eal_interrupts.c revision 8e6d9d11
1/*-
2 *   BSD LICENSE
3 *
4 *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
5 *   All rights reserved.
6 *
7 *   Redistribution and use in source and binary forms, with or without
8 *   modification, are permitted provided that the following conditions
9 *   are met:
10 *
11 *     * Redistributions of source code must retain the above copyright
12 *       notice, this list of conditions and the following disclaimer.
13 *     * Redistributions in binary form must reproduce the above copyright
14 *       notice, this list of conditions and the following disclaimer in
15 *       the documentation and/or other materials provided with the
16 *       distribution.
17 *     * Neither the name of Intel Corporation nor the names of its
18 *       contributors may be used to endorse or promote products derived
19 *       from this software without specific prior written permission.
20 *
21 *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
32 */
33
34#include <stdio.h>
35#include <stdint.h>
36#include <stdlib.h>
37#include <pthread.h>
38#include <sys/queue.h>
39#include <stdarg.h>
40#include <unistd.h>
41#include <string.h>
42#include <errno.h>
43#include <inttypes.h>
44#include <sys/epoll.h>
45#include <sys/signalfd.h>
46#include <sys/ioctl.h>
47#include <sys/eventfd.h>
48#include <assert.h>
49
50#include <rte_common.h>
51#include <rte_interrupts.h>
52#include <rte_memory.h>
53#include <rte_memzone.h>
54#include <rte_launch.h>
55#include <rte_eal.h>
56#include <rte_per_lcore.h>
57#include <rte_lcore.h>
58#include <rte_atomic.h>
59#include <rte_branch_prediction.h>
60#include <rte_debug.h>
61#include <rte_log.h>
62#include <rte_pci.h>
63#include <rte_malloc.h>
64#include <rte_errno.h>
65#include <rte_spinlock.h>
66
67#include "eal_private.h"
68#include "eal_vfio.h"
69#include "eal_thread.h"
70
71#define EAL_INTR_EPOLL_WAIT_FOREVER (-1)
72#define NB_OTHER_INTR               1
73
74static RTE_DEFINE_PER_LCORE(int, _epfd) = -1; /**< epoll fd per thread */
75
76/**
77 * union for pipe fds.
78 */
79union intr_pipefds{
80	struct {
81		int pipefd[2];
82	};
83	struct {
84		int readfd;
85		int writefd;
86	};
87};
88
89/**
90 * union buffer for reading on different devices
91 */
92union rte_intr_read_buffer {
93	int uio_intr_count;              /* for uio device */
94#ifdef VFIO_PRESENT
95	uint64_t vfio_intr_count;        /* for vfio device */
96#endif
97	uint64_t timerfd_num;            /* for timerfd */
98	char charbuf[16];                /* for others */
99};
100
101TAILQ_HEAD(rte_intr_cb_list, rte_intr_callback);
102TAILQ_HEAD(rte_intr_source_list, rte_intr_source);
103
104struct rte_intr_callback {
105	TAILQ_ENTRY(rte_intr_callback) next;
106	rte_intr_callback_fn cb_fn;  /**< callback address */
107	void *cb_arg;                /**< parameter for callback */
108};
109
110struct rte_intr_source {
111	TAILQ_ENTRY(rte_intr_source) next;
112	struct rte_intr_handle intr_handle; /**< interrupt handle */
113	struct rte_intr_cb_list callbacks;  /**< user callbacks */
114	uint32_t active;
115};
116
117/* global spinlock for interrupt data operation */
118static rte_spinlock_t intr_lock = RTE_SPINLOCK_INITIALIZER;
119
120/* union buffer for pipe read/write */
121static union intr_pipefds intr_pipe;
122
123/* interrupt sources list */
124static struct rte_intr_source_list intr_sources;
125
126/* interrupt handling thread */
127static pthread_t intr_thread;
128
129/* VFIO interrupts */
130#ifdef VFIO_PRESENT
131
132#define IRQ_SET_BUF_LEN  (sizeof(struct vfio_irq_set) + sizeof(int))
133/* irq set buffer length for queue interrupts and LSC interrupt */
134#define MSIX_IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + \
135			      sizeof(int) * (RTE_MAX_RXTX_INTR_VEC_ID + 1))
136
137/* enable legacy (INTx) interrupts */
138static int
139vfio_enable_intx(struct rte_intr_handle *intr_handle) {
140	struct vfio_irq_set *irq_set;
141	char irq_set_buf[IRQ_SET_BUF_LEN];
142	int len, ret;
143	int *fd_ptr;
144
145	len = sizeof(irq_set_buf);
146
147	/* enable INTx */
148	irq_set = (struct vfio_irq_set *) irq_set_buf;
149	irq_set->argsz = len;
150	irq_set->count = 1;
151	irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
152	irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
153	irq_set->start = 0;
154	fd_ptr = (int *) &irq_set->data;
155	*fd_ptr = intr_handle->fd;
156
157	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
158
159	if (ret) {
160		RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
