Context Navigation

setup.cfa@ d60d30e

Visit:

ADT arm-eh ast-experimental enum forall-pointer-decay jacob/cs343-translation new-ast-unique-expr pthread-emulation qualifiedEnum

Last change on this file since d60d30e was 78da4ab, checked in by Thierry Delisle <tdelisle@…>, 5 years ago
New implementation of io based on instance burrowing. Trying to avoid the unbounded growth of the previous flat combining approach.
Property mode set to `100644`
File size: 15.7 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/setup.cfa --
8	//
9	// Author : Thierry Delisle
10	// Created On : Fri Jul 31 16:25:51 2020
11	// Last Modified By :
12	// Last Modified On :
13	// Update Count :
14	//
15
16	#define __cforall_thread__
17	#define _GNU_SOURCE /* See feature_test_macros(7) */
18
19	#if defined(__CFA_DEBUG__)
20	// #define __CFA_DEBUG_PRINT_IO__
21	// #define __CFA_DEBUG_PRINT_IO_CORE__
22	#endif
23
24	#include "io/types.hfa"
25	#include "kernel.hfa"
26
27	#if !defined(CFA_HAVE_LINUX_IO_URING_H)
28	void __kernel_io_startup() {
29	// Nothing to do without io_uring
30	}
31
32	void __kernel_io_shutdown() {
33	// Nothing to do without io_uring
34	}
35
36	void ?{}(io_context_params & this) {}
37
38	void ?{}($io_context & this, struct cluster & cl) {}
39	void ^?{}($io_context & this) {}
40
41	$io_arbiter * create(void) { return 0p; }
42	void destroy($io_arbiter *) {}
43
44	#else
45	#include <errno.h>
46	#include <stdint.h>
47	#include <string.h>
48	#include <signal.h>
49	#include <unistd.h>
50
51	extern "C" {
52	#include <pthread.h>
53	#include <sys/epoll.h>
54	#include <sys/eventfd.h>
55	#include <sys/mman.h>
56	#include <sys/syscall.h>
57
58	#include <linux/io_uring.h>
59	}
60
61	#include "bitmanip.hfa"
62	#include "kernel_private.hfa"
63	#include "thread.hfa"
64
65	void ?{}(io_context_params & this) {
66	this.num_entries = 256;
67	}
68
69	static void * __io_poller_slow( void * arg );
70
71	// Weirdly, some systems that do support io_uring don't actually define these
72	#ifdef __alpha__
73	/*
74	* alpha is the only exception, all other architectures
75	* have common numbers for new system calls.
76	*/
77	#ifndef __NR_io_uring_setup
78	#define __NR_io_uring_setup 535
79	#endif
80	#ifndef __NR_io_uring_enter
81	#define __NR_io_uring_enter 536
82	#endif
83	#ifndef __NR_io_uring_register
84	#define __NR_io_uring_register 537
85	#endif
86	#else /* !__alpha__ */
87	#ifndef __NR_io_uring_setup
88	#define __NR_io_uring_setup 425
89	#endif
90	#ifndef __NR_io_uring_enter
91	#define __NR_io_uring_enter 426
92	#endif
93	#ifndef __NR_io_uring_register
94	#define __NR_io_uring_register 427
95	#endif
96	#endif
97
98	//=============================================================================================
99	// I/O Startup / Shutdown logic + Master Poller
100	//=============================================================================================
101
102	// IO Master poller loop forward
103	static void * iopoll_loop( __attribute__((unused)) void * args );
104
105	static struct {
106	pthread_t thrd; // pthread handle to io poller thread
107	void * stack; // pthread stack for io poller thread
108	int epollfd; // file descriptor to the epoll instance
109	volatile bool run; // Whether or not to continue
110	volatile bool stopped; // Whether the poller has finished running
111	volatile uint64_t epoch; // Epoch used for memory reclamation
112	} iopoll;
113
114	void __kernel_io_startup(void) {
115	__cfadbg_print_safe(io_core, "Kernel : Creating EPOLL instance\n" );
116
117	iopoll.epollfd = epoll_create1(0);
118	if (iopoll.epollfd == -1) {
119	abort( "internal error, epoll_create1\n");
120	}
121
122	__cfadbg_print_safe(io_core, "Kernel : Starting io poller thread\n" );
123
124	iopoll.stack = __create_pthread( &iopoll.thrd, iopoll_loop, 0p );
125	iopoll.run = true;
126	iopoll.stopped = false;
127	iopoll.epoch = 0;
128	}
129
130	void __kernel_io_shutdown(void) {
131	// Notify the io poller thread of the shutdown
132	iopoll.run = false;
133	sigval val = { 1 };
134	pthread_sigqueue( iopoll.thrd, SIGUSR1, val );
135
136	// Wait for the io poller thread to finish
137
138	__destroy_pthread( iopoll.thrd, iopoll.stack, 0p );
139
140	int ret = close(iopoll.epollfd);
141	if (ret == -1) {
