ViewVC Help
View File | Revision Log | Show Annotations | Download File
/cvs/libev/ev_linuxaio.c
(Generate patch)

Comparing libev/ev_linuxaio.c (file contents):
Revision 1.26 by root, Tue Jun 25 05:10:44 2019 UTC vs.
Revision 1.53 by root, Fri Dec 27 16:12:55 2019 UTC

… …
56 * POLLOUT|POLLIN, but polling for POLLIN fails. just great, 56 * POLLOUT|POLLIN, but polling for POLLIN fails. just great,
57 * so we have to fall back to something else (hello, epoll), 57 * so we have to fall back to something else (hello, epoll),
58 * but at least the fallback can be slow, because these are 58 * but at least the fallback can be slow, because these are
59 * exceptional cases, right? 59 * exceptional cases, right?
60 * d) hmm, you have to tell the kernel the maximum number of watchers 60 * d) hmm, you have to tell the kernel the maximum number of watchers
61 * you want to queue when initialiasing the aio context. but of 61 * you want to queue when initialising the aio context. but of
62 * course the real limit is magically calculated in the kernel, and 62 * course the real limit is magically calculated in the kernel, and
63 * is often higher then we asked for. so we just have to destroy 63 * is often higher then we asked for. so we just have to destroy
64 * the aio context and re-create it a bit larger if we hit the limit. 64 * the aio context and re-create it a bit larger if we hit the limit.
65 * (starts to remind you of epoll? well, it's a bit more deterministic 65 * (starts to remind you of epoll? well, it's a bit more deterministic
66 * and less gambling, but still ugly as hell). 66 * and less gambling, but still ugly as hell).
… …
68 * limit. or the kernel simply doesn't want to handle your watchers. 68 * limit. or the kernel simply doesn't want to handle your watchers.
69 * what the fuck do we do then? you guessed it, in the middle 69 * what the fuck do we do then? you guessed it, in the middle
70 * of event handling we have to switch to 100% epoll polling. and 70 * of event handling we have to switch to 100% epoll polling. and
71 * that better is as fast as normal epoll polling, so you practically 71 * that better is as fast as normal epoll polling, so you practically
72 * have to use the normal epoll backend with all its quirks. 72 * have to use the normal epoll backend with all its quirks.
73 * f) end result of this trainwreck: it inherits all the disadvantages 73 * f) end result of this train wreck: it inherits all the disadvantages
74 * from epoll, while adding a number on its own. why even bother to use 74 * from epoll, while adding a number on its own. why even bother to use
75 * it? because if conditions are right and your fds are supported and you 75 * it? because if conditions are right and your fds are supported and you
76 * don't hit a limit, this backend is actually faster, doesn't gamble with 76 * don't hit a limit, this backend is actually faster, doesn't gamble with
77 * your fds, batches watchers and events and doesn't require costly state 77 * your fds, batches watchers and events and doesn't require costly state
78 * recreates. well, until it does. 78 * recreates. well, until it does.
79 * g) all of this makes this backend use almost twice as much code as epoll. 79 * g) all of this makes this backend use almost twice as much code as epoll.
80 * which in turn uses twice as much code as poll. and thats not counting 80 * which in turn uses twice as much code as poll. and that#s not counting
81 * the fact that this backend also depends on the epoll backend, making 81 * the fact that this backend also depends on the epoll backend, making
82 * it three times as much code as poll, or kqueue. 82 * it three times as much code as poll, or kqueue.
83 * h) bleah. why can't linux just do kqueue. sure kqueue is ugly, but by now 83 * h) bleah. why can't linux just do kqueue. sure kqueue is ugly, but by now
84 * it's clear that whetaver linux comes up with is far, far, far worse. 84 * it's clear that whatever linux comes up with is far, far, far worse.
85 */ 85 */
86 86
87#include <sys/time.h> /* actually linux/time.h, but we must assume they are compatible */ 87#include <sys/time.h> /* actually linux/time.h, but we must assume they are compatible */
88#include <poll.h> 88#include <poll.h>
89#include <linux/aio_abi.h> 89#include <linux/aio_abi.h>
… …
100 * not only is this totally undocumented, not even the source code 100 * not only is this totally undocumented, not even the source code
101 * can tell you what the future semantics of compat_features and 101 * can tell you what the future semantics of compat_features and
102 * incompat_features are, or what header_length actually is for. 102 * incompat_features are, or what header_length actually is for.
