source: libcfa/src/concurrency/kernel/cluster.cfa @ 4ecc35a

ADTast-experimentalenumpthread-emulationqualifiedEnum
Last change on this file since 4ecc35a was adb3ea1, checked in by Thierry Delisle <tdelisle@…>, 2 years ago

Some more incremental work towards using timestamps for io fairness
  • Property mode set to 100644
File size: 17.7 KB
Line 
1//
2// Cforall Version 1.0.0 Copyright (C) 2022 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// cluster.cfa -- file that includes helpers for subsystem that need cluster wide support
8//
9// Author           : Thierry Delisle
10// Created On       : Fri Mar 11 12:39:24 2022
11// Last Modified By :
12// Last Modified On :
13// Update Count     :
14//
15
16#define __cforall_thread__
17#define _GNU_SOURCE
18
19#include "bits/defs.hfa"
20#include "device/cpu.hfa"
21#include "kernel/cluster.hfa"
22#include "kernel/private.hfa"
23
24#include "stdlib.hfa"
25#include "limits.hfa"
26#include "math.hfa"
27
28#include "ready_subqueue.hfa"
29
30#include <errno.h>
31#include <unistd.h>
32
33extern "C" {
34        #include <sys/syscall.h>  // __NR_xxx
35}
36
37// No overriden function, no environment variable, no define
38// fall back to a magic number
39#ifndef __CFA_MAX_PROCESSORS__
40        #define __CFA_MAX_PROCESSORS__ 1024
41#endif
42
43#if !defined(__CFA_NO_STATISTICS__)
44        #define __STATS(...) __VA_ARGS__
45#else
46        #define __STATS(...)
47#endif
48
49// returns the maximum number of processors the RWLock support
50__attribute__((weak)) unsigned __max_processors() {
51        const char * max_cores_s = getenv("CFA_MAX_PROCESSORS");
52        if(!max_cores_s) {
53                __cfadbg_print_nolock(ready_queue, "No CFA_MAX_PROCESSORS in ENV\n");
54                return __CFA_MAX_PROCESSORS__;
55        }
56
57        char * endptr = 0p;
58        long int max_cores_l = strtol(max_cores_s, &endptr, 10);
59        if(max_cores_l < 1 || max_cores_l > 65535) {
60                __cfadbg_print_nolock(ready_queue, "CFA_MAX_PROCESSORS out of range : %ld\n", max_cores_l);
61                return __CFA_MAX_PROCESSORS__;
62        }
63        if('\0' != *endptr) {
64                __cfadbg_print_nolock(ready_queue, "CFA_MAX_PROCESSORS not a decimal number : %s\n", max_cores_s);
65                return __CFA_MAX_PROCESSORS__;
66        }
67
68        return max_cores_l;
69}
70
71#if   defined(CFA_HAVE_LINUX_LIBRSEQ)
72        // No forward declaration needed
73        #define __kernel_rseq_register rseq_register_current_thread
74        #define __kernel_rseq_unregister rseq_unregister_current_thread
75#elif defined(CFA_HAVE_LINUX_RSEQ_H)
76        static void __kernel_raw_rseq_register  (void);
77        static void __kernel_raw_rseq_unregister(void);
78
79        #define __kernel_rseq_register __kernel_raw_rseq_register
80        #define __kernel_rseq_unregister __kernel_raw_rseq_unregister
81#else
82        // No forward declaration needed
83        // No initialization needed
84        static inline void noop(void) {}
85
86        #define __kernel_rseq_register noop
87        #define __kernel_rseq_unregister noop
88#endif
89
90//=======================================================================
91// Cluster wide reader-writer lock
92//=======================================================================
93void  ?{}(__scheduler_RWLock_t & this) {
94        this.max   = __max_processors();
95        this.alloc = 0;
96        this.ready = 0;
97        this.data  = alloc(this.max);
98        this.write_lock  = false;
99
100        /*paranoid*/ verify(__atomic_is_lock_free(sizeof(this.alloc), &this.alloc));
101        /*paranoid*/ verify(__atomic_is_lock_free(sizeof(this.ready), &this.ready));
102
103}
