Context Navigation

io.cfa @ e46c753

ADTarm-ehast-experimentalenumforall-pointer-decayjacob/cs343-translationnew-astnew-ast-unique-exprpthread-emulationqualifiedEnum

Last change on this file since e46c753 was e46c753, checked in by Thierry Delisle <tdelisle@…>, 4 years ago
Added new io algorithm that eagerly submits while still helping
Property mode set to `100644`
File size: 27.7 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	if( (io_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS) && (io_flags & CFA_CLUSTER_IO_EAGER_SUBMITS) ) {
185	abort("CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS and CFA_CLUSTER_IO_EAGER_SUBMITS cannot be mixed\n");
186	}
187
188	this.io = malloc();
189
190	// Step 1 : call to setup
191	struct io_uring_params params;
192	memset(&params, 0, sizeof(params));
193
194	uint32_t nentries = entries_per_cluster();
195
196	int fd = syscall(__NR_io_uring_setup, nentries, &params );
197	if(fd < 0) {
198	abort("KERNEL ERROR: IO_URING SETUP - %s\n", strerror(errno));
199	}
200
201	// Step 2 : mmap result
202	memset( this.io, 0, sizeof(struct __io_data) );
203	struct __submition_data & sq = this.io->submit_q;
204	struct __completion_data & cq = this.io->completion_q;
205
206	// calculate the right ring size
207	sq.ring_sz = params.sq_off.array + (params.sq_entries * sizeof(unsigned) );
208	cq.ring_sz = params.cq_off.cqes + (params.cq_entries * sizeof(struct io_uring_cqe));
209
210	// Requires features
211	#if defined(IORING_FEAT_SINGLE_MMAP)
212	// adjust the size according to the parameters
213	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
214	cq->ring_sz = sq->ring_sz = max(cq->ring_sz, sq->ring_sz);
215	}
216	#endif
217
218	// mmap the Submit Queue into existence
219	sq.ring_ptr = mmap(0, sq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQ_RING);
220	if (sq.ring_ptr == (void*)MAP_FAILED) {
221	abort("KERNEL ERROR: IO_URING MMAP1 - %s\n", strerror(errno));
222	}
223
224	// Requires features
225	#if defined(IORING_FEAT_SINGLE_MMAP)
226	// mmap the Completion Queue into existence (may or may not be needed)
227	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
228	cq->ring_ptr = sq->ring_ptr;
229	}
230	else
231	#endif
232	{
233	// We need multiple call to MMAP
234	cq.ring_ptr = mmap(0, cq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_CQ_RING);
235	if (cq.ring_ptr == (void*)MAP_FAILED) {
236	munmap(sq.ring_ptr, sq.ring_sz);
237	abort("KERNEL ERROR: IO_URING MMAP2 - %s\n", strerror(errno));
238	}
239	}
240
241	// mmap the submit queue entries
242	size_t size = params.sq_entries * sizeof(struct io_uring_sqe);
243	sq.sqes = (struct io_uring_sqe *)mmap(0, size, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQES);
244	if (sq.sqes == (struct io_uring_sqe *)MAP_FAILED) {
245	munmap(sq.ring_ptr, sq.ring_sz);
246	if (cq.ring_ptr != sq.ring_ptr) munmap(cq.ring_ptr, cq.ring_sz);
247	abort("KERNEL ERROR: IO_URING MMAP3 - %s\n", strerror(errno));
248	}
249
250	// Get the pointers from the kernel to fill the structure
251	// submit queue
252	sq.head = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.head);
253	sq.tail = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail);
254	sq.mask = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask);
255	sq.num = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries);
256	sq.flags = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags);
257	sq.dropped = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped);
258	sq.array = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.array);
259
260	{
261	const uint32_t num = *sq.num;
262	for( i; num ) {
263	sq.sqes[i].user_data = 0ul64;
264	}
265	}
266
267	if( io_flags & ( CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS \| CFA_CLUSTER_IO_EAGER_SUBMITS ) ) {
268	/* paranoid */ verify( is_pow2( io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET ) \|\| ((io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET) < 8) );
269	sq.ready_cnt = max(io_flags >> CFA_CLUSTER_IO_BUFFLEN_OFFSET, 8);
270	sq.ready = alloc_align( 64, sq.ready_cnt );
