source: libcfa/src/concurrency/io.cfa@ 9f350c4

ADT arm-eh ast-experimental enum forall-pointer-decay jacob/cs343-translation new-ast new-ast-unique-expr pthread-emulation qualifiedEnum
Last change on this file since 9f350c4 was 4ec028d, checked in by Thierry Delisle <tdelisle@…>, 5 years ago

Merge branch 'master' of plg.uwaterloo.ca:software/cfa/cfa-cc
  • Property mode set to 100644
File size: 26.1 KB
Line 
1//
2// Cforall Version 1.0.0 Copyright (C) 2020 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// io.cfa --
8//
9// Author : Thierry Delisle
10// Created On : Thu Apr 23 17:31:00 2020
11// Last Modified By :
12// Last Modified On :
13// Update Count :
14//
15
16// #define __CFA_DEBUG_PRINT_IO__
17// #define __CFA_DEBUG_PRINT_IO_CORE__
18
19#include "kernel.hfa"
20#include "bitmanip.hfa"
21
22#if !defined(HAVE_LINUX_IO_URING_H)
23 void __kernel_io_startup( cluster &, unsigned, bool ) {
24 // Nothing to do without io_uring
25 }
26
27 void __kernel_io_finish_start( cluster & ) {
28 // Nothing to do without io_uring
29 }
30
31 void __kernel_io_prepare_stop( cluster & ) {
32 // Nothing to do without io_uring
33 }
34
35 void __kernel_io_shutdown( cluster &, bool ) {
36 // Nothing to do without io_uring
37 }
38
39#else
40 #define _GNU_SOURCE /* See feature_test_macros(7) */
41 #include <errno.h>
42 #include <stdint.h>
43 #include <string.h>
44 #include <unistd.h>
45 #include <sys/mman.h>
46
47 extern "C" {
48 #include <sys/syscall.h>
49
50 #include <linux/io_uring.h>
51 }
52
53 #include "bits/signal.hfa"
54 #include "kernel_private.hfa"
55 #include "thread.hfa"
56
57 uint32_t entries_per_cluster() {
58 return 256;
59 }
60
61 static void * __io_poller_slow( void * arg );
62
63 // Weirdly, some systems that do support io_uring don't actually define these
64 #ifdef __alpha__
65 /*
66 * alpha is the only exception, all other architectures
67 * have common numbers for new system calls.
68 */
69 #ifndef __NR_io_uring_setup
70 #define __NR_io_uring_setup 535
71 #endif
72 #ifndef __NR_io_uring_enter
73 #define __NR_io_uring_enter 536
74 #endif
75 #ifndef __NR_io_uring_register
76 #define __NR_io_uring_register 537
77 #endif
78 #else /* !__alpha__ */
79 #ifndef __NR_io_uring_setup
80 #define __NR_io_uring_setup 425
81 #endif
82 #ifndef __NR_io_uring_enter
83 #define __NR_io_uring_enter 426
84 #endif
85 #ifndef __NR_io_uring_register
86 #define __NR_io_uring_register 427
87 #endif
88 #endif
89
90 // Fast poller user-thread
91 // Not using the "thread" keyword because we want to control
92 // more carefully when to start/stop it
93 struct __io_poller_fast {
94 struct __io_data * ring;
95 $thread thrd;
96 };
97
98 void ?{}( __io_poller_fast & this, struct cluster & cltr ) {
99 this.ring = cltr.io;
100 (this.thrd){ "Fast I/O Poller", cltr };
101 }
102 void ^?{}( __io_poller_fast & mutex this );
103 void main( __io_poller_fast & this );
104 static inline $thread * get_thread( __io_poller_fast & this ) { return &this.thrd; }
105 void ^?{}( __io_poller_fast & mutex this ) {}
106
107 struct __submition_data {
108 // Head and tail of the ring (associated with array)
109 volatile uint32_t * head;
110 volatile uint32_t * tail;
111
112 // The actual kernel ring which uses head/tail
113 // indexes into the sqes arrays
114 uint32_t * array;
115
116 // number of entries and mask to go with it
117 const uint32_t * num;
118 const uint32_t * mask;
119
120 // Submission flags (Not sure what for)
121 uint32_t * flags;
122
123 // number of sqes not submitted (whatever that means)
124 uint32_t * dropped;
125
126 // Like head/tail but not seen by the kernel
127 volatile uint32_t * ready;
128 uint32_t ready_cnt;
129
130 __spinlock_t lock;
131
132 // A buffer of sqes (not the actual ring)
133 struct io_uring_sqe * sqes;
134
