source: libcfa/src/concurrency/kernel.cfa @ 986cb99

arm-ehjacob/cs343-translationnew-ast-unique-expr
Last change on this file since 986cb99 was 431cd4f, checked in by Thierry Delisle <tdelisle@…>, 6 months ago

Added alternative to relaxed-fifo scheduler.
Disabled by default

  • Property mode set to 100644
File size: 25.2 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 : Mon Aug 31 07:08:20 2020
13// Update Count     : 71
14//
15
16#define __cforall_thread__
17// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
18
19//C Includes
20#include <errno.h>
21#include <stdio.h>
22#include <signal.h>
23#include <unistd.h>
24extern "C" {
25        #include <sys/eventfd.h>
26}
27
28//CFA Includes
29#include "kernel_private.hfa"
30#include "preemption.hfa"
31
32//Private includes
33#define __CFA_INVOKE_PRIVATE__
34#include "invoke.h"
35
36
37//-----------------------------------------------------------------------------
38// Some assembly required
39#if defined( __i386 )
40        // mxcr : SSE Status and Control bits (control bits are preserved across function calls)
41        // fcw  : X87 FPU control word (preserved across function calls)
42        #define __x87_store         \
43                uint32_t __mxcr;      \
44                uint16_t __fcw;       \
45                __asm__ volatile (    \
46                        "stmxcsr %0\n"  \
47                        "fnstcw  %1\n"  \
48                        : "=m" (__mxcr),\
49                                "=m" (__fcw)  \
50                )
51
52        #define __x87_load         \
53                __asm__ volatile (   \
54                        "fldcw  %1\n"  \
55                        "ldmxcsr %0\n" \
56                        ::"m" (__mxcr),\
57                                "m" (__fcw)  \
58                )
59
60#elif defined( __x86_64 )
61        #define __x87_store         \
62                uint32_t __mxcr;      \
63                uint16_t __fcw;       \
64                __asm__ volatile (    \
65                        "stmxcsr %0\n"  \
66                        "fnstcw  %1\n"  \
67                        : "=m" (__mxcr),\
68                                "=m" (__fcw)  \
69                )
70
71        #define __x87_load          \
72                __asm__ volatile (    \
73                        "fldcw  %1\n"   \
74                        "ldmxcsr %0\n"  \
75                        :: "m" (__mxcr),\
76                                "m" (__fcw)  \
77                )
78
79#elif defined( __arm__ )
80        #define __x87_store
81        #define __x87_load
82
83#elif defined( __aarch64__ )
84        #define __x87_store              \
85                uint32_t __fpcntl[2];    \
86                __asm__ volatile (    \
87                        "mrs x9, FPCR\n" \
88                        "mrs x10, FPSR\n"  \
89                        "stp x9, x10, %0\n"  \
90                        : "=m" (__fpcntl) : : "x9", "x10" \
91                )
92
93        #define __x87_load         \
94                __asm__ volatile (    \
95                        "ldp x9, x10, %0\n"  \
96                        "msr FPSR, x10\n"  \
97                        "msr FPCR, x9\n" \
98                : "=m" (__fpcntl) : : "x9", "x10" \
99                )
100
101#else
102        #error unsupported hardware architecture
103#endif
104
105extern $thread * mainThread;
106extern processor * mainProcessor;
107
108//-----------------------------------------------------------------------------
109// Kernel Scheduling logic
110static $thread * __next_thread(cluster * this);
111static $thread * __next_thread_slow(cluster * this);
112static void __run_thread(processor * this, $thread * dst);
113static void __wake_one(cluster * cltr);
114
115static void mark_idle (__cluster_proc_list & idles, processor & proc);
116static void mark_awake(__cluster_proc_list & idles, processor & proc);
117static [unsigned idle, unsigned total, * processor] query_idles( & __cluster_proc_list idles );
118
119extern void __cfa_io_start( processor * );
120extern void __cfa_io_drain( processor * );
121extern void __cfa_io_flush( processor * );
122extern void __cfa_io_stop ( processor * );
123static inline void __maybe_io_drain( processor * );
124
125extern void __disable_interrupts_hard();
126extern void __enable_interrupts_hard();
127
128//=============================================================================================
129// Kernel Scheduling logic
130//=============================================================================================
131//Main of the processor contexts
132void main(processorCtx_t & runner) {
133        // Because of a bug, we couldn't initialized the seed on construction
134        // Do it here
135        __cfaabi_tls.rand_seed ^= rdtscl();
136        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
137        __tls_rand_advance_bck();
138
139        processor * this = runner.proc;
140        verify(this);
141
142        __cfa_io_start( this );
143
144        __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
145        #if !defined(__CFA_NO_STATISTICS__)
146                if( this->print_halts ) {
147                        __cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->id, this->name, (void*)this);
148                }
149        #endif
150
151        {
152                // Setup preemption data
153                preemption_scope scope = { this };
154
155                #if !defined(__CFA_NO_STATISTICS__)
156                        unsigned long long last_tally = rdtscl();
157                #endif
158
159                // if we need to run some special setup, now is the time to do it.
