source: libcfa/src/concurrency/kernel.cfa @ 059ad16

ADTast-experimentalenumforall-pointer-decaypthread-emulationqualifiedEnum
Last change on this file since 059ad16 was 059ad16, checked in by Thierry Delisle <tdelisle@…>, 2 years ago

Flush now supports blocking until at least one I/O op terminates.

  • Property mode set to 100644
File size: 30.0 KB
RevLine 
[8118303]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
[75f3522]10// Created On       : Tue Jan 17 12:27:26 2017
[6b0b624]11// Last Modified By : Peter A. Buhr
[0190480]12// Last Modified On : Mon Aug 31 07:08:20 2020
13// Update Count     : 71
[8118303]14//
15
[2026bb6]16#define __cforall_thread__
[43784ac]17#define _GNU_SOURCE
18
[4069faad]19// #define __CFA_DEBUG_PRINT_RUNTIME_CORE__
[2026bb6]20
[8118303]21//C Includes
[214e8da]22#include <errno.h>
[9d944b2]23#include <stdio.h>
[445f984]24#include <string.h>
[58b6d1b]25#include <signal.h>
[9d944b2]26#include <unistd.h>
[dddb3dd0]27extern "C" {
28        #include <sys/eventfd.h>
29}
[8118303]30
31//CFA Includes
[73abe95]32#include "kernel_private.hfa"
33#include "preemption.hfa"
[445f984]34#include "strstream.hfa"
35#include "device/cpu.hfa"
[8118303]36
37//Private includes
38#define __CFA_INVOKE_PRIVATE__
39#include "invoke.h"
40
[89eff25]41#if !defined(__CFA_NO_STATISTICS__)
42        #define __STATS( ...) __VA_ARGS__
43#else
44        #define __STATS( ...)
45#endif
[4069faad]46
[deca0f5]47//-----------------------------------------------------------------------------
48// Some assembly required
[1805b1b]49#if defined( __i386 )
[deca0f5]50        // mxcr : SSE Status and Control bits (control bits are preserved across function calls)
51        // fcw  : X87 FPU control word (preserved across function calls)
52        #define __x87_store         \
53                uint32_t __mxcr;      \
54                uint16_t __fcw;       \
55                __asm__ volatile (    \
56                        "stmxcsr %0\n"  \
57                        "fnstcw  %1\n"  \
58                        : "=m" (__mxcr),\
59                                "=m" (__fcw)  \
60                )
61
62        #define __x87_load         \
63                __asm__ volatile (   \
64                        "fldcw  %1\n"  \
65                        "ldmxcsr %0\n" \
66                        ::"m" (__mxcr),\
67                                "m" (__fcw)  \
68                )
69
70#elif defined( __x86_64 )
71        #define __x87_store         \
72                uint32_t __mxcr;      \
73                uint16_t __fcw;       \
74                __asm__ volatile (    \
75                        "stmxcsr %0\n"  \
76                        "fnstcw  %1\n"  \
77                        : "=m" (__mxcr),\
78                                "=m" (__fcw)  \
79                )
80
81        #define __x87_load          \
82                __asm__ volatile (    \
83                        "fldcw  %1\n"   \
84                        "ldmxcsr %0\n"  \
85                        :: "m" (__mxcr),\
86                                "m" (__fcw)  \
87                )
88
[0190480]89#elif defined( __arm__ )
90        #define __x87_store
91        #define __x87_load
92
93#elif defined( __aarch64__ )
[74f5c83]94        #define __x87_store              \
95                uint32_t __fpcntl[2];    \
96                __asm__ volatile (    \
97                        "mrs x9, FPCR\n" \
98                        "mrs x10, FPSR\n"  \
99                        "stp x9, x10, %0\n"  \
100                        : "=m" (__fpcntl) : : "x9", "x10" \
101                )
102
103        #define __x87_load         \
104                __asm__ volatile (    \
105                        "ldp x9, x10, %0\n"  \
106                        "msr FPSR, x10\n"  \
107                        "msr FPCR, x9\n" \
108                : "=m" (__fpcntl) : : "x9", "x10" \
109                )
110
[deca0f5]111#else
[0190480]112        #error unsupported hardware architecture
[deca0f5]113#endif
114
[e84ab3d]115extern thread$ * mainThread;
[e660761]116extern processor * mainProcessor;
[2ac095d]117
[92e7631]118//-----------------------------------------------------------------------------
119// Kernel Scheduling logic
[e84ab3d]120static thread$ * __next_thread(cluster * this);
121static thread$ * __next_thread_slow(cluster * this);
122static inline bool __must_unpark( thread$ * thrd ) __attribute((nonnull(1)));
123static void __run_thread(processor * this, thread$ * dst);
[e873838]124static void __wake_one(cluster * cltr);
[1eb239e4]125
[1757f98]126static void idle_sleep(processor * proc);
[5f5a729]127static bool mark_idle (__cluster_proc_list & idles, processor & proc);
[6a9b12b]128static void mark_awake(__cluster_proc_list & idles, processor & proc);
[1eb239e4]129
[dddb3dd0]130extern void __cfa_io_start( processor * );
[c1c95b1]131extern bool __cfa_io_drain( processor * );
[059ad16]132extern bool __cfa_io_flush( processor *, bool wait );
[dddb3dd0]133extern void __cfa_io_stop ( processor * );
[c1c95b1]134static inline bool __maybe_io_drain( processor * );
[dddb3dd0]135
136extern void __disable_interrupts_hard();
137extern void __enable_interrupts_hard();
[c84e80a]138
[a3821fa]139static inline void __disable_interrupts_checked() {
140        /* paranoid */ verify( __preemption_enabled() );