135 // The location and size of the mmaped area
136 void * ring_ptr;
137 size_t ring_sz;
138 };
139
140 struct __completion_data {
141 // Head and tail of the ring
142 volatile uint32_t * head;
143 volatile uint32_t * tail;
144
145 // number of entries and mask to go with it
146 const uint32_t * mask;
147 const uint32_t * num;
148
149 // number of cqes not submitted (whatever that means)
150 uint32_t * overflow;
151
152 // the kernel ring
153 struct io_uring_cqe * cqes;
154
155 // The location and size of the mmaped area
156 void * ring_ptr;
157 size_t ring_sz;
158 };
159
160 struct __io_data {
161 struct __submition_data submit_q;
162 struct __completion_data completion_q;
163 uint32_t ring_flags;
164 int cltr_flags;
165 int fd;
166 semaphore submit;
167 volatile bool done;
168 struct {
169 struct {
170 __processor_id_t id;
171 void * stack;
172 pthread_t kthrd;
173 volatile bool blocked;
174 } slow;
175 __io_poller_fast fast;
176 __bin_sem_t sem;
177 } poller;
178 };
179
180//=============================================================================================
181// I/O Startup / Shutdown logic
182//=============================================================================================
183 void __kernel_io_startup( cluster & this, unsigned io_flags, bool main_cluster ) {
184 this.io = malloc();
185
186 // Step 1 : call to setup
187 struct io_uring_params params;
188 memset(&params, 0, sizeof(params));
189
190 uint32_t nentries = entries_per_cluster();
191
192 int fd = syscall(__NR_io_uring_setup, nentries, &params );
193 if(fd < 0) {
194 abort("KERNEL ERROR: IO_URING SETUP - %s\n", strerror(errno));
195 }
196
197 // Step 2 : mmap result
198 memset( this.io, 0, sizeof(struct __io_data) );
199 struct __submition_data & sq = this.io->submit_q;
200 struct __completion_data & cq = this.io->completion_q;
201
202 // calculate the right ring size
203 sq.ring_sz = params.sq_off.array + (params.sq_entries * sizeof(unsigned) );
204 cq.ring_sz = params.cq_off.cqes + (params.cq_entries * sizeof(struct io_uring_cqe));
205
206 // Requires features
207 #if defined(IORING_FEAT_SINGLE_MMAP)
208 // adjust the size according to the parameters
209 if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
210 cq->ring_sz = sq->ring_sz = max(cq->ring_sz, sq->ring_sz);
211 }
212 #endif
213
214 // mmap the Submit Queue into existence
215 sq.ring_ptr = mmap(0, sq.ring_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_SQ_RING);
216 if (sq.ring_ptr == (void*)MAP_FAILED) {
217 abort("KERNEL ERROR: IO_URING MMAP1 - %s\n", strerror(errno));
218 }
219
220 // Requires features
221 #if defined(IORING_FEAT_SINGLE_MMAP)
222 // mmap the Completion Queue into existence (may or may not be needed)
223 if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
224 cq->ring_ptr = sq->ring_ptr;
225 }
226 else
227 #endif
228 {
229 // We need multiple call to MMAP
230 cq.ring_ptr = mmap(0, cq.ring_sz, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_CQ_RING);
231 if (cq.ring_ptr == (void*)MAP_FAILED) {
232 munmap(sq.ring_ptr, sq.ring_sz);
233 abort("KERNEL ERROR: IO_URING MMAP2 - %s\n", strerror(errno));
234 }
235 }
236
237 // mmap the submit queue entries
238 size_t size = params.sq_entries * sizeof(struct io_uring_sqe);
239 sq.sqes = (struct io_uring_sqe *)mmap(0, size, PROT_READ | PROT_WRITE, MAP_SHARED | MAP_POPULATE, fd, IORING_OFF_SQES);
240 if (sq.sqes == (struct io_uring_sqe *)MAP_FAILED) {
241 munmap(sq.ring_ptr, sq.ring_sz);
242 if (cq.ring_ptr != sq.ring_ptr) munmap(cq.ring_ptr, cq.ring_sz);
243 abort("KERNEL ERROR: IO_URING MMAP3 - %s\n", strerror(errno));
244 }
245
246 // Get the pointers from the kernel to fill the structure
247 // submit queue
248 sq.head = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.head);