160                if(this->init.thrd) {
161                        this->init.thrd->curr_cluster = this->cltr;
162                        __run_thread(this, this->init.thrd);
163                }
164
165                __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
166
167                $thread * readyThread = 0p;
168                MAIN_LOOP:
169                for() {
170                        // Check if there is pending io
171                        __maybe_io_drain( this );
172
173                        // Try to get the next thread
174                        readyThread = __next_thread( this->cltr );
175
176                        if( !readyThread ) {
177                                __cfa_io_flush( this );
178                                readyThread = __next_thread_slow( this->cltr );
179                        }
180
181                        HALT:
182                        if( !readyThread ) {
183                                // Don't block if we are done
184                                if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
185
186                                #if !defined(__CFA_NO_STATISTICS__)
187                                        __tls_stats()->ready.sleep.halts++;
188                                #endif
189
190                                // Push self to idle stack
191                                mark_idle(this->cltr->procs, * this);
192
193                                // Confirm the ready-queue is empty
194                                readyThread = __next_thread_slow( this->cltr );
195                                if( readyThread ) {
196                                        // A thread was found, cancel the halt
197                                        mark_awake(this->cltr->procs, * this);
198
199                                        #if !defined(__CFA_NO_STATISTICS__)
200                                                __tls_stats()->ready.sleep.cancels++;
201                                        #endif
202
203                                        // continue the mai loop
204                                        break HALT;
205                                }
206
207                                #if !defined(__CFA_NO_STATISTICS__)
208                                        if(this->print_halts) {
209                                                __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->id, rdtscl());
210                                        }
211                                #endif
212
213                                __cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle);
214
215                                __disable_interrupts_hard();
216                                eventfd_t val;
217                                eventfd_read( this->idle, &val );
218                                __enable_interrupts_hard();
219
220                                #if !defined(__CFA_NO_STATISTICS__)
221                                        if(this->print_halts) {
222                                                __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->id, rdtscl());
223                                        }
224                                #endif
225
226                                // We were woken up, remove self from idle
227                                mark_awake(this->cltr->procs, * this);
228
229                                // DON'T just proceed, start looking again
230                                continue MAIN_LOOP;
231                        }
232
233                        /* paranoid */ verify( readyThread );
234
235                        // Reset io dirty bit
236                        this->io.dirty = false;
237
238                        // We found a thread run it
239                        __run_thread(this, readyThread);
240
241                        // Are we done?
