Context Navigation

setup.cfa@ c44d652

Visit:

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 c44d652 was c44d652, checked in by Thierry Delisle <tdelisle@…>, 5 years ago
Fixed broken merge
Property mode set to `100644`
File size: 15.8 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	#include "io/types.hfa"
20	#include "kernel.hfa"
21
22	#if !defined(CFA_HAVE_LINUX_IO_URING_H)
23	void __kernel_io_startup() {
24	// Nothing to do without io_uring
25	}
26
27	void __kernel_io_shutdown() {
28	// Nothing to do without io_uring
29	}
30
31	void ?{}(io_context & this, struct cluster & cl) {}
32	void ?{}(io_context & this, struct cluster & cl, const io_context_params & params) {}
33
34	void ^?{}(io_context & this) {}
35	void ^?{}(io_context & this, bool cluster_context) {}
36
37	#else
38	#include <errno.h>
39	#include <stdint.h>
40	#include <string.h>
41	#include <signal.h>
42	#include <unistd.h>
43
44	extern "C" {
45	#include <pthread.h>
46	#include <sys/epoll.h>
47	#include <sys/mman.h>
48	#include <sys/syscall.h>
49
50	#include <linux/io_uring.h>
51	}
52
53	#include "bitmanip.hfa"
54	#include "kernel_private.hfa"
55	#include "thread.hfa"
56
57	void ?{}(io_context_params & this) {
58	this.num_entries = 256;
59	this.num_ready = 256;
60	this.submit_aff = -1;
61	this.eager_submits = false;
62	this.poller_submits = false;
63	this.poll_submit = false;
64	this.poll_complete = false;
65	}
66
67	static void * __io_poller_slow( void * arg );
68
69	// Weirdly, some systems that do support io_uring don't actually define these
70	#ifdef __alpha__
71	/*
72	* alpha is the only exception, all other architectures
73	* have common numbers for new system calls.
74	*/
75	#ifndef __NR_io_uring_setup
76	#define __NR_io_uring_setup 535
77	#endif
78	#ifndef __NR_io_uring_enter
79	#define __NR_io_uring_enter 536
80	#endif
81	#ifndef __NR_io_uring_register
82	#define __NR_io_uring_register 537
83	#endif
84	#else /* !__alpha__ */
85	#ifndef __NR_io_uring_setup
86	#define __NR_io_uring_setup 425
87	#endif
88	#ifndef __NR_io_uring_enter
89	#define __NR_io_uring_enter 426
90	#endif
91	#ifndef __NR_io_uring_register
92	#define __NR_io_uring_register 427
93	#endif
94	#endif
95
96	//=============================================================================================
97	// I/O Startup / Shutdown logic + Master Poller
98	//=============================================================================================
99
100	// IO Master poller loop forward
101	static void * iopoll_loop( __attribute__((unused)) void * args );
102
103	static struct {
104	pthread_t thrd; // pthread handle to io poller thread
105	void * stack; // pthread stack for io poller thread
106	int epollfd; // file descriptor to the epoll instance
107	volatile bool run; // Whether or not to continue
108	} iopoll;
109
110	void __kernel_io_startup(void) {
111	__cfaabi_dbg_print_safe( "Kernel : Creating EPOLL instance\n" );
112
113	iopoll.epollfd = epoll_create1(0);
114	if (iopoll.epollfd == -1) {
115	abort( "internal error, epoll_create1\n");
116	}
117
118	__cfaabi_dbg_print_safe( "Kernel : Starting io poller thread\n" );
119
120	iopoll.run = true;
121	iopoll.stack = __create_pthread( &iopoll.thrd, iopoll_loop, 0p );
122	}
123
124	void __kernel_io_shutdown(void) {
125	// Notify the io poller thread of the shutdown
126	iopoll.run = false;
127	sigval val = { 1 };
128	pthread_sigqueue( iopoll.thrd, SIGUSR1, val );
129
130	// Wait for the io poller thread to finish
131
132	pthread_join( iopoll.thrd, 0p );
133	free( iopoll.stack );
134
135	int ret = close(iopoll.epollfd);
136	if (ret == -1) {
137	abort( "internal error, close epoll\n");
138	}
139
140	// Io polling is now fully stopped
141
142	__cfaabi_dbg_print_safe( "Kernel : IO poller stopped\n" );
143	}
144
145	static void * iopoll_loop( __attribute__((unused)) void * args ) {
146	__processor_id_t id;
147	id.id = doregister(&id);
148	__cfaabi_dbg_print_safe( "Kernel : IO poller thread starting\n" );
149