249 sq.tail = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail);
250 sq.mask = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask);
251 sq.num = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries);
252 sq.flags = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags);
253 sq.dropped = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped);
254 sq.array = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.array);
255
256 {
257 const uint32_t num = *sq.num;
258 for( i; num ) {
259 sq.sqes[i].user_data = 0ul64;
260 }
261 }
262
263 if( io_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) {
264 /* paranoid */ verify( is_pow2( io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET ) || ((io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET) < 8) );
265 sq.ready_cnt = max(io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET, 8);
266 sq.ready = alloc_align( 64, sq.ready_cnt );
267 for(i; sq.ready_cnt) {
268 sq.ready[i] = -1ul32;
269 }
270 }
271 else {
272 sq.ready_cnt = 0;
273 sq.ready = 0p;
274 }
275
276 // completion queue
277 cq.head = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.head);
278 cq.tail = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail);
279 cq.mask = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask);
280 cq.num = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries);
281 cq.overflow = ( uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow);
282 cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes);
283
284 // some paranoid checks
285 /* paranoid */ verifyf( (*cq.mask) == ((*cq.num) - 1ul32), "IO_URING Expected mask to be %u (%u entries), was %u", (*cq.num) - 1ul32, *cq.num, *cq.mask );
286 /* paranoid */ verifyf( (*cq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *cq.num );
287 /* paranoid */ verifyf( (*cq.head) == 0, "IO_URING Expected head to be 0, got %u", *cq.head );
288 /* paranoid */ verifyf( (*cq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *cq.tail );
289
290 /* paranoid */ verifyf( (*sq.mask) == ((*sq.num) - 1ul32), "IO_URING Expected mask to be %u (%u entries), was %u", (*sq.num) - 1ul32, *sq.num, *sq.mask );
291 /* paranoid */ verifyf( (*sq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *sq.num );
292 /* paranoid */ verifyf( (*sq.head) == 0, "IO_URING Expected head to be 0, got %u", *sq.head );
293 /* paranoid */ verifyf( (*sq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *sq.tail );
294
295 // Update the global ring info
296 this.io->ring_flags = params.flags;
297 this.io->cltr_flags = io_flags;
298 this.io->fd = fd;
299 this.io->done = false;
300 (this.io->submit){ min(*sq.num, *cq.num) };
301
302 if(!main_cluster) {
303 __kernel_io_finish_start( this );
304 }
305 }
306
307 void __kernel_io_finish_start( cluster & this ) {
308 if( this.io->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
309 __cfadbg_print_safe(io_core, "Kernel I/O : Creating fast poller for cluter %p\n", &this);
310 (this.io->poller.fast){ this };
311 __thrd_start( this.io->poller.fast, main );
312 }
313
314 // Create the poller thread
315 __cfadbg_print_safe(io_core, "Kernel I/O : Creating slow poller for cluter %p\n", &this);
316 this.io->poller.slow.blocked = false;
317 this.io->poller.slow.stack = __create_pthread( &this.io->poller.slow.kthrd, __io_poller_slow, &this );
318 }
319
320 void __kernel_io_prepare_stop( cluster & this ) {
321 __cfadbg_print_safe(io_core, "Kernel I/O : Stopping pollers for cluster\n", &this);
322 // Notify the poller thread of the shutdown
323 __atomic_store_n(&this.io->done, true, __ATOMIC_SEQ_CST);
324
325 // Stop the IO Poller
326 sigval val = { 1 };
327 pthread_sigqueue( this.io->poller.slow.kthrd, SIGUSR1, val );
328 post( this.io->poller.sem );
329