242                        if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
243
244                        #if !defined(__CFA_NO_STATISTICS__)
245                                unsigned long long curr = rdtscl();
246                                if(curr > (last_tally + 500000000)) {
247                                        __tally_stats(this->cltr->stats, __cfaabi_tls.this_stats);
248                                        last_tally = curr;
249                                }
250                        #endif
251
252                        if(this->io.pending && !this->io.dirty) {
253                                __cfa_io_flush( this );
254                        }
255                }
256
257                __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
258        }
259
260        __cfa_io_stop( this );
261
262        post( this->terminated );
263
264
265        if(this == mainProcessor) {
266                // HACK : the coroutine context switch expects this_thread to be set
267                // and it make sense for it to be set in all other cases except here
268                // fake it
269                __cfaabi_tls.this_thread = mainThread;
270        }
271
272        __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
273}
274
275static int * __volatile_errno() __attribute__((noinline));
276static int * __volatile_errno() { asm(""); return &errno; }
277
278// KERNEL ONLY
279// runThread runs a thread by context switching
280// from the processor coroutine to the target thread
281static void __run_thread(processor * this, $thread * thrd_dst) {
282        /* paranoid */ verify( ! __preemption_enabled() );
283        /* paranoid */ verifyf( thrd_dst->state == Ready || thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
284        /* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
285        __builtin_prefetch( thrd_dst->context.SP );
286
287        __cfadbg_print_safe(runtime_core, "Kernel : core %p running thread %p (%s)\n", this, thrd_dst, thrd_dst->self_cor.name);
288
289        $coroutine * proc_cor = get_coroutine(this->runner);
290
291        // set state of processor coroutine to inactive
292        verify(proc_cor->state == Active);
293        proc_cor->state = Blocked;
294
295        // Actually run the thread
296        RUNNING:  while(true) {
297                thrd_dst->preempted = __NO_PREEMPTION;
298                thrd_dst->state = Active;
299
300                // Update global state
301                kernelTLS().this_thread = thrd_dst;
302
303                /* paranoid */ verify( ! __preemption_enabled() );
304                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
305                /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
306                /* paranoid */ verify( thrd_dst->context.SP );
307                /* paranoid */ verify( thrd_dst->state != Halted );
308                /* 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
309                /* 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
310                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
311
312
313
314                // set context switch to the thread that the processor is executing
315                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
316                // when __cfactx_switch returns we are back in the processor coroutine
317
318                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
319                /* 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 );
320                /* 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 );
321                /* paranoid */ verify( thrd_dst->context.SP );
322                /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
323                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
324                /* paranoid */ verify( ! __preemption_enabled() );
325
326                // Reset global state
327                kernelTLS().this_thread = 0p;
328
329                // We just finished running a thread, there are a few things that could have happened.
330                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
331                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
332                // 4 - Preempted
333                // In case 1, we may have won a race so we can't write to the state again.
334                // In case 2, we lost the race so we now own the thread.
335
336                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
337                        // The thread was preempted, reschedule it and reset the flag
338                        __schedule_thread( thrd_dst );
339                        break RUNNING;
340                }
341
342                if(unlikely(thrd_dst->state == Halting)) {
343                        // The thread has halted, it should never be scheduled/run again
344                        // finish the thread
345                        __thread_finish( thrd_dst );
346                        break RUNNING;
347                }
348
349                /* paranoid */ verify( thrd_dst->state == Active );
350                thrd_dst->state = Blocked;
351
352                // set state of processor coroutine to active and the thread to inactive
353                int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST);
354                switch(old_ticket) {
355                        case TICKET_RUNNING:
356                                // This is case 1, the regular case, nothing more is needed
357                                break RUNNING;
358                        case TICKET_UNBLOCK:
359                                #if !defined(__CFA_NO_STATISTICS__)
360                                        __tls_stats()->ready.threads.threads++;
361                                        __push_stat( __tls_stats(), __tls_stats()->ready.threads.threads, false, "Processor", this );
362                                #endif
363                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
364                                // In this case, just run it again.
