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"
71 #define EAL_INTR_EPOLL_WAIT_FOREVER (-1)
72 #define NB_OTHER_INTR 1
74 static RTE_DEFINE_PER_LCORE(int, _epfd) = -1; /**< epoll fd per thread */
90 * union buffer for reading on different devices
92 union rte_intr_read_buffer {
93 int uio_intr_count; /* for uio device */
95 uint64_t vfio_intr_count; /* for vfio device */
97 uint64_t timerfd_num; /* for timerfd */
98 char charbuf[16]; /* for others */
101 TAILQ_HEAD(rte_intr_cb_list, rte_intr_callback);
102 TAILQ_HEAD(rte_intr_source_list, rte_intr_source);
104 struct 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 */
110 struct 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 */
117 /* global spinlock for interrupt data operation */
118 static rte_spinlock_t intr_lock = RTE_SPINLOCK_INITIALIZER;
120 /* union buffer for pipe read/write */
121 static union intr_pipefds intr_pipe;
123 /* interrupt sources list */
124 static struct rte_intr_source_list intr_sources;
126 /* interrupt handling thread */
127 static pthread_t intr_thread;
129 /* VFIO interrupts */
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))
137 /* enable legacy (INTx) interrupts */
139 vfio_enable_intx(struct rte_intr_handle *intr_handle) {
140 struct vfio_irq_set *irq_set;
141 char irq_set_buf[IRQ_SET_BUF_LEN];
145 len = sizeof(irq_set_buf);
148 irq_set = (struct vfio_irq_set *) irq_set_buf;
149 irq_set->argsz = len;
151 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
152 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
154 fd_ptr = (int *) &irq_set->data;
155 *fd_ptr = intr_handle->fd;
157 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
160 RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
165 /* unmask INTx after enabling */
166 memset(irq_set, 0, len);
167 len = sizeof(struct vfio_irq_set);
168 irq_set->argsz = len;
170 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
171 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
174 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
177 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
184 /* disable legacy (INTx) interrupts */
186 vfio_disable_intx(struct rte_intr_handle *intr_handle) {
187 struct vfio_irq_set *irq_set;
188 char irq_set_buf[IRQ_SET_BUF_LEN];
191 len = sizeof(struct vfio_irq_set);
193 /* mask interrupts before disabling */
194 irq_set = (struct vfio_irq_set *) irq_set_buf;
195 irq_set->argsz = len;
197 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
198 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
201 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
204 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
210 memset(irq_set, 0, len);
211 irq_set->argsz = len;
213 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
214 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
217 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
221 "Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
227 /* enable MSI interrupts */
229 vfio_enable_msi(struct rte_intr_handle *intr_handle) {
231 char irq_set_buf[IRQ_SET_BUF_LEN];
232 struct vfio_irq_set *irq_set;
235 len = sizeof(irq_set_buf);
237 irq_set = (struct vfio_irq_set *) irq_set_buf;
238 irq_set->argsz = len;
240 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
241 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
243 fd_ptr = (int *) &irq_set->data;
244 *fd_ptr = intr_handle->fd;
246 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
249 RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
256 /* disable MSI interrupts */
258 vfio_disable_msi(struct rte_intr_handle *intr_handle) {
259 struct vfio_irq_set *irq_set;
260 char irq_set_buf[IRQ_SET_BUF_LEN];
263 len = sizeof(struct vfio_irq_set);
265 irq_set = (struct vfio_irq_set *) irq_set_buf;
266 irq_set->argsz = len;
268 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
269 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
272 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
276 "Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
281 /* enable MSI-X interrupts */
283 vfio_enable_msix(struct rte_intr_handle *intr_handle) {
285 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
286 struct vfio_irq_set *irq_set;
289 len = sizeof(irq_set_buf);