161						intr_handle->fd);
162		return -1;
163	}
164
165	/* unmask INTx after enabling */
166	memset(irq_set, 0, len);
167	len = sizeof(struct vfio_irq_set);
168	irq_set->argsz = len;
169	irq_set->count = 1;
170	irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
171	irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
172	irq_set->start = 0;
173
174	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
175
176	if (ret) {
177		RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
178						intr_handle->fd);
179		return -1;
180	}
181	return 0;
182}
183
184/* disable legacy (INTx) interrupts */
185static int
186vfio_disable_intx(struct rte_intr_handle *intr_handle) {
187	struct vfio_irq_set *irq_set;
188	char irq_set_buf[IRQ_SET_BUF_LEN];
189	int len, ret;
190
191	len = sizeof(struct vfio_irq_set);
192
193	/* mask interrupts before disabling */
194	irq_set = (struct vfio_irq_set *) irq_set_buf;
195	irq_set->argsz = len;
196	irq_set->count = 1;
197	irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_MASK;
198	irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
199	irq_set->start = 0;
200
201	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
202
203	if (ret) {
204		RTE_LOG(ERR, EAL, "Error masking INTx interrupts for fd %d\n",
205						intr_handle->fd);
206		return -1;
207	}
208
209	/* disable INTx*/
210	memset(irq_set, 0, len);
211	irq_set->argsz = len;
212	irq_set->count = 0;
213	irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
214	irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
215	irq_set->start = 0;
216
217	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
218
219	if (ret) {
220		RTE_LOG(ERR, EAL,
221			"Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
222		return -1;
223	}
224	return 0;
225}
226
227/* enable MSI interrupts */
228static int
229vfio_enable_msi(struct rte_intr_handle *intr_handle) {
230	int len, ret;
231	char irq_set_buf[IRQ_SET_BUF_LEN];
232	struct vfio_irq_set *irq_set;
233	int *fd_ptr;
234
235	len = sizeof(irq_set_buf);
236
237	irq_set = (struct vfio_irq_set *) irq_set_buf;
238	irq_set->argsz = len;
239	irq_set->count = 1;
240	irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
241	irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
242	irq_set->start = 0;
243	fd_ptr = (int *) &irq_set->data;
244	*fd_ptr = intr_handle->fd;
245
246	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
247
248	if (ret) {
249		RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
250						intr_handle->fd);
251		return -1;
252	}
253	return 0;
254}
255
256/* disable MSI interrupts */
257static int
258vfio_disable_msi(struct rte_intr_handle *intr_handle) {
259	struct vfio_irq_set *irq_set;
260	char irq_set_buf[IRQ_SET_BUF_LEN];
261	int len, ret;
262
263	len = sizeof(struct vfio_irq_set);
264
265	irq_set = (struct vfio_irq_set *) irq_set_buf;
266	irq_set->argsz = len;
267	irq_set->count = 0;
268	irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
269	irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
270	irq_set->start = 0;
271
272	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
273
274	if (ret)
275		RTE_LOG(ERR, EAL,
276			"Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
277
278	return ret;
279}
280
281/* enable MSI-X interrupts */
282static int
283vfio_enable_msix(struct rte_intr_handle *intr_handle) {
284	int len, ret;
285	char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
286	struct vfio_irq_set *irq_set;
287	int *fd_ptr;
288
289	len = sizeof(irq_set_buf);
290
291	irq_set = (struct vfio_irq_set *) irq_set_buf;
292	irq_set->argsz = len;
293	if (!intr_handle->max_intr)
294		intr_handle->max_intr = 1;
295	else if (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
296		intr_handle->max_intr = RTE_MAX_RXTX_INTR_VEC_ID + 1;
297
298	irq_set->count = intr_handle->max_intr;
299	irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
300	irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
301	irq_set->start = 0;
302	fd_ptr = (int *) &irq_set->data;
303	/* INTR vector offset 0 reserve for non-efds mapping */
304	fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
305	memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
306		sizeof(*intr_handle->efds) * intr_handle->nb_efd);
307
308	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
309
310	if (ret) {
311		RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
312						intr_handle->fd);
313		return -1;
314	}
315
316	return 0;
317}
318