103 */ 103 */
104#define AIO_RING_MAGIC 0xa10a10a1 104#define AIO_RING_MAGIC 0xa10a10a1
105#define AIO_RING_INCOMPAT_FEATURES 0 105#define EV_AIO_RING_INCOMPAT_FEATURES 0
106struct aio_ring 106struct aio_ring
107{ 107{
108 unsigned id; /* kernel internal index number */ 108 unsigned id; /* kernel internal index number */
109 unsigned nr; /* number of io_events */ 109 unsigned nr; /* number of io_events */
110 unsigned head; /* Written to by userland or by kernel. */ 110 unsigned head; /* Written to by userland or by kernel. */
… …
120 120
121inline_size 121inline_size
122int 122int
123evsys_io_setup (unsigned nr_events, aio_context_t *ctx_idp) 123evsys_io_setup (unsigned nr_events, aio_context_t *ctx_idp)
124{ 124{
125 return syscall (SYS_io_setup, nr_events, ctx_idp); 125 return ev_syscall2 (SYS_io_setup, nr_events, ctx_idp);
126} 126}
127 127
128inline_size 128inline_size
129int 129int
130evsys_io_destroy (aio_context_t ctx_id) 130evsys_io_destroy (aio_context_t ctx_id)
131{ 131{
132 return syscall (SYS_io_destroy, ctx_id); 132 return ev_syscall1 (SYS_io_destroy, ctx_id);
133} 133}
134 134
135inline_size 135inline_size
136int 136int
137evsys_io_submit (aio_context_t ctx_id, long nr, struct iocb *cbp[]) 137evsys_io_submit (aio_context_t ctx_id, long nr, struct iocb *cbp[])
138{ 138{
139 return syscall (SYS_io_submit, ctx_id, nr, cbp); 139 return ev_syscall3 (SYS_io_submit, ctx_id, nr, cbp);
140} 140}
141 141
142inline_size 142inline_size
143int 143int
144evsys_io_cancel (aio_context_t ctx_id, struct iocb *cbp, struct io_event *result) 144evsys_io_cancel (aio_context_t ctx_id, struct iocb *cbp, struct io_event *result)
145{ 145{
146 return syscall (SYS_io_cancel, ctx_id, cbp, result); 146 return ev_syscall3 (SYS_io_cancel, ctx_id, cbp, result);
147} 147}
148 148
149inline_size 149inline_size
150int 150int
151evsys_io_getevents (aio_context_t ctx_id, long min_nr, long nr, struct io_event *events, struct timespec *timeout) 151evsys_io_getevents (aio_context_t ctx_id, long min_nr, long nr, struct io_event *events, struct timespec *timeout)
152{ 152{
153 return syscall (SYS_io_getevents, ctx_id, min_nr, nr, events, timeout); 153 return ev_syscall5 (SYS_io_getevents, ctx_id, min_nr, nr, events, timeout);
154} 154}
155 155
156/*****************************************************************************/ 156/*****************************************************************************/
157/* actual backed implementation */ 157/* actual backed implementation */
158 158
… …
190 requests = requests / one_page * one_page + first_page; 190 requests = requests / one_page * one_page + first_page;
191 191
192 return requests; 192 return requests;
193} 193}
194 194
195/* we use out own wrapper structure in acse we ever want to do something "clever" */ 195/* we use out own wrapper structure in case we ever want to do something "clever" */
196typedef struct aniocb 196typedef struct aniocb
197{ 197{
198 struct iocb io; 198 struct iocb io;
199 /*int inuse;*/ 199 /*int inuse;*/
200} *ANIOCBP; 200} *ANIOCBP;
… …
203void 203void
204linuxaio_array_needsize_iocbp (ANIOCBP *base, int offset, int count) 204linuxaio_array_needsize_iocbp (ANIOCBP *base, int offset, int count)
205{ 205{
206 while (count--) 206 while (count--)
207 { 207 {
208 /* TODO: quite the overhead to allocate every iocb separately, maybe use our own alocator? */ 208 /* TODO: quite the overhead to allocate every iocb separately, maybe use our own allocator? */
209 ANIOCBP iocb = (ANIOCBP)ev_malloc (sizeof (*iocb)); 209 ANIOCBP iocb = (ANIOCBP)ev_malloc (sizeof (*iocb));
210 210
211 /* full zero initialise is probably not required at the moment, but 211 /* full zero initialise is probably not required at the moment, but
212 * this is not well documented, so we better do it. 212 * this is not well documented, so we better do it.