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;
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;
302 fd_ptr = (int *) &irq_set->data;
303 memcpy(fd_ptr, intr_handle->efds, sizeof(intr_handle->efds));
304 fd_ptr[intr_handle->max_intr - 1] = intr_handle->fd;
306 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
309 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
317 /* disable MSI-X interrupts */
319 vfio_disable_msix(struct rte_intr_handle *intr_handle) {
320 struct vfio_irq_set *irq_set;
321 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
324 len = sizeof(struct vfio_irq_set);
326 irq_set = (struct vfio_irq_set *) irq_set_buf;
327 irq_set->argsz = len;
329 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
330 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
333 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
337 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
344 uio_intx_intr_disable(struct rte_intr_handle *intr_handle)
346 unsigned char command_high;
348 /* use UIO config file descriptor for uio_pci_generic */
349 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
351 "Error reading interrupts status for fd %d\n",
352 intr_handle->uio_cfg_fd);
355 /* disable interrupts */
357 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
359 "Error disabling interrupts for fd %d\n",
360 intr_handle->uio_cfg_fd);
368 uio_intx_intr_enable(struct rte_intr_handle *intr_handle)
370 unsigned char command_high;
372 /* use UIO config file descriptor for uio_pci_generic */
373 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
375 "Error reading interrupts status for fd %d\n",
376 intr_handle->uio_cfg_fd);
379 /* enable interrupts */
380 command_high &= ~0x4;
381 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
383 "Error enabling interrupts for fd %d\n",
384 intr_handle->uio_cfg_fd);
392 uio_intr_disable(struct rte_intr_handle *intr_handle)
396 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
398 "Error disabling interrupts for fd %d (%s)\n",
399 intr_handle->fd, strerror(errno));
406 uio_intr_enable(struct rte_intr_handle *intr_handle)
410 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
412 "Error enabling interrupts for fd %d (%s)\n",
413 intr_handle->fd, strerror(errno));
420 rte_intr_callback_register(struct rte_intr_handle *intr_handle,
421 rte_intr_callback_fn cb, void *cb_arg)
423 int ret, wake_thread;
424 struct rte_intr_source *src;
425 struct rte_intr_callback *callback;
429 /* first do parameter checking */
430 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
432 "Registering with invalid input parameter\n");
436 /* allocate a new interrupt callback entity */
437 callback = rte_zmalloc("interrupt callback list",
438 sizeof(*callback), 0);
439 if (callback == NULL) {
440 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
443 callback->cb_fn = cb;
444 callback->cb_arg = cb_arg;
446 rte_spinlock_lock(&intr_lock);
448 /* check if there is at least one callback registered for the fd */
449 TAILQ_FOREACH(src, &intr_sources, next) {
450 if (src->intr_handle.fd == intr_handle->fd) {
451 /* we had no interrupts for this */
452 if TAILQ_EMPTY(&src->callbacks)
455 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
461 /* no existing callbacks for this - add new source */
463 if ((src = rte_zmalloc("interrupt source list",
464 sizeof(*src), 0)) == NULL) {
465 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
469 src->intr_handle = *intr_handle;
470 TAILQ_INIT(&src->callbacks);
471 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
472 TAILQ_INSERT_TAIL(&intr_sources, src, next);
478 rte_spinlock_unlock(&intr_lock);
481 * check if need to notify the pipe fd waited by epoll_wait to
482 * rebuild the wait list.
485 if (write(intr_pipe.writefd, "1", 1) < 0)
492 rte_intr_callback_unregister(struct rte_intr_handle *intr_handle,
493 rte_intr_callback_fn cb_fn, void *cb_arg)
496 struct rte_intr_source *src;
497 struct rte_intr_callback *cb, *next;
499 /* do parameter checking first */
500 if (intr_handle == NULL || intr_handle->fd < 0) {
502 "Unregistering with invalid input parameter\n");
506 rte_spinlock_lock(&intr_lock);
508 /* check if the insterrupt source for the fd is existent */
509 TAILQ_FOREACH(src, &intr_sources, next)