365                                continue RUNNING;
366                        default:
367                                // This makes no sense, something is wrong abort
368                                abort();
369                }
370        }
371
372        // Just before returning to the processor, set the processor coroutine to active
373        proc_cor->state = Active;
374
375        __cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
376
377        #if !defined(__CFA_NO_STATISTICS__)
378                __tls_stats()->ready.threads.threads--;
379                __push_stat( __tls_stats(), __tls_stats()->ready.threads.threads, false, "Processor", this );
380        #endif
381
382        /* paranoid */ verify( ! __preemption_enabled() );
383}
384
385// KERNEL_ONLY
386void returnToKernel() {
387        /* paranoid */ verify( ! __preemption_enabled() );
388        $coroutine * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
389        $thread * thrd_src = kernelTLS().this_thread;
390
391        #if !defined(__CFA_NO_STATISTICS__)
392                struct processor * last_proc = kernelTLS().this_processor;
393        #endif
394
395        // Run the thread on this processor
396        {
397                int local_errno = *__volatile_errno();
398                #if defined( __i386 ) || defined( __x86_64 )
399                        __x87_store;
400                #endif
401                /* paranoid */ verify( proc_cor->context.SP );
402                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
403                __cfactx_switch( &thrd_src->context, &proc_cor->context );
404                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
405                #if defined( __i386 ) || defined( __x86_64 )
406                        __x87_load;
407                #endif
408                *__volatile_errno() = local_errno;
409        }
410
411        #if !defined(__CFA_NO_STATISTICS__)
412                if(last_proc != kernelTLS().this_processor) {
413                        __tls_stats()->ready.threads.migration++;
414                }
415        #endif
416
417        /* paranoid */ verify( ! __preemption_enabled() );
418        /* 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 );
419        /* 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 );
420}
421
422//-----------------------------------------------------------------------------
423// Scheduler routines
424// KERNEL ONLY
425void __schedule_thread( $thread * thrd ) {
426        /* paranoid */ verify( ! __preemption_enabled() );
427        /* paranoid */ verify( kernelTLS().this_proc_id );
428        /* paranoid */ verify( thrd );
429        /* paranoid */ verify( thrd->state != Halted );
430        /* paranoid */ verify( thrd->curr_cluster );
431        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
432        /* paranoid */  if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
433                                        "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
434        /* paranoid */  if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active,
435                                        "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
436        /* paranoid */ #endif
437        /* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
438        /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
439
440
441        if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
442
443        // Dereference the thread now because once we push it, there is not guaranteed it's still valid.
444        struct cluster * cl = thrd->curr_cluster;
445
446        ready_schedule_lock();
447                // push the thread to the cluster ready-queue
448                push( cl, thrd );
449
450                // variable thrd is no longer safe to use
451
452                // wake the cluster using the save variable.
453                __wake_one( cl );
454        ready_schedule_unlock();
455
456        #if !defined(__CFA_NO_STATISTICS__)
457                if( kernelTLS().this_stats ) {
458                        __tls_stats()->ready.threads.threads++;
459                        __push_stat( __tls_stats(), __tls_stats()->ready.threads.threads, false, "Processor", kernelTLS().this_processor );
460                }
461                else {
462                        __atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
463                        __push_stat( cl->stats, cl->stats->ready.threads.threads, true, "Cluster", cl );
464                }
465        #endif
466
467        /* paranoid */ verify( ! __preemption_enabled() );
468}
469
470// KERNEL ONLY
471static inline $thread * __next_thread(cluster * this) with( *this ) {
472        /* paranoid */ verify( ! __preemption_enabled() );
473        /* paranoid */ verify( kernelTLS().this_proc_id );
474
475        ready_schedule_lock();
476                $thread * thrd = pop_fast( this );
477        ready_schedule_unlock();
478
479        /* paranoid */ verify( kernelTLS().this_proc_id );
480        /* paranoid */ verify( ! __preemption_enabled() );
481        return thrd;
482}
483
484// KERNEL ONLY
485static inline $thread * __next_thread_slow(cluster * this) with( *this ) {
486        /* paranoid */ verify( ! __preemption_enabled() );
487        /* paranoid */ verify( kernelTLS().this_proc_id );
488
489        ready_schedule_lock();
490                $thread * thrd = pop_slow( this );
491        ready_schedule_unlock();
492
493        /* paranoid */ verify( kernelTLS().this_proc_id );
494        /* paranoid */ verify( ! __preemption_enabled() );
495        return thrd;
496}
497
498void unpark( $thread * thrd ) {
499        if( !thrd ) return;
500
501        int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
502        switch(old_ticket) {
503                case TICKET_RUNNING:
504                        // Wake won the race, the thread will reschedule/rerun itself
505                        break;
506                case TICKET_BLOCKED:
507                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
508                        /* paranoid */ verify( thrd->state == Blocked );
509
510                        {
511                                /* paranoid */ verify( publicTLS_get(this_proc_id) );
512                                disable_interrupts();
513
514                                /* paranoid */ verify( ! __preemption_enabled() );
515
516                                // Wake lost the race,
517                                __schedule_thread( thrd );
518
519                                /* paranoid */ verify( ! __preemption_enabled() );
520
521                                enable_interrupts_noPoll();
522                                /* paranoid */ verify( publicTLS_get(this_proc_id) );
523                        }
524
525                        break;
526                default:
527                        // This makes no sense, something is wrong abort
528                        abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
529        }
530}
531
532void park( void ) {
533        /* paranoid */ verify( __preemption_enabled() );
534        disable_interrupts();
535        /* paranoid */ verify( ! __preemption_enabled() );
536        /* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
537
538        returnToKernel();
539
540        /* paranoid */ verify( ! __preemption_enabled() );
541        enable_interrupts( __cfaabi_dbg_ctx );
542        /* paranoid */ verify( __preemption_enabled() );
543
544}
545
546extern "C" {
547        // Leave the thread monitor
548        // last routine called by a thread.