213 */ 213 */
214 memset (iocb, 0, sizeof (*iocb)); 214 memset (iocb, 0, sizeof (*iocb));
215 215
216 iocb->io.aio_lio_opcode = IOCB_CMD_POLL; 216 iocb->io.aio_lio_opcode = IOCB_CMD_POLL;
217 iocb->io.aio_data = offset;
218 iocb->io.aio_fildes = offset; 217 iocb->io.aio_fildes = offset;
219 218
220 base [offset++] = iocb; 219 base [offset++] = iocb;
221 } 220 }
222} 221}
… …
234static void 233static void
235linuxaio_modify (EV_P_ int fd, int oev, int nev) 234linuxaio_modify (EV_P_ int fd, int oev, int nev)
236{ 235{
237 array_needsize (ANIOCBP, linuxaio_iocbps, linuxaio_iocbpmax, fd + 1, linuxaio_array_needsize_iocbp); 236 array_needsize (ANIOCBP, linuxaio_iocbps, linuxaio_iocbpmax, fd + 1, linuxaio_array_needsize_iocbp);
238 ANIOCBP iocb = linuxaio_iocbps [fd]; 237 ANIOCBP iocb = linuxaio_iocbps [fd];
238 ANFD *anfd = &anfds [fd];
239 239
240 if (iocb->io.aio_reqprio < 0) 240 if (ecb_expect_false (iocb->io.aio_reqprio < 0))
241 { 241 {
242 /* we handed this fd over to epoll, so undo this first */ 242 /* we handed this fd over to epoll, so undo this first */
243 /* we do it manually becvause the optimisations on epoll_modfy won't do us any good */ 243 /* we do it manually because the optimisations on epoll_modify won't do us any good */
244 epoll_ctl (backend_fd, EPOLL_CTL_DEL, fd, 0); 244 epoll_ctl (backend_fd, EPOLL_CTL_DEL, fd, 0);
245 anfd->emask = 0;
245 iocb->io.aio_reqprio = 0; 246 iocb->io.aio_reqprio = 0;
246 } 247 }
247 248 else if (ecb_expect_false (iocb->io.aio_buf))
248 if (iocb->io.aio_buf) 249 {
249 /* io_cancel always returns some error on relevant kernels, but works */ 250 /* iocb active, so cancel it first before resubmit */
251 /* this assumes we only ever get one call per fd per loop iteration */
252 for (;;)
253 {
254 /* on all relevant kernels, io_cancel fails with EINPROGRESS on "success" */
250 evsys_io_cancel (linuxaio_ctx, &iocb->io, (struct io_event *)0); 255 if (ecb_expect_false (evsys_io_cancel (linuxaio_ctx, &iocb->io, (struct io_event *)0) == 0))
256 break;
257
258 if (ecb_expect_true (errno == EINPROGRESS))
259 break;
260
261 /* the EINPROGRESS test is for nicer error message. clumsy. */
262 if (errno != EINTR)
263 {
264 assert (("libev: linuxaio unexpected io_cancel failed", errno != EINTR && errno != EINPROGRESS));
265 break;
266 }
267 }
268
269 /* increment generation counter to avoid handling old events */
270 ++anfd->egen;
271 }
272
273 iocb->io.aio_buf = (nev & EV_READ ? POLLIN : 0)
274 | (nev & EV_WRITE ? POLLOUT : 0);
251 275
252 if (nev) 276 if (nev)
253 { 277 {
254 iocb->io.aio_buf = 278 iocb->io.aio_data = (uint32_t)fd | ((__u64)(uint32_t)anfd->egen << 32);
255 (nev & EV_READ ? POLLIN : 0)
256 | (nev & EV_WRITE ? POLLOUT : 0);
257 279
258 /* queue iocb up for io_submit */ 280 /* queue iocb up for io_submit */
259 /* this assumes we only ever get one call per fd per loop iteration */ 281 /* this assumes we only ever get one call per fd per loop iteration */
260 ++linuxaio_submitcnt; 282 ++linuxaio_submitcnt;
261 array_needsize (struct iocb *, linuxaio_submits, linuxaio_submitmax, linuxaio_submitcnt, array_needsize_noinit); 283 array_needsize (struct iocb *, linuxaio_submits, linuxaio_submitmax, linuxaio_submitcnt, array_needsize_noinit);