330 // Wait for the poller thread to finish
331 pthread_join( this.io->poller.slow.kthrd, 0p );
332 free( this.io->poller.slow.stack );
333
334 __cfadbg_print_safe(io_core, "Kernel I/O : Slow poller stopped for cluster\n", &this);
335
336 if( this.io->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
337 with( this.io->poller.fast ) {
338 /* paranoid */ verify( this.nprocessors == 0 || &this == mainCluster );
339 /* paranoid */ verify( !ready_mutate_islocked() );
340
341 // We need to adjust the clean-up based on where the thread is
342 if( thrd.state == Ready || thrd.preempted != __NO_PREEMPTION ) {
343
344 ready_schedule_lock( (struct __processor_id_t *)active_processor() );
345
346 // This is the tricky case
347 // The thread was preempted and now it is on the ready queue
348 // The thread should be the last on the list
349 /* paranoid */ verify( thrd.link.next != 0p );
350
351 // Remove the thread from the ready queue of this cluster
352 __attribute__((unused)) bool removed = remove_head( &this, &thrd );
353 /* paranoid */ verify( removed );
354 thrd.link.next = 0p;
355 thrd.link.prev = 0p;
356 __cfaabi_dbg_debug_do( thrd.unpark_stale = true );
357
358 // Fixup the thread state
359 thrd.state = Blocked;
360 thrd.ticket = 0;
361 thrd.preempted = __NO_PREEMPTION;
362
363 ready_schedule_unlock( (struct __processor_id_t *)active_processor() );
364
365 // Pretend like the thread was blocked all along
366 }
367 // !!! This is not an else if !!!
368 if( thrd.state == Blocked ) {
369
370 // This is the "easy case"
371 // The thread is parked and can easily be moved to active cluster
372 verify( thrd.curr_cluster != active_cluster() || thrd.curr_cluster == mainCluster );
373 thrd.curr_cluster = active_cluster();
374
375 // unpark the fast io_poller
376 unpark( &thrd __cfaabi_dbg_ctx2 );
377 }
378 else {
379
380 // The thread is in a weird state
381 // I don't know what to do here
382 abort("Fast poller thread is in unexpected state, cannot clean-up correctly\n");
383 }
384
385 }
386
387 ^(this.io->poller.fast){};
388
389 __cfadbg_print_safe(io_core, "Kernel I/O : Fast poller stopped for cluster\n", &this);
390 }
391 }
392
393 void __kernel_io_shutdown( cluster & this, bool main_cluster ) {
394 if(!main_cluster) {
395 __kernel_io_prepare_stop( this );
396 }
397
398 // Shutdown the io rings
399 struct __submition_data & sq = this.io->submit_q;
400 struct __completion_data & cq = this.io->completion_q;
401
402 // unmap the submit queue entries
403 munmap(sq.sqes, (*sq.num) * sizeof(struct io_uring_sqe));
404
405 // unmap the Submit Queue ring
406 munmap(sq.ring_ptr, sq.ring_sz);
407
408 // unmap the Completion Queue ring, if it is different
409 if (cq.ring_ptr != sq.ring_ptr) {
410 munmap(cq.ring_ptr, cq.ring_sz);
411 }
412
413 // close the file descriptor
414 close(this.io->fd);
415
416 free( this.io->submit_q.ready ); // Maybe null, doesn't matter
417 free( this.io );
418 }
419
420//=============================================================================================
421// I/O Polling
422//=============================================================================================
423 // Process a single completion message from the io_uring
424 // This is NOT thread-safe
425 static [int, bool] __drain_io( & struct __io_data ring, * sigset_t mask, int waitcnt, bool in_kernel ) {
426 unsigned to_submit = 0;
427 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) {
428
429 // If the poller thread also submits, then we need to aggregate the submissions which are ready
430 uint32_t tail = *ring.submit_q.tail;
431 const uint32_t mask = *ring.submit_q.mask;
432
433 // Go through the list of ready submissions
434 for( i; ring.submit_q.ready_cnt ) {
435 // replace any submission with the sentinel, to consume it.
