4 * Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
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
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.
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.
38 #include <sys/queue.h>
44 #include <sys/epoll.h>
45 #include <sys/signalfd.h>
46 #include <sys/ioctl.h>
47 #include <sys/eventfd.h>
49 #include <rte_common.h>
50 #include <rte_interrupts.h>
51 #include <rte_memory.h>
52 #include <rte_memzone.h>
53 #include <rte_launch.h>
55 #include <rte_per_lcore.h>
56 #include <rte_lcore.h>
57 #include <rte_atomic.h>
58 #include <rte_branch_prediction.h>
60 #include <rte_debug.h>
62 #include <rte_mempool.h>
64 #include <rte_malloc.h>
65 #include <rte_errno.h>
66 #include <rte_spinlock.h>
68 #include "eal_private.h"
70 #include "eal_thread.h"
72 #define EAL_INTR_EPOLL_WAIT_FOREVER (-1)
73 #define NB_OTHER_INTR 1
75 static RTE_DEFINE_PER_LCORE(int, _epfd) = -1; /**< epoll fd per thread */
91 * union buffer for reading on different devices
93 union rte_intr_read_buffer {
94 int uio_intr_count; /* for uio device */
96 uint64_t vfio_intr_count; /* for vfio device */
98 uint64_t timerfd_num; /* for timerfd */
99 char charbuf[16]; /* for others */
102 TAILQ_HEAD(rte_intr_cb_list, rte_intr_callback);
103 TAILQ_HEAD(rte_intr_source_list, rte_intr_source);
105 struct rte_intr_callback {
106 TAILQ_ENTRY(rte_intr_callback) next;
107 rte_intr_callback_fn cb_fn; /**< callback address */
108 void *cb_arg; /**< parameter for callback */
111 struct rte_intr_source {
112 TAILQ_ENTRY(rte_intr_source) next;
113 struct rte_intr_handle intr_handle; /**< interrupt handle */
114 struct rte_intr_cb_list callbacks; /**< user callbacks */
118 /* global spinlock for interrupt data operation */
119 static rte_spinlock_t intr_lock = RTE_SPINLOCK_INITIALIZER;
121 /* union buffer for pipe read/write */
122 static union intr_pipefds intr_pipe;
124 /* interrupt sources list */
125 static struct rte_intr_source_list intr_sources;
127 /* interrupt handling thread */
128 static pthread_t intr_thread;
130 /* VFIO interrupts */
133 #define IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + sizeof(int))
134 /* irq set buffer length for queue interrupts and LSC interrupt */
135 #define MSIX_IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + \
136 sizeof(int) * (RTE_MAX_RXTX_INTR_VEC_ID + 1))
138 /* enable legacy (INTx) interrupts */
140 vfio_enable_intx(struct rte_intr_handle *intr_handle) {
141 struct vfio_irq_set *irq_set;
142 char irq_set_buf[IRQ_SET_BUF_LEN];
146 len = sizeof(irq_set_buf);
149 irq_set = (struct vfio_irq_set *) irq_set_buf;
150 irq_set->argsz = len;
152 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
153 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
155 fd_ptr = (int *) &irq_set->data;
156 *fd_ptr = intr_handle->fd;
158 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
161 RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
166 /* unmask INTx after enabling */
167 memset(irq_set, 0, len);
168 len = sizeof(struct vfio_irq_set);
169 irq_set->argsz = len;
171 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
172 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
175 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
178 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
185 /* disable legacy (INTx) interrupts */
187 vfio_disable_intx(struct rte_intr_handle *intr_handle) {
188 struct vfio_irq_set *irq_set;
189 char irq_set_buf[IRQ_SET_BUF_LEN];
192 len = sizeof(struct vfio_irq_set);
194 /* mask interrupts before disabling */
195 irq_set = (struct vfio_irq_set *) irq_set_buf;
196 irq_set->argsz = len;
198 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
199 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
202 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
205 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
211 memset(irq_set, 0, len);
212 irq_set->argsz = len;
214 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
215 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
218 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
222 "Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
228 /* enable MSI interrupts */
230 vfio_enable_msi(struct rte_intr_handle *intr_handle) {
232 char irq_set_buf[IRQ_SET_BUF_LEN];
233 struct vfio_irq_set *irq_set;
236 len = sizeof(irq_set_buf);
238 irq_set = (struct vfio_irq_set *) irq_set_buf;
239 irq_set->argsz = len;
241 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
242 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
244 fd_ptr = (int *) &irq_set->data;
245 *fd_ptr = intr_handle->fd;
247 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
250 RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
257 /* disable MSI interrupts */
259 vfio_disable_msi(struct rte_intr_handle *intr_handle) {