319/* disable MSI-X interrupts */
320static int
321vfio_disable_msix(struct rte_intr_handle *intr_handle) {
322	struct vfio_irq_set *irq_set;
323	char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
324	int len, ret;
325
326	len = sizeof(struct vfio_irq_set);
327
328	irq_set = (struct vfio_irq_set *) irq_set_buf;
329	irq_set->argsz = len;
330	irq_set->count = 0;
331	irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
332	irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
333	irq_set->start = 0;
334
335	ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
336
337	if (ret)
338		RTE_LOG(ERR, EAL,
339			"Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
340
341	return ret;
342}
343#endif
344
345static int
346uio_intx_intr_disable(struct rte_intr_handle *intr_handle)
347{
348	unsigned char command_high;
349
350	/* use UIO config file descriptor for uio_pci_generic */
351	if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
352		RTE_LOG(ERR, EAL,
353			"Error reading interrupts status for fd %d\n",
354			intr_handle->uio_cfg_fd);
355		return -1;
356	}
357	/* disable interrupts */
358	command_high |= 0x4;
359	if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
360		RTE_LOG(ERR, EAL,
361			"Error disabling interrupts for fd %d\n",
362			intr_handle->uio_cfg_fd);
363		return -1;
364	}
365
366	return 0;
367}
368
369static int
370uio_intx_intr_enable(struct rte_intr_handle *intr_handle)
371{
372	unsigned char command_high;
373
374	/* use UIO config file descriptor for uio_pci_generic */
375	if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
376		RTE_LOG(ERR, EAL,
377			"Error reading interrupts status for fd %d\n",
378			intr_handle->uio_cfg_fd);
379		return -1;
380	}
381	/* enable interrupts */
382	command_high &= ~0x4;
383	if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
384		RTE_LOG(ERR, EAL,
385			"Error enabling interrupts for fd %d\n",
386			intr_handle->uio_cfg_fd);
387		return -1;
388	}
389
390	return 0;
391}
392
393static int
394uio_intr_disable(struct rte_intr_handle *intr_handle)
395{
396	const int value = 0;
397
398	if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
399		RTE_LOG(ERR, EAL,
400			"Error disabling interrupts for fd %d (%s)\n",
401			intr_handle->fd, strerror(errno));
402		return -1;
403	}
404	return 0;
405}
406
407static int
408uio_intr_enable(struct rte_intr_handle *intr_handle)
409{
410	const int value = 1;
411
412	if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
413		RTE_LOG(ERR, EAL,
414			"Error enabling interrupts for fd %d (%s)\n",
415			intr_handle->fd, strerror(errno));
416		return -1;
417	}
418	return 0;
419}
420
421int
422rte_intr_callback_register(struct rte_intr_handle *intr_handle,
423			rte_intr_callback_fn cb, void *cb_arg)
424{
425	int ret, wake_thread;
426	struct rte_intr_source *src;
427	struct rte_intr_callback *callback;
428
429	wake_thread = 0;
430
431	/* first do parameter checking */
432	if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
433		RTE_LOG(ERR, EAL,
434			"Registering with invalid input parameter\n");
435		return -EINVAL;
436	}
437
438	/* allocate a new interrupt callback entity */
439	callback = rte_zmalloc("interrupt callback list",
440				sizeof(*callback), 0);
441	if (callback == NULL) {
442		RTE_LOG(ERR, EAL, "Can not allocate memory\n");
443		return -ENOMEM;
444	}
445	callback->cb_fn = cb;
446	callback->cb_arg = cb_arg;
447
448	rte_spinlock_lock(&intr_lock);
449
450	/* check if there is at least one callback registered for the fd */
451	TAILQ_FOREACH(src, &intr_sources, next) {
452		if (src->intr_handle.fd == intr_handle->fd) {
453			/* we had no interrupts for this */
454			if (TAILQ_EMPTY(&src->callbacks))
455				wake_thread = 1;
456
457			TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
458			ret = 0;
459			break;
460		}
461	}
462
463	/* no existing callbacks for this - add new source */
464	if (src == NULL) {
465		if ((src = rte_zmalloc("interrupt source list",
466				sizeof(*src), 0)) == NULL) {
467			RTE_LOG(ERR, EAL, "Can not allocate memory\n");
468			rte_free(callback);
469			ret = -ENOMEM;
470		} else {
471			src->intr_handle = *intr_handle;
472			TAILQ_INIT(&src->callbacks);
473			TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
474			TAILQ_INSERT_TAIL(&intr_sources, src, next);
475			wake_thread = 1;
476			ret = 0;
477		}
478	}
479
480	rte_spinlock_unlock(&intr_lock);
481
482	/**
483	 * check if need to notify the pipe fd waited by epoll_wait to
484	 * rebuild the wait list.