104void ^?{}(__scheduler_RWLock_t & this) {
105        free(this.data);
106}
107
108
109//=======================================================================
110// Lock-Free registering/unregistering of threads
111unsigned register_proc_id( void ) with(*__scheduler_lock) {
112        __kernel_rseq_register();
113
114        bool * handle = (bool *)&kernelTLS().sched_lock;
115
116        // Step - 1 : check if there is already space in the data
117        uint_fast32_t s = ready;
118
119        // Check among all the ready
120        for(uint_fast32_t i = 0; i < s; i++) {
121                bool * volatile * cell = (bool * volatile *)&data[i]; // Cforall is bugged and the double volatiles causes problems
122                /* paranoid */ verify( handle != *cell );
123
124                bool * null = 0p; // Re-write every loop since compare thrashes it
125                if( __atomic_load_n(cell, (int)__ATOMIC_RELAXED) == null
126                        && __atomic_compare_exchange_n( cell, &null, handle, false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST)) {
127                        /* paranoid */ verify(i < ready);
128                        /* paranoid */ verify( (kernelTLS().sched_id = i, true) );
129                        return i;
130                }
131        }
132
133        if(max <= alloc) abort("Trying to create more than %ud processors", __scheduler_lock->max);
134
135        // Step - 2 : F&A to get a new spot in the array.
136        uint_fast32_t n = __atomic_fetch_add(&alloc, 1, __ATOMIC_SEQ_CST);
137        if(max <= n) abort("Trying to create more than %ud processors", __scheduler_lock->max);
138
139        // Step - 3 : Mark space as used and then publish it.
140        data[n] = handle;
141        while() {
142                unsigned copy = n;
143                if( __atomic_load_n(&ready, __ATOMIC_RELAXED) == n
144                        && __atomic_compare_exchange_n(&ready, &copy, n + 1, true, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
145                        break;
146                Pause();
147        }
148
149        // Return new spot.
150        /* paranoid */ verify(n < ready);
151        /* paranoid */ verify( (kernelTLS().sched_id = n, true) );
152        return n;
153}
154
155void unregister_proc_id( unsigned id ) with(*__scheduler_lock) {
156        /* paranoid */ verify(id < ready);
157        /* paranoid */ verify(id == kernelTLS().sched_id);
158        /* paranoid */ verify(data[id] == &kernelTLS().sched_lock);
159
160        bool * volatile * cell = (bool * volatile *)&data[id]; // Cforall is bugged and the double volatiles causes problems
161
162        __atomic_store_n(cell, 0p, __ATOMIC_RELEASE);
163
164        __kernel_rseq_unregister();
165}
166
167//-----------------------------------------------------------------------
168// Writer side : acquire when changing the ready queue, e.g. adding more
169//  queues or removing them.
170uint_fast32_t ready_mutate_lock( void ) with(*__scheduler_lock) {
171        /* paranoid */ verify( ! __preemption_enabled() );
172
173        // Step 1 : lock global lock
174        // It is needed to avoid processors that register mid Critical-Section
175        //   to simply lock their own lock and enter.
176        __atomic_acquire( &write_lock );
177
178        // Make sure we won't deadlock ourself
179        // Checking before acquiring the writer lock isn't safe
180        // because someone else could have locked us.
181        /* paranoid */ verify( ! kernelTLS().sched_lock );
182
183        // Step 2 : lock per-proc lock
184        // Processors that are currently being registered aren't counted
185        //   but can't be in read_lock or in the critical section.