260 struct vfio_irq_set *irq_set;
261 char irq_set_buf[IRQ_SET_BUF_LEN];
264 len = sizeof(struct vfio_irq_set);
266 irq_set = (struct vfio_irq_set *) irq_set_buf;
267 irq_set->argsz = len;
269 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
270 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
273 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
277 "Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
282 /* enable MSI-X interrupts */
284 vfio_enable_msix(struct rte_intr_handle *intr_handle) {
286 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
287 struct vfio_irq_set *irq_set;
290 len = sizeof(irq_set_buf);
292 irq_set = (struct vfio_irq_set *) irq_set_buf;
293 irq_set->argsz = len;
294 if (!intr_handle->max_intr)
295 intr_handle->max_intr = 1;
296 else if (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
297 intr_handle->max_intr = RTE_MAX_RXTX_INTR_VEC_ID + 1;
299 irq_set->count = intr_handle->max_intr;
300 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
301 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
303 fd_ptr = (int *) &irq_set->data;
304 /* INTR vector offset 0 reserve for non-efds mapping */
305 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
306 memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
307 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
309 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
312 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
320 /* disable MSI-X interrupts */
322 vfio_disable_msix(struct rte_intr_handle *intr_handle) {
323 struct vfio_irq_set *irq_set;
324 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
327 len = sizeof(struct vfio_irq_set);
329 irq_set = (struct vfio_irq_set *) irq_set_buf;
330 irq_set->argsz = len;
332 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
333 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
336 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
340 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
347 uio_intx_intr_disable(struct rte_intr_handle *intr_handle)
349 unsigned char command_high;
351 /* use UIO config file descriptor for uio_pci_generic */
352 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
354 "Error reading interrupts status for fd %d\n",
355 intr_handle->uio_cfg_fd);
358 /* disable interrupts */
360 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
362 "Error disabling interrupts for fd %d\n",
363 intr_handle->uio_cfg_fd);
371 uio_intx_intr_enable(struct rte_intr_handle *intr_handle)
373 unsigned char command_high;
375 /* use UIO config file descriptor for uio_pci_generic */
376 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
378 "Error reading interrupts status for fd %d\n",
379 intr_handle->uio_cfg_fd);
382 /* enable interrupts */
383 command_high &= ~0x4;
384 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
386 "Error enabling interrupts for fd %d\n",
387 intr_handle->uio_cfg_fd);
395 uio_intr_disable(struct rte_intr_handle *intr_handle)
399 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
401 "Error disabling interrupts for fd %d (%s)\n",
402 intr_handle->fd, strerror(errno));
409 uio_intr_enable(struct rte_intr_handle *intr_handle)
413 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
415 "Error enabling interrupts for fd %d (%s)\n",
416 intr_handle->fd, strerror(errno));
423 rte_intr_callback_register(struct rte_intr_handle *intr_handle,
424 rte_intr_callback_fn cb, void *cb_arg)
426 int ret, wake_thread;
427 struct rte_intr_source *src;
428 struct rte_intr_callback *callback;
432 /* first do parameter checking */
433 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
435 "Registering with invalid input parameter\n");
439 /* allocate a new interrupt callback entity */
440 callback = rte_zmalloc("interrupt callback list",
441 sizeof(*callback), 0);
442 if (callback == NULL) {
443 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
446 callback->cb_fn = cb;
447 callback->cb_arg = cb_arg;
449 rte_spinlock_lock(&intr_lock);
451 /* check if there is at least one callback registered for the fd */
452 TAILQ_FOREACH(src, &intr_sources, next) {
453 if (src->intr_handle.fd == intr_handle->fd) {
454 /* we had no interrupts for this */
455 if TAILQ_EMPTY(&src->callbacks)
458 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
464 /* no existing callbacks for this - add new source */
466 if ((src = rte_zmalloc("interrupt source list",
467 sizeof(*src), 0)) == NULL) {
468 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
472 src->intr_handle = *intr_handle;
473 TAILQ_INIT(&src->callbacks);
474 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
475 TAILQ_INSERT_TAIL(&intr_sources, src, next);
481 rte_spinlock_unlock(&intr_lock);
484 * check if need to notify the pipe fd waited by epoll_wait to
485 * rebuild the wait list.