271	for(i; sq.ready_cnt) {
272	sq.ready[i] = -1ul32;
273	}
274	}
275	else {
276	sq.ready_cnt = 0;
277	sq.ready = 0p;
278	}
279
280	// completion queue
281	cq.head = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.head);
282	cq.tail = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail);
283	cq.mask = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask);
284	cq.num = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries);
285	cq.overflow = ( uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow);
286	cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes);
287
288	// some paranoid checks
289	/* 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 );
290	/* paranoid / verifyf( (cq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *cq.num );
291	/* paranoid / verifyf( (cq.head) == 0, "IO_URING Expected head to be 0, got %u", *cq.head );
292	/* paranoid / verifyf( (cq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *cq.tail );
293
294	/* 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 );
295	/* paranoid / verifyf( (sq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *sq.num );
296	/* paranoid / verifyf( (sq.head) == 0, "IO_URING Expected head to be 0, got %u", *sq.head );
297	/* paranoid / verifyf( (sq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *sq.tail );
298
299	// Update the global ring info
300	this.io->ring_flags = params.flags;
301	this.io->cltr_flags = io_flags;
302	this.io->fd = fd;
303	this.io->done = false;
304	(this.io->submit){ min(sq.num, cq.num) };
305
306	if(!main_cluster) {
307	__kernel_io_finish_start( this );
308	}
309	}
310
311	void __kernel_io_finish_start( cluster & this ) {
312	if( this.io->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
313	__cfadbg_print_safe(io_core, "Kernel I/O : Creating fast poller for cluter %p\n", &this);
314	(this.io->poller.fast){ this };
315	__thrd_start( this.io->poller.fast, main );
316	}
317
318	// Create the poller thread
319	__cfadbg_print_safe(io_core, "Kernel I/O : Creating slow poller for cluter %p\n", &this);
320	this.io->poller.slow.blocked = false;
321	this.io->poller.slow.stack = __create_pthread( &this.io->poller.slow.kthrd, __io_poller_slow, &this );
322	}
323
324	void __kernel_io_prepare_stop( cluster & this ) {
325	__cfadbg_print_safe(io_core, "Kernel I/O : Stopping pollers for cluster\n", &this);
326	// Notify the poller thread of the shutdown
327	__atomic_store_n(&this.io->done, true, __ATOMIC_SEQ_CST);
328
329	// Stop the IO Poller
330	sigval val = { 1 };
331	pthread_sigqueue( this.io->poller.slow.kthrd, SIGUSR1, val );
332	post( this.io->poller.sem );
333
334	// Wait for the poller thread to finish
335	pthread_join( this.io->poller.slow.kthrd, 0p );
336	free( this.io->poller.slow.stack );
337
338	__cfadbg_print_safe(io_core, "Kernel I/O : Slow poller stopped for cluster\n", &this);
339
340	if( this.io->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
341	with( this.io->poller.fast ) {
342	/* paranoid */ verify( this.nprocessors == 0 \|\| &this == mainCluster );
343	/* paranoid */ verify( !ready_mutate_islocked() );
344
345	// We need to adjust the clean-up based on where the thread is
346	if( thrd.state == Ready \|\| thrd.preempted != __NO_PREEMPTION ) {
347
348	ready_schedule_lock( (struct __processor_id_t *)active_processor() );
349
350	// This is the tricky case
351	// The thread was preempted and now it is on the ready queue
352	// The thread should be the last on the list
353	/* paranoid */ verify( thrd.link.next != 0p );
354
355	// Remove the thread from the ready queue of this cluster
356	__attribute__((unused)) bool removed = remove_head( &this, &thrd );
357	/* paranoid */ verify( removed );
358	thrd.link.next = 0p;
359	thrd.link.prev = 0p;
360	__cfaabi_dbg_debug_do( thrd.unpark_stale = true );
361
362	// Fixup the thread state
363	thrd.state = Blocked;
364	thrd.ticket = 0;
365	thrd.preempted = __NO_PREEMPTION;
366
367	ready_schedule_unlock( (struct __processor_id_t *)active_processor() );
368
369	// Pretend like the thread was blocked all along
370	}
371	// !!! This is not an else if !!!