549        // Should never return
550        void __cfactx_thrd_leave() {
551                $thread * thrd = active_thread();
552                $monitor * this = &thrd->self_mon;
553
554                // Lock the monitor now
555                lock( this->lock __cfaabi_dbg_ctx2 );
556
557                disable_interrupts();
558
559                /* paranoid */ verify( ! __preemption_enabled() );
560                /* paranoid */ verify( thrd->state == Active );
561                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
562                /* paranoid */ verify( kernelTLS().this_thread == thrd );
563                /* paranoid */ verify( thrd->context.SP );
564                /* paranoid */ verifyf( ((uintptr_t)thrd->context.SP) > ((uintptr_t)__get_stack(thrd->curr_cor)->limit), "ERROR : $thread %p has been corrupted.\n StackPointer too large.\n", thrd );
565                /* paranoid */ verifyf( ((uintptr_t)thrd->context.SP) < ((uintptr_t)__get_stack(thrd->curr_cor)->base ), "ERROR : $thread %p has been corrupted.\n StackPointer too small.\n", thrd );
566
567                thrd->state = Halting;
568                if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); }
569                if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); }
570                if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); }
571
572                // Leave the thread
573                returnToKernel();
574
575                // Control flow should never reach here!
576                abort();
577        }
578}
579
580// KERNEL ONLY
581bool force_yield( __Preemption_Reason reason ) {
582        /* paranoid */ verify( __preemption_enabled() );
583        disable_interrupts();
584        /* paranoid */ verify( ! __preemption_enabled() );
585
586        $thread * thrd = kernelTLS().this_thread;
587        /* paranoid */ verify(thrd->state == Active);
588
589        // SKULLDUGGERY: It is possible that we are preempting this thread just before
590        // it was going to park itself. If that is the case and it is already using the
591        // intrusive fields then we can't use them to preempt the thread
592        // If that is the case, abandon the preemption.
593        bool preempted = false;
594        if(thrd->link.next == 0p) {
595                preempted = true;
596                thrd->preempted = reason;
597                returnToKernel();
598        }
599
600        /* paranoid */ verify( ! __preemption_enabled() );
601        enable_interrupts_noPoll();
602        /* paranoid */ verify( __preemption_enabled() );
603
604        return preempted;
605}
606
607//=============================================================================================
608// Kernel Idle Sleep
609//=============================================================================================
610// Wake a thread from the front if there are any
611static void __wake_one(cluster * this) {
612        /* paranoid */ verify( ! __preemption_enabled() );
613        /* paranoid */ verify( ready_schedule_islocked() );
614
615        // Check if there is a sleeping processor
616        processor * p;
617        unsigned idle;
618        unsigned total;
619        [idle, total, p] = query_idles(this->procs);
620
621        // If no one is sleeping, we are done
622        if( idle == 0 ) return;
623
624        // We found a processor, wake it up
625        eventfd_t val;
626        val = 1;
627        eventfd_write( p->idle, val );
628
629        #if !defined(__CFA_NO_STATISTICS__)
630                if( kernelTLS().this_stats ) {
631                        __tls_stats()->ready.sleep.wakes++;
632                }
633                else {
634                        __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED);
635                }
636        #endif
637
638        /* paranoid */ verify( ready_schedule_islocked() );
639        /* paranoid */ verify( ! __preemption_enabled() );
640
641        return;
642}
643
644// Unconditionnaly wake a thread
645void __wake_proc(processor * this) {
646        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
647
648        disable_interrupts();
649                /* paranoid */ verify( ! __preemption_enabled() );
650                eventfd_t val;
651                val = 1;
652                eventfd_write( this->idle, val );
653        enable_interrupts( __cfaabi_dbg_ctx );
654}
655
656static void mark_idle(__cluster_proc_list & this, processor & proc) {
657        /* paranoid */ verify( ! __preemption_enabled() );
658        lock( this );
659                this.idle++;
660                /* paranoid */ verify( this.idle <= this.total );
661                remove(proc);
662                insert_first(this.idles, proc);
663        unlock( this );
664        /* paranoid */ verify( ! __preemption_enabled() );
665}
666
667static void mark_awake(__cluster_proc_list & this, processor & proc) {