488 if (write(intr_pipe.writefd, "1", 1) < 0)
495 rte_intr_callback_unregister(struct rte_intr_handle *intr_handle,
496 rte_intr_callback_fn cb_fn, void *cb_arg)
499 struct rte_intr_source *src;
500 struct rte_intr_callback *cb, *next;
502 /* do parameter checking first */
503 if (intr_handle == NULL || intr_handle->fd < 0) {
505 "Unregistering with invalid input parameter\n");
509 rte_spinlock_lock(&intr_lock);
511 /* check if the insterrupt source for the fd is existent */
512 TAILQ_FOREACH(src, &intr_sources, next)
513 if (src->intr_handle.fd == intr_handle->fd)
516 /* No interrupt source registered for the fd */
520 /* interrupt source has some active callbacks right now. */
521 } else if (src->active != 0) {
528 /*walk through the callbacks and remove all that match. */
529 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
531 next = TAILQ_NEXT(cb, next);
533 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
534 cb->cb_arg == cb_arg)) {
535 TAILQ_REMOVE(&src->callbacks, cb, next);
541 /* all callbacks for that source are removed. */
542 if (TAILQ_EMPTY(&src->callbacks)) {
543 TAILQ_REMOVE(&intr_sources, src, next);
548 rte_spinlock_unlock(&intr_lock);
550 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
551 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
559 rte_intr_enable(struct rte_intr_handle *intr_handle)
561 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
564 switch (intr_handle->type){
565 /* write to the uio fd to enable the interrupt */
566 case RTE_INTR_HANDLE_UIO:
567 if (uio_intr_enable(intr_handle))
570 case RTE_INTR_HANDLE_UIO_INTX:
571 if (uio_intx_intr_enable(intr_handle))
574 /* not used at this moment */
575 case RTE_INTR_HANDLE_ALARM:
578 case RTE_INTR_HANDLE_VFIO_MSIX:
579 if (vfio_enable_msix(intr_handle))
582 case RTE_INTR_HANDLE_VFIO_MSI:
583 if (vfio_enable_msi(intr_handle))
586 case RTE_INTR_HANDLE_VFIO_LEGACY:
587 if (vfio_enable_intx(intr_handle))
591 /* unknown handle type */
594 "Unknown handle type of fd %d\n",
603 rte_intr_disable(struct rte_intr_handle *intr_handle)
605 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
608 switch (intr_handle->type){
609 /* write to the uio fd to disable the interrupt */
610 case RTE_INTR_HANDLE_UIO:
611 if (uio_intr_disable(intr_handle))
614 case RTE_INTR_HANDLE_UIO_INTX:
615 if (uio_intx_intr_disable(intr_handle))
618 /* not used at this moment */
619 case RTE_INTR_HANDLE_ALARM:
622 case RTE_INTR_HANDLE_VFIO_MSIX:
623 if (vfio_disable_msix(intr_handle))
626 case RTE_INTR_HANDLE_VFIO_MSI:
627 if (vfio_disable_msi(intr_handle))
630 case RTE_INTR_HANDLE_VFIO_LEGACY:
631 if (vfio_disable_intx(intr_handle))
635 /* unknown handle type */
638 "Unknown handle type of fd %d\n",
647 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
650 struct rte_intr_source *src;
651 struct rte_intr_callback *cb;
652 union rte_intr_read_buffer buf;
653 struct rte_intr_callback active_cb;
655 for (n = 0; n < nfds; n++) {
658 * if the pipe fd is ready to read, return out to
659 * rebuild the wait list.