510 if (src->intr_handle.fd == intr_handle->fd)
513 /* No interrupt source registered for the fd */
517 /* interrupt source has some active callbacks right now. */
518 } else if (src->active != 0) {
525 /*walk through the callbacks and remove all that match. */
526 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
528 next = TAILQ_NEXT(cb, next);
530 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
531 cb->cb_arg == cb_arg)) {
532 TAILQ_REMOVE(&src->callbacks, cb, next);
538 /* all callbacks for that source are removed. */
539 if (TAILQ_EMPTY(&src->callbacks)) {
540 TAILQ_REMOVE(&intr_sources, src, next);
545 rte_spinlock_unlock(&intr_lock);
547 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
548 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
556 rte_intr_enable(struct rte_intr_handle *intr_handle)
558 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
561 switch (intr_handle->type){
562 /* write to the uio fd to enable the interrupt */
563 case RTE_INTR_HANDLE_UIO:
564 if (uio_intr_enable(intr_handle))
567 case RTE_INTR_HANDLE_UIO_INTX:
568 if (uio_intx_intr_enable(intr_handle))
571 /* not used at this moment */
572 case RTE_INTR_HANDLE_ALARM:
575 case RTE_INTR_HANDLE_VFIO_MSIX:
576 if (vfio_enable_msix(intr_handle))
579 case RTE_INTR_HANDLE_VFIO_MSI:
580 if (vfio_enable_msi(intr_handle))
583 case RTE_INTR_HANDLE_VFIO_LEGACY:
584 if (vfio_enable_intx(intr_handle))
588 /* unknown handle type */
591 "Unknown handle type of fd %d\n",
600 rte_intr_disable(struct rte_intr_handle *intr_handle)
602 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
605 switch (intr_handle->type){
606 /* write to the uio fd to disable the interrupt */
607 case RTE_INTR_HANDLE_UIO:
608 if (uio_intr_disable(intr_handle))
611 case RTE_INTR_HANDLE_UIO_INTX:
612 if (uio_intx_intr_disable(intr_handle))
615 /* not used at this moment */
616 case RTE_INTR_HANDLE_ALARM:
619 case RTE_INTR_HANDLE_VFIO_MSIX:
620 if (vfio_disable_msix(intr_handle))
623 case RTE_INTR_HANDLE_VFIO_MSI:
624 if (vfio_disable_msi(intr_handle))
627 case RTE_INTR_HANDLE_VFIO_LEGACY:
628 if (vfio_disable_intx(intr_handle))
632 /* unknown handle type */
635 "Unknown handle type of fd %d\n",
644 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
647 struct rte_intr_source *src;
648 struct rte_intr_callback *cb;
649 union rte_intr_read_buffer buf;
650 struct rte_intr_callback active_cb;
652 for (n = 0; n < nfds; n++) {
655 * if the pipe fd is ready to read, return out to
656 * rebuild the wait list.
658 if (events[n].data.fd == intr_pipe.readfd){
659 int r = read(intr_pipe.readfd, buf.charbuf,
660 sizeof(buf.charbuf));
664 rte_spinlock_lock(&intr_lock);
665 TAILQ_FOREACH(src, &intr_sources, next)
666 if (src->intr_handle.fd ==
670 rte_spinlock_unlock(&intr_lock);
674 /* mark this interrupt source as active and release the lock. */
676 rte_spinlock_unlock(&intr_lock);
678 /* set the length to be read dor different handle type */
679 switch (src->intr_handle.type) {
680 case RTE_INTR_HANDLE_UIO:
681 case RTE_INTR_HANDLE_UIO_INTX:
682 bytes_read = sizeof(buf.uio_intr_count);
684 case RTE_INTR_HANDLE_ALARM:
685 bytes_read = sizeof(buf.timerfd_num);
688 case RTE_INTR_HANDLE_VFIO_MSIX:
689 case RTE_INTR_HANDLE_VFIO_MSI:
690 case RTE_INTR_HANDLE_VFIO_LEGACY:
691 bytes_read = sizeof(buf.vfio_intr_count);
700 * read out to clear the ready-to-be-read flag
703 bytes_read = read(events[n].data.fd, &buf, bytes_read);
704 if (bytes_read < 0) {
705 if (errno == EINTR || errno == EWOULDBLOCK)
708 RTE_LOG(ERR, EAL, "Error reading from file "
709 "descriptor %d: %s\n", events[n].data.fd,
711 } else if (bytes_read == 0)
712 RTE_LOG(ERR, EAL, "Read nothing from file "
713 "descriptor %d\n", events[n].data.fd);
715 /* grab a lock, again to call callbacks and update status. */
716 rte_spinlock_lock(&intr_lock);
718 if (bytes_read > 0) {
720 /* Finally, call all callbacks. */
721 TAILQ_FOREACH(cb, &src->callbacks, next) {
723 /* make a copy and unlock. */
725 rte_spinlock_unlock(&intr_lock);
727 /* call the actual callback */
728 active_cb.cb_fn(&src->intr_handle,
731 /*get the lock back. */
732 rte_spinlock_lock(&intr_lock);
736 /* we done with that interrupt source, release it. */
738 rte_spinlock_unlock(&intr_lock);
745 * It handles all the interrupts.