150	// Block signals to control when they arrive
151	sigset_t mask;
152	sigfillset(&mask);
153	if ( pthread_sigmask( SIG_BLOCK, &mask, 0p ) == -1 ) {
154	abort( "internal error, pthread_sigmask" );
155	}
156
157	sigdelset( &mask, SIGUSR1 );
158
159	// Create sufficient events
160	struct epoll_event events[10];
161	// Main loop
162	while( iopoll.run ) {
163	// Wait for events
164	int nfds = epoll_pwait( iopoll.epollfd, events, 10, -1, &mask );
165
166	// Check if an error occured
167	if (nfds == -1) {
168	if( errno == EINTR ) continue;
169	abort( "internal error, pthread_sigmask" );
170	}
171
172	for(i; nfds) {
173	$io_ctx_thread * io_ctx = ($io_ctx_thread *)(uintptr_t)events[i].data.u64;
174	/* paranoid */ verify( io_ctx );
175	__cfadbg_print_safe(io_core, "Kernel I/O : Unparking io poller %p\n", io_ctx);
176	#if !defined( __CFA_NO_STATISTICS__ )
177	kernelTLS.this_stats = io_ctx->self.curr_cluster->stats;
178	#endif
179	__post( io_ctx->sem, &id );
180	}
181	}
182
183	__cfaabi_dbg_print_safe( "Kernel : IO poller thread stopping\n" );
184	unregister(&id);
185	return 0p;
186	}
187
188	//=============================================================================================
189	// I/O Context Constrution/Destruction
190	//=============================================================================================
191
192	void ?{}($io_ctx_thread & this, struct cluster & cl) { (this.self){ "IO Poller", cl }; }
193	void main( $io_ctx_thread & this );
194	static inline $thread * get_thread( $io_ctx_thread & this ) { return &this.self; }
195	void ^?{}( $io_ctx_thread & mutex this ) {}
196
197	static void __io_create ( __io_data & this, const io_context_params & params_in );
198	static void __io_destroy( __io_data & this );
199
200	void ?{}(io_context & this, struct cluster & cl, const io_context_params & params) {
201	(this.thrd){ cl };
202	this.thrd.ring = malloc();
203	__cfadbg_print_safe(io_core, "Kernel I/O : Creating ring for io_context %p\n", &this);
204	__io_create( *this.thrd.ring, params );
205
206	__cfadbg_print_safe(io_core, "Kernel I/O : Starting poller thread for io_context %p\n", &this);
207	this.thrd.done = false;
208	__thrd_start( this.thrd, main );
209
210	__cfadbg_print_safe(io_core, "Kernel I/O : io_context %p ready\n", &this);
211	}
212
213	void ?{}(io_context & this, struct cluster & cl) {
214	io_context_params params;
215	(this){ cl, params };
216	}
217
218	void ^?{}(io_context & this, bool cluster_context) {
219	__cfadbg_print_safe(io_core, "Kernel I/O : tearing down io_context %p\n", &this);
220
221	// Notify the thread of the shutdown
222	__atomic_store_n(&this.thrd.done, true, __ATOMIC_SEQ_CST);
223
224	// If this is an io_context within a cluster, things get trickier
225	$thread & thrd = this.thrd.self;
226	if( cluster_context ) {
227	cluster & cltr = *thrd.curr_cluster;
228	/* paranoid */ verify( cltr.nprocessors == 0 \|\| &cltr == mainCluster );
229	/* paranoid */ verify( !ready_mutate_islocked() );
230
231	// We need to adjust the clean-up based on where the thread is
232	if( thrd.state == Ready \|\| thrd.preempted != __NO_PREEMPTION ) {
233
234	ready_schedule_lock( (struct __processor_id_t *)active_processor() );
235
236	// This is the tricky case
237	// The thread was preempted and now it is on the ready queue
238	// The thread should be the last on the list
239	/* paranoid */ verify( thrd.link.next != 0p );
240
241	// Remove the thread from the ready queue of this cluster
242	__attribute__((unused)) bool removed = remove_head( &cltr, &thrd );
243	/* paranoid */ verify( removed );
244	thrd.link.next = 0p;
245	thrd.link.prev = 0p;
246	__cfaabi_dbg_debug_do( thrd.unpark_stale = true );
247
248	// Fixup the thread state
249	thrd.state = Blocked;
250	thrd.ticket = 0;
251	thrd.preempted = __NO_PREEMPTION;
252
253	ready_schedule_unlock( (struct __processor_id_t *)active_processor() );
254
255	// Pretend like the thread was blocked all along
256	}
257	// !!! This is not an else if !!!