372	if( thrd.state == Blocked ) {
373
374	// This is the "easy case"
375	// The thread is parked and can easily be moved to active cluster
376	verify( thrd.curr_cluster != active_cluster() \|\| thrd.curr_cluster == mainCluster );
377	thrd.curr_cluster = active_cluster();
378
379	// unpark the fast io_poller
380	unpark( &thrd __cfaabi_dbg_ctx2 );
381	}
382	else {
383
384	// The thread is in a weird state
385	// I don't know what to do here
386	abort("Fast poller thread is in unexpected state, cannot clean-up correctly\n");
387	}
388
389	}
390
391	^(this.io->poller.fast){};
392
393	__cfadbg_print_safe(io_core, "Kernel I/O : Fast poller stopped for cluster\n", &this);
394	}
395	}
396
397	void __kernel_io_shutdown( cluster & this, bool main_cluster ) {
398	if(!main_cluster) {
399	__kernel_io_prepare_stop( this );
400	}
401
402	// Shutdown the io rings
403	struct __submition_data & sq = this.io->submit_q;
404	struct __completion_data & cq = this.io->completion_q;
405
406	// unmap the submit queue entries
407	munmap(sq.sqes, (sq.num) sizeof(struct io_uring_sqe));
408
409	// unmap the Submit Queue ring
410	munmap(sq.ring_ptr, sq.ring_sz);
411
412	// unmap the Completion Queue ring, if it is different
413	if (cq.ring_ptr != sq.ring_ptr) {
414	munmap(cq.ring_ptr, cq.ring_sz);
415	}
416
417	// close the file descriptor
418	close(this.io->fd);
419
420	free( this.io->submit_q.ready ); // Maybe null, doesn't matter
421	free( this.io );
422	}
423
424	//=============================================================================================
425	// I/O Polling
426	//=============================================================================================
427	// Process a single completion message from the io_uring
428	// This is NOT thread-safe
429	static unsigned __collect_submitions( struct __io_data & ring );
430	static [int, bool] __drain_io( & struct __io_data ring, * sigset_t mask, int waitcnt, bool in_kernel ) {
431	/* paranoid */ verify( !kernelTLS.preemption_state.enabled );
432
433	unsigned to_submit = 0;
434	if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) {
435	// If the poller thread also submits, then we need to aggregate the submissions which are ready
436	to_submit = __collect_submitions( ring );
437	}
438
439	const uint32_t smask = *ring.submit_q.mask;
440	uint32_t shead = *ring.submit_q.head;
441	int ret = syscall( __NR_io_uring_enter, ring.fd, to_submit, waitcnt, IORING_ENTER_GETEVENTS, mask, _NSIG / 8);
442	if( ret < 0 ) {
443	switch((int)errno) {
444	case EAGAIN:
445	case EINTR:
446	return [0, true];
447	default:
448	abort( "KERNEL ERROR: IO_URING WAIT - %s\n", strerror(errno) );
449	}
450	}
451
452	// Release the consumed SQEs
453	for( i; ret ) {
454	uint32_t idx = ring.submit_q.array[ (i + shead) & smask ];
455	ring.submit_q.sqes[ idx ].user_data = 0;
456	}
457
458	// update statistics
459	#if !defined(__CFA_NO_STATISTICS__)
460	if( to_submit > 0 ) {
461	__tls_stats()->io.submit_q.submit_avg.rdy += to_submit;
462	__tls_stats()->io.submit_q.submit_avg.csm += ret;
463	__tls_stats()->io.submit_q.submit_avg.cnt += 1;
464	}
465	#endif
466
467	// Drain the queue
468	unsigned head = *ring.completion_q.head;
469	unsigned tail = *ring.completion_q.tail;
470	const uint32_t mask = *ring.completion_q.mask;
471
472	// Memory barrier
473	__atomic_thread_fence( __ATOMIC_SEQ_CST );
474
475	// Nothing was new return 0
476	if (head == tail) {
477	return [0, to_submit > 0];
478	}
479
480	uint32_t count = tail - head;
481	for(i; count) {
482	unsigned idx = (head + i) & mask;
483	struct io_uring_cqe & cqe = ring.completion_q.cqes[idx];
484
485	/* paranoid */ verify(&cqe);
486
487	struct __io_user_data_t * data = (struct __io_user_data_t *)(uintptr_t)cqe.user_data;
488	__cfadbg_print_safe( io, "Kernel I/O : Performed reading io cqe %p, result %d for %p\n", data, cqe.res, data->thrd );
489
490	data->result = cqe.res;
491	if(!in_kernel) { unpark( data->thrd __cfaabi_dbg_ctx2 ); }
492	else { __unpark( &ring.poller.slow.id, data->thrd __cfaabi_dbg_ctx2 ); }
493	}
494
495	// Allow new submissions to happen
496	// V(ring.submit, count);
497
498	// Mark to the kernel that the cqe has been seen
499	// Ensure that the kernel only sees the new value of the head index after the CQEs have been read.