485	 */
486	if (wake_thread)
487		if (write(intr_pipe.writefd, "1", 1) < 0)
488			return -EPIPE;
489
490	return ret;
491}
492
493int
494rte_intr_callback_unregister(struct rte_intr_handle *intr_handle,
495			rte_intr_callback_fn cb_fn, void *cb_arg)
496{
497	int ret;
498	struct rte_intr_source *src;
499	struct rte_intr_callback *cb, *next;
500
501	/* do parameter checking first */
502	if (intr_handle == NULL || intr_handle->fd < 0) {
503		RTE_LOG(ERR, EAL,
504		"Unregistering with invalid input parameter\n");
505		return -EINVAL;
506	}
507
508	rte_spinlock_lock(&intr_lock);
509
510	/* check if the insterrupt source for the fd is existent */
511	TAILQ_FOREACH(src, &intr_sources, next)
512		if (src->intr_handle.fd == intr_handle->fd)
513			break;
514
515	/* No interrupt source registered for the fd */
516	if (src == NULL) {
517		ret = -ENOENT;
518
519	/* interrupt source has some active callbacks right now. */
520	} else if (src->active != 0) {
521		ret = -EAGAIN;
522
523	/* ok to remove. */
524	} else {
525		ret = 0;
526
527		/*walk through the callbacks and remove all that match. */
528		for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
529
530			next = TAILQ_NEXT(cb, next);
531
532			if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
533					cb->cb_arg == cb_arg)) {
534				TAILQ_REMOVE(&src->callbacks, cb, next);
535				rte_free(cb);
536				ret++;
537			}
538		}
539
540		/* all callbacks for that source are removed. */
541		if (TAILQ_EMPTY(&src->callbacks)) {
542			TAILQ_REMOVE(&intr_sources, src, next);
543			rte_free(src);
544		}
545	}
546
547	rte_spinlock_unlock(&intr_lock);
548
549	/* notify the pipe fd waited by epoll_wait to rebuild the wait list */
550	if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
551		ret = -EPIPE;
552	}
553
554	return ret;
555}
556
557int
558rte_intr_enable(struct rte_intr_handle *intr_handle)
559{
560	if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
561		return -1;
562
563	switch (intr_handle->type){
564	/* write to the uio fd to enable the interrupt */
565	case RTE_INTR_HANDLE_UIO:
566		if (uio_intr_enable(intr_handle))
567			return -1;
568		break;
569	case RTE_INTR_HANDLE_UIO_INTX:
570		if (uio_intx_intr_enable(intr_handle))
571			return -1;
572		break;
573	/* not used at this moment */
574	case RTE_INTR_HANDLE_ALARM:
575		return -1;
576#ifdef VFIO_PRESENT
577	case RTE_INTR_HANDLE_VFIO_MSIX:
578		if (vfio_enable_msix(intr_handle))
579			return -1;
580		break;
581	case RTE_INTR_HANDLE_VFIO_MSI:
582		if (vfio_enable_msi(intr_handle))
583			return -1;
584		break;
585	case RTE_INTR_HANDLE_VFIO_LEGACY:
586		if (vfio_enable_intx(intr_handle))
587			return -1;
588		break;
589#endif
590	/* unknown handle type */
591	default:
592		RTE_LOG(ERR, EAL,
593			"Unknown handle type of fd %d\n",
594					intr_handle->fd);
595		return -1;
596	}
597
598	return 0;
599}
600
601int
602rte_intr_disable(struct rte_intr_handle *intr_handle)
603{
604	if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
605		return -1;
606
607	switch (intr_handle->type){
608	/* write to the uio fd to disable the interrupt */
609	case RTE_INTR_HANDLE_UIO:
610		if (uio_intr_disable(intr_handle))
611			return -1;
612		break;
613	case RTE_INTR_HANDLE_UIO_INTX:
614		if (uio_intx_intr_disable(intr_handle))
615			return -1;
616		break;
617	/* not used at this moment */
618	case RTE_INTR_HANDLE_ALARM:
619		return -1;
620#ifdef VFIO_PRESENT
621	case RTE_INTR_HANDLE_VFIO_MSIX:
622		if (vfio_disable_msix(intr_handle))
623			return -1;
624		break;
625	case RTE_INTR_HANDLE_VFIO_MSI:
626		if (vfio_disable_msi(intr_handle))
627			return -1;
628		break;
629	case RTE_INTR_HANDLE_VFIO_LEGACY:
630		if (vfio_disable_intx(intr_handle))
631			return -1;
632		break;
633#endif
634	/* unknown handle type */
635	default:
636		RTE_LOG(ERR, EAL,
637			"Unknown handle type of fd %d\n",
638					intr_handle->fd);
639		return -1;
640	}
641
642	return 0;
643}
644
645static int
646eal_intr_process_interrupts(struct epoll_event *events, int nfds)
647{
648	int n, bytes_read;
649	struct rte_intr_source *src;
650	struct rte_intr_callback *cb;
651	union rte_intr_read_buffer buf;
652	struct rte_intr_callback active_cb;
653
654	for (n = 0; n < nfds; n++) {
655
656		/**
657		 * if the pipe fd is ready to read, return out to
658		 * rebuild the wait list.