258	if( thrd.state == Blocked ) {
259
260	// This is the "easy case"
261	// The thread is parked and can easily be moved to active cluster
262	verify( thrd.curr_cluster != active_cluster() \|\| thrd.curr_cluster == mainCluster );
263	thrd.curr_cluster = active_cluster();
264
265	// unpark the fast io_poller
266	unpark( &thrd __cfaabi_dbg_ctx2 );
267	}
268	else {
269
270	// The thread is in a weird state
271	// I don't know what to do here
272	abort("io_context poller thread is in unexpected state, cannot clean-up correctly\n");
273	}
274	} else {
275	unpark( &thrd __cfaabi_dbg_ctx2 );
276	}
277
278	^(this.thrd){};
279	__cfadbg_print_safe(io_core, "Kernel I/O : Stopped poller thread for io_context %p\n", &this);
280
281	__io_destroy( *this.thrd.ring );
282	__cfadbg_print_safe(io_core, "Kernel I/O : Destroyed ring for io_context %p\n", &this);
283
284	free(this.thrd.ring);
285	}
286
287	void ^?{}(io_context & this) {
288	^(this){ false };
289	}
290
291	static void __io_create( __io_data & this, const io_context_params & params_in ) {
292	// Step 1 : call to setup
293	struct io_uring_params params;
294	memset(&params, 0, sizeof(params));
295	if( params_in.poll_submit ) params.flags \|= IORING_SETUP_SQPOLL;
296	if( params_in.poll_complete ) params.flags \|= IORING_SETUP_IOPOLL;
297
298	uint32_t nentries = params_in.num_entries;
299
300	int fd = syscall(__NR_io_uring_setup, nentries, &params );
301	if(fd < 0) {
302	abort("KERNEL ERROR: IO_URING SETUP - %s\n", strerror(errno));
303	}
304
305	// Step 2 : mmap result
306	memset( &this, 0, sizeof(struct __io_data) );
307	struct __submition_data & sq = this.submit_q;
308	struct __completion_data & cq = this.completion_q;
309
310	// calculate the right ring size
311	sq.ring_sz = params.sq_off.array + (params.sq_entries * sizeof(unsigned) );
312	cq.ring_sz = params.cq_off.cqes + (params.cq_entries * sizeof(struct io_uring_cqe));
313
314	// Requires features
315	#if defined(IORING_FEAT_SINGLE_MMAP)
316	// adjust the size according to the parameters
317	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
318	cq.ring_sz = sq.ring_sz = max(cq.ring_sz, sq.ring_sz);
319	}
320	#endif
321
322	// mmap the Submit Queue into existence
323	sq.ring_ptr = mmap(0, sq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQ_RING);
324	if (sq.ring_ptr == (void*)MAP_FAILED) {
325	abort("KERNEL ERROR: IO_URING MMAP1 - %s\n", strerror(errno));
326	}
327
328	// Requires features
329	#if defined(IORING_FEAT_SINGLE_MMAP)
330	// mmap the Completion Queue into existence (may or may not be needed)
331	if ((params.features & IORING_FEAT_SINGLE_MMAP) != 0) {
332	cq.ring_ptr = sq.ring_ptr;
333	}
334	else
335	#endif
336	{
337	// We need multiple call to MMAP
338	cq.ring_ptr = mmap(0, cq.ring_sz, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_CQ_RING);
339	if (cq.ring_ptr == (void*)MAP_FAILED) {
340	munmap(sq.ring_ptr, sq.ring_sz);
341	abort("KERNEL ERROR: IO_URING MMAP2 - %s\n", strerror(errno));
342	}
343	}
344
345	// mmap the submit queue entries
346	size_t size = params.sq_entries * sizeof(struct io_uring_sqe);
347	sq.sqes = (struct io_uring_sqe *)mmap(0, size, PROT_READ \| PROT_WRITE, MAP_SHARED \| MAP_POPULATE, fd, IORING_OFF_SQES);
348	if (sq.sqes == (struct io_uring_sqe *)MAP_FAILED) {
349	munmap(sq.ring_ptr, sq.ring_sz);
350	if (cq.ring_ptr != sq.ring_ptr) munmap(cq.ring_ptr, cq.ring_sz);
351	abort("KERNEL ERROR: IO_URING MMAP3 - %s\n", strerror(errno));
352	}
353
354	// Get the pointers from the kernel to fill the structure
355	// submit queue
356	sq.head = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.head);
357	sq.tail = (volatile uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.tail);
358	sq.mask = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_mask);
359	sq.num = ( const uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.ring_entries);
360	sq.flags = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.flags);
361	sq.dropped = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.dropped);
362	sq.array = ( uint32_t *)(((intptr_t)sq.ring_ptr) + params.sq_off.array);
363	sq.prev_head = *sq.head;
364
365	{
366	const uint32_t num = *sq.num;
367	for( i; num ) {
368	sq.sqes[i].user_data = 0ul64;
369	}
370	}
371
372	(sq.lock){};
373	(sq.release_lock){};
374