436 uint32_t idx = __atomic_exchange_n( &ring.submit_q.ready[i], -1ul32, __ATOMIC_RELAXED);
437
438 // If it was already the sentinel, then we are done
439 if( idx == -1ul32 ) continue;
440
441 // If we got a real submission, append it to the list
442 ring.submit_q.array[ (tail + to_submit) & mask ] = idx & mask;
443 to_submit++;
444 }
445
446 // Increment the tail based on how many we are ready to submit
447 __atomic_fetch_add(ring.submit_q.tail, to_submit, __ATOMIC_SEQ_CST);
448 }
449
450 const uint32_t smask = *ring.submit_q.mask;
451 uint32_t shead = *ring.submit_q.head;
452 int ret = syscall( __NR_io_uring_enter, ring.fd, to_submit, waitcnt, IORING_ENTER_GETEVENTS, mask, _NSIG / 8);
453 if( ret < 0 ) {
454 switch((int)errno) {
455 case EAGAIN:
456 case EINTR:
457 return -EAGAIN;
458 default:
459 abort( "KERNEL ERROR: IO_URING WAIT - %s\n", strerror(errno) );
460 }
461 }
462
463 // Release the consumed SQEs
464 for( i; ret ) {
465 uint32_t idx = ring.submit_q.array[ (i + shead) & smask ];
466 ring.submit_q.sqes[ idx ].user_data = 0;
467 }
468
469 uint32_t avail = 0;
470 uint32_t sqe_num = *ring.submit_q.num;
471 for(i; sqe_num) {
472 if( ring.submit_q.sqes[ i ].user_data == 0 ) avail++;
473 }
474
475 // update statistics
476 #if !defined(__CFA_NO_STATISTICS__)
477 __tls_stats()->io.submit_q.submit_avg.rdy += to_submit;
478 __tls_stats()->io.submit_q.submit_avg.csm += ret;
479 __tls_stats()->io.submit_q.submit_avg.avl += avail;
480 __tls_stats()->io.submit_q.submit_avg.cnt += 1;
481 #endif
482
483 // Drain the queue
484 unsigned head = *ring.completion_q.head;
485 unsigned tail = *ring.completion_q.tail;
486 const uint32_t mask = *ring.completion_q.mask;
487
488 // Memory barrier
489 __atomic_thread_fence( __ATOMIC_SEQ_CST );
490
491 // Nothing was new return 0
492 if (head == tail) {
493 return 0;
494 }
495
496 uint32_t count = tail - head;
497 for(i; count) {
498 unsigned idx = (head + i) & mask;
499 struct io_uring_cqe & cqe = ring.completion_q.cqes[idx];
500
501 /* paranoid */ verify(&cqe);
502
503 struct __io_user_data_t * data = (struct __io_user_data_t *)(uintptr_t)cqe.user_data;
504 __cfadbg_print_safe( io, "Kernel I/O : Performed reading io cqe %p, result %d for %p\n", data, cqe.res, data->thrd );
505
506 data->result = cqe.res;
507 if(!in_kernel) { unpark( data->thrd __cfaabi_dbg_ctx2 ); }
508 else { __unpark( &ring.poller.slow.id, data->thrd __cfaabi_dbg_ctx2 ); }
509 }
510
511 // Allow new submissions to happen
512 // V(ring.submit, count);
513
514 // Mark to the kernel that the cqe has been seen
515 // Ensure that the kernel only sees the new value of the head index after the CQEs have been read.