659		 */
660		if (events[n].data.fd == intr_pipe.readfd){
661			int r = read(intr_pipe.readfd, buf.charbuf,
662					sizeof(buf.charbuf));
663			RTE_SET_USED(r);
664			return -1;
665		}
666		rte_spinlock_lock(&intr_lock);
667		TAILQ_FOREACH(src, &intr_sources, next)
668			if (src->intr_handle.fd ==
669					events[n].data.fd)
670				break;
671		if (src == NULL){
672			rte_spinlock_unlock(&intr_lock);
673			continue;
674		}
675
676		/* mark this interrupt source as active and release the lock. */
677		src->active = 1;
678		rte_spinlock_unlock(&intr_lock);
679
680		/* set the length to be read dor different handle type */
681		switch (src->intr_handle.type) {
682		case RTE_INTR_HANDLE_UIO:
683		case RTE_INTR_HANDLE_UIO_INTX:
684			bytes_read = sizeof(buf.uio_intr_count);
685			break;
686		case RTE_INTR_HANDLE_ALARM:
687			bytes_read = sizeof(buf.timerfd_num);
688			break;
689#ifdef VFIO_PRESENT
690		case RTE_INTR_HANDLE_VFIO_MSIX:
691		case RTE_INTR_HANDLE_VFIO_MSI:
692		case RTE_INTR_HANDLE_VFIO_LEGACY:
693			bytes_read = sizeof(buf.vfio_intr_count);
694			break;
695#endif
696		case RTE_INTR_HANDLE_EXT:
697		default:
698			bytes_read = 1;
699			break;
700		}
701
702		if (src->intr_handle.type != RTE_INTR_HANDLE_EXT) {
703			/**
704			 * read out to clear the ready-to-be-read flag
705			 * for epoll_wait.
706			 */
707			bytes_read = read(events[n].data.fd, &buf, bytes_read);
708			if (bytes_read < 0) {
709				if (errno == EINTR || errno == EWOULDBLOCK)
710					continue;
711
712				RTE_LOG(ERR, EAL, "Error reading from file "
713					"descriptor %d: %s\n",
714					events[n].data.fd,
715					strerror(errno));
716			} else if (bytes_read == 0)
717				RTE_LOG(ERR, EAL, "Read nothing from file "
718					"descriptor %d\n", events[n].data.fd);
719		}
720
721		/* grab a lock, again to call callbacks and update status. */
722		rte_spinlock_lock(&intr_lock);
723
724		if (bytes_read > 0) {
725
726			/* Finally, call all callbacks. */
727			TAILQ_FOREACH(cb, &src->callbacks, next) {
728
729				/* make a copy and unlock. */
730				active_cb = *cb;
731				rte_spinlock_unlock(&intr_lock);
732
733				/* call the actual callback */
734				active_cb.cb_fn(&src->intr_handle,
735					active_cb.cb_arg);
736
737				/*get the lock back. */
738				rte_spinlock_lock(&intr_lock);
739			}
740		}
741
742		/* we done with that interrupt source, release it. */
743		src->active = 0;
744		rte_spinlock_unlock(&intr_lock);
745	}
746
747	return 0;
748}
749
750/**
751 * It handles all the interrupts.
752 *
753 * @param pfd
754 *  epoll file descriptor.
755 * @param totalfds
756 *  The number of file descriptors added in epoll.