661 if (events[n].data.fd == intr_pipe.readfd){
662 int r = read(intr_pipe.readfd, buf.charbuf,
663 sizeof(buf.charbuf));
667 rte_spinlock_lock(&intr_lock);
668 TAILQ_FOREACH(src, &intr_sources, next)
669 if (src->intr_handle.fd ==
673 rte_spinlock_unlock(&intr_lock);
677 /* mark this interrupt source as active and release the lock. */
679 rte_spinlock_unlock(&intr_lock);
681 /* set the length to be read dor different handle type */
682 switch (src->intr_handle.type) {
683 case RTE_INTR_HANDLE_UIO:
684 case RTE_INTR_HANDLE_UIO_INTX:
685 bytes_read = sizeof(buf.uio_intr_count);
687 case RTE_INTR_HANDLE_ALARM:
688 bytes_read = sizeof(buf.timerfd_num);
691 case RTE_INTR_HANDLE_VFIO_MSIX:
692 case RTE_INTR_HANDLE_VFIO_MSI:
693 case RTE_INTR_HANDLE_VFIO_LEGACY:
694 bytes_read = sizeof(buf.vfio_intr_count);
697 case RTE_INTR_HANDLE_EXT:
703 if (src->intr_handle.type != RTE_INTR_HANDLE_EXT) {
705 * read out to clear the ready-to-be-read flag
708 bytes_read = read(events[n].data.fd, &buf, bytes_read);
709 if (bytes_read < 0) {
710 if (errno == EINTR || errno == EWOULDBLOCK)
713 RTE_LOG(ERR, EAL, "Error reading from file "
714 "descriptor %d: %s\n",
717 } else if (bytes_read == 0)
718 RTE_LOG(ERR, EAL, "Read nothing from file "
719 "descriptor %d\n", events[n].data.fd);
722 /* grab a lock, again to call callbacks and update status. */
723 rte_spinlock_lock(&intr_lock);
725 if (bytes_read > 0) {
727 /* Finally, call all callbacks. */
728 TAILQ_FOREACH(cb, &src->callbacks, next) {
730 /* make a copy and unlock. */
732 rte_spinlock_unlock(&intr_lock);
734 /* call the actual callback */
735 active_cb.cb_fn(&src->intr_handle,
738 /*get the lock back. */
739 rte_spinlock_lock(&intr_lock);
743 /* we done with that interrupt source, release it. */
745 rte_spinlock_unlock(&intr_lock);
752 * It handles all the interrupts.
755 * epoll file descriptor.
757 * The number of file descriptors added in epoll.
763 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
765 struct epoll_event events[totalfds];
769 nfds = epoll_wait(pfd, events, totalfds,
770 EAL_INTR_EPOLL_WAIT_FOREVER);
771 /* epoll_wait fail */
776 "epoll_wait returns with fail\n");
779 /* epoll_wait timeout, will never happens here */
782 /* epoll_wait has at least one fd ready to read */
783 if (eal_intr_process_interrupts(events, nfds) < 0)
789 * It builds/rebuilds up the epoll file descriptor with all the
790 * file descriptors being waited on. Then handles the interrupts.
798 static __attribute__((noreturn)) void *
799 eal_intr_thread_main(__rte_unused void *arg)
801 struct epoll_event ev;
803 /* host thread, never break out */
805 /* build up the epoll fd with all descriptors we are to
806 * wait on then pass it to the handle_interrupts function
808 static struct epoll_event pipe_event = {
809 .events = EPOLLIN | EPOLLPRI,
811 struct rte_intr_source *src;
814 /* create epoll fd */
815 int pfd = epoll_create(1);
817 rte_panic("Cannot create epoll instance\n");
819 pipe_event.data.fd = intr_pipe.readfd;
821 * add pipe fd into wait list, this pipe is used to
822 * rebuild the wait list.
824 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
826 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
827 intr_pipe.readfd, strerror(errno));
831 rte_spinlock_lock(&intr_lock);
833 TAILQ_FOREACH(src, &intr_sources, next) {
834 if (src->callbacks.tqh_first == NULL)
835 continue; /* skip those with no callbacks */
836 ev.events = EPOLLIN | EPOLLPRI;
837 ev.data.fd = src->intr_handle.fd;
840 * add all the uio device file descriptor
843 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
844 src->intr_handle.fd, &ev) < 0){
845 rte_panic("Error adding fd %d epoll_ctl, %s\n",
846 src->intr_handle.fd, strerror(errno));
851 rte_spinlock_unlock(&intr_lock);
852 /* serve the interrupt */
853 eal_intr_handle_interrupts(pfd, numfds);
856 * when we return, we need to rebuild the
857 * list of fds to monitor.
864 rte_eal_intr_init(void)
866 int ret = 0, ret_1 = 0;
867 char thread_name[RTE_MAX_THREAD_NAME_LEN];
869 /* init the global interrupt source head */
870 TAILQ_INIT(&intr_sources);
873 * create a pipe which will be waited by epoll and notified to
874 * rebuild the wait list of epoll.