516 __atomic_thread_fence( __ATOMIC_SEQ_CST );
517 __atomic_fetch_add( ring.completion_q.head, count, __ATOMIC_RELAXED );
518
519 return [count, count > 0 || to_submit > 0];
520 }
521
522 static void * __io_poller_slow( void * arg ) {
523 #if !defined( __CFA_NO_STATISTICS__ )
524 __stats_t local_stats;
525 __init_stats( &local_stats );
526 kernelTLS.this_stats = &local_stats;
527 #endif
528
529 cluster * cltr = (cluster *)arg;
530 struct __io_data & ring = *cltr->io;
531
532 ring.poller.slow.id.id = doregister( &ring.poller.slow.id );
533
534 sigset_t mask;
535 sigfillset(&mask);
536 if ( pthread_sigmask( SIG_BLOCK, &mask, 0p ) == -1 ) {
537 abort( "KERNEL ERROR: IO_URING - pthread_sigmask" );
538 }
539
540 sigdelset( &mask, SIGUSR1 );
541
542 verify( (*ring.submit_q.head) == (*ring.submit_q.tail) );
543 verify( (*ring.completion_q.head) == (*ring.completion_q.tail) );
544
545 __cfadbg_print_safe(io_core, "Kernel I/O : Slow poller for ring %p ready\n", &ring);
546
547 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
548 while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) {
549
550 __atomic_store_n( &ring.poller.slow.blocked, true, __ATOMIC_SEQ_CST );
551
552 // In the user-thread approach drain and if anything was drained,
553 // batton pass to the user-thread
554 int count;
555 bool again;
556 [count, again] = __drain_io( ring, &mask, 1, true );
557
558 __atomic_store_n( &ring.poller.slow.blocked, false, __ATOMIC_SEQ_CST );
559
560 // Update statistics
561 #if !defined(__CFA_NO_STATISTICS__)
562 __tls_stats()->io.complete_q.completed_avg.val += count;
563 __tls_stats()->io.complete_q.completed_avg.slow_cnt += 1;
564 #endif
565
566 if(again) {
567 __cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to fast poller\n", &ring);
568 __unpark( &ring.poller.slow.id, &ring.poller.fast.thrd __cfaabi_dbg_ctx2 );
569 wait( ring.poller.sem );
570 }
571 }
572 }
573 else {
574 while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) {
575 //In the naive approach, just poll the io completion queue directly
576 int count;
577 bool again;
578 [count, again] = __drain_io( ring, &mask, 0, true );
579
580 // Update statistics
581 #if !defined(__CFA_NO_STATISTICS__)
582 __tls_stats()->io.complete_q.completed_avg.val += count;
583 __tls_stats()->io.complete_q.completed_avg.slow_cnt += 1;
584 #endif
585 }
586 }
587
588 __cfadbg_print_safe(io_core, "Kernel I/O : Slow poller for ring %p stopping\n", &ring);
589
590 unregister( &ring.poller.slow.id );
591
592 #if !defined(__CFA_NO_STATISTICS__)
593 __tally_stats(cltr->stats, &local_stats);
594 #endif
595
596 return 0p;
597 }
598
599 void main( __io_poller_fast & this ) {
600 verify( this.ring->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD );
601
602 // Start parked
603 park( __cfaabi_dbg_ctx );
604
605 __cfadbg_print_safe(io_core, "Kernel I/O : Fast poller for ring %p ready\n", &this.ring);
606
607 int reset = 0;
608
609 // Then loop until we need to start
610 while(!__atomic_load_n(&this.ring->done, __ATOMIC_SEQ_CST)) {
611
612 // Drain the io
613 int count;
614 bool again;
615 disable_interrupts();
616 [count, again] = __drain_io( *this.ring, 0p, 0, false );
617
618 if(!again) reset++;
619
620 // Update statistics
621 #if !defined(__CFA_NO_STATISTICS__)
622 __tls_stats()->io.complete_q.completed_avg.val += count;
623 __tls_stats()->io.complete_q.completed_avg.fast_cnt += 1;
624 #endif
625 enable_interrupts( __cfaabi_dbg_ctx );