500	__atomic_thread_fence( __ATOMIC_SEQ_CST );
501	__atomic_fetch_add( ring.completion_q.head, count, __ATOMIC_RELAXED );
502
503	return [count, count > 0 \|\| to_submit > 0];
504	}
505
506	static void * __io_poller_slow( void * arg ) {
507	#if !defined( __CFA_NO_STATISTICS__ )
508	__stats_t local_stats;
509	__init_stats( &local_stats );
510	kernelTLS.this_stats = &local_stats;
511	#endif
512
513	cluster * cltr = (cluster *)arg;
514	struct __io_data & ring = *cltr->io;
515
516	ring.poller.slow.id.id = doregister( &ring.poller.slow.id );
517
518	sigset_t mask;
519	sigfillset(&mask);
520	if ( pthread_sigmask( SIG_BLOCK, &mask, 0p ) == -1 ) {
521	abort( "KERNEL ERROR: IO_URING - pthread_sigmask" );
522	}
523
524	sigdelset( &mask, SIGUSR1 );
525
526	verify( (ring.submit_q.head) == (ring.submit_q.tail) );
527	verify( (ring.completion_q.head) == (ring.completion_q.tail) );
528
529	__cfadbg_print_safe(io_core, "Kernel I/O : Slow poller for ring %p ready\n", &ring);
530
531	if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD ) {
532	while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) {
533
534	__atomic_store_n( &ring.poller.slow.blocked, true, __ATOMIC_SEQ_CST );
535
536	// In the user-thread approach drain and if anything was drained,
537	// batton pass to the user-thread
538	int count;
539	bool again;
540	[count, again] = __drain_io( ring, &mask, 1, true );
541
542	__atomic_store_n( &ring.poller.slow.blocked, false, __ATOMIC_SEQ_CST );
543
544	// Update statistics
545	#if !defined(__CFA_NO_STATISTICS__)
546	__tls_stats()->io.complete_q.completed_avg.val += count;
547	__tls_stats()->io.complete_q.completed_avg.slow_cnt += 1;
548	#endif
549
550	if(again) {
551	__cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to fast poller\n", &ring);
552	__unpark( &ring.poller.slow.id, &ring.poller.fast.thrd __cfaabi_dbg_ctx2 );
553	wait( ring.poller.sem );
554	}
555	}
556	}
557	else {
558	while(!__atomic_load_n(&ring.done, __ATOMIC_SEQ_CST)) {
559	//In the naive approach, just poll the io completion queue directly
560	int count;
561	bool again;
562	[count, again] = __drain_io( ring, &mask, 1, true );
563
564	// Update statistics
565	#if !defined(__CFA_NO_STATISTICS__)
566	__tls_stats()->io.complete_q.completed_avg.val += count;
567	__tls_stats()->io.complete_q.completed_avg.slow_cnt += 1;
568	#endif
569	}
570	}
571
572	__cfadbg_print_safe(io_core, "Kernel I/O : Slow poller for ring %p stopping\n", &ring);
573
574	unregister( &ring.poller.slow.id );
575
576	#if !defined(__CFA_NO_STATISTICS__)
577	__tally_stats(cltr->stats, &local_stats);
578	#endif
579
580	return 0p;
581	}
582
583	void main( __io_poller_fast & this ) {
584	verify( this.ring->cltr_flags & CFA_CLUSTER_IO_POLLER_USER_THREAD );