186        // All other processors are counted
187        uint_fast32_t s = ready;
188        for(uint_fast32_t i = 0; i < s; i++) {
189                volatile bool * llock = data[i];
190                if(llock) __atomic_acquire( llock );
191        }
192
193        /* paranoid */ verify( ! __preemption_enabled() );
194        return s;
195}
196
197void ready_mutate_unlock( uint_fast32_t last_s ) with(*__scheduler_lock) {
198        /* paranoid */ verify( ! __preemption_enabled() );
199
200        // Step 1 : release local locks
201        // This must be done while the global lock is held to avoid
202        //   threads that where created mid critical section
203        //   to race to lock their local locks and have the writer
204        //   immidiately unlock them
205        // Alternative solution : return s in write_lock and pass it to write_unlock
206        for(uint_fast32_t i = 0; i < last_s; i++) {
207                volatile bool * llock = data[i];
208                if(llock) __atomic_store_n(llock, (bool)false, __ATOMIC_RELEASE);
209        }
210
211        // Step 2 : release global lock
212        /*paranoid*/ assert(true == write_lock);
213        __atomic_store_n(&write_lock, (bool)false, __ATOMIC_RELEASE);
214
215        /* paranoid */ verify( ! __preemption_enabled() );
216}
217
218//=======================================================================
219// Cluster growth
220static const unsigned __readyq_single_shard = 2;
221
222//-----------------------------------------------------------------------
223// Check that all the intrusive queues in the data structure are still consistent
224static void check_readyQ( cluster * cltr ) with (cltr->sched) {
225        #if defined(__CFA_WITH_VERIFY__)
226                {
227                        const unsigned lanes_count = readyQ.count;
228                        for( idx ; lanes_count ) {
229                                __intrusive_lane_t & sl = readyQ.data[idx];
230                                assert(!readyQ.data[idx].lock);
231
232                                        if(is_empty(sl)) {
233                                                assert( sl.anchor.next == 0p );
234                                                assert( sl.anchor.ts   == -1llu );
235                                                assert( mock_head(sl)  == sl.prev );
236                                        } else {
237                                                assert( sl.anchor.next != 0p );
238                                                assert( sl.anchor.ts   != -1llu );
239                                                assert( mock_head(sl)  != sl.prev );
240                                        }
241                        }
242                }
243        #endif
244}
245
246// Call this function of the intrusive list was moved using memcpy
247// fixes the list so that the pointers back to anchors aren't left dangling
248static inline void fix(__intrusive_lane_t & ll) {
249        if(is_empty(ll)) {
250                verify(ll.anchor.next == 0p);
251                ll.prev = mock_head(ll);
252        }
253}
254
255static void assign_list(unsigned & valrq, unsigned & valio, dlist(processor) & list, unsigned count) {
256        processor * it = &list`first;
257        for(unsigned i = 0; i < count; i++) {
258                /* paranoid */ verifyf( it, "Unexpected null iterator, at index %u of %u\n", i, count);
259                it->rdq.id = valrq;
260                it->rdq.target = MAX;
261                it->io.id = valio;
262                it->io.target = MAX;
263                valrq += __shard_factor.readyq;
264                valio += __shard_factor.io;
265                it = &(*it)`next;
266        }
267}
268
269static void reassign_cltr_id(struct cluster * cltr) {
270        unsigned prefrq = 0;
271        unsigned prefio = 0;
272        assign_list(prefrq, prefio, cltr->procs.actives, cltr->procs.total - cltr->procs.idle);
273        assign_list(prefrq, prefio, cltr->procs.idles  , cltr->procs.idle );
274}
275
276static void assign_io($io_context ** data, size_t count, dlist(processor) & list) {
277        processor * it = &list`first;
278        while(it) {
279                /* paranoid */ verifyf( it, "Unexpected null iterator\n");
280                /* paranoid */ verifyf( it->io.id < count, "Processor %p has id %u above count %zu\n", it, it->rdq.id, count);
281                data[it->io.id] = it->io.ctx;
282                it = &(*it)`next;
283        }
284}
285
286static void reassign_cltr_io(struct cluster * cltr) {
287        assign_io(cltr->sched.io.data, cltr->sched.io.count, cltr->procs.actives);
288        assign_io(cltr->sched.io.data, cltr->sched.io.count, cltr->procs.idles  );
289}
290
291static void fix_times( __timestamp_t * volatile & tscs, unsigned count ) {
292        tscs = alloc(count, tscs`realloc);
293        for(i; count) {
294                tscs[i].tv = rdtscl();
295                tscs[i].ma = 0;
296        }
297}
298
299// Grow the ready queue
300void ready_queue_grow(struct cluster * cltr) {
301        int target = cltr->procs.total;
302
303        /* paranoid */ verify( ready_mutate_islocked() );
304        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue\n");
305
306        // Make sure that everything is consistent
307        /* paranoid */ check_readyQ( cltr );
308
309
310        // Find new count
311        // Make sure we always have atleast 1 list
312        size_t ocount = cltr->sched.readyQ.count;
313        size_t ncount = max(target * __shard_factor.readyq, __readyq_single_shard);
314
315        // Do we have to do anything?