626
627 // If we got something, just yield and check again
628 if(reset < 5) {
629 yield();
630 }
631 // We didn't get anything baton pass to the slow poller
632 else {
633 __cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to slow poller\n", &this.ring);
634 reset = 0;
635
636 // wake up the slow poller
637 post( this.ring->poller.sem );
638
639 // park this thread
640 park( __cfaabi_dbg_ctx );
641 }
642 }
643
644 __cfadbg_print_safe(io_core, "Kernel I/O : Fast poller for ring %p stopping\n", &this.ring);
645 }
646
647 static inline void __wake_poller( struct __io_data & ring ) __attribute__((artificial));
648 static inline void __wake_poller( struct __io_data & ring ) {
649 if(!__atomic_load_n( &ring.poller.slow.blocked, __ATOMIC_SEQ_CST)) return;
650
651 sigval val = { 1 };
652 pthread_sigqueue( ring.poller.slow.kthrd, SIGUSR1, val );
653 }
654
655//=============================================================================================
656// I/O Submissions
657//=============================================================================================
658
659// Submition steps :
660// 1 - We need to make sure we don't overflow any of the buffer, P(ring.submit) to make sure
661// entries are available. The semaphore make sure that there is no more operations in
662// progress then the number of entries in the buffer. This probably limits concurrency
663// more than necessary since submitted but not completed operations don't need any
664// entries in user space. However, I don't know what happens if we overflow the buffers
665// because too many requests completed at once. This is a safe approach in all cases.
666// Furthermore, with hundreds of entries, this may be okay.
667//
668// 2 - Allocate a queue entry. The ring already has memory for all entries but only the ones
669// listed in sq.array are visible by the kernel. For those not listed, the kernel does not
670// offer any assurance that an entry is not being filled by multiple flags. Therefore, we
671// need to write an allocator that allows allocating concurrently.
672//
673// 3 - Actually fill the submit entry, this is the only simple and straightforward step.
674//
675// 4 - Append the entry index to the array and adjust the tail accordingly. This operation
676// needs to arrive to two concensus at the same time:
677// A - The order in which entries are listed in the array: no two threads must pick the
678// same index for their entries
679// B - When can the tail be update for the kernel. EVERY entries in the array between
680// head and tail must be fully filled and shouldn't ever be touched again.
681//
682
683 [* struct io_uring_sqe, uint32_t] __submit_alloc( struct __io_data & ring, uint64_t data ) {
684 verify( data != 0 );
685
686
687 // Prepare the data we need
688 __attribute((unused)) int len = 0;
689 __attribute((unused)) int block = 0;
690 uint32_t cnt = *ring.submit_q.num;
691 uint32_t mask = *ring.submit_q.mask;
692
693 disable_interrupts();
694 uint32_t off = __tls_rand();
695 enable_interrupts( __cfaabi_dbg_ctx );
696
697 // Loop around looking for an available spot
698 for() {
699 // Look through the list starting at some offset
700 for(i; cnt) {
701 uint64_t expected = 0;
702 uint32_t idx = (i + off) & mask;
703 struct io_uring_sqe * sqe = &ring.submit_q.sqes[idx];
704 volatile uint64_t * udata = &sqe->user_data;
705
706 if( *udata == expected &&