142	abort( "internal error, close epoll\n");
143	}
144
145	// Io polling is now fully stopped
146
147	__cfadbg_print_safe(io_core, "Kernel : IO poller stopped\n" );
148	}
149
150	static void * iopoll_loop( __attribute__((unused)) void * args ) {
151	__processor_id_t id;
152	id.full_proc = false;
153	id.id = doregister(&id);
154	__cfaabi_tls.this_proc_id = &id;
155	__cfadbg_print_safe(io_core, "Kernel : IO poller thread starting\n" );
156
157	// Block signals to control when they arrive
158	sigset_t mask;
159	sigfillset(&mask);
160	if ( pthread_sigmask( SIG_BLOCK, &mask, 0p ) == -1 ) {
161	abort( "internal error, pthread_sigmask" );
162	}
163
164	sigdelset( &mask, SIGUSR1 );
165
166	// Create sufficient events
167	struct epoll_event events[10];
168	// Main loop
169	while( iopoll.run ) {
170	__cfadbg_print_safe(io_core, "Kernel I/O - epoll : waiting on io_uring contexts\n");
171
172	// increment the epoch to notify any deleters we are starting a new cycle
173	__atomic_fetch_add(&iopoll.epoch, 1, __ATOMIC_SEQ_CST);
174
175	// Wait for events
176	int nfds = epoll_pwait( iopoll.epollfd, events, 10, -1, &mask );
177
178	__cfadbg_print_safe(io_core, "Kernel I/O - epoll : %d io contexts events, waking up\n", nfds);
179
180	// Check if an error occured
181	if (nfds == -1) {
182	if( errno == EINTR ) continue;
183	abort( "internal error, pthread_sigmask" );
184	}
185
186	for(i; nfds) {
187	$io_context * io_ctx = ($io_context *)(uintptr_t)events[i].data.u64;
188	/* paranoid */ verify( io_ctx );
189	__cfadbg_print_safe(io_core, "Kernel I/O - epoll : Unparking io poller %d (%p)\n", io_ctx->fd, io_ctx);
190	#if !defined( __CFA_NO_STATISTICS__ )
191	__cfaabi_tls.this_stats = io_ctx->self.curr_cluster->stats;
192	#endif
193
194	eventfd_t v;
195	eventfd_read(io_ctx->efd, &v);
196
197	post( io_ctx->sem );
198	}
199	}
200
201	__atomic_store_n(&iopoll.stopped, true, __ATOMIC_SEQ_CST);
202
203	__cfadbg_print_safe(io_core, "Kernel : IO poller thread stopping\n" );
204	unregister(&id);
205	return 0p;
206	}
207
208	//=============================================================================================
209	// I/O Context Constrution/Destruction
210	//=============================================================================================
211
212	static void __io_uring_setup ( $io_context & this, const io_context_params & params_in );
213	static void __io_uring_teardown( $io_context & this );
214	static void __epoll_register($io_context & ctx);
215	static void __epoll_unregister($io_context & ctx);
216	void __ioarbiter_register( $io_arbiter & mutex, $io_context & ctx );
217	void __ioarbiter_unregister( $io_arbiter & mutex, $io_context & ctx );
218
219	void ?{}($io_context & this, struct cluster & cl) {
220	(this.self){ "IO Poller", cl };
221	this.ext_sq.empty = true;
222	__io_uring_setup( this, cl.io.params );
223	__cfadbg_print_safe(io_core, "Kernel I/O : Created ring for io_context %u (%p)\n", this.fd, &this);
224
225	__epoll_register(this);
226
227	__ioarbiter_register(*cl.io.arbiter, this);
228
229	__thrd_start( this, main );
230	__cfadbg_print_safe(io_core, "Kernel I/O : Started poller thread for io_context %u\n", this.fd);
231	}
232
233	void ^?{}($io_context & mutex this) {
234	__cfadbg_print_safe(io_core, "Kernel I/O : tearing down io_context %u\n", this.fd);
235
236	^(this.self){};
237	__cfadbg_print_safe(io_core, "Kernel I/O : Stopped poller thread for io_context %u\n", this.fd);
238
239	__ioarbiter_unregister(*this.arbiter, this);
240
241	__epoll_unregister(this);
242
243	__io_uring_teardown( this );
244	__cfadbg_print_safe(io_core, "Kernel I/O : Destroyed ring for io_context %u\n", this.fd);
245	}
246
247	void ?{}(io_context & this, struct cluster & cl) {
248	// this.ctx = new(cl);
249	this.ctx = alloc();
250	(*this.ctx){ cl };
251
252	__cfadbg_print_safe(io_core, "Kernel I/O : io_context %u ready\n", this.ctx->fd);
253	}
254
255	void ^?{}(io_context & this) {
256	post( this.ctx->sem );
257
258	delete(this.ctx);
259	}
260
261	extern void __disable_interrupts_hard();
262	extern void __enable_interrupts_hard();
263
264	static void __io_uring_setup( $io_context & this, const io_context_params & params_in ) {
265	// Step 1 : call to setup
266	struct io_uring_params params;