… …
267linuxaio_epoll_cb (EV_P_ struct ev_io *w, int revents) 289linuxaio_epoll_cb (EV_P_ struct ev_io *w, int revents)
268{ 290{
269 epoll_poll (EV_A_ 0); 291 epoll_poll (EV_A_ 0);
270} 292}
271 293
272static void 294inline_speed
295void
273linuxaio_fd_rearm (EV_P_ int fd) 296linuxaio_fd_rearm (EV_P_ int fd)
274{ 297{
275 anfds [fd].events = 0; 298 anfds [fd].events = 0;
276 linuxaio_iocbps [fd]->io.aio_buf = 0; 299 linuxaio_iocbps [fd]->io.aio_buf = 0;
277 fd_change (EV_A_ fd, EV_ANFD_REIFY); 300 fd_change (EV_A_ fd, EV_ANFD_REIFY);
… …
280static void 303static void
281linuxaio_parse_events (EV_P_ struct io_event *ev, int nr) 304linuxaio_parse_events (EV_P_ struct io_event *ev, int nr)
282{ 305{
283 while (nr) 306 while (nr)
284 { 307 {
285 int fd = ev->data; 308 int fd = ev->data & 0xffffffff;
309 uint32_t gen = ev->data >> 32;
286 int res = ev->res; 310 int res = ev->res;
287 311
288 assert (("libev: iocb fd must be in-bounds", fd >= 0 && fd < anfdmax)); 312 assert (("libev: iocb fd must be in-bounds", fd >= 0 && fd < anfdmax));
289 313
314 /* only accept events if generation counter matches */
315 if (ecb_expect_true (gen == (uint32_t)anfds [fd].egen))
316 {
290 /* feed events, we do not expect or handle POLLNVAL */ 317 /* feed events, we do not expect or handle POLLNVAL */
291 fd_event ( 318 fd_event (
292 EV_A_ 319 EV_A_
293 fd, 320 fd,
294 (res & (POLLOUT | POLLERR | POLLHUP) ? EV_WRITE : 0) 321 (res & (POLLOUT | POLLERR | POLLHUP) ? EV_WRITE : 0)
295 | (res & (POLLIN | POLLERR | POLLHUP) ? EV_READ : 0) 322 | (res & (POLLIN | POLLERR | POLLHUP) ? EV_READ : 0)
296 ); 323 );
297 324
298 /* linux aio is oneshot: rearm fd. TODO: this does more work than needed */ 325 /* linux aio is oneshot: rearm fd. TODO: this does more work than strictly needed */
299 linuxaio_fd_rearm (EV_A_ fd); 326 linuxaio_fd_rearm (EV_A_ fd);
327 }
300 328
301 --nr; 329 --nr;
302 ++ev; 330 ++ev;
303 } 331 }
304} 332}
305 333
306/* get any events from ringbuffer, return true if any were handled */ 334/* get any events from ring buffer, return true if any were handled */
307static int 335static int
308linuxaio_get_events_from_ring (EV_P) 336linuxaio_get_events_from_ring (EV_P)
309{ 337{
310 struct aio_ring *ring = (struct aio_ring *)linuxaio_ctx; 338 struct aio_ring *ring = (struct aio_ring *)linuxaio_ctx;
339 unsigned head, tail;
311 340
312 /* the kernel reads and writes both of these variables, */ 341 /* the kernel reads and writes both of these variables, */
313 /* as a C extension, we assume that volatile use here */ 342 /* as a C extension, we assume that volatile use here */
314 /* both makes reads atomic and once-only */ 343 /* both makes reads atomic and once-only */
315 unsigned head = *(volatile unsigned *)&ring->head; 344 head = *(volatile unsigned *)&ring->head;
345 ECB_MEMORY_FENCE_ACQUIRE;
316 unsigned tail = *(volatile unsigned *)&ring->tail; 346 tail = *(volatile unsigned *)&ring->tail;
317 347
318 if (head == tail) 348 if (head == tail)
319 return 0; 349 return 0;
320 350
321 /* bail out if the ring buffer doesn't match the expected layout */
322 if (expect_false (ring->magic != AIO_RING_MAGIC)
323 || ring->incompat_features != AIO_RING_INCOMPAT_FEATURES