141        disable_interrupts();
142        /* paranoid */ verify( ! __preemption_enabled() );
143}
144
145static inline void __enable_interrupts_checked( bool poll = true ) {
146        /* paranoid */ verify( ! __preemption_enabled() );
147        enable_interrupts( poll );
148        /* paranoid */ verify( __preemption_enabled() );
149}
150
[75f3522]151//=============================================================================================
152// Kernel Scheduling logic
153//=============================================================================================
[8fcbb4c]154//Main of the processor contexts
[83a071f9]155void main(processorCtx_t & runner) {
[21184e3]156        // Because of a bug, we couldn't initialized the seed on construction
157        // Do it here
[8fc652e0]158        __cfaabi_tls.rand_seed ^= rdtscl();
159        __cfaabi_tls.ready_rng.fwd_seed = 25214903917_l64u * (rdtscl() ^ (uintptr_t)&runner);
[f2384c9a]160        __tls_rand_advance_bck();
[21184e3]161
[83a071f9]162        processor * this = runner.proc;
[094476d]163        verify(this);
[c81ebf9]164
[dddb3dd0]165        __cfa_io_start( this );
166
[4069faad]167        __cfadbg_print_safe(runtime_core, "Kernel : core %p starting\n", this);
[28d73c1]168        #if !defined(__CFA_NO_STATISTICS__)
169                if( this->print_halts ) {
[c993b15]170                        __cfaabi_bits_print_safe( STDOUT_FILENO, "Processor : %d - %s (%p)\n", this->unique_id, this->name, (void*)this);
[28d73c1]171                }
172        #endif
[b798713]173
[75f3522]174        {
[c81ebf9]175                // Setup preemption data
176                preemption_scope scope = { this };
177
[a5e7233]178                // if we need to run some special setup, now is the time to do it.
179                if(this->init.thrd) {
180                        this->init.thrd->curr_cluster = this->cltr;
181                        __run_thread(this, this->init.thrd);
182                }
[325e6ea]183
[4069faad]184                __cfadbg_print_safe(runtime_core, "Kernel : core %p started\n", this);
[8118303]185
[e84ab3d]186                thread$ * readyThread = 0p;
[1eb239e4]187                MAIN_LOOP:
188                for() {
[57f70ab]189                        #define OLD_MAIN 1
190                        #if OLD_MAIN
[dddb3dd0]191                        // Check if there is pending io
192                        __maybe_io_drain( this );
193
[92e7631]194                        // Try to get the next thread
[8c50aed]195                        readyThread = __next_thread( this->cltr );
[75f3522]196
[1eb239e4]197                        if( !readyThread ) {
[c33c2af]198                                ready_schedule_lock();
[059ad16]199                                __cfa_io_flush( this, false );
[c33c2af]200                                ready_schedule_unlock();
201
[1eb239e4]202                                readyThread = __next_thread_slow( this->cltr );
203                        }
[4e6fb8e]204
[1eb239e4]205                        HALT:
206                        if( !readyThread ) {
207                                // Don't block if we are done
208                                if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[c81ebf9]209
[1eb239e4]210                                #if !defined(__CFA_NO_STATISTICS__)
211                                        __tls_stats()->ready.sleep.halts++;
212                                #endif
[398e8e9]213
[1eb239e4]214                                // Push self to idle stack
[5f5a729]215                                if(!mark_idle(this->cltr->procs, * this)) continue MAIN_LOOP;
[398e8e9]216
[1eb239e4]217                                // Confirm the ready-queue is empty
218                                readyThread = __next_thread_slow( this->cltr );
219                                if( readyThread ) {
220                                        // A thread was found, cancel the halt
[6a9b12b]221                                        mark_awake(this->cltr->procs, * this);
[1eb239e4]222
223                                        #if !defined(__CFA_NO_STATISTICS__)
224                                                __tls_stats()->ready.sleep.cancels++;
225                                        #endif
226
227                                        // continue the mai loop
228                                        break HALT;
229                                }
230
[1757f98]231                                idle_sleep( this );
[1eb239e4]232
233                                // We were woken up, remove self from idle
[6a9b12b]234                                mark_awake(this->cltr->procs, * this);
[1eb239e4]235
236                                // DON'T just proceed, start looking again
237                                continue MAIN_LOOP;
[64a7146]238                        }
[1eb239e4]239
240                        /* paranoid */ verify( readyThread );
241
[dddb3dd0]242                        // Reset io dirty bit
243                        this->io.dirty = false;
244
[1eb239e4]245                        // We found a thread run it
246                        __run_thread(this, readyThread);
247
248                        // Are we done?
