068        push_front(cltr->threads, thrd);
1069        unlock    (cltr->thread_list_lock);
1070}
1071
1072void unregister( cluster * cltr, $thread & thrd ) {
1073        lock  (cltr->thread_list_lock __cfaabi_dbg_ctx2);
1074        remove(cltr->threads, thrd );
1075        cltr->nthreads -= 1;
1076        unlock(cltr->thread_list_lock);
1077}
1078
1079void doregister( cluster * cltr, processor * proc ) {
1080        lock      (cltr->idle_lock __cfaabi_dbg_ctx2);
1081        cltr->nprocessors += 1;
1082        push_front(cltr->procs, *proc);
1083        unlock    (cltr->idle_lock);
1084}
1085
1086void unregister( cluster * cltr, processor * proc ) {
1087        lock  (cltr->idle_lock __cfaabi_dbg_ctx2);
1088        remove(cltr->procs, *proc );
1089        cltr->nprocessors -= 1;
1090        unlock(cltr->idle_lock);
1091}
1092
1093//-----------------------------------------------------------------------------
1094// Debug
1095__cfaabi_dbg_debug_do(
1096        extern "C" {
1097                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
1098                        this.prev_name = prev_name;
1099                        this.prev_thrd = kernelTLS.this_thread;
1100                }
1101
1102                void __cfaabi_dbg_record_thrd($thread & this, bool park, const char prev_name[]) {
1103                        if(park) {
1104                                this.park_caller   = prev_name;
1105                                this.park_stale    = false;
1106                        }
1107                        else {
1108                                this.unpark_caller = prev_name;
1109                                this.unpark_stale  = false;
1110                        }
1111                }
1112        }
1113)
1114
1115//-----------------------------------------------------------------------------
1116// Debug
1117bool threading_enabled(void) __attribute__((const)) {
1118        return true;
1119}
1120// Local Variables: //
1121// mode: c //
1122// tab-width: 4 //
1123// End: //
Note: See TracBrowser for help on using the repository browser.