324 || ring->header_length != sizeof (struct aio_ring)) /* TODO: or use it to find io_event[0]? */
325 return 0;
326
327 /* make sure the events up to tail are visible */
328 ECB_MEMORY_FENCE_ACQUIRE;
329
330 /* parse all available events, but only once, to avoid starvation */ 351 /* parse all available events, but only once, to avoid starvation */
331 if (tail > head) /* normal case around */ 352 if (ecb_expect_true (tail > head)) /* normal case around */
332 linuxaio_parse_events (EV_A_ ring->io_events + head, tail - head); 353 linuxaio_parse_events (EV_A_ ring->io_events + head, tail - head);
333 else /* wrapped around */ 354 else /* wrapped around */
334 { 355 {
335 linuxaio_parse_events (EV_A_ ring->io_events + head, ring->nr - head); 356 linuxaio_parse_events (EV_A_ ring->io_events + head, ring->nr - head);
336 linuxaio_parse_events (EV_A_ ring->io_events, tail); 357 linuxaio_parse_events (EV_A_ ring->io_events, tail);
337 } 358 }
338 359
339 ECB_MEMORY_FENCE_RELAXED; 360 ECB_MEMORY_FENCE_RELEASE;
340 /* as an extension to C, we hope that the volatile will make this atomic and once-only */ 361 /* as an extension to C, we hope that the volatile will make this atomic and once-only */
341 *(volatile unsigned *)&ring->head = tail; 362 *(volatile unsigned *)&ring->head = tail;
342 /* make sure kernel can see our new head value - probably not required */
343 ECB_MEMORY_FENCE_RELEASE;
344 363
345 return 1; 364 return 1;
365}
366
367inline_size
368int
369linuxaio_ringbuf_valid (EV_P)
370{
371 struct aio_ring *ring = (struct aio_ring *)linuxaio_ctx;
372
373 return ecb_expect_true (ring->magic == AIO_RING_MAGIC)
374 && ring->incompat_features == EV_AIO_RING_INCOMPAT_FEATURES
375 && ring->header_length == sizeof (struct aio_ring); /* TODO: or use it to find io_event[0]? */
346} 376}
347 377
348/* read at least one event from kernel, or timeout */ 378/* read at least one event from kernel, or timeout */
349inline_size 379inline_size
350void 380void
351linuxaio_get_events (EV_P_ ev_tstamp timeout) 381linuxaio_get_events (EV_P_ ev_tstamp timeout)
352{ 382{
353 struct timespec ts; 383 struct timespec ts;
354 struct io_event ioev[1]; 384 struct io_event ioev[8]; /* 256 octet stack space */
355 int res; 385 int want = 1; /* how many events to request */
386 int ringbuf_valid = linuxaio_ringbuf_valid (EV_A);
356 387
388 if (ecb_expect_true (ringbuf_valid))
389 {
390 /* if the ring buffer has any events, we don't wait or call the kernel at all */
357 if (linuxaio_get_events_from_ring (EV_A)) 391 if (linuxaio_get_events_from_ring (EV_A))
358 return; 392 return;
359 393
360 /* no events, so wait for at least one, then poll ring buffer again */ 394 /* if the ring buffer is empty, and we don't have a timeout, then don't call the kernel */
361 /* this degrades to one event per loop iteration */ 395 if (!timeout)
362 /* if the ring buffer changes layout, but so be it */ 396 return;
397 }
398 else
399 /* no ringbuffer, request slightly larger batch */
400 want = sizeof (ioev) / sizeof (ioev [0]);
363 401
402 /* no events, so wait for some
403 * for fairness reasons, we do this in a loop, to fetch all events
404 */
405 for (;;)
406 {
407 int res;
408
364 EV_RELEASE_CB; 409 EV_RELEASE_CB;