757 *
758 * @return
759 *  void
760 */
761static void
762eal_intr_handle_interrupts(int pfd, unsigned totalfds)
763{
764	struct epoll_event events[totalfds];
765	int nfds = 0;
766
767	for(;;) {
768		nfds = epoll_wait(pfd, events, totalfds,
769			EAL_INTR_EPOLL_WAIT_FOREVER);
770		/* epoll_wait fail */
771		if (nfds < 0) {
772			if (errno == EINTR)
773				continue;
774			RTE_LOG(ERR, EAL,
775				"epoll_wait returns with fail\n");
776			return;
777		}
778		/* epoll_wait timeout, will never happens here */
779		else if (nfds == 0)
780			continue;
781		/* epoll_wait has at least one fd ready to read */
782		if (eal_intr_process_interrupts(events, nfds) < 0)
783			return;
784	}
785}
786
787/**
788 * It builds/rebuilds up the epoll file descriptor with all the
789 * file descriptors being waited on. Then handles the interrupts.
790 *
791 * @param arg
792 *  pointer. (unused)
793 *
794 * @return
795 *  never return;
796 */
797static __attribute__((noreturn)) void *
798eal_intr_thread_main(__rte_unused void *arg)
799{
800	struct epoll_event ev;
801
802	/* host thread, never break out */
803	for (;;) {
804		/* build up the epoll fd with all descriptors we are to
805		 * wait on then pass it to the handle_interrupts function
806		 */
807		static struct epoll_event pipe_event = {
808			.events = EPOLLIN | EPOLLPRI,
809		};
810		struct rte_intr_source *src;
811		unsigned numfds = 0;
812
813		/* create epoll fd */
814		int pfd = epoll_create(1);
815		if (pfd < 0)
816			rte_panic("Cannot create epoll instance\n");
817
818		pipe_event.data.fd = intr_pipe.readfd;
819		/**
820		 * add pipe fd into wait list, this pipe is used to
821		 * rebuild the wait list.
822		 */
823		if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
824						&pipe_event) < 0) {
825			rte_panic("Error adding fd to %d epoll_ctl, %s\n",
826					intr_pipe.readfd, strerror(errno));
827		}
828		numfds++;
829
830		rte_spinlock_lock(&intr_lock);
831
832		TAILQ_FOREACH(src, &intr_sources, next) {
833			if (src->callbacks.tqh_first == NULL)
834				continue; /* skip those with no callbacks */
835			ev.events = EPOLLIN | EPOLLPRI;
836			ev.data.fd = src->intr_handle.fd;
837
838			/**
839			 * add all the uio device file descriptor
840			 * into wait list.
841			 */
842			if (epoll_ctl(pfd, EPOLL_CTL_ADD,
843					src->intr_handle.fd, &ev) < 0){
844				rte_panic("Error adding fd %d epoll_ctl, %s\n",
845					src->intr_handle.fd, strerror(errno));
846			}
847			else
848				numfds++;
849		}
850		rte_spinlock_unlock(&intr_lock);
851		/* serve the interrupt */
852		eal_intr_handle_interrupts(pfd, numfds);
853
854		/**
855		 * when we return, we need to rebuild the
856		 * list of fds to monitor.
857		 */
858		close(pfd);
859	}
860}
861
862int
863rte_eal_intr_init(void)
864{
865	int ret = 0, ret_1 = 0;
866	char thread_name[RTE_MAX_THREAD_NAME_LEN];
867
868	/* init the global interrupt source head */
869	TAILQ_INIT(&intr_sources);
870
871	/**
872	 * create a pipe which will be waited by epoll and notified to
873	 * rebuild the wait list of epoll.
874	 */
875	if (pipe(intr_pipe.pipefd) < 0)
876		return -1;
877
878	/* create the host thread to wait/handle the interrupt */
879	ret = pthread_create(&intr_thread, NULL,
880			eal_intr_thread_main, NULL);
881	if (ret != 0) {
882		RTE_LOG(ERR, EAL,
883			"Failed to create thread for interrupt handling\n");
884	} else {
885		/* Set thread_name for aid in debugging. */
886		snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
887			"eal-intr-thread");
888		ret_1 = rte_thread_setname(intr_thread, thread_name);
889		if (ret_1 != 0)
890			RTE_LOG(DEBUG, EAL,
891			"Failed to set thread name for interrupt handling\n");
892	}
893
894	return -ret;
895}
896
897static void
898eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
899{
900	union rte_intr_read_buffer buf;
901	int bytes_read = 1;
902	int nbytes;
903
904	switch (intr_handle->type) {
905	case RTE_INTR_HANDLE_UIO:
906	case RTE_INTR_HANDLE_UIO_INTX:
907		bytes_read = sizeof(buf.uio_intr_count);
908		break;
909#ifdef VFIO_PRESENT
910	case RTE_INTR_HANDLE_VFIO_MSIX:
911	case RTE_INTR_HANDLE_VFIO_MSI:
912	case RTE_INTR_HANDLE_VFIO_LEGACY:
913		bytes_read = sizeof(buf.vfio_intr_count);
914		break;
915#endif
916	default:
917		bytes_read = 1;
918		RTE_LOG(INFO, EAL, "unexpected intr type\n");
919		break;
920	}
921
922	/**
923	 * read out to clear the ready-to-be-read flag
924	 * for epoll_wait.