585
586	// Start parked
587	park( __cfaabi_dbg_ctx );
588
589	__cfadbg_print_safe(io_core, "Kernel I/O : Fast poller for ring %p ready\n", &this.ring);
590
591	int reset = 0;
592
593	// Then loop until we need to start
594	while(!__atomic_load_n(&this.ring->done, __ATOMIC_SEQ_CST)) {
595
596	// Drain the io
597	int count;
598	bool again;
599	disable_interrupts();
600	[count, again] = __drain_io( *this.ring, 0p, 0, false );
601
602	if(!again) reset++;
603
604	// Update statistics
605	#if !defined(__CFA_NO_STATISTICS__)
606	__tls_stats()->io.complete_q.completed_avg.val += count;
607	__tls_stats()->io.complete_q.completed_avg.fast_cnt += 1;
608	#endif
609	enable_interrupts( __cfaabi_dbg_ctx );
610
611	// If we got something, just yield and check again
612	if(reset < 5) {
613	yield();
614	}
615	// We didn't get anything baton pass to the slow poller
616	else {
617	__cfadbg_print_safe(io_core, "Kernel I/O : Moving to ring %p to slow poller\n", &this.ring);
618	reset = 0;
619
620	// wake up the slow poller
621	post( this.ring->poller.sem );
622
623	// park this thread
624	park( __cfaabi_dbg_ctx );
625	}
626	}
627
628	__cfadbg_print_safe(io_core, "Kernel I/O : Fast poller for ring %p stopping\n", &this.ring);
629	}
630
631	static inline void __wake_poller( struct __io_data & ring ) __attribute__((artificial));
632	static inline void __wake_poller( struct __io_data & ring ) {
633	if(!__atomic_load_n( &ring.poller.slow.blocked, __ATOMIC_SEQ_CST)) return;
634
635	sigval val = { 1 };
636	pthread_sigqueue( ring.poller.slow.kthrd, SIGUSR1, val );
637	}
638
639	//=============================================================================================
640	// I/O Submissions
641	//=============================================================================================
642
643	// Submition steps :
644	// 1 - Allocate a queue entry. The ring already has memory for all entries but only the ones
645	// listed in sq.array are visible by the kernel. For those not listed, the kernel does not
646	// offer any assurance that an entry is not being filled by multiple flags. Therefore, we
647	// need to write an allocator that allows allocating concurrently.
648	//
649	// 2 - Actually fill the submit entry, this is the only simple and straightforward step.
650	//
651	// 3 - Append the entry index to the array and adjust the tail accordingly. This operation
652	// needs to arrive to two concensus at the same time:
653	// A - The order in which entries are listed in the array: no two threads must pick the
654	// same index for their entries
655	// B - When can the tail be update for the kernel. EVERY entries in the array between
656	// head and tail must be fully filled and shouldn't ever be touched again.