249                        if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
[325e6ea]250
[dddb3dd0]251                        if(this->io.pending && !this->io.dirty) {
[c33c2af]252                                ready_schedule_lock();
[059ad16]253                                __cfa_io_flush( this, false );
[c33c2af]254                                ready_schedule_unlock();
[dddb3dd0]255                        }
[89eff25]256
[34b2796]257                        #else
[57f70ab]258                                #warning new kernel loop
[34b2796]259                        SEARCH: {
260                                /* paranoid */ verify( ! __preemption_enabled() );
261
262                                // First, lock the scheduler since we are searching for a thread
263                                ready_schedule_lock();
264
265                                // Try to get the next thread
266                                readyThread = pop_fast( this->cltr );
267                                if(readyThread) { ready_schedule_unlock(); break SEARCH; }
268
269                                // If we can't find a thread, might as well flush any outstanding I/O
[059ad16]270                                if(this->io.pending) { __cfa_io_flush( this, false ); }
[34b2796]271
272                                // Spin a little on I/O, just in case
[12daa43]273                                for(5) {
[34b2796]274                                        __maybe_io_drain( this );
275                                        readyThread = pop_fast( this->cltr );
276                                        if(readyThread) { ready_schedule_unlock(); break SEARCH; }
277                                }
278
279                                // no luck, try stealing a few times
[12daa43]280                                for(5) {
[34b2796]281                                        if( __maybe_io_drain( this ) ) {
282                                                readyThread = pop_fast( this->cltr );
283                                        } else {
284                                                readyThread = pop_slow( this->cltr );
285                                        }
286                                        if(readyThread) { ready_schedule_unlock(); break SEARCH; }
287                                }
288
289                                // still no luck, search for a thread
290                                readyThread = pop_search( this->cltr );
291                                if(readyThread) { ready_schedule_unlock(); break SEARCH; }
292
293                                // Don't block if we are done
[c33c2af]294                                if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) {
295                                        ready_schedule_unlock();
296                                        break MAIN_LOOP;
297                                }
[34b2796]298
299                                __STATS( __tls_stats()->ready.sleep.halts++; )
300
301                                // Push self to idle stack
302                                ready_schedule_unlock();
[5f5a729]303                                if(!mark_idle(this->cltr->procs, * this)) goto SEARCH;
[34b2796]304                                ready_schedule_lock();
305
306                                // Confirm the ready-queue is empty
307                                __maybe_io_drain( this );
308                                readyThread = pop_search( this->cltr );
309                                ready_schedule_unlock();
310
311                                if( readyThread ) {
312                                        // A thread was found, cancel the halt
313                                        mark_awake(this->cltr->procs, * this);
314
315                                        __STATS( __tls_stats()->ready.sleep.cancels++; )
316
317                                        // continue the main loop
318                                        break SEARCH;
319                                }
320
[abcae55]321                                __STATS( if(this->print_halts) __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->unique_id, rdtscl()); )
[1757f98]322                                __cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle_fd);
[34b2796]323
[ec421636]324                                {
325                                        eventfd_t val;
[1757f98]326                                        ssize_t ret = read( this->idle_fd, &val, sizeof(val) );
[ec421636]327                                        if(ret < 0) {
328                                                switch((int)errno) {
329                                                case EAGAIN:
330                                                #if EAGAIN != EWOULDBLOCK
331                                                        case EWOULDBLOCK:
332                                                #endif
333                                                case EINTR:
334                                                        // No need to do anything special here, just assume it's a legitimate wake-up
335                                                        break;
336                                                default:
337                                                        abort( "KERNEL : internal error, read failure on idle eventfd, error(%d) %s.", (int)errno, strerror( (int)errno ) );
338                                                }
339                                        }
340                                }
[34b2796]341
[fb4ccdf]342                                        __STATS( if(this->print_halts) __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->unique_id, rdtscl()); )
[34b2796]343
344                                // We were woken up, remove self from idle
345                                mark_awake(this->cltr->procs, * this);
346
347                                // DON'T just proceed, start looking again
348                                continue MAIN_LOOP;
349                        }
350
351                RUN_THREAD:
352                        /* paranoid */ verify( ! __preemption_enabled() );
353                        /* paranoid */ verify( readyThread );
354
355                        // Reset io dirty bit
356                        this->io.dirty = false;
357
358                        // We found a thread run it
359                        __run_thread(this, readyThread);
360
361                        // Are we done?
362                        if( __atomic_load_n(&this->do_terminate, __ATOMIC_SEQ_CST) ) break MAIN_LOOP;
363
364                        if(this->io.pending && !this->io.dirty) {
[059ad16]365                                __cfa_io_flush( this, false );
[34b2796]366                        }
367
368                        ready_schedule_lock();