748 * epoll file descriptor.
750 * The number of file descriptors added in epoll.
756 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
758 struct epoll_event events[totalfds];
762 nfds = epoll_wait(pfd, events, totalfds,
763 EAL_INTR_EPOLL_WAIT_FOREVER);
764 /* epoll_wait fail */
769 "epoll_wait returns with fail\n");
772 /* epoll_wait timeout, will never happens here */
775 /* epoll_wait has at least one fd ready to read */
776 if (eal_intr_process_interrupts(events, nfds) < 0)
782 * It builds/rebuilds up the epoll file descriptor with all the
783 * file descriptors being waited on. Then handles the interrupts.
791 static __attribute__((noreturn)) void *
792 eal_intr_thread_main(__rte_unused void *arg)
794 struct epoll_event ev;
796 /* host thread, never break out */
798 /* build up the epoll fd with all descriptors we are to
799 * wait on then pass it to the handle_interrupts function
801 static struct epoll_event pipe_event = {
802 .events = EPOLLIN | EPOLLPRI,
804 struct rte_intr_source *src;
807 /* create epoll fd */
808 int pfd = epoll_create(1);
810 rte_panic("Cannot create epoll instance\n");
812 pipe_event.data.fd = intr_pipe.readfd;
814 * add pipe fd into wait list, this pipe is used to
815 * rebuild the wait list.
817 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
819 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
820 intr_pipe.readfd, strerror(errno));
824 rte_spinlock_lock(&intr_lock);
826 TAILQ_FOREACH(src, &intr_sources, next) {
827 if (src->callbacks.tqh_first == NULL)
828 continue; /* skip those with no callbacks */
829 ev.events = EPOLLIN | EPOLLPRI;
830 ev.data.fd = src->intr_handle.fd;
833 * add all the uio device file descriptor
836 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
837 src->intr_handle.fd, &ev) < 0){
838 rte_panic("Error adding fd %d epoll_ctl, %s\n",
839 src->intr_handle.fd, strerror(errno));
844 rte_spinlock_unlock(&intr_lock);
845 /* serve the interrupt */
846 eal_intr_handle_interrupts(pfd, numfds);
849 * when we return, we need to rebuild the
850 * list of fds to monitor.
857 rte_eal_intr_init(void)
861 /* init the global interrupt source head */
862 TAILQ_INIT(&intr_sources);
865 * create a pipe which will be waited by epoll and notified to
866 * rebuild the wait list of epoll.
868 if (pipe(intr_pipe.pipefd) < 0)
871 /* create the host thread to wait/handle the interrupt */
872 ret = pthread_create(&intr_thread, NULL,
873 eal_intr_thread_main, NULL);
876 "Failed to create thread for interrupt handling\n");
882 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
884 union rte_intr_read_buffer buf;
887 switch (intr_handle->type) {
888 case RTE_INTR_HANDLE_UIO:
889 case RTE_INTR_HANDLE_UIO_INTX:
890 bytes_read = sizeof(buf.uio_intr_count);
893 case RTE_INTR_HANDLE_VFIO_MSIX:
894 case RTE_INTR_HANDLE_VFIO_MSI:
895 case RTE_INTR_HANDLE_VFIO_LEGACY:
896 bytes_read = sizeof(buf.vfio_intr_count);
901 RTE_LOG(INFO, EAL, "unexpected intr type\n");
906 * read out to clear the ready-to-be-read flag
910 bytes_read = read(fd, &buf, bytes_read);
911 if (bytes_read < 0) {
912 if (errno == EINTR || errno == EWOULDBLOCK ||
916 "Error reading from fd %d: %s\n",
917 fd, strerror(errno));
918 } else if (bytes_read == 0)
919 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
925 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
926 struct rte_epoll_event *events)
928 unsigned int i, count = 0;
929 struct rte_epoll_event *rev;
931 for (i = 0; i < n; i++) {
932 rev = evs[i].data.ptr;
933 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
937 events[count].status = RTE_EPOLL_VALID;
938 events[count].fd = rev->fd;
939 events[count].epfd = rev->epfd;
940 events[count].epdata.event = rev->epdata.event;
941 events[count].epdata.data = rev->epdata.data;
942 if (rev->epdata.cb_fun)
943 rev->epdata.cb_fun(rev->fd,
946 rte_compiler_barrier();
947 rev->status = RTE_EPOLL_VALID;
954 eal_init_tls_epfd(void)
956 int pfd = epoll_create(255);
960 "Cannot create epoll instance\n");
967 rte_intr_tls_epfd(void)
969 if (RTE_PER_LCORE(_epfd) == -1)