657	//
658
659	[* struct io_uring_sqe, uint32_t] __submit_alloc( struct __io_data & ring, uint64_t data ) {
660	/* paranoid */ verify( data != 0 );
661
662	// Prepare the data we need
663	__attribute((unused)) int len = 0;
664	__attribute((unused)) int block = 0;
665	uint32_t cnt = *ring.submit_q.num;
666	uint32_t mask = *ring.submit_q.mask;
667
668	disable_interrupts();
669	uint32_t off = __tls_rand();
670	enable_interrupts( __cfaabi_dbg_ctx );
671
672	// Loop around looking for an available spot
673	for() {
674	// Look through the list starting at some offset
675	for(i; cnt) {
676	uint64_t expected = 0;
677	uint32_t idx = (i + off) & mask;
678	struct io_uring_sqe * sqe = &ring.submit_q.sqes[idx];
679	volatile uint64_t * udata = &sqe->user_data;
680
681	if( *udata == expected &&
682	__atomic_compare_exchange_n( udata, &expected, data, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) )
683	{
684	// update statistics
685	#if !defined(__CFA_NO_STATISTICS__)
686	disable_interrupts();
687	__tls_stats()->io.submit_q.alloc_avg.val += len;
688	__tls_stats()->io.submit_q.alloc_avg.block += block;
689	__tls_stats()->io.submit_q.alloc_avg.cnt += 1;
690	enable_interrupts( __cfaabi_dbg_ctx );
691	#endif
692
693
694	// Success return the data
695	return [sqe, idx];
696	}
697	verify(expected != data);
698
699	len ++;
700	}
701
702	block++;
703	yield();
704	}
705	}
706
707	static inline uint32_t __submit_to_ready_array( struct __io_data & ring, uint32_t idx, const uint32_t mask ) {
708	/* paranoid */ verify( idx <= mask );
709	/* paranoid */ verify( idx != -1ul32 );
710
711	// We need to find a spot in the ready array
712	__attribute((unused)) int len = 0;
713	__attribute((unused)) int block = 0;
714	uint32_t ready_mask = ring.submit_q.ready_cnt - 1;
715
716	disable_interrupts();
717	uint32_t off = __tls_rand();
718	enable_interrupts( __cfaabi_dbg_ctx );
719
720	uint32_t picked;
721	LOOKING: for() {
722	for(i; ring.submit_q.ready_cnt) {
723	picked = (i + off) & ready_mask;
724	uint32_t expected = -1ul32;
725	if( __atomic_compare_exchange_n( &ring.submit_q.ready[picked], &expected, idx, true, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED ) ) {
726	break LOOKING;
727	}
728	verify(expected != idx);
729
730	len ++;
731	}
732
733	block++;
734	yield();
735	}
736
737	// update statistics
738	#if !defined(__CFA_NO_STATISTICS__)
739	disable_interrupts();
740	__tls_stats()->io.submit_q.look_avg.val += len;
741	__tls_stats()->io.submit_q.look_avg.block += block;
742	__tls_stats()->io.submit_q.look_avg.cnt += 1;
743	enable_interrupts( __cfaabi_dbg_ctx );
744	#endif
745
746	return picked;
747	}
748
749	void __submit( struct __io_data & ring, uint32_t idx ) {
750	// Get now the data we definetely need
751	uint32_t * const tail = ring.submit_q.tail;
752	const uint32_t mask = *ring.submit_q.mask;
753
754	// There are 2 submission schemes, check which one we are using
755	if( ring.cltr_flags & CFA_CLUSTER_IO_POLLER_THREAD_SUBMITS ) {
756	// If the poller thread submits, then we just need to add this to the ready array
757	__submit_to_ready_array( ring, idx, mask );
758
759	__wake_poller( ring );
760
761	__cfadbg_print_safe( io, "Kernel I/O : Added %u to ready for %p\n", idx, active_thread() );
762	}
763	else if( ring.cltr_flags & CFA_CLUSTER_IO_EAGER_SUBMITS ) {
764	uint32_t picked = __submit_to_ready_array( ring, idx, mask );
765
766	for() {
767	yield();
768
769	// If some one else collected our index, we are done
770	if( ring.submit_q.ready[picked] != idx ) {
771	#if !defined(__CFA_NO_STATISTICS__)