267	memset(&params, 0, sizeof(params));
268	// if( params_in.poll_submit ) params.flags \|= IORING_SETUP_SQPOLL;
269	// if( params_in.poll_complete ) params.flags \|= IORING_SETUP_IOPOLL;
270
271	__u32 nentries = params_in.num_entries != 0 ? params_in.num_entries : 256;
272	if( !is_pow2(nentries) ) {
273	abort("ERROR: I/O setup 'num_entries' must be a power of 2\n");
274	}
275
276	int fd = syscall(__NR_io_uring_setup, nentries, &params );
277	if(fd < 0) {
278	abort("KERNEL ERROR: IO_URING SETUP - %s\n", strerror(errno));
279	}
280
281	// Step 2 : mmap result
282	struct __sub_ring_t & sq = this.sq;
283	struct __cmp_ring_t & cq = this.cq;
284
285	// calculate the right ring size
286	sq.ring_sz = params.sq_off.array + (params.sq_entries * sizeof(unsigned) );
287	cq.ring_sz = params.cq_off.cqes + (params.cq_entries * sizeof(struct io_uring_cqe));
288
289	// Requires features
290	#if defined(IORING_FEAT_SINGLE_MMAP)
291	// adjust the size according to the parameters
292	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
293	cq.ring_sz = sq.ring_sz = max(cq.ring_sz, sq.ring_sz);
294	}
295	#endif
296
297	// mmap the Submit Queue into existence
298	sq.ring_ptr = mmap(0, sq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQ_RING);
299	if (sq.ring_ptr == (void*)MAP_FAILED) {
300	abort("KERNEL ERROR: IO_URING MMAP1 - %s\n", strerror(errno));
301	}
302
303	// Requires features
304	#if defined(IORING_FEAT_SINGLE_MMAP)
305	// mmap the Completion Queue into existence (may or may not be needed)
306	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
307	cq.ring_ptr = sq.ring_ptr;
308	}
309	else
310	#endif
311	{
312	// We need multiple call to MMAP
313	cq.ring_ptr = mmap(0, cq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_CQ_RING);
314	if (cq.ring_ptr == (void*)MAP_FAILED) {
315	munmap(sq.ring_ptr, sq.ring_sz);
316	abort("KERNEL ERROR: IO_URING MMAP2 - %s\n", strerror(errno));
317	}
318	}
319
320	// mmap the submit queue entries
321	size_t size = params.sq_entries * sizeof(struct io_uring_sqe);
322	sq.sqes = (struct io_uring_sqe *)mmap(0, size, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQES);
323	if (sq.sqes == (struct io_uring_sqe *)MAP_FAILED) {
324	munmap(sq.ring_ptr, sq.ring_sz);
325	if (cq.ring_ptr != sq.ring_ptr) munmap(cq.ring_ptr, cq.ring_sz);
326	abort("KERNEL ERROR: IO_URING MMAP3 - %s\n", strerror(errno));
327	}
328
329	// Step 3 : Initialize the data structure
330	// Get the pointers from the kernel to fill the structure
331	// submit queue
332	sq.kring.head = (volatile __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.head);
333	sq.kring.tail = (volatile __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail);
334	sq.kring.array = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.array);
335	sq.mask = ( const __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask);
336	sq.num = ( const __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries);
337	sq.flags = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags);
338	sq.dropped = ( __u32 *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped);
339
340	sq.kring.ready = 0;
341	sq.kring.released = 0;
342
343	sq.free_ring.head = 0;
344	sq.free_ring.tail = *sq.num;
345	sq.free_ring.array = alloc( *sq.num, 128`align );
346	for(i; (__u32)*sq.num) {
347	sq.free_ring.array[i] = i;
348	}
349
350	sq.to_submit = 0;
351
352	// completion queue
353	cq.head = (volatile __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.head);
354	cq.tail = (volatile __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail);
355	cq.mask = ( const __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask);
356	cq.num = ( const __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries);
357	cq.overflow = ( __u32 *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow);
358	cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes);
359
360	// Step 4 : eventfd
361	// io_uring_register is so f*cking slow on some machine that it
362	// will never succeed if preemption isn't hard blocked
363	__disable_interrupts_hard();
364
365	int efd = eventfd(0, 0);
366	if (efd < 0) {
367	abort("KERNEL ERROR: IO_URING EVENTFD - %s\n", strerror(errno));
368	}
369