375	if( params_in.poller_submits \|\| params_in.eager_submits ) {
376	/* paranoid */ verify( is_pow2( params_in.num_ready ) \|\| (params_in.num_ready < 8) );
377	sq.ready_cnt = max( params_in.num_ready, 8 );
378	sq.ready = alloc_align( 64, sq.ready_cnt );
379	for(i; sq.ready_cnt) {
380	sq.ready[i] = -1ul32;
381	}
382	}
383	else {
384	sq.ready_cnt = 0;
385	sq.ready = 0p;
386	}
387
388	// completion queue
389	cq.head = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.head);
390	cq.tail = (volatile uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.tail);
391	cq.mask = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_mask);
392	cq.num = ( const uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.ring_entries);
393	cq.overflow = ( uint32_t *)(((intptr_t)cq.ring_ptr) + params.cq_off.overflow);
394	cq.cqes = (struct io_uring_cqe *)(((intptr_t)cq.ring_ptr) + params.cq_off.cqes);
395
396	// some paranoid checks
397	/* 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 );
398	/* paranoid / verifyf( (cq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *cq.num );
399	/* paranoid / verifyf( (cq.head) == 0, "IO_URING Expected head to be 0, got %u", *cq.head );
400	/* paranoid / verifyf( (cq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *cq.tail );
401
402	/* 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 );
403	/* paranoid / verifyf( (sq.num) >= nentries, "IO_URING Expected %u entries, got %u", nentries, *sq.num );
404	/* paranoid / verifyf( (sq.head) == 0, "IO_URING Expected head to be 0, got %u", *sq.head );
405	/* paranoid / verifyf( (sq.tail) == 0, "IO_URING Expected tail to be 0, got %u", *sq.tail );
406
407	// Update the global ring info
408	this.ring_flags = params.flags;
409	this.fd = fd;
410	this.eager_submits = params_in.eager_submits;
411	this.poller_submits = params_in.poller_submits;
412	}
413
414	static void __io_destroy( __io_data & this ) {
415	// Shutdown the io rings
416	struct __submition_data & sq = this.submit_q;
417	struct __completion_data & cq = this.completion_q;
418
419	// unmap the submit queue entries
420	munmap(sq.sqes, (sq.num) sizeof(struct io_uring_sqe));
421
422	// unmap the Submit Queue ring
423	munmap(sq.ring_ptr, sq.ring_sz);
424
425	// unmap the Completion Queue ring, if it is different
426	if (cq.ring_ptr != sq.ring_ptr) {
427	munmap(cq.ring_ptr, cq.ring_sz);
428	}
429
430	// close the file descriptor
431	close(this.fd);
432
433	free( this.submit_q.ready ); // Maybe null, doesn't matter
434	}
435
436	//=============================================================================================
437	// I/O Context Sleep
438	//=============================================================================================
439
440	void __ioctx_register($io_ctx_thread & ctx, struct epoll_event & ev) {
441	ev.events = EPOLLIN \| EPOLLONESHOT;
442	ev.data.u64 = (uint64_t)&ctx;
443	int ret = epoll_ctl(iopoll.epollfd, EPOLL_CTL_ADD, ctx.ring->fd, &ev);
444	if (ret < 0) {
445	abort( "KERNEL ERROR: EPOLL ADD - (%d) %s\n", (int)errno, strerror(errno) );
446	}
447	}
448
449	void __ioctx_prepare_block($io_ctx_thread & ctx, struct epoll_event & ev) {
450	int ret = epoll_ctl(iopoll.epollfd, EPOLL_CTL_MOD, ctx.ring->fd, &ev);
451	if (ret < 0) {
452	abort( "KERNEL ERROR: EPOLL REARM - (%d) %s\n", (int)errno, strerror(errno) );
453	}
454	}
455
456	//=============================================================================================
457	// I/O Context Misc Setup
458	//=============================================================================================
459	void register_fixed_files( io_context & ctx, int * files, unsigned count ) {
460	int ret = syscall( __NR_io_uring_register, ctx.thrd.ring->fd, IORING_REGISTER_FILES, files, count );
461	if( ret < 0 ) {
462	abort( "KERNEL ERROR: IO_URING SYSCALL - (%d) %s\n", (int)errno, strerror(errno) );
463	}
464
465	__cfadbg_print_safe( io_core, "Kernel I/O : Performed io_register for %p, returned %d\n", active_thread(), ret );
466	}
467
468	void register_fixed_files( cluster & cltr, int * files, unsigned count ) {
469	for(i; cltr.io.cnt) {
470	register_fixed_files( cltr.io.ctxs[i], files, count );
471	}
472	}
473	#endif

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

Download in other formats:

Original Format