772	disable_interrupts();
773	__tls_stats()->io.submit_q.helped += 1;
774	enable_interrupts( __cfaabi_dbg_ctx );
775	#endif
776	return;
777	}
778
779	if( try_lock(ring.submit_q.lock __cfaabi_dbg_ctx2) ) {
780	#if !defined(__CFA_NO_STATISTICS__)
781	__tls_stats()->io.submit_q.leader += 1;
782	#endif
783	break;
784	}
785	}
786
787	// We got the lock
788	unsigned to_submit = __collect_submitions( ring );
789	// /* paranoid */ verify( to_submit > 0 );
790	if( to_submit == 0 ) abort();
791
792	const uint32_t smask = *ring.submit_q.mask;
793	uint32_t shead = *ring.submit_q.head;
794	int ret = syscall( __NR_io_uring_enter, ring.fd, to_submit, 0, 0, 0p, _NSIG / 8);
795	if( ret < 0 ) {
796	switch((int)errno) {
797	case EAGAIN:
798	case EINTR:
799	unlock(ring.submit_q.lock);
800	return;
801	default:
802	abort( "KERNEL ERROR: IO_URING WAIT - %s\n", strerror(errno) );
803	}
804	}
805
806	/* paranoid */ verify( ret > 0 );
807
808	// Release the consumed SQEs
809	for( i; ret ) {
810	uint32_t idx = ring.submit_q.array[ (i + shead) & smask ];
811	ring.submit_q.sqes[ idx ].user_data = 0;
812	}
813
814	// update statistics
815	#if !defined(__CFA_NO_STATISTICS__)
816	__tls_stats()->io.submit_q.submit_avg.rdy += to_submit;
817	__tls_stats()->io.submit_q.submit_avg.csm += ret;
818	__tls_stats()->io.submit_q.submit_avg.cnt += 1;
819	#endif
820
821	unlock(ring.submit_q.lock);
822	}
823	else {
824	// get mutual exclusion
825	lock(ring.submit_q.lock __cfaabi_dbg_ctx2);
826
827	// Append to the list of ready entries
828
829	/* paranoid */ verify( idx <= mask );
830
831	ring.submit_q.array[ (*tail) & mask ] = idx & mask;
832	__atomic_fetch_add(tail, 1ul32, __ATOMIC_SEQ_CST);
833
834	// Submit however, many entries need to be submitted
835	int ret = syscall( __NR_io_uring_enter, ring.fd, 1, 0, 0, 0p, 0);
836	if( ret < 0 ) {
837	switch((int)errno) {
838	default:
839	abort( "KERNEL ERROR: IO_URING SUBMIT - %s\n", strerror(errno) );
840	}
841	}
842
843	// update statistics
844	#if !defined(__CFA_NO_STATISTICS__)
845	disable_interrupts();
846	abort();
847	__tls_stats()->io.submit_q.submit_avg.csm += 1;
848	__tls_stats()->io.submit_q.submit_avg.cnt += 1;
849	enable_interrupts( __cfaabi_dbg_ctx );
850	#endif
851
852	ring.submit_q.sqes[ idx & mask ].user_data = 0;
853
854	unlock(ring.submit_q.lock);
855
856	__cfadbg_print_safe( io, "Kernel I/O : Performed io_submit for %p, returned %d\n", active_thread(), ret );
857	}
858	}
859
860	static unsigned __collect_submitions( struct __io_data & ring ) {
861	/* paranoid */ verify( ring.submit_q.ready != 0p );
862	/* paranoid */ verify( ring.submit_q.ready_cnt > 0 );
863
864	unsigned to_submit = 0;
865	uint32_t tail = *ring.submit_q.tail;
866	const uint32_t mask = *ring.submit_q.mask;
867
868	// Go through the list of ready submissions
869	for( i; ring.submit_q.ready_cnt ) {
870	// replace any submission with the sentinel, to consume it.
871	uint32_t idx = __atomic_exchange_n( &ring.submit_q.ready[i], -1ul32, __ATOMIC_RELAXED);
872
873	// If it was already the sentinel, then we are done
874	if( idx == -1ul32 ) continue;
875
876	// If we got a real submission, append it to the list
877	ring.submit_q.array[ (tail + to_submit) & mask ] = idx & mask;
878	to_submit++;
879	}
880
881	// Increment the tail based on how many we are ready to submit
882	__atomic_fetch_add(ring.submit_q.tail, to_submit, __ATOMIC_SEQ_CST);
883
884	return to_submit;
885	}
886	#endif

Note: See TracBrowser for help on using the repository browser.

Download in other formats:

Original Format