668        /* paranoid */ verify( ! __preemption_enabled() );
669        lock( this );
670                this.idle--;
671                /* paranoid */ verify( this.idle >= 0 );
672                remove(proc);
673                insert_last(this.actives, proc);
674        unlock( this );
675        /* paranoid */ verify( ! __preemption_enabled() );
676}
677
678static [unsigned idle, unsigned total, * processor] query_idles( & __cluster_proc_list this ) {
679        /* paranoid */ verify( ! __preemption_enabled() );
680        /* paranoid */ verify( ready_schedule_islocked() );
681
682        for() {
683                uint64_t l = __atomic_load_n(&this.lock, __ATOMIC_SEQ_CST);
684                if( 1 == (l % 2) ) { Pause(); continue; }
685                unsigned idle    = this.idle;
686                unsigned total   = this.total;
687                processor * proc = &this.idles`first;
688                // Compiler fence is unnecessary, but gcc-8 and older incorrectly reorder code without it
689                asm volatile("": : :"memory");
690                if(l != __atomic_load_n(&this.lock, __ATOMIC_SEQ_CST)) { Pause(); continue; }
691                return [idle, total, proc];
692        }
693
694        /* paranoid */ verify( ready_schedule_islocked() );
695        /* paranoid */ verify( ! __preemption_enabled() );
696}
697
698//=============================================================================================
699// Unexpected Terminating logic
700//=============================================================================================
701void __kernel_abort_msg( char * abort_text, int abort_text_size ) {
702        $thread * thrd = __cfaabi_tls.this_thread;
703
704        if(thrd) {
705                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
706                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
707
708                if ( &thrd->self_cor != thrd->curr_cor ) {
709                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
710                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
711                }
712                else {
713                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
714                }
715        }
716        else {
717                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
718                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
719        }
720}
721
722int __kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
723        return get_coroutine(__cfaabi_tls.this_thread) == get_coroutine(mainThread) ? 4 : 2;
724}
725
726static __spinlock_t kernel_debug_lock;
727
728extern "C" {
729        void __cfaabi_bits_acquire() {
730                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
731        }
732
733        void __cfaabi_bits_release() {
734                unlock( kernel_debug_lock );
735        }
736}
737
738//=============================================================================================
739// Kernel Utilities
740//=============================================================================================
741#if defined(CFA_HAVE_LINUX_IO_URING_H)
742#include "io/types.hfa"
743#endif
744
745static inline void __maybe_io_drain( processor * proc ) {
746        #if defined(CFA_HAVE_LINUX_IO_URING_H)
747                __cfadbg_print_safe(runtime_core, "Kernel : core %p checking io for ring %d\n", proc, proc->io.ctx->fd);
748
749                // Check if we should drain the queue
750                $io_context * ctx = proc->io.ctx;
751                unsigned head = *ctx->cq.head;
752                unsigned tail = *ctx->cq.tail;
753                if(head != tail) __cfa_io_drain( proc );
754        #endif
755}
756
757//-----------------------------------------------------------------------------
758// Debug
759__cfaabi_dbg_debug_do(
760        extern "C" {
761                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
762                        this.prev_name = prev_name;
763                        this.prev_thrd = kernelTLS().this_thread;
764                }
765        }
766)
767
768//-----------------------------------------------------------------------------
769// Debug
770bool threading_enabled(void) __attribute__((const)) {
771        return true;
772}
773
774//-----------------------------------------------------------------------------
775// Statistics
776#if !defined(__CFA_NO_STATISTICS__)
777        void print_halts( processor & this ) {
778                this.print_halts = true;
779        }
780
781        void print_stats_now( cluster & this, int flags ) {
782                __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
783        }
784#endif
785// Local Variables: //
786// mode: c //
787// tab-width: 4 //
788// End: //
Note: See TracBrowser for help on using the repository browser.