369                        __maybe_io_drain( this );
370                        ready_schedule_unlock();
371                        #endif
[c81ebf9]372                }
373
[4069faad]374                __cfadbg_print_safe(runtime_core, "Kernel : core %p stopping\n", this);
[c84e80a]375        }
[8118303]376
[dddb3dd0]377        __cfa_io_stop( this );
378
[454f478]379        post( this->terminated );
[bdeba0b]380
[28d73c1]381        if(this == mainProcessor) {
[6a490b2]382                // HACK : the coroutine context switch expects this_thread to be set
383                // and it make sense for it to be set in all other cases except here
384                // fake it
[8fc652e0]385                __cfaabi_tls.this_thread = mainThread;
[6a490b2]386        }
[7768b8d]387
[4069faad]388        __cfadbg_print_safe(runtime_core, "Kernel : core %p terminated\n", this);
[c84e80a]389}
390
[5c1a531]391static int * __volatile_errno() __attribute__((noinline));
392static int * __volatile_errno() { asm(""); return &errno; }
393
[14a61b5]394// KERNEL ONLY
[1c273d0]395// runThread runs a thread by context switching
396// from the processor coroutine to the target thread
[e84ab3d]397static void __run_thread(processor * this, thread$ * thrd_dst) {
[8fc652e0]398        /* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]399        /* paranoid */ verifyf( thrd_dst->state == Ready || thrd_dst->preempted != __NO_PREEMPTION, "state : %d, preempted %d\n", thrd_dst->state, thrd_dst->preempted);
400        /* paranoid */ verifyf( thrd_dst->link.next == 0p, "Expected null got %p", thrd_dst->link.next );
401        __builtin_prefetch( thrd_dst->context.SP );
402
[dddb3dd0]403        __cfadbg_print_safe(runtime_core, "Kernel : core %p running thread %p (%s)\n", this, thrd_dst, thrd_dst->self_cor.name);
404
[e84ab3d]405        coroutine$ * proc_cor = get_coroutine(this->runner);
[8fcbb4c]406
[9f575ea]407        // set state of processor coroutine to inactive
408        verify(proc_cor->state == Active);
[ae7be7a]409        proc_cor->state = Blocked;
[e8e457e]410
[9f575ea]411        // Actually run the thread
[3381ed7]412        RUNNING:  while(true) {
[ff79d5e]413                thrd_dst->preempted = __NO_PREEMPTION;
414                thrd_dst->state = Active;
[e8e457e]415
[58d64a4]416                // Update global state
[8fc652e0]417                kernelTLS().this_thread = thrd_dst;
[75f3522]418
[8fc652e0]419                /* paranoid */ verify( ! __preemption_enabled() );
420                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
[9d6e1b8a]421                /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[b4b63e8]422                /* paranoid */ verify( thrd_dst->context.SP );
[5afb49a]423                /* paranoid */ verify( thrd_dst->state != Halted );
[e84ab3d]424                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) || thrd_dst->curr_cor == proc_cor || thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_dst ); // add escape condition if we are setting up the processor
425                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) || thrd_dst->curr_cor == proc_cor || thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_dst ); // add escape condition if we are setting up the processor
[ac12f1f]426                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[b4b63e8]427
[3381ed7]428
[58d64a4]429
[9f575ea]430                // set context switch to the thread that the processor is executing
[c7a900a]431                __cfactx_switch( &proc_cor->context, &thrd_dst->context );
432                // when __cfactx_switch returns we are back in the processor coroutine
[9f575ea]433
[50871b4]434
435
[ac12f1f]436                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_dst->canary );
[e84ab3d]437                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) > ((uintptr_t)__get_stack(thrd_dst->curr_cor)->limit) || thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_dst );
438                /* paranoid */ verifyf( ((uintptr_t)thrd_dst->context.SP) < ((uintptr_t)__get_stack(thrd_dst->curr_cor)->base ) || thrd_dst->corctx_flag, "ERROR : Destination thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_dst );
[b4b63e8]439                /* paranoid */ verify( thrd_dst->context.SP );
[9d6e1b8a]440                /* paranoid */ verify( thrd_dst->curr_cluster == this->cltr );
[8fc652e0]441                /* paranoid */ verify( kernelTLS().this_thread == thrd_dst );
442                /* paranoid */ verify( ! __preemption_enabled() );
[75f3522]443
[58d64a4]444                // Reset global state
[8fc652e0]445                kernelTLS().this_thread = 0p;
[3381ed7]446
447                // We just finished running a thread, there are a few things that could have happened.
448                // 1 - Regular case : the thread has blocked and now one has scheduled it yet.
449                // 2 - Racy case    : the thread has blocked but someone has already tried to schedule it.
450                // 4 - Preempted
451                // In case 1, we may have won a race so we can't write to the state again.
452                // In case 2, we lost the race so we now own the thread.
453
454                if(unlikely(thrd_dst->preempted != __NO_PREEMPTION)) {
455                        // The thread was preempted, reschedule it and reset the flag
[24e321c]456                        schedule_thread$( thrd_dst, UNPARK_LOCAL );
[3381ed7]457                        break RUNNING;
458                }
[75f3522]459
[3ea8ad1]460                if(unlikely(thrd_dst->state == Halting)) {
[ff79d5e]461                        // The thread has halted, it should never be scheduled/run again
[5afb49a]462                        // finish the thread
463                        __thread_finish( thrd_dst );
[ff79d5e]464                        break RUNNING;
465                }
466
467                /* paranoid */ verify( thrd_dst->state == Active );
468                thrd_dst->state = Blocked;
469
[3381ed7]470                // set state of processor coroutine to active and the thread to inactive
[ff79d5e]471                int old_ticket = __atomic_fetch_sub(&thrd_dst->ticket, 1, __ATOMIC_SEQ_CST);
472                switch(old_ticket) {
[6a77224]473                        case TICKET_RUNNING:
[3381ed7]474                                // This is case 1, the regular case, nothing more is needed
475                                break RUNNING;
[6a77224]476                        case TICKET_UNBLOCK:
[ec43cf9]477                                #if !defined(__CFA_NO_STATISTICS__)
478                                        __tls_stats()->ready.threads.threads++;
479                                #endif
[3381ed7]480                                // This is case 2, the racy case, someone tried to run this thread before it finished blocking
481                                // In this case, just run it again.