365 410
366 ts.tv_sec = (long)timeout; 411 EV_TS_SET (ts, timeout);
367 ts.tv_nsec = (long)((timeout - ts.tv_sec) * 1e9); 412 res = evsys_io_getevents (linuxaio_ctx, 1, want, ioev, &ts);
368 413
369 res = evsys_io_getevents (linuxaio_ctx, 1, sizeof (ioev) / sizeof (ioev [0]), ioev, &ts);
370
371 EV_ACQUIRE_CB; 414 EV_ACQUIRE_CB;
372 415
373 if (res < 0) 416 if (res < 0)
374 if (errno == EINTR) 417 if (errno == EINTR)
375 /* ignored */; 418 /* ignored, retry */;
376 else 419 else
377 ev_syserr ("(libev) linuxaio io_getevents"); 420 ev_syserr ("(libev) linuxaio io_getevents");
378 else if (res) 421 else if (res)
379 { 422 {
380 /* at least one event received, handle it and any remaining ones in the ring buffer */ 423 /* at least one event available, handle them */
381 linuxaio_parse_events (EV_A_ ioev, res); 424 linuxaio_parse_events (EV_A_ ioev, res);
425
426 if (ecb_expect_true (ringbuf_valid))
427 {
428 /* if we have a ring buffer, handle any remaining events in it */
382 linuxaio_get_events_from_ring (EV_A); 429 linuxaio_get_events_from_ring (EV_A);
383 }
384}
385 430
386static int 431 /* at this point, we should have handled all outstanding events */
432 break;
433 }
434 else if (res < want)
435 /* otherwise, if there were fewere events than we wanted, we assume there are no more */
436 break;
437 }
438 else
439 break; /* no events from the kernel, we are done */
440
441 timeout = EV_TS_CONST (0.); /* only wait in the first iteration */
442 }
443}
444
445inline_size
446int
387linuxaio_io_setup (EV_P) 447linuxaio_io_setup (EV_P)
388{ 448{
389 linuxaio_ctx = 0; 449 linuxaio_ctx = 0;
390 return evsys_io_setup (linuxaio_nr_events (EV_A), &linuxaio_ctx); 450 return evsys_io_setup (linuxaio_nr_events (EV_A), &linuxaio_ctx);
391} 451}
… …
397 457
398 /* first phase: submit new iocbs */ 458 /* first phase: submit new iocbs */
399 459
400 /* io_submit might return less than the requested number of iocbs */ 460 /* io_submit might return less than the requested number of iocbs */
401 /* this is, afaics, only because of errors, but we go by the book and use a loop, */ 461 /* this is, afaics, only because of errors, but we go by the book and use a loop, */
402 /* which allows us to pinpoint the errornous iocb */ 462 /* which allows us to pinpoint the erroneous iocb */
403 for (submitted = 0; submitted < linuxaio_submitcnt; ) 463 for (submitted = 0; submitted < linuxaio_submitcnt; )
404 { 464 {
405 int res = evsys_io_submit (linuxaio_ctx, linuxaio_submitcnt - submitted, linuxaio_submits + submitted); 465 int res = evsys_io_submit (linuxaio_ctx, linuxaio_submitcnt - submitted, linuxaio_submits + submitted);
406 466
407 if (expect_false (res < 0)) 467 if (ecb_expect_false (res < 0))
408 if (errno == EINVAL) 468 if (errno == EINVAL)
409 { 469 {
410 /* This happens for unsupported fds, officially, but in my testing, 470 /* This happens for unsupported fds, officially, but in my testing,
411 * also randomly happens for supported fds. We fall back to good old 471 * also randomly happens for supported fds. We fall back to good old
412 * poll() here, under the assumption that this is a very rare case. 472 * poll() here, under the assumption that this is a very rare case.