316        if( ocount != ncount ) {
317
318                // grow the ready queue
319                with( cltr->sched ) {
320
321                        // Allocate new array (uses realloc and memcpies the data)
322                        readyQ.data = alloc( ncount, readyQ.data`realloc );
323
324                        // Fix the moved data
325                        for( idx; ocount ) {
326                                fix(readyQ.data[idx]);
327                        }
328
329                        // Construct new data
330                        for( idx; ocount ~ ncount) {
331                                (readyQ.data[idx]){};
332                        }
333
334                        // Update original count
335                        readyQ.count = ncount;
336                }
337
338
339                fix_times(cltr->sched.readyQ.tscs, cltr->sched.readyQ.count);
340        }
341
342        // Fix the io times
343        cltr->sched.io.count = target * __shard_factor.io;
344        fix_times(cltr->sched.io.tscs, cltr->sched.io.count);
345
346        // realloc the caches
347        cltr->sched.caches = alloc( target, cltr->sched.caches`realloc );
348
349        // reassign the clusters.
350        reassign_cltr_id(cltr);
351
352        cltr->sched.io.data = alloc( cltr->sched.io.count, cltr->sched.io.data`realloc );
353        reassign_cltr_io(cltr);
354
355        // Make sure that everything is consistent
356        /* paranoid */ check_readyQ( cltr );
357        /* paranoid */ verify( (target == 0) == (cltr->sched.caches == 0p) );
358
359        __cfadbg_print_safe(ready_queue, "Kernel : Growing ready queue done\n");
360
361        /* paranoid */ verify( ready_mutate_islocked() );
362}
363
364// Shrink the ready queue
365void ready_queue_shrink(struct cluster * cltr) {
366        /* paranoid */ verify( ready_mutate_islocked() );
367        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue\n");
368
369        // Make sure that everything is consistent
370        /* paranoid */ check_readyQ( cltr );
371
372        int target = cltr->procs.total;
373
374        with( cltr->sched ) {
375                // Remember old count
376                size_t ocount = readyQ.count;
377
378                // Find new count
379                // Make sure we always have atleast 1 list
380                size_t ncount = max(target * __shard_factor.readyq, __readyq_single_shard);
381                /* paranoid */ verifyf( ocount >= ncount, "Error in shrinking size calculation, %zu >= %zu", ocount, ncount );
382                /* paranoid */ verifyf( ncount == target * __shard_factor.readyq || ncount == __readyq_single_shard,
383                /* paranoid */          "Error in shrinking size calculation, expected %u or %u, got %zu", target * __shard_factor.readyq, __readyq_single_shard, ncount );
384
385                readyQ.count = ncount;
386
387                // for printing count the number of displaced threads
388                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
389                        __attribute__((unused)) size_t displaced = 0;
390                #endif
391
392                // redistribute old data
393                for( idx; ncount ~ ocount) {
394                        // Lock is not strictly needed but makes checking invariants much easier
395                        __attribute__((unused)) bool locked = __atomic_try_acquire(&readyQ.data[idx].lock);
396                        verify(locked);
397
398                        // As long as we can pop from this lane to push the threads somewhere else in the queue
399                        while(!is_empty(readyQ.data[idx])) {