482                                continue RUNNING;
483                        default:
484                                // This makes no sense, something is wrong abort
[ff79d5e]485                                abort();
[3381ed7]486                }
[9f575ea]487        }
[e8e457e]488
[9f575ea]489        // Just before returning to the processor, set the processor coroutine to active
[e8e457e]490        proc_cor->state = Active;
[1eb239e4]491
[dddb3dd0]492        __cfadbg_print_safe(runtime_core, "Kernel : core %p finished running thread %p\n", this, thrd_dst);
493
[ec43cf9]494        #if !defined(__CFA_NO_STATISTICS__)
495                __tls_stats()->ready.threads.threads--;
496        #endif
497
[8fc652e0]498        /* paranoid */ verify( ! __preemption_enabled() );
[82c948c]499}
500
[14a61b5]501// KERNEL_ONLY
[b0c7419]502void returnToKernel() {
[8fc652e0]503        /* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]504        coroutine$ * proc_cor = get_coroutine(kernelTLS().this_processor->runner);
505        thread$ * thrd_src = kernelTLS().this_thread;
[e8e457e]506
[89eff25]507        __STATS( thrd_src->last_proc = kernelTLS().this_processor; )
[29cb302]508
[9f575ea]509        // Run the thread on this processor
510        {
511                int local_errno = *__volatile_errno();
512                #if defined( __i386 ) || defined( __x86_64 )
513                        __x87_store;
514                #endif
[b4b63e8]515                /* paranoid */ verify( proc_cor->context.SP );
[ac12f1f]516                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[c7a900a]517                __cfactx_switch( &thrd_src->context, &proc_cor->context );
[ac12f1f]518                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd_src->canary );
[9f575ea]519                #if defined( __i386 ) || defined( __x86_64 )
520                        __x87_load;
521                #endif
522                *__volatile_errno() = local_errno;
[8fcbb4c]523        }
[deca0f5]524
[29cb302]525        #if !defined(__CFA_NO_STATISTICS__)
[89eff25]526                /* paranoid */ verify( thrd_src->last_proc != 0p );
527                if(thrd_src->last_proc != kernelTLS().this_processor) {
[29cb302]528                        __tls_stats()->ready.threads.migration++;
529                }
530        #endif
531
[8fc652e0]532        /* paranoid */ verify( ! __preemption_enabled() );
[e84ab3d]533        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) < ((uintptr_t)__get_stack(thrd_src->curr_cor)->base ) || thrd_src->corctx_flag, "ERROR : Returning thread$ %p has been corrupted.\n StackPointer too small.\n", thrd_src );
534        /* paranoid */ verifyf( ((uintptr_t)thrd_src->context.SP) > ((uintptr_t)__get_stack(thrd_src->curr_cor)->limit) || thrd_src->corctx_flag, "ERROR : Returning thread$ %p has been corrupted.\n StackPointer too large.\n", thrd_src );
[c84e80a]535}
536
[8def349]537//-----------------------------------------------------------------------------
538// Scheduler routines
[14a61b5]539// KERNEL ONLY
[24e321c]540static void __schedule_thread( thread$ * thrd, unpark_hint hint ) {
[8fc652e0]541        /* paranoid */ verify( ! __preemption_enabled() );
[254ad1b]542        /* paranoid */ verify( ready_schedule_islocked());
[6a490b2]543        /* paranoid */ verify( thrd );
544        /* paranoid */ verify( thrd->state != Halted );
[9d6e1b8a]545        /* paranoid */ verify( thrd->curr_cluster );
[3381ed7]546        /* paranoid */ #if defined( __CFA_WITH_VERIFY__ )
[504a7dc]547        /* paranoid */  if( thrd->state == Blocked || thrd->state == Start ) assertf( thrd->preempted == __NO_PREEMPTION,
548                                        "Error inactive thread marked as preempted, state %d, preemption %d\n", thrd->state, thrd->preempted );
[ff79d5e]549        /* paranoid */  if( thrd->preempted != __NO_PREEMPTION ) assertf(thrd->state == Active,
[504a7dc]550                                        "Error preempted thread marked as not currently running, state %d, preemption %d\n", thrd->state, thrd->preempted );
[3381ed7]551        /* paranoid */ #endif
[6a490b2]552        /* paranoid */ verifyf( thrd->link.next == 0p, "Expected null got %p", thrd->link.next );
[ac12f1f]553        /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
[b4b63e8]554
[b808625]555        const bool local = thrd->state != Start;
[ae7be7a]556        if (thrd->preempted == __NO_PREEMPTION) thrd->state = Ready;
[6b4cdd3]557
[ec43cf9]558        // Dereference the thread now because once we push it, there is not guaranteed it's still valid.
559        struct cluster * cl = thrd->curr_cluster;
[24e321c]560        __STATS(bool outside = hint == UNPARK_LOCAL && thrd->last_proc && thrd->last_proc != kernelTLS().this_processor; )
[32a8b61]561
[254ad1b]562        // push the thread to the cluster ready-queue
[24e321c]563        push( cl, thrd, hint );
[32a8b61]564
[254ad1b]565        // variable thrd is no longer safe to use
[734908c]566        thrd = 0xdeaddeaddeaddeadp;
[32a8b61]567
[254ad1b]568        // wake the cluster using the save variable.