… …
421 res = 1; /* skip this iocb - another iocb, another chance */ 481 res = 1; /* skip this iocb - another iocb, another chance */
422 } 482 }
423 else if (errno == EAGAIN) 483 else if (errno == EAGAIN)
424 { 484 {
425 /* This happens when the ring buffer is full, or some other shit we 485 /* This happens when the ring buffer is full, or some other shit we
426 * dont' know and isn't documented. Most likely because we have too 486 * don't know and isn't documented. Most likely because we have too
427 * many requests and linux aio can't be assed to handle them. 487 * many requests and linux aio can't be assed to handle them.
428 * In this case, we try to allocate a larger ring buffer, freeing 488 * In this case, we try to allocate a larger ring buffer, freeing
429 * ours first. This might fail, in which case we have to fall back to 100% 489 * ours first. This might fail, in which case we have to fall back to 100%
430 * epoll. 490 * epoll.
431 * God, how I hate linux not getting its act together. Ever. 491 * God, how I hate linux not getting its act together. Ever.
… …
442 } 502 }
443 503
444 ++linuxaio_iteration; 504 ++linuxaio_iteration;
445 if (linuxaio_io_setup (EV_A) < 0) 505 if (linuxaio_io_setup (EV_A) < 0)
446 { 506 {
507 /* TODO: rearm all and recreate epoll backend from scratch */
508 /* TODO: might be more prudent? */
509
447 /* to bad, we can't get a new aio context, go 100% epoll */ 510 /* to bad, we can't get a new aio context, go 100% epoll */
448 linuxaio_free_iocbp (EV_A); 511 linuxaio_free_iocbp (EV_A);
449 ev_io_stop (EV_A_ &linuxaio_epoll_w); 512 ev_io_stop (EV_A_ &linuxaio_epoll_w);
450 ev_ref (EV_A); 513 ev_ref (EV_A);
451 linuxaio_ctx = 0; 514 linuxaio_ctx = 0;
515
516 backend = EVBACKEND_EPOLL;
452 backend_modify = epoll_modify; 517 backend_modify = epoll_modify;
453 backend_poll = epoll_poll; 518 backend_poll = epoll_poll;
454 } 519 }
455 520
456 timeout = 0; 521 timeout = EV_TS_CONST (0.);
457 /* it's easiest to handle this mess in another iteration */ 522 /* it's easiest to handle this mess in another iteration */
458 return; 523 return;
459 } 524 }
460 else if (errno == EBADF) 525 else if (errno == EBADF)
461 { 526 {
527 assert (("libev: event loop rejected bad fd", errno != EBADF));
462 fd_kill (EV_A_ linuxaio_submits [submitted]->aio_fildes); 528 fd_kill (EV_A_ linuxaio_submits [submitted]->aio_fildes);
463 529
464 res = 1; /* skip this iocb */ 530 res = 1; /* skip this iocb */
465 } 531 }
532 else if (errno == EINTR) /* not seen in reality, not documented */
533 res = 0; /* silently ignore and retry */
466 else 534 else
535 {
467 ev_syserr ("(libev) linuxaio io_submit"); 536 ev_syserr ("(libev) linuxaio io_submit");
537 res = 0;
538 }
468 539
469 submitted += res; 540 submitted += res;
470 } 541 }
471 542
472 linuxaio_submitcnt = 0; 543 linuxaio_submitcnt = 0;
… …
480int 551int
481linuxaio_init (EV_P_ int flags) 552linuxaio_init (EV_P_ int flags)
482{ 553{
483 /* would be great to have a nice test for IOCB_CMD_POLL instead */ 554 /* would be great to have a nice test for IOCB_CMD_POLL instead */
484 /* also: test some semi-common fd types, such as files and ttys in recommended_backends */ 555 /* also: test some semi-common fd types, such as files and ttys in recommended_backends */
