Changes in / [730f4f1:70ac8d0]
- Files:
-
- 5 edited
-
benchmark/io/readv.cfa (modified) (3 diffs)
-
libcfa/src/concurrency/io.cfa (modified) (21 diffs)
-
libcfa/src/concurrency/kernel.cfa (modified) (1 diff)
-
libcfa/src/concurrency/kernel.hfa (modified) (2 diffs)
-
libcfa/src/concurrency/kernel_private.hfa (modified) (1 diff)
Legend:
- Unmodified
- Added
- Removed
-
benchmark/io/readv.cfa
r730f4f1 r70ac8d0 59 59 unsigned long int nthreads = 2; 60 60 unsigned long int nprocs = 1; 61 int flags = 0; 61 unsigned flags = 0; 62 unsigned sublen = 16; 62 63 63 64 arg_loop: 64 65 for(;;) { 65 66 static struct option options[] = { 66 {"duration", required_argument, 0, 'd'}, 67 {"nthreads", required_argument, 0, 't'}, 68 {"nprocs", required_argument, 0, 'p'}, 69 {"bufsize", required_argument, 0, 'b'}, 70 {"userthread", no_argument , 0, 'u'}, 67 {"duration", required_argument, 0, 'd'}, 68 {"nthreads", required_argument, 0, 't'}, 69 {"nprocs", required_argument, 0, 'p'}, 70 {"bufsize", required_argument, 0, 'b'}, 71 {"userthread", no_argument , 0, 'u'}, 72 {"submitthread", no_argument , 0, 's'}, 73 {"submitlength", required_argument, 0, 'l'}, 71 74 {0, 0, 0, 0} 72 75 }; 73 76 74 77 int idx = 0; 75 int opt = getopt_long(argc, argv, "d:t:p:b:u ", options, &idx);78 int opt = getopt_long(argc, argv, "d:t:p:b:usl:", options, &idx); 76 79 77 80 const char * arg = optarg ? optarg : ""; … … 113 116 flags |= CFA_CLUSTER_IO_POLLER_USER_THREAD; 114 117 break; 118 case 's': 119 flags |= CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS; 120 break; 121 case 'l': 122 sublen = strtoul(arg, &end, 10); 123 if(*end != '\0' && sublen < 16) { 124 fprintf(stderr, "Submit length must be at least 16, was %s\n", arg); 125 goto usage; 126 } 127 flags |= (sublen << CFA_CLUSTER_IO_BUFFLEN_OFFSET); 128 break; 115 129 // Other cases 116 130 default: /* ? */ … … 123 137 fprintf(stderr, " -p, --nprocs=NPROCS Number of kernel threads\n"); 124 138 fprintf(stderr, " -b, --buflen=SIZE Number of bytes to read per request\n"); 139 fprintf(stderr, " -u, --userthread If set, cluster uses user-thread to poll I/O\n"); 140 fprintf(stderr, " -s, --submitthread If set, cluster uses polling thread to submit I/O\n"); 125 141 exit(EXIT_FAILURE); 126 142 } -
libcfa/src/concurrency/io.cfa
r730f4f1 r70ac8d0 18 18 19 19 #include "kernel.hfa" 20 #include "bitmanip.hfa" 20 21 21 22 #if !defined(HAVE_LINUX_IO_URING_H) 22 void __kernel_io_startup( cluster &, int, bool ) {23 void __kernel_io_startup( cluster &, unsigned, bool ) { 23 24 // Nothing to do without io_uring 24 25 } … … 91 92 struct __io_poller_fast { 92 93 struct __io_data * ring; 93 bool waiting;94 94 $thread thrd; 95 95 }; … … 97 97 void ?{}( __io_poller_fast & this, struct cluster & cltr ) { 98 98 this.ring = cltr.io; 99 this.waiting = true;100 99 (this.thrd){ "Fast I/O Poller", cltr }; 101 100 } … … 126 125 // Like head/tail but not seen by the kernel 127 126 volatile uint32_t alloc; 128 volatile uint32_t ready; 127 volatile uint32_t * ready; 128 uint32_t ready_cnt; 129 129 130 130 __spinlock_t lock; … … 145 145 volatile unsigned long long int block; 146 146 } submit_avg; 147 struct { 148 volatile unsigned long long int val; 149 volatile unsigned long long int cnt; 150 volatile unsigned long long int block; 151 } look_avg; 147 152 } stats; 148 153 #endif … … 192 197 void * stack; 193 198 pthread_t kthrd; 199 volatile bool blocked; 194 200 } slow; 195 201 __io_poller_fast fast; … … 201 207 // I/O Startup / Shutdown logic 202 208 //============================================================================================= 203 void __kernel_io_startup( cluster & this, intio_flags, bool main_cluster ) {209 void __kernel_io_startup( cluster & this, unsigned io_flags, bool main_cluster ) { 204 210 this.io = malloc(); 205 211 … … 274 280 sq.array = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.array); 275 281 sq.alloc = *sq.tail; 276 sq.ready = *sq.tail; 282 283 if( io_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) { 284 /* paranoid */ verify( is_pow2( io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET ) || ((io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET) < 8) ); 285 sq.ready_cnt = max(io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET, 8); 286 sq.ready = alloc_align( 64, sq.ready_cnt ); 287 for(i; sq.ready_cnt) { 288 sq.ready[i] = -1ul32; 289 } 290 } 291 else { 292 sq.ready_cnt = 0; 293 sq.ready = 0p; 294 } 277 295 278 296 // completion queue … … 307 325 this.io->submit_q.stats.submit_avg.cnt = 0; 308 326 this.io->submit_q.stats.submit_avg.block = 0; 327 this.io->submit_q.stats.look_avg.val = 0; 328 this.io->submit_q.stats.look_avg.cnt = 0; 329 this.io->submit_q.stats.look_avg.block = 0; 309 330 this.io->completion_q.stats.completed_avg.val = 0; 310 331 this.io->completion_q.stats.completed_avg.slow_cnt = 0; … … 326 347 // Create the poller thread 327 348 __cfadbg_print_safe(io_core, "Kernel I/O : Creating slow poller for cluter %p\n", &this); 349 this.io->poller.slow.blocked = false; 328 350 this.io->poller.slow.stack = __create_pthread( &this.io->poller.slow.kthrd, __io_poller_slow, &this ); 329 351 } … … 347 369 if( this.io->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) { 348 370 with( this.io->poller.fast ) { 349 /* paranoid */ verify( waiting ); // The thread shouldn't be in a system call350 371 /* paranoid */ verify( this.procs.head == 0p || &this == mainCluster ); 351 372 /* paranoid */ verify( this.idles.head == 0p || &this == mainCluster ); 352 373 353 374 // We need to adjust the clean-up based on where the thread is 354 if( thrd. preempted != __NO_PREEMPTION ) {375 if( thrd.state == Ready || thrd.preempted != __NO_PREEMPTION ) { 355 376 356 377 // This is the tricky case 357 378 // The thread was preempted and now it is on the ready queue 358 /* paranoid */ verify( thrd.state == Active ); // The thread better be in this state359 379 /* paranoid */ verify( thrd.next == 1p ); // The thread should be the last on the list 360 380 /* paranoid */ verify( this.ready_queue.head == &thrd ); // The thread should be the only thing on the list … … 405 425 if(this.print_stats) { 406 426 with(this.io->submit_q.stats, this.io->completion_q.stats) { 427 double lavgv = 0; 428 double lavgb = 0; 429 if(look_avg.cnt != 0) { 430 lavgv = ((double)look_avg.val ) / look_avg.cnt; 431 lavgb = ((double)look_avg.block) / look_avg.cnt; 432 } 433 407 434 __cfaabi_bits_print_safe( STDERR_FILENO, 408 435 "----- I/O uRing Stats -----\n" 409 "- total submit calls : %'15llu\n" 410 "- avg submit : %'18.2lf\n" 411 "- pre-submit block %% : %'18.2lf\n" 412 "- total wait calls : %'15llu (%'llu slow, %'llu fast)\n" 413 "- avg completion/wait : %'18.2lf\n", 436 "- total submit calls : %'15llu\n" 437 "- avg submit : %'18.2lf\n" 438 "- pre-submit block %% : %'18.2lf\n" 439 "- total ready search : %'15llu\n" 440 "- avg ready search len : %'18.2lf\n" 441 "- avg ready search block : %'18.2lf\n" 442 "- total wait calls : %'15llu (%'llu slow, %'llu fast)\n" 443 "- avg completion/wait : %'18.2lf\n", 414 444 submit_avg.cnt, 415 445 ((double)submit_avg.val) / submit_avg.cnt, 416 446 (100.0 * submit_avg.block) / submit_avg.cnt, 447 look_avg.cnt, 448 lavgv, 449 lavgb, 417 450 completed_avg.slow_cnt + completed_avg.fast_cnt, 418 451 completed_avg.slow_cnt, completed_avg.fast_cnt, … … 441 474 close(this.io->fd); 442 475 476 free( this.io->submit_q.ready ); // Maybe null, doesn't matter 443 477 free( this.io ); 444 478 } … … 454 488 // Process a single completion message from the io_uring 455 489 // This is NOT thread-safe 456 static int __drain_io( struct __io_data & ring, sigset_t * mask, int waitcnt, bool in_kernel ) { 457 int ret = syscall( __NR_io_uring_enter, ring.fd, 0, waitcnt, IORING_ENTER_GETEVENTS, mask, _NSIG / 8); 490 static [int, bool] __drain_io( & struct __io_data ring, * sigset_t mask, int waitcnt, bool in_kernel ) { 491 unsigned to_submit = 0; 492 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) { 493 494 // If the poller thread also submits, then we need to aggregate the submissions which are ready 495 uint32_t * tail = ring.submit_q.tail; 496 const uint32_t mask = *ring.submit_q.mask; 497 498 // Go through the list of ready submissions 499 for( i; ring.submit_q.ready_cnt ) { 500 // replace any submission with the sentinel, to consume it. 501 uint32_t idx = __atomic_exchange_n( &ring.submit_q.ready[i], -1ul32, __ATOMIC_RELAXED); 502 503 // If it was already the sentinel, then we are done 504 if( idx == -1ul32 ) continue; 505 506 // If we got a real submission, append it to the list 507 ring.submit_q.array[ ((*tail) + to_submit) & mask ] = idx & mask; 508 to_submit++; 509 } 510 511 // Increment the tail based on how many we are ready to submit 512 __atomic_fetch_add(tail, to_submit, __ATOMIC_SEQ_CST); 513 514 // update statistics 515 #if !defined(__CFA_NO_STATISTICS__) 516 ring.submit_q.stats.submit_avg.val += to_submit; 517 ring.submit_q.stats.submit_avg.cnt += 1; 518 #endif 519 } 520 521 int ret = syscall( __NR_io_uring_enter, ring.fd, to_submit, waitcnt, IORING_ENTER_GETEVENTS, mask, _NSIG / 8); 458 522 if( ret < 0 ) { 459 523 switch((int)errno) { … … 497 561 __atomic_fetch_add( ring.completion_q.head, count, __ATOMIC_RELAXED ); 498 562 499 return count;563 return [count, count > 0 || to_submit > 0]; 500 564 } 501 565 … … 519 583 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) { 520 584 while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) { 585 586 __atomic_store_n( &ring.poller.slow.blocked, true, __ATOMIC_SEQ_CST ); 587 521 588 // In the user-thread approach drain and if anything was drained, 522 589 // batton pass to the user-thread 523 int count = __drain_io( ring, &mask, 1, true ); 590 int count; 591 bool again; 592 [count, again] = __drain_io( ring, &mask, 1, true ); 593 594 __atomic_store_n( &ring.poller.slow.blocked, false, __ATOMIC_SEQ_CST ); 524 595 525 596 // Update statistics … … 529 600 #endif 530 601 531 if( count > 0) {602 if(again) { 532 603 __cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to fast poller\n", &ring); 533 604 __unpark( &ring.poller.fast.thrd __cfaabi_dbg_ctx2 ); … … 539 610 while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) { 540 611 //In the naive approach, just poll the io completion queue directly 541 int count = __drain_io( ring, &mask, 1, true ); 612 int count; 613 bool again; 614 [count, again] = __drain_io( ring, &mask, 1, true ); 542 615 543 616 // Update statistics … … 566 639 // Then loop until we need to start 567 640 while(!__atomic_load_n(&this.ring->done, __ATOMIC_SEQ_CST)) { 641 568 642 // Drain the io 569 this.waiting = false; 570 int count = __drain_io( *this.ring, 0p, 0, false ); 571 reset += count > 0 ? 1 : 0; 643 int count; 644 bool again; 645 [count, again] = __drain_io( *this.ring, 0p, 0, false ); 646 647 if(!again) reset++; 572 648 573 649 // Update statistics … … 577 653 #endif 578 654 579 this.waiting = true;655 // If we got something, just yield and check again 580 656 if(reset < 5) { 581 // If we got something, just yield and check again582 657 yield(); 583 658 } 659 // We didn't get anything baton pass to the slow poller 584 660 else { 585 // We didn't get anything baton pass to the slow poller586 661 __cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to slow poller\n", &this.ring); 662 reset = 0; 663 664 // wake up the slow poller 587 665 post( this.ring->poller.sem ); 666 667 // park this thread 588 668 park( __cfaabi_dbg_ctx ); 589 reset = 0;590 669 } 591 670 } 592 671 593 672 __cfadbg_print_safe(io_core, "Kernel I/O : Fast poller for ring %p stopping\n", &this.ring); 673 } 674 675 static inline void __wake_poller( struct __io_data & ring ) __attribute__((artificial)); 676 static inline void __wake_poller( struct __io_data & ring ) { 677 if(!