569        __wake_one( cl );
[1c273d0]570
[ec43cf9]571        #if !defined(__CFA_NO_STATISTICS__)
572                if( kernelTLS().this_stats ) {
573                        __tls_stats()->ready.threads.threads++;
[89eff25]574                        if(outside) {
575                                __tls_stats()->ready.threads.extunpark++;
576                        }
[ec43cf9]577                }
578                else {
579                        __atomic_fetch_add(&cl->stats->ready.threads.threads, 1, __ATOMIC_RELAXED);
[89eff25]580                        __atomic_fetch_add(&cl->stats->ready.threads.extunpark, 1, __ATOMIC_RELAXED);
[ec43cf9]581                }
582        #endif
583
[254ad1b]584        /* paranoid */ verify( ready_schedule_islocked());
[8fc652e0]585        /* paranoid */ verify( ! __preemption_enabled() );
[db6f06a]586}
587
[24e321c]588void schedule_thread$( thread$ * thrd, unpark_hint hint ) {
[254ad1b]589        ready_schedule_lock();
[24e321c]590                __schedule_thread( thrd, hint );
[254ad1b]591        ready_schedule_unlock();
592}
593
[14a61b5]594// KERNEL ONLY
[e84ab3d]595static inline thread$ * __next_thread(cluster * this) with( *this ) {
[8fc652e0]596        /* paranoid */ verify( ! __preemption_enabled() );
[7768b8d]597
[e873838]598        ready_schedule_lock();
[e84ab3d]599                thread$ * thrd = pop_fast( this );
[e873838]600        ready_schedule_unlock();
[7768b8d]601
[8fc652e0]602        /* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]603        return thrd;
[eb2e723]604}
605
[64a7146]606// KERNEL ONLY
[e84ab3d]607static inline thread$ * __next_thread_slow(cluster * this) with( *this ) {
[8fc652e0]608        /* paranoid */ verify( ! __preemption_enabled() );
[64a7146]609
[e873838]610        ready_schedule_lock();
[e84ab3d]611                thread$ * thrd;
[fc59df78]612                for(25) {
613                        thrd = pop_slow( this );
614                        if(thrd) goto RET;
615                }
616                thrd = pop_search( this );
617
618                RET:
[e873838]619        ready_schedule_unlock();
[64a7146]620
[8fc652e0]621        /* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]622        return thrd;
[64a7146]623}
624
[e84ab3d]625static inline bool __must_unpark( thread$ * thrd ) {
[ff79d5e]626        int old_ticket = __atomic_fetch_add(&thrd->ticket, 1, __ATOMIC_SEQ_CST);
627        switch(old_ticket) {
[6a77224]628                case TICKET_RUNNING:
[3381ed7]629                        // Wake won the race, the thread will reschedule/rerun itself
[c6c7e6c]630                        return false;
[6a77224]631                case TICKET_BLOCKED:
[3381ed7]632                        /* paranoid */ verify( ! thrd->preempted != __NO_PREEMPTION );
[ff79d5e]633                        /* paranoid */ verify( thrd->state == Blocked );
[c6c7e6c]634                        return true;
[3381ed7]635                default:
636                        // This makes no sense, something is wrong abort
[7ee8153]637                        abort("Thread %p (%s) has mismatch park/unpark\n", thrd, thrd->self_cor.name);
[de6319f]638        }
[eb2e723]639}
640
[24e321c]641void __kernel_unpark( thread$ * thrd, unpark_hint hint ) {
[e9c0b4c]642        /* paranoid */ verify( ! __preemption_enabled() );
643        /* paranoid */ verify( ready_schedule_islocked());
644
645        if( !thrd ) return;
646
647        if(__must_unpark(thrd)) {
648                // Wake lost the race,
[24e321c]649                __schedule_thread( thrd, hint );
[e9c0b4c]650        }
651
652        /* paranoid */ verify( ready_schedule_islocked());
653        /* paranoid */ verify( ! __preemption_enabled() );
654}
655
[24e321c]656void unpark( thread$ * thrd, unpark_hint hint ) {
[c6c7e6c]657        if( !thrd ) return;
[0b33412]658
[c6c7e6c]659        if(__must_unpark(thrd)) {
660                disable_interrupts();
[a3821fa]661                        // Wake lost the race,
[24e321c]662                        schedule_thread$( thrd, hint );
[a3821fa]663                enable_interrupts(false);
[de6319f]664        }
[eb2e723]665}
[0b33412]666
[e235429]667void park( void ) {
[a3821fa]668        __disable_interrupts_checked();
669                /* paranoid */ verify( kernelTLS().this_thread->preempted == __NO_PREEMPTION );
670                returnToKernel();
671        __enable_interrupts_checked();
[0c78741]672
673}
[09800e9]674
[5afb49a]675extern "C" {
676        // Leave the thread monitor
677        // last routine called by a thread.
678        // Should never return
679        void __cfactx_thrd_leave() {
[e84ab3d]680                thread$ * thrd = active_thread();
681                monitor$ * this = &thrd->self_mon;
[5afb49a]682
683                // Lock the monitor now
684                lock( this->lock __cfaabi_dbg_ctx2 );
685
686                disable_interrupts();
687
[9d6e1b8a]688                /* paranoid */ verify( ! __preemption_enabled() );
689                /* paranoid */ verify( thrd->state == Active );
690                /* paranoid */ verify( 0x0D15EA5E0D15EA5Ep == thrd->canary );
691                /* paranoid */ verify( kernelTLS().this_thread == thrd );
692                /* paranoid */ verify( thrd->context.SP );
[e84ab3d]693                /* 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 );
694                /* 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 );
[9d6e1b8a]695
[3ea8ad1]696                thrd->state = Halting;
[58688bf]697                if( TICKET_RUNNING != thrd->ticket ) { abort( "Thread terminated with pending unpark" ); }
[9d6e1b8a]698                if( thrd != this->owner ) { abort( "Thread internal monitor has incorrect owner" ); }
699                if( this->recursion != 1) { abort( "Thread internal monitor has unbalanced recursion" ); }
[5afb49a]700
701                // Leave the thread
702                returnToKernel();
703
704                // Control flow should never reach here!