925	 */
926	do {
927		nbytes = read(fd, &buf, bytes_read);
928		if (nbytes < 0) {
929			if (errno == EINTR || errno == EWOULDBLOCK ||
930			    errno == EAGAIN)
931				continue;
932			RTE_LOG(ERR, EAL,
933				"Error reading from fd %d: %s\n",
934				fd, strerror(errno));
935		} else if (nbytes == 0)
936			RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
937		return;
938	} while (1);
939}
940
941static int
942eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
943			struct rte_epoll_event *events)
944{
945	unsigned int i, count = 0;
946	struct rte_epoll_event *rev;
947
948	for (i = 0; i < n; i++) {
949		rev = evs[i].data.ptr;
950		if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
951						 RTE_EPOLL_EXEC))
952			continue;
953
954		events[count].status        = RTE_EPOLL_VALID;
955		events[count].fd            = rev->fd;
956		events[count].epfd          = rev->epfd;
957		events[count].epdata.event  = rev->epdata.event;
958		events[count].epdata.data   = rev->epdata.data;
959		if (rev->epdata.cb_fun)
960			rev->epdata.cb_fun(rev->fd,
961					   rev->epdata.cb_arg);
962
963		rte_compiler_barrier();
964		rev->status = RTE_EPOLL_VALID;
965		count++;
966	}
967	return count;
968}
969
970static inline int
971eal_init_tls_epfd(void)
972{
973	int pfd = epoll_create(255);
974
975	if (pfd < 0) {
976		RTE_LOG(ERR, EAL,
977			"Cannot create epoll instance\n");
978		return -1;
979	}
980	return pfd;
981}
982
983int
984rte_intr_tls_epfd(void)
985{
986	if (RTE_PER_LCORE(_epfd) == -1)
987		RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
988
989	return RTE_PER_LCORE(_epfd);
990}
991
992int
993rte_epoll_wait(int epfd, struct rte_epoll_event *events,
994	       int maxevents, int timeout)
995{
996	struct epoll_event evs[maxevents];
997	int rc;
998
999	if (!events) {
1000		RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1001		return -1;
1002	}
1003
1004	/* using per thread epoll fd */
1005	if (epfd == RTE_EPOLL_PER_THREAD)
1006		epfd = rte_intr_tls_epfd();
1007
1008	while (1) {
1009		rc = epoll_wait(epfd, evs, maxevents, timeout);
1010		if (likely(rc > 0)) {
1011			/* epoll_wait has at least one fd ready to read */
1012			rc = eal_epoll_process_event(evs, rc, events);
1013			break;
1014		} else if (rc < 0) {
1015			if (errno == EINTR)
1016				continue;
1017			/* epoll_wait fail */
1018			RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1019				strerror(errno));
1020			rc = -1;
1021			break;
1022		} else {
1023			/* rc == 0, epoll_wait timed out */
1024			break;
1025		}
1026	}
1027
1028	return rc;
1029}
1030
1031static inline void
1032eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1033{
1034	while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1035				    RTE_EPOLL_INVALID))
1036		while (ev->status != RTE_EPOLL_VALID)
1037			rte_pause();
1038	memset(&ev->epdata, 0, sizeof(ev->epdata));
1039	ev->fd = -1;
1040	ev->epfd = -1;
1041}
1042
1043int
1044rte_epoll_ctl(int epfd, int op, int fd,
1045	      struct rte_epoll_event *event)
1046{
1047	struct epoll_event ev;
1048
1049	if (!event) {
1050		RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1051		return -1;
1052	}
1053
1054	/* using per thread epoll fd */
1055	if (epfd == RTE_EPOLL_PER_THREAD)
1056		epfd = rte_intr_tls_epfd();
1057
1058	if (op == EPOLL_CTL_ADD) {
1059		event->status = RTE_EPOLL_VALID;
1060		event->fd = fd;  /* ignore fd in event */
1061		event->epfd = epfd;
1062		ev.data.ptr = (void *)event;
1063	}
1064
1065	ev.events = event->epdata.event;
1066	if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1067		RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1068			op, fd, strerror(errno));
1069		if (op == EPOLL_CTL_ADD)
1070			/* rollback status when CTL_ADD fail */
1071			event->status = RTE_EPOLL_INVALID;
1072		return -1;
1073	}
1074
1075	if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1076		eal_epoll_data_safe_free(event);