707 __atomic_compare_exchange_n( udata, &expected, data, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) )
708 {
709 // update statistics
710 #if !defined(__CFA_NO_STATISTICS__)
711 disable_interrupts();
712 __tls_stats()->io.submit_q.alloc_avg.val += len;
713 __tls_stats()->io.submit_q.alloc_avg.block += block;
714 __tls_stats()->io.submit_q.alloc_avg.cnt += 1;
715 enable_interrupts( __cfaabi_dbg_ctx );
716 #endif
717
718
719 // Success return the data
720 return [sqe, idx];
721 }
722 verify(expected != data);
723
724 len ++;
725 }
726
727 block++;
728 yield();
729 }
730 }
731
732 static inline uint32_t __submit_to_ready_array( struct __io_data & ring, uint32_t idx, const uint32_t mask ) {
733 /* paranoid */ verify( idx <= mask );
734 /* paranoid */ verify( idx != -1ul32 );
735
736 // We need to find a spot in the ready array
737 __attribute((unused)) int len = 0;
738 __attribute((unused)) int block = 0;
739 uint32_t ready_mask = ring.submit_q.ready_cnt - 1;
740
741 disable_interrupts();
742 uint32_t off = __tls_rand();
743 enable_interrupts( __cfaabi_dbg_ctx );
744
745 uint32_t picked;
746 LOOKING: for() {
747 for(i; ring.submit_q.ready_cnt) {
748 picked = (i + off) & ready_mask;
749 uint32_t expected = -1ul32;
750 if( __atomic_compare_exchange_n( &ring.submit_q.ready[picked], &expected, idx, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) ) {
751 break LOOKING;
752 }
753 verify(expected != idx);
754
755 len ++;
756 }
757
758 block++;
759 yield();
760 }
761
762 // update statistics
763 #if !defined(__CFA_NO_STATISTICS__)
764 disable_interrupts();
765 __tls_stats()->io.submit_q.look_avg.val += len;
766 __tls_stats()->io.submit_q.look_avg.block += block;
767 __tls_stats()->io.submit_q.look_avg.cnt += 1;
768 enable_interrupts( __cfaabi_dbg_ctx );
769 #endif
770
771 return picked;
772 }
773
774 void __submit( struct __io_data & ring, uint32_t idx ) {
775 // Get now the data we definetely need
776 uint32_t * const tail = ring.submit_q.tail;
777 const uint32_t mask = *ring.submit_q.mask;
778
779 // There are 2 submission schemes, check which one we are using
780 if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) {
781 // If the poller thread submits, then we just need to add this to the ready array
782
783 __submit_to_ready_array( ring, idx, mask );
784
785 __wake_poller( ring );
786
787 __cfadbg_print_safe( io, "Kernel I/O : Added %u to ready for %p\n", idx, active_thread() );
788 }
789 else {
790 // get mutual exclusion
791 lock(ring.submit_q.lock __cfaabi_dbg_ctx2);
792
793 // Append to the list of ready entries
794
795 /* paranoid */ verify( idx <= mask );
796
797 ring.submit_q.array[ (*tail) & mask ] = idx & mask;
798 __atomic_fetch_add(tail, 1ul32, __ATOMIC_SEQ_CST);
799
800 // Submit however, many entries need to be submitted
801 int ret = syscall( __NR_io_uring_enter, ring.fd, 1, 0, 0, 0p, 0);
802 if( ret < 0 ) {
803 switch((int)errno) {
804 default:
805 abort( "KERNEL ERROR: IO_URING SUBMIT - %s\n", strerror(errno) );
806 }
807 }
808
809 // update statistics
810 #if !defined(__CFA_NO_STATISTICS__)
811 __tls_stats()->io.submit_q.submit_avg.csm += 1;
812 __tls_stats()->io.submit_q.submit_avg.cnt += 1;
813 #endif
814
815 ring.submit_q.sqes[ idx & mask ].user_data = 0;
816
817 unlock(ring.submit_q.lock);
818
819 __cfadbg_print_safe( io, "Kernel I/O : Performed io_submit for %p, returned %d\n", active_thread(), ret );
820 }
821 }
822#endif
Note: See TracBrowser for help on using the repository browser.