[9d6e1b8a]705                abort();
[5afb49a]706        }
[09800e9]707}
708
[14a61b5]709// KERNEL ONLY
[3381ed7]710bool force_yield( __Preemption_Reason reason ) {
[a3821fa]711        __disable_interrupts_checked();
[e84ab3d]712                thread$ * thrd = kernelTLS().this_thread;
[a3821fa]713                /* paranoid */ verify(thrd->state == Active);
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->link.next == 0p) {
721                        preempted = true;
722                        thrd->preempted = reason;
723                        returnToKernel();
724                }
725        __enable_interrupts_checked( false );
[3381ed7]726        return preempted;
[f2b12406]727}
728
[14a61b5]729//=============================================================================================
[92e7631]730// Kernel Idle Sleep
[14a61b5]731//=============================================================================================
[64a7146]732// Wake a thread from the front if there are any
[e873838]733static void __wake_one(cluster * this) {
[8fc652e0]734        /* paranoid */ verify( ! __preemption_enabled() );
[e873838]735        /* paranoid */ verify( ready_schedule_islocked() );
[14a61b5]736
[64a7146]737        // Check if there is a sleeping processor
[34b8cb7]738        int fd = __atomic_load_n(&this->procs.fd, __ATOMIC_SEQ_CST);
[14a61b5]739
[64a7146]740        // If no one is sleeping, we are done
[34b8cb7]741        if( fd == 0 ) return;
[14a61b5]742
[64a7146]743        // We found a processor, wake it up
[dddb3dd0]744        eventfd_t val;
745        val = 1;
[34b8cb7]746        eventfd_write( fd, val );
[14a61b5]747
[1eb239e4]748        #if !defined(__CFA_NO_STATISTICS__)
[5cb51502]749                if( kernelTLS().this_stats ) {
750                        __tls_stats()->ready.sleep.wakes++;
751                }
752                else {
753                        __atomic_fetch_add(&this->stats->ready.sleep.wakes, 1, __ATOMIC_RELAXED);
754                }
[1eb239e4]755        #endif
756
[e873838]757        /* paranoid */ verify( ready_schedule_islocked() );
[8fc652e0]758        /* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]759
760        return;
[64a7146]761}
[14a61b5]762
[64a7146]763// Unconditionnaly wake a thread
[1eb239e4]764void __wake_proc(processor * this) {
[64a7146]765        __cfadbg_print_safe(runtime_core, "Kernel : waking Processor %p\n", this);
[14a61b5]766
[a3821fa]767        __disable_interrupts_checked();
[8fc652e0]768                /* paranoid */ verify( ! __preemption_enabled() );
[dddb3dd0]769                eventfd_t val;
770                val = 1;
[1757f98]771                eventfd_write( this->idle_fd, val );
[a3821fa]772        __enable_interrupts_checked();
[92e7631]773}
774
[1757f98]775static void idle_sleep(processor * this) {
776        #if !defined(__CFA_NO_STATISTICS__)
777                if(this->print_halts) {
778                        __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 0\n", this->unique_id, rdtscl());
779                }
780        #endif
781
782        __cfadbg_print_safe(runtime_core, "Kernel : core %p waiting on eventfd %d\n", this, this->idle_fd);
783
784        {
785                eventfd_t val;
786                ssize_t ret = read( this->idle_fd, &val, sizeof(val) );
787                if(ret < 0) {
788                        switch((int)errno) {
789                        case EAGAIN:
790                        #if EAGAIN != EWOULDBLOCK
791                                case EWOULDBLOCK:
792                        #endif
793                        case EINTR:
794                                // No need to do anything special here, just assume it's a legitimate wake-up
795                                break;
796                        default:
797                                abort( "KERNEL : internal error, read failure on idle eventfd, error(%d) %s.", (int)errno, strerror( (int)errno ) );
798                        }
799                }
800        }
801
802        #if !defined(__CFA_NO_STATISTICS__)
803                if(this->print_halts) {
804                        __cfaabi_bits_print_safe( STDOUT_FILENO, "PH:%d - %lld 1\n", this->unique_id, rdtscl());
805                }
806        #endif
807}
808
[5f5a729]809static bool mark_idle(__cluster_proc_list & this, processor & proc) {
[8fc652e0]810        /* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]811        if(!try_lock( this )) return false;
[1eb239e4]812                this.idle++;
813                /* paranoid */ verify( this.idle <= this.total );
[fc59b580]814                remove(proc);
[6a9b12b]815                insert_first(this.idles, proc);
[34b8cb7]816
[1757f98]817                __atomic_store_n(&this.fd, proc.idle_fd, __ATOMIC_SEQ_CST);
[1eb239e4]818        unlock( this );
[8fc652e0]819        /* paranoid */ verify( ! __preemption_enabled() );
[5f5a729]820
821        return true;
[1eb239e4]822}
[8e16177]823
[6a9b12b]824static void mark_awake(__cluster_proc_list & this, processor & proc) {
[8fc652e0]825        /* paranoid */ verify( ! __preemption_enabled() );
[1eb239e4]826        lock( this );
827                this.idle--;
828                /* paranoid */ verify( this.idle >= 0 );
829                remove(proc);
[fc59b580]830                insert_last(this.actives, proc);
[6a9b12b]831
[a633f6f]832                {
833                        int fd = 0;
[1757f98]834                        if(!this.idles`isEmpty) fd = this.idles`first.idle_fd;
[a633f6f]835                        __atomic_store_n(&this.fd, fd, __ATOMIC_SEQ_CST);
836                }
837
[34b8cb7]838        unlock( this );
[6a9b12b]839        /* paranoid */ verify( ! __preemption_enabled() );
[6b4cdd3]840}
841
[dbe9b08]842//=============================================================================================
843// Unexpected Terminating logic
844//=============================================================================================