__atomic_load_n( &ring.poller.slow.blocked, __ATOMIC_SEQ_CST)) return; 678 679 sigval val = { 1 }; 680 pthread_sigqueue( ring.poller.slow.kthrd, SIGUSR1, val ); 594 681 } 595 682 … … 632 719 uint32_t idx = __atomic_fetch_add(&ring.submit_q.alloc, 1ul32, __ATOMIC_SEQ_CST); 633 720 634 // Validate that we didn't overflow anything 635 // Check that nothing overflowed 636 /* paranoid */ verify( true ); 637 638 // Check that it goes head -> tail -> alloc and never head -> alloc -> tail 639 /* paranoid */ verify( true ); 721 // Mask the idx now to allow make everything easier to check 722 idx &= *ring.submit_q.mask; 640 723 641 724 // Return the sqe 642 return [&ring.submit_q.sqes[ idx & (*ring.submit_q.mask)], idx];725 return [&ring.submit_q.sqes[ idx ], idx]; 643 726 } 644 727 645 728 static inline void __submit( struct __io_data & ring, uint32_t idx ) { 646 // get mutual exclusion 647 lock(ring.submit_q.lock __cfaabi_dbg_ctx2); 648 649 // Append to the list of ready entries 650 uint32_t * tail = ring.submit_q.tail; 729 // Get now the data we definetely need 730 uint32_t * const tail = ring.submit_q.tail; 651 731 const uint32_t mask = *ring.submit_q.mask; 652 732 653 ring.submit_q.array[ (*tail) & mask ] = idx & mask; 654 __atomic_fetch_add(tail, 1ul32, __ATOMIC_SEQ_CST); 655 656 // Submit however, many entries need to be submitted 657 int ret = syscall( __NR_io_uring_enter, ring.fd, 1, 0, 0, 0p, 0); 658 if( ret < 0 ) { 659 switch((int)errno) { 660 default: 661 abort( "KERNEL ERROR: IO_URING SUBMIT - %s\n", strerror(errno) ); 662 } 663 } 664 665 // update statistics 666 #if !defined(__CFA_NO_STATISTICS__) 667 ring.submit_q.stats.submit_avg.val += 1; 668 ring.submit_q.stats.submit_avg.cnt += 1; 669 #endif 670 671 unlock(ring.submit_q.lock); 672 // Make sure that idx was submitted 673 // Be careful to not get false positive if we cycled the entire list or that someone else submitted for us 674 __cfadbg_print_safe( io, "Kernel I/O : Performed io_submit for %p, returned %d\n", active_thread(), ret ); 733 // There are 2 submission schemes, check which one we are using 734 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) { 735 // If the poller thread submits, then we just need to add this to the ready array 736 737 /* paranoid */ verify( idx <= mask ); 738 /* paranoid */ verify( idx != -1ul32 ); 739 740 // We need to find a spot in the ready array 741 __attribute((unused)) int len = 0; 742 __attribute((unused)) int block = 0; 743 uint32_t expected = -1ul32; 744 uint32_t ready_mask = ring.submit_q.ready_cnt - 1; 745 uint32_t off = __tls_rand(); 746 LOOKING: for() { 747 for(i; ring.submit_q.ready_cnt) { 748 uint32_t ii = (i + off) & ready_mask; 749 if( __atomic_compare_exchange_n( &ring.submit_q.ready[ii], &expected, idx, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) ) { 750 break LOOKING; 751 } 752 753 len ++; 754 } 755 756 block++; 757 yield(); 758 } 759 760 __wake_poller( ring ); 761 762 // update statistics 763 #if !defined(__CFA_NO_STATISTICS__) 764 __atomic_fetch_add( &ring.submit_q.stats.look_avg.val, len, __ATOMIC_RELAXED ); 765 __atomic_fetch_add( &ring.submit_q.stats.look_avg.block, block, __ATOMIC_RELAXED ); 766 __atomic_fetch_add( &ring.submit_q.stats.look_avg.cnt, 1, __ATOMIC_RELAXED ); 767 #endif 768 769 __cfadbg_print_safe( io, "Kernel I/O : Added %u to ready for %p\n", idx, active_thread() ); 770 } 771 else { 772 // get mutual exclusion 773 lock(ring.submit_q.lock __cfaabi_dbg_ctx2); 774 775 // Append to the list of ready entries 776 777 /* paranoid */ verify( idx <= mask ); 778 779 ring.submit_q.array[ (*tail) & mask ] = idx & mask; 780 __atomic_fetch_add(tail, 1ul32, __ATOMIC_SEQ_CST); 781 782 // Submit however, many entries need to be submitted 783 int ret = syscall( __NR_io_uring_enter, ring.fd, 1, 0, 0, 0p, 0); 784 if( ret < 0 ) { 785 switch((int)errno) { 786 default: 787 abort( "KERNEL ERROR: IO_URING SUBMIT - %s\n", strerror(errno) ); 788 } 789 } 790 791 // update statistics 792 #if !defined(__CFA_NO_STATISTICS__) 793 ring.submit_q.stats.submit_avg.val += 1; 794 ring.submit_q.stats.submit_avg.cnt += 1; 795 #endif 796 797 unlock(ring.submit_q.lock); 798 799 __cfadbg_print_safe( io, "Kernel I/O : Performed io_submit for %p, returned %d\n", active_thread(), ret ); 800 } 675 801 } 676 802 -
libcfa/src/concurrency/kernel.cfa
r730f4f1 r70ac8d0 256 256 } 257 257 258 void ?{}(cluster & this, const char name[], Duration preemption_rate, intio_flags) with( this ) {258 void ?{}(cluster & this, const char name[], Duration preemption_rate, unsigned io_flags) with( this ) { 259 259 this.name = name; 260 260 this.preemption_rate = preemption_rate; -
libcfa/src/concurrency/kernel.hfa
r730f4f1 r70ac8d0 116 116 struct __io_data; 117 117 118 #define CFA_CLUSTER_IO_POLLER_USER_THREAD 1 << 0 119 // #define CFA_CLUSTER_IO_POLLER_KERNEL_SIDE 1 << 1 118 #define CFA_CLUSTER_IO_POLLER_USER_THREAD 1 << 0 // 0x1 119 #define CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS 1 << 1 // 0x2 120 // #define CFA_CLUSTER_IO_POLLER_KERNEL_SIDE 1 << 2 // 0x4 121 #define CFA_CLUSTER_IO_BUFFLEN_OFFSET 16 120 122 121 123 //----------------------------------------------------------------------------- … … 159 161 extern Duration default_preemption(); 160 162 161 void ?{} (cluster & this, const char name[], Duration preemption_rate, intflags);163 void ?{} (cluster & this, const char name[], Duration preemption_rate, unsigned flags); 162 164 void ^?{}(cluster & this); 163 165 164 static inline void ?{} (cluster & this) { this{"Anonymous Cluster", default_preemption(), 0}; }165 static inline void ?{} (cluster & this, Duration preemption_rate) { this{"Anonymous Cluster", preemption_rate, 0}; }166 static inline void ?{} (cluster & this, const char name[]) { this{name, default_preemption(), 0}; }167 static inline void ?{} (cluster & this, intflags) { this{"Anonymous Cluster", default_preemption(), flags}; }168 static inline void ?{} (cluster & this, Duration preemption_rate, intflags) { this{"Anonymous Cluster", preemption_rate, flags}; }169 static inline void ?{} (cluster & this, const char name[], intflags) { this{name, default_preemption(), flags}; }166 static inline void ?{} (cluster & this) { this{"Anonymous Cluster", default_preemption(), 0}; } 167 static inline void ?{} (cluster & this, Duration preemption_rate) { this{"Anonymous Cluster", preemption_rate, 0}; } 168 static inline void ?{} (cluster & this, const char name[]) { this{name, default_preemption(), 0}; } 169 static inline void ?{} (cluster & this, unsigned flags) { this{"Anonymous Cluster", default_preemption(), flags}; } 170 static inline void ?{} (cluster & this, Duration preemption_rate, unsigned flags) { this{"Anonymous Cluster", preemption_rate, flags}; } 171 static inline void ?{} (cluster & this, const char name[], unsigned flags) { this{name, default_preemption(), flags}; } 170 172 171 173 static inline [cluster *&, cluster *& ] __get( cluster & this ) __attribute__((const)) { return this.node.[next, prev]; } -
libcfa/src/concurrency/kernel_private.hfa
r730f4f1 r70ac8d0 77 77 //----------------------------------------------------------------------------- 78 78 // I/O 79 void __kernel_io_startup ( cluster &, int, bool );79 void __kernel_io_startup ( cluster &, unsigned, bool ); 80 80 void __kernel_io_finish_start( cluster & ); 81 81 void __kernel_io_prepare_stop( cluster & );
Note:
See TracChangeset
for help on using the changeset viewer.