400                                struct thread$ * thrd;
401                                unsigned long long _;
402                                [thrd, _] = pop(readyQ.data[idx]);
403
404                                push(cltr, thrd, true);
405
406                                // for printing count the number of displaced threads
407                                #if defined(__CFA_DEBUG_PRINT__) || defined(__CFA_DEBUG_PRINT_READY_QUEUE__)
408                                        displaced++;
409                                #endif
410                        }
411
412                        // Unlock the lane
413                        __atomic_unlock(&readyQ.data[idx].lock);
414
415                        // TODO print the queue statistics here
416
417                        ^(readyQ.data[idx]){};
418                }
419
420                __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue displaced %zu threads\n", displaced);
421
422                // Allocate new array (uses realloc and memcpies the data)
423                readyQ.data = alloc( ncount, readyQ.data`realloc );
424
425                // Fix the moved data
426                for( idx; ncount ) {
427                        fix(readyQ.data[idx]);
428                }
429
430                fix_times(readyQ.tscs, ncount);
431        }
432        cltr->sched.caches = alloc( target, cltr->sched.caches`realloc );
433
434        // Fix the io times
435        cltr->sched.io.count = target * __shard_factor.io;
436        fix_times(cltr->sched.io.tscs, cltr->sched.io.count);
437
438        reassign_cltr_id(cltr);
439
440        cltr->sched.io.data = alloc( cltr->sched.io.count, cltr->sched.io.data`realloc );
441        reassign_cltr_io(cltr);
442
443        // Make sure that everything is consistent
444        /* paranoid */ verify( (target == 0) == (cltr->sched.caches == 0p) );
445        /* paranoid */ check_readyQ( cltr );
446
447        __cfadbg_print_safe(ready_queue, "Kernel : Shrinking ready queue done\n");
448        /* paranoid */ verify( ready_mutate_islocked() );
449}
450
451void ready_queue_close(struct cluster * cltr) {
452        free( cltr->sched.readyQ.data );
453        free( cltr->sched.readyQ.tscs );
454        cltr->sched.readyQ.data = 0p;
455        cltr->sched.readyQ.tscs = 0p;
456        cltr->sched.readyQ.count = 0;
457
458        free( cltr->sched.io.tscs );
459        free( cltr->sched.caches );
460}
461
462// Ctor
463void ?{}( __intrusive_lane_t & this ) {
464        this.lock = false;
465        this.prev = mock_head(this);
466        this.anchor.next = 0p;
467        this.anchor.ts   = -1llu;
468        #if !defined(__CFA_NO_STATISTICS__)
469                this.cnt  = 0;
470        #endif
471
472        // We add a boat-load of assertions here because the anchor code is very fragile
473        /* paranoid */ _Static_assert( offsetof( thread$, link ) == offsetof(__intrusive_lane_t, anchor) );
474        /* paranoid */ verify( offsetof( thread$, link ) == offsetof(__intrusive_lane_t, anchor) );
475        /* paranoid */ verify( ((uintptr_t)( mock_head(this) ) + offsetof( thread$, link )) == (uintptr_t)(&this.anchor) );
476        /* paranoid */ verify( &mock_head(this)->link.next == &this.anchor.next );
477        /* paranoid */ verify( &mock_head(this)->link.ts   == &this.anchor.ts   );
478        /* paranoid */ verify( mock_head(this)->link.next == 0p );
479        /* paranoid */ verify( mock_head(this)->link.ts   == -1llu  );
480        /* paranoid */ verify( mock_head(this) == this.prev );
481        /* paranoid */ verify( __alignof__(__intrusive_lane_t) == 128 );
482        /* paranoid */ verify( __alignof__(this) == 128 );