876 if (pipe(intr_pipe.pipefd) < 0)
879 /* create the host thread to wait/handle the interrupt */
880 ret = pthread_create(&intr_thread, NULL,
881 eal_intr_thread_main, NULL);
884 "Failed to create thread for interrupt handling\n");
886 /* Set thread_name for aid in debugging. */
887 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
889 ret_1 = pthread_setname_np(intr_thread, thread_name);
892 "Failed to set thread name for interrupt handling\n");
899 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
901 union rte_intr_read_buffer buf;
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);
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);
918 RTE_LOG(INFO, EAL, "unexpected intr type\n");
923 * read out to clear the ready-to-be-read flag
927 bytes_read = read(fd, &buf, bytes_read);
928 if (bytes_read < 0) {
929 if (errno == EINTR || errno == EWOULDBLOCK ||
933 "Error reading from fd %d: %s\n",
934 fd, strerror(errno));
935 } else if (bytes_read == 0)
936 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
942 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
943 struct rte_epoll_event *events)
945 unsigned int i, count = 0;
946 struct rte_epoll_event *rev;
948 for (i = 0; i < n; i++) {
949 rev = evs[i].data.ptr;
950 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
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,
963 rte_compiler_barrier();
964 rev->status = RTE_EPOLL_VALID;
971 eal_init_tls_epfd(void)
973 int pfd = epoll_create(255);
977 "Cannot create epoll instance\n");
984 rte_intr_tls_epfd(void)
986 if (RTE_PER_LCORE(_epfd) == -1)
987 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
989 return RTE_PER_LCORE(_epfd);
993 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
994 int maxevents, int timeout)
996 struct epoll_event evs[maxevents];
1000 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1004 /* using per thread epoll fd */
1005 if (epfd == RTE_EPOLL_PER_THREAD)
1006 epfd = rte_intr_tls_epfd();
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);
1014 } else if (rc < 0) {
1017 /* epoll_wait fail */
1018 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1023 /* rc == 0, epoll_wait timed out */
1032 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1034 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1036 while (ev->status != RTE_EPOLL_VALID)
1038 memset(&ev->epdata, 0, sizeof(ev->epdata));
1044 rte_epoll_ctl(int epfd, int op, int fd,
1045 struct rte_epoll_event *event)
1047 struct epoll_event ev;
1050 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1054 /* using per thread epoll fd */
1055 if (epfd == RTE_EPOLL_PER_THREAD)
1056 epfd = rte_intr_tls_epfd();
1058 if (op == EPOLL_CTL_ADD) {
1059 event->status = RTE_EPOLL_VALID;
1060 event->fd = fd; /* ignore fd in event */
1062 ev.data.ptr = (void *)event;
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;
1075 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1076 eal_epoll_data_safe_free(event);
1082 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1083 int op, unsigned int vec, void *data)
1085 struct rte_epoll_event *rev;
1086 struct rte_epoll_data *epdata;
1088 unsigned int efd_idx;
1091 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1092 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
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");
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");
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);
1119 "efd %d associated with vec %d added on epfd %d"
1120 "\n", rev->fd, vec, epfd);
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");
1132 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1137 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1145 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1149 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1151 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1152 for (i = 0; i < n; i++) {
1153 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1156 "can't setup eventfd, error %i (%s)\n",
1157 errno, strerror(errno));
1160 intr_handle->efds[i] = fd;
1162 intr_handle->nb_efd = n;
1163 intr_handle->max_intr = NB_OTHER_INTR + n;
1165 intr_handle->efds[0] = intr_handle->fd;
1166 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1167 intr_handle->max_intr = NB_OTHER_INTR;
1174 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1177 struct rte_epoll_event *rev;
1179 for (i = 0; i < intr_handle->nb_efd; i++) {
1180 rev = &intr_handle->elist[i];
1181 if (rev->status == RTE_EPOLL_INVALID)
1183 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1184 /* force free if the entry valid */
1185 eal_epoll_data_safe_free(rev);
1186 rev->status = RTE_EPOLL_INVALID;
1190 if (intr_handle->max_intr > intr_handle->nb_efd) {
1191 for (i = 0; i < intr_handle->nb_efd; i++)
1192 close(intr_handle->efds[i]);
1194 intr_handle->nb_efd = 0;
1195 intr_handle->max_intr = 0;
1199 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1201 return !(!intr_handle->nb_efd);
1205 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1207 return !!(intr_handle->max_intr - intr_handle->nb_efd);