485 /* 4.18 introduced IOCB_CMD_POLL, 4.19 made epoll work */ 556 /* 4.18 introduced IOCB_CMD_POLL, 4.19 made epoll work, and we need that */
486 if (ev_linux_version () < 0x041300) 557 if (ev_linux_version () < 0x041300)
487 return 0; 558 return 0;
488 559
489 if (!epoll_init (EV_A_ 0)) 560 if (!epoll_init (EV_A_ 0))
490 return 0; 561 return 0;
… …
495 { 566 {
496 epoll_destroy (EV_A); 567 epoll_destroy (EV_A);
497 return 0; 568 return 0;
498 } 569 }
499 570
500 ev_io_init (EV_A_ &linuxaio_epoll_w, linuxaio_epoll_cb, backend_fd, EV_READ); 571 ev_io_init (&linuxaio_epoll_w, linuxaio_epoll_cb, backend_fd, EV_READ);
501 ev_set_priority (&linuxaio_epoll_w, EV_MAXPRI); 572 ev_set_priority (&linuxaio_epoll_w, EV_MAXPRI);
502 ev_io_start (EV_A_ &linuxaio_epoll_w); 573 ev_io_start (EV_A_ &linuxaio_epoll_w);
503 ev_unref (EV_A); /* watcher should not keep loop alive */ 574 ev_unref (EV_A); /* watcher should not keep loop alive */
504 575
505 backend_modify = linuxaio_modify; 576 backend_modify = linuxaio_modify;
506 backend_poll = linuxaio_poll; 577 backend_poll = linuxaio_poll;
507 578
508 linuxaio_iocbpmax = 0; 579 linuxaio_iocbpmax = 0;
509 linuxaio_iocbps = 0; 580 linuxaio_iocbps = 0;
510 581
511 linuxaio_submits = 0; 582 linuxaio_submits = 0;
… …
519void 590void
520linuxaio_destroy (EV_P) 591linuxaio_destroy (EV_P)
521{ 592{
522 epoll_destroy (EV_A); 593 epoll_destroy (EV_A);
523 linuxaio_free_iocbp (EV_A); 594 linuxaio_free_iocbp (EV_A);
524 evsys_io_destroy (linuxaio_ctx); 595 evsys_io_destroy (linuxaio_ctx); /* fails in child, aio context is destroyed */
525} 596}
526 597
527inline_size 598ecb_cold
528void 599static void
529linuxaio_fork (EV_P) 600linuxaio_fork (EV_P)
530{ 601{
531 /* this frees all iocbs, which is very heavy-handed */
532 linuxaio_destroy (EV_A);
533 linuxaio_submitcnt = 0; /* all pointers were invalidated */ 602 linuxaio_submitcnt = 0; /* all pointers were invalidated */
603 linuxaio_free_iocbp (EV_A); /* this frees all iocbs, which is very heavy-handed */
604 evsys_io_destroy (linuxaio_ctx); /* fails in child, aio context is destroyed */
534 605
535 linuxaio_iteration = 0; /* we start over in the child */ 606 linuxaio_iteration = 0; /* we start over in the child */
536 607
537 while (linuxaio_io_setup (EV_A) < 0) 608 while (linuxaio_io_setup (EV_A) < 0)
538 ev_syserr ("(libev) linuxaio io_setup"); 609 ev_syserr ("(libev) linuxaio io_setup");
539 610
611 /* forking epoll should also effectively unregister all fds from the backend */
540 epoll_fork (EV_A); 612 epoll_fork (EV_A);
613 /* epoll_fork already did this. hopefully */
614 /*fd_rearm_all (EV_A);*/
541 615
542 ev_io_stop (EV_A_ &linuxaio_epoll_w); 616 ev_io_stop (EV_A_ &linuxaio_epoll_w);
543 ev_io_set (EV_A_ &linuxaio_epoll_w, backend_fd, EV_READ); 617 ev_io_set (EV_A_ &linuxaio_epoll_w, backend_fd, EV_READ);
544 ev_io_start (EV_A_ &linuxaio_epoll_w); 618 ev_io_start (EV_A_ &linuxaio_epoll_w);
545
546 /* epoll_fork already did this. hopefully */
547 /*fd_rearm_all (EV_A);*/
548} 619}
549 620

Diff Legend

– Removed lines
+ Added lines
< Changed lines
> Changed lines