483        /* paranoid */ verifyf( ((intptr_t)(&this) % 128) == 0, "Expected address to be aligned %p %% 128 == %zd", &this, ((intptr_t)(&this) % 128) );
484}
485
486// Dtor is trivial
487void ^?{}( __intrusive_lane_t & this ) {
488        // Make sure the list is empty
489        /* paranoid */ verify( this.anchor.next == 0p );
490        /* paranoid */ verify( this.anchor.ts   == -1llu );
491        /* paranoid */ verify( mock_head(this)  == this.prev );
492}
493
494#if   defined(CFA_HAVE_LINUX_LIBRSEQ)
495        // No definition needed
496#elif defined(CFA_HAVE_LINUX_RSEQ_H)
497
498        #if defined( __x86_64 ) || defined( __i386 )
499                #define RSEQ_SIG        0x53053053
500        #elif defined( __ARM_ARCH )
501                #ifdef __ARMEB__
502                #define RSEQ_SIG    0xf3def5e7      /* udf    #24035    ; 0x5de3 (ARMv6+) */
503                #else
504                #define RSEQ_SIG    0xe7f5def3      /* udf    #24035    ; 0x5de3 */
505                #endif
506        #endif
507
508        extern void __disable_interrupts_hard();
509        extern void __enable_interrupts_hard();
510
511        static void __kernel_raw_rseq_register  (void) {
512                /* paranoid */ verify( __cfaabi_rseq.cpu_id == RSEQ_CPU_ID_UNINITIALIZED );
513
514                // int ret = syscall(__NR_rseq, &__cfaabi_rseq, sizeof(struct rseq), 0, (sigset_t *)0p, _NSIG / 8);
515                int ret = syscall(__NR_rseq, &__cfaabi_rseq, sizeof(struct rseq), 0, RSEQ_SIG);
516                if(ret != 0) {
517                        int e = errno;
518                        switch(e) {
519                        case EINVAL: abort("KERNEL ERROR: rseq register invalid argument");
520                        case ENOSYS: abort("KERNEL ERROR: rseq register no supported");
521                        case EFAULT: abort("KERNEL ERROR: rseq register with invalid argument");
522                        case EBUSY : abort("KERNEL ERROR: rseq register already registered");
523                        case EPERM : abort("KERNEL ERROR: rseq register sig  argument  on unregistration does not match the signature received on registration");
524                        default: abort("KERNEL ERROR: rseq register unexpected return %d", e);
525                        }
526                }
527        }
528
529        static void __kernel_raw_rseq_unregister(void) {
530                /* paranoid */ verify( __cfaabi_rseq.cpu_id >= 0 );
531
532                // int ret = syscall(__NR_rseq, &__cfaabi_rseq, sizeof(struct rseq), RSEQ_FLAG_UNREGISTER, (sigset_t *)0p, _NSIG / 8);
533                int ret = syscall(__NR_rseq, &__cfaabi_rseq, sizeof(struct rseq), RSEQ_FLAG_UNREGISTER, RSEQ_SIG);
534                if(ret != 0) {
535                        int e = errno;
536                        switch(e) {
537                        case EINVAL: abort("KERNEL ERROR: rseq unregister invalid argument");
538                        case ENOSYS: abort("KERNEL ERROR: rseq unregister no supported");
539                        case EFAULT: abort("KERNEL ERROR: rseq unregister with invalid argument");
540                        case EBUSY : abort("KERNEL ERROR: rseq unregister already registered");
541                        case EPERM : abort("KERNEL ERROR: rseq unregister sig  argument  on unregistration does not match the signature received on registration");
542                        default: abort("KERNEL ERROR: rseq unregisteunexpected return %d", e);
543                        }
544                }
545        }
546#else
547        // No definition needed
548#endif
Note: See TracBrowser for help on using the repository browser.