370	int ret = syscall( __NR_io_uring_register, fd, IORING_REGISTER_EVENTFD, &efd, 1);
371	if (ret < 0) {
372	abort("KERNEL ERROR: IO_URING EVENTFD REGISTER - %s\n", strerror(errno));
373	}
374
375	__enable_interrupts_hard();
376
377	// some paranoid checks
378	/* 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 );
379	/* paranoid / verifyf( (cq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *cq.num );
380	/* paranoid / verifyf( (cq.head) == 0, "IO_URING Expected head to be 0, got %u", *cq.head );
381	/* paranoid / verifyf( (cq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *cq.tail );
382
383	/* 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 );
384	/* paranoid / verifyf( (sq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *sq.num );
385	/* paranoid / verifyf( (sq.kring.head) == 0, "IO_URING Expected head to be 0, got %u", *sq.kring.head );
386	/* paranoid / verifyf( (sq.kring.tail) == 0, "IO_URING Expected tail to be 0, got %u", *sq.kring.tail );
387
388	// Update the global ring info
389	this.ring_flags = 0;
390	this.fd = fd;
391	this.efd = efd;
392	}
393
394	static void __io_uring_teardown( $io_context & this ) {
395	// Shutdown the io rings
396	struct __sub_ring_t & sq = this.sq;
397	struct __cmp_ring_t & cq = this.cq;
398
399	// unmap the submit queue entries
400	munmap(sq.sqes, (sq.num) sizeof(struct io_uring_sqe));
401
402	// unmap the Submit Queue ring
403	munmap(sq.ring_ptr, sq.ring_sz);
404
405	// unmap the Completion Queue ring, if it is different
406	if (cq.ring_ptr != sq.ring_ptr) {
407	munmap(cq.ring_ptr, cq.ring_sz);
408	}
409
410	// close the file descriptor
411	close(this.fd);
412	close(this.efd);
413
414	free( this.sq.free_ring.array ); // Maybe null, doesn't matter
415	}
416
417	//=============================================================================================
418	// I/O Context Sleep
419	//=============================================================================================
420	static inline void __epoll_ctl($io_context & ctx, int op, const char * error) {
421	struct epoll_event ev;
422	ev.events = EPOLLIN \| EPOLLONESHOT;
423	ev.data.u64 = (__u64)&ctx;
424	int ret = epoll_ctl(iopoll.epollfd, op, ctx.efd, &ev);
425	if (ret < 0) {
426	abort( "KERNEL ERROR: EPOLL %s - (%d) %s\n", error, (int)errno, strerror(errno) );
427	}
428	}
429
430	static void __epoll_register($io_context & ctx) {
431	__epoll_ctl(ctx, EPOLL_CTL_ADD, "ADD");
432	}
433
434	static void __epoll_unregister($io_context & ctx) {
435	// Read the current epoch so we know when to stop
436	size_t curr = __atomic_load_n(&iopoll.epoch, __ATOMIC_SEQ_CST);
437
438	// Remove the fd from the iopoller
439	__epoll_ctl(ctx, EPOLL_CTL_DEL, "REMOVE");
440
441	// Notify the io poller thread of the shutdown
442	iopoll.run = false;
443	sigval val = { 1 };
444	pthread_sigqueue( iopoll.thrd, SIGUSR1, val );
445
446	// Make sure all this is done
447	__atomic_thread_fence(__ATOMIC_SEQ_CST);
448
449	// Wait for the next epoch
450	while(curr == iopoll.epoch && !iopoll.stopped) Pause();
451	}
452
453	void __ioctx_prepare_block($io_context & ctx) {
454	__cfadbg_print_safe(io_core, "Kernel I/O - epoll : Re-arming io poller %d (%p)\n", ctx.fd, &ctx);
455	__epoll_ctl(ctx, EPOLL_CTL_MOD, "REARM");
456	}
457
458
459	//=============================================================================================
460	// I/O Context Misc Setup
461	//=============================================================================================
462	void ?{}( $io_arbiter & this ) {
463	this.pending.flag = false;
464	}
465
466	void ^?{}( $io_arbiter & mutex this ) {
467	/* paranoid */ verify( empty(this.assigned) );
468	/* paranoid */ verify( empty(this.available) );
469	/* paranoid */ verify( is_empty(this.pending.blocked) );
470	}
471
472	$io_arbiter * create(void) {
473	return new();
474	}
475	void destroy($io_arbiter * arbiter) {
476	delete(arbiter);
477	}
478
479	//=============================================================================================
480	// I/O Context Misc Setup
481	//=============================================================================================
482
483	#endif

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

Download in other formats:

Original Format