[92bfda0]845void __kernel_abort_msg( char * abort_text, int abort_text_size ) {
[e84ab3d]846        thread$ * thrd = __cfaabi_tls.this_thread;
[9d944b2]847
[de94a60]848        if(thrd) {
849                int len = snprintf( abort_text, abort_text_size, "Error occurred while executing thread %.256s (%p)", thrd->self_cor.name, thrd );
[1c40091]850                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]851
[212c2187]852                if ( &thrd->self_cor != thrd->curr_cor ) {
853                        len = snprintf( abort_text, abort_text_size, " in coroutine %.256s (%p).\n", thrd->curr_cor->name, thrd->curr_cor );
[1c40091]854                        __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[de94a60]855                }
856                else {
[1c40091]857                        __cfaabi_bits_write( STDERR_FILENO, ".\n", 2 );
[de94a60]858                }
[1c273d0]859        }
[9d944b2]860        else {
[de94a60]861                int len = snprintf( abort_text, abort_text_size, "Error occurred outside of any thread.\n" );
[1c40091]862                __cfaabi_bits_write( STDERR_FILENO, abort_text, len );
[9d944b2]863        }
864}
865
[92bfda0]866int __kernel_abort_lastframe( void ) __attribute__ ((__nothrow__)) {
867        return get_coroutine(__cfaabi_tls.this_thread) == get_coroutine(mainThread) ? 4 : 2;
[2b8bc41]868}
869
[de94a60]870static __spinlock_t kernel_debug_lock;
871
[9d944b2]872extern "C" {
[1c40091]873        void __cfaabi_bits_acquire() {
[36982fc]874                lock( kernel_debug_lock __cfaabi_dbg_ctx2 );
[9d944b2]875        }
876
[1c40091]877        void __cfaabi_bits_release() {
[ea7d2b0]878                unlock( kernel_debug_lock );
[9d944b2]879        }
[8118303]880}
881
[fa21ac9]882//=============================================================================================
883// Kernel Utilities
884//=============================================================================================
[dddb3dd0]885#if defined(CFA_HAVE_LINUX_IO_URING_H)
886#include "io/types.hfa"
887#endif
888
[c1c95b1]889static inline bool __maybe_io_drain( processor * proc ) {
[e9c0b4c]890        bool ret = false;
[dddb3dd0]891        #if defined(CFA_HAVE_LINUX_IO_URING_H)
892                __cfadbg_print_safe(runtime_core, "Kernel : core %p checking io for ring %d\n", proc, proc->io.ctx->fd);
893
894                // Check if we should drain the queue
895                $io_context * ctx = proc->io.ctx;
896                unsigned head = *ctx->cq.head;
897                unsigned tail = *ctx->cq.tail;
[c1c95b1]898                if(head == tail) return false;
[57f70ab]899                #if OLD_MAIN
[34b8cb7]900                        ready_schedule_lock();
901                        ret = __cfa_io_drain( proc );
902                        ready_schedule_unlock();
[57f70ab]903                #else
904                        ret = __cfa_io_drain( proc );
[34b8cb7]905                #endif
[dddb3dd0]906        #endif
[e9c0b4c]907        return ret;
[dddb3dd0]908}
909
[de94a60]910//-----------------------------------------------------------------------------
911// Debug
912__cfaabi_dbg_debug_do(
[1997b4e]913        extern "C" {
[ae66348]914                void __cfaabi_dbg_record_lock(__spinlock_t & this, const char prev_name[]) {
[1997b4e]915                        this.prev_name = prev_name;
[8fc652e0]916                        this.prev_thrd = kernelTLS().this_thread;
[1997b4e]917                }
[9181f1d]918        }
[f7d6bb0]919)
[2026bb6]920
921//-----------------------------------------------------------------------------
922// Debug
[8c50aed]923bool threading_enabled(void) __attribute__((const)) {
[2026bb6]924        return true;
925}
[c34ebf2]926
927//-----------------------------------------------------------------------------
928// Statistics
929#if !defined(__CFA_NO_STATISTICS__)
930        void print_halts( processor & this ) {
931                this.print_halts = true;
932        }
[58688bf]933
[69914cbc]934        static void crawl_list( cluster * cltr, dlist(processor) & list, unsigned count ) {
[8464edf]935                /* paranoid */ verify( cltr->stats );
936
937                processor * it = &list`first;
938                for(unsigned i = 0; i < count; i++) {
939                        /* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
940                        /* paranoid */ verify( it->local_data->this_stats );
[c33c2af]941                        // __print_stats( it->local_data->this_stats, cltr->print_stats, "Processor", it->name, (void*)it );
[8464edf]942                        __tally_stats( cltr->stats, it->local_data->this_stats );
943                        it = &(*it)`next;
944                }
945        }
946
947        void crawl_cluster_stats( cluster & this ) {
948                // Stop the world, otherwise stats could get really messed-up
949                // this doesn't solve all problems but does solve many
950                // so it's probably good enough
[c33c2af]951                disable_interrupts();
[8464edf]952                uint_fast32_t last_size = ready_mutate_lock();
953
954                        crawl_list(&this, this.procs.actives, this.procs.total - this.procs.idle);
955                        crawl_list(&this, this.procs.idles  , this.procs.idle );
956
957                // Unlock the RWlock
958                ready_mutate_unlock( last_size );
[c33c2af]959                enable_interrupts();
[8464edf]960        }
961
962
[58688bf]963        void print_stats_now( cluster & this, int flags ) {
[8464edf]964                crawl_cluster_stats( this );
[1b033b8]965                __print_stats( this.stats, this.print_stats, "Cluster", this.name, (void*)&this );
966        }
[c34ebf2]967#endif
[8118303]968// Local Variables: //
969// mode: c //
970// tab-width: 4 //
971// End: //
Note: See TracBrowser for help on using the repository browser.