1077
1078	return 0;
1079}
1080
1081int
1082rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1083		int op, unsigned int vec, void *data)
1084{
1085	struct rte_epoll_event *rev;
1086	struct rte_epoll_data *epdata;
1087	int epfd_op;
1088	unsigned int efd_idx;
1089	int rc = 0;
1090
1091	efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1092		(vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1093
1094	if (!intr_handle || intr_handle->nb_efd == 0 ||
1095	    efd_idx >= intr_handle->nb_efd) {
1096		RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1097		return -EPERM;
1098	}
1099
1100	switch (op) {
1101	case RTE_INTR_EVENT_ADD:
1102		epfd_op = EPOLL_CTL_ADD;
1103		rev = &intr_handle->elist[efd_idx];
1104		if (rev->status != RTE_EPOLL_INVALID) {
1105			RTE_LOG(INFO, EAL, "Event already been added.\n");
1106			return -EEXIST;
1107		}
1108
1109		/* attach to intr vector fd */
1110		epdata = &rev->epdata;
1111		epdata->event  = EPOLLIN | EPOLLPRI | EPOLLET;
1112		epdata->data   = data;
1113		epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1114		epdata->cb_arg = (void *)intr_handle;
1115		rc = rte_epoll_ctl(epfd, epfd_op,
1116				   intr_handle->efds[efd_idx], rev);
1117		if (!rc)
1118			RTE_LOG(DEBUG, EAL,
1119				"efd %d associated with vec %d added on epfd %d"
1120				"\n", rev->fd, vec, epfd);
1121		else
1122			rc = -EPERM;
1123		break;
1124	case RTE_INTR_EVENT_DEL:
1125		epfd_op = EPOLL_CTL_DEL;
1126		rev = &intr_handle->elist[efd_idx];
1127		if (rev->status == RTE_EPOLL_INVALID) {
1128			RTE_LOG(INFO, EAL, "Event does not exist.\n");
1129			return -EPERM;
1130		}
1131
1132		rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1133		if (rc)
1134			rc = -EPERM;
1135		break;
1136	default:
1137		RTE_LOG(ERR, EAL, "event op type mismatch\n");
1138		rc = -EPERM;
1139	}
1140
1141	return rc;
1142}
1143
1144int
1145rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1146{
1147	uint32_t i;
1148	int fd;
1149	uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1150
1151	assert(nb_efd != 0);
1152
1153	if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1154		for (i = 0; i < n; i++) {
1155			fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1156			if (fd < 0) {
1157				RTE_LOG(ERR, EAL,
1158					"can't setup eventfd, error %i (%s)\n",
1159					errno, strerror(errno));
1160				return -1;
1161			}
1162			intr_handle->efds[i] = fd;
1163		}
1164		intr_handle->nb_efd   = n;
1165		intr_handle->max_intr = NB_OTHER_INTR + n;
1166	} else {
1167		intr_handle->efds[0]  = intr_handle->fd;
1168		intr_handle->nb_efd   = RTE_MIN(nb_efd, 1U);
1169		intr_handle->max_intr = NB_OTHER_INTR;
1170	}
1171
1172	return 0;
1173}
1174
1175void
1176rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1177{
1178	uint32_t i;
1179	struct rte_epoll_event *rev;
1180
1181	for (i = 0; i < intr_handle->nb_efd; i++) {
1182		rev = &intr_handle->elist[i];
1183		if (rev->status == RTE_EPOLL_INVALID)
1184			continue;
1185		if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1186			/* force free if the entry valid */
1187			eal_epoll_data_safe_free(rev);
1188			rev->status = RTE_EPOLL_INVALID;
1189		}
1190	}
1191
1192	if (intr_handle->max_intr > intr_handle->nb_efd) {
1193		for (i = 0; i < intr_handle->nb_efd; i++)
1194			close(intr_handle->efds[i]);
1195	}
1196	intr_handle->nb_efd = 0;
1197	intr_handle->max_intr = 0;
1198}
1199
1200int
1201rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1202{
1203	return !(!intr_handle->nb_efd);
1204}
1205
1206int
1207rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1208{
1209	if (!rte_intr_dp_is_en(intr_handle))
1210		return 1;
1211	else
1212		return !!(intr_handle->max_intr - intr_handle->nb_efd);
1213}
1214
1215int
1216rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1217{
1218	if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)
1219		return 1;
1220
1221	return 0;
1222}
1223