970 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
972 return RTE_PER_LCORE(_epfd);
976 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
977 int maxevents, int timeout)
979 struct epoll_event evs[maxevents];
983 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
987 /* using per thread epoll fd */
988 if (epfd == RTE_EPOLL_PER_THREAD)
989 epfd = rte_intr_tls_epfd();
992 rc = epoll_wait(epfd, evs, maxevents, timeout);
993 if (likely(rc > 0)) {
994 /* epoll_wait has at least one fd ready to read */
995 rc = eal_epoll_process_event(evs, rc, events);
1000 /* epoll_wait fail */
1001 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1012 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1014 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1016 while (ev->status != RTE_EPOLL_VALID)
1018 memset(&ev->epdata, 0, sizeof(ev->epdata));
1024 rte_epoll_ctl(int epfd, int op, int fd,
1025 struct rte_epoll_event *event)
1027 struct epoll_event ev;
1030 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1034 /* using per thread epoll fd */
1035 if (epfd == RTE_EPOLL_PER_THREAD)
1036 epfd = rte_intr_tls_epfd();
1038 if (op == EPOLL_CTL_ADD) {
1039 event->status = RTE_EPOLL_VALID;
1040 event->fd = fd; /* ignore fd in event */
1042 ev.data.ptr = (void *)event;
1045 ev.events = event->epdata.event;
1046 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1047 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1048 op, fd, strerror(errno));
1049 if (op == EPOLL_CTL_ADD)
1050 /* rollback status when CTL_ADD fail */
1051 event->status = RTE_EPOLL_INVALID;
1055 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1056 eal_epoll_data_safe_free(event);
1062 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1063 int op, unsigned int vec, void *data)
1065 struct rte_epoll_event *rev;
1066 struct rte_epoll_data *epdata;
1070 if (!intr_handle || intr_handle->nb_efd == 0 ||
1071 vec >= intr_handle->nb_efd) {
1072 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1077 case RTE_INTR_EVENT_ADD:
1078 epfd_op = EPOLL_CTL_ADD;
1079 rev = &intr_handle->elist[vec];
1080 if (rev->status != RTE_EPOLL_INVALID) {
1081 RTE_LOG(INFO, EAL, "Event already been added.\n");
1085 /* attach to intr vector fd */
1086 epdata = &rev->epdata;
1087 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1088 epdata->data = data;
1089 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1090 epdata->cb_arg = (void *)intr_handle;
1091 rc = rte_epoll_ctl(epfd, epfd_op, intr_handle->efds[vec], rev);
1094 "efd %d associated with vec %d added on epfd %d"
1095 "\n", rev->fd, vec, epfd);
1099 case RTE_INTR_EVENT_DEL:
1100 epfd_op = EPOLL_CTL_DEL;
1101 rev = &intr_handle->elist[vec];
1102 if (rev->status == RTE_EPOLL_INVALID) {
1103 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1107 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1112 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1120 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1124 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1126 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1127 for (i = 0; i < n; i++) {
1128 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1131 "can't setup eventfd, error %i (%s)\n",
1132 errno, strerror(errno));
1135 intr_handle->efds[i] = fd;
1137 intr_handle->nb_efd = n;
1138 intr_handle->max_intr = NB_OTHER_INTR + n;
1140 intr_handle->efds[0] = intr_handle->fd;
1141 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1142 intr_handle->max_intr = NB_OTHER_INTR;
1149 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1152 struct rte_epoll_event *rev;
1154 for (i = 0; i < intr_handle->nb_efd; i++) {
1155 rev = &intr_handle->elist[i];
1156 if (rev->status == RTE_EPOLL_INVALID)
1158 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1159 /* force free if the entry valid */
1160 eal_epoll_data_safe_free(rev);
1161 rev->status = RTE_EPOLL_INVALID;
1165 if (intr_handle->max_intr > intr_handle->nb_efd) {
1166 for (i = 0; i < intr_handle->nb_efd; i++)
1167 close(intr_handle->efds[i]);
1169 intr_handle->nb_efd = 0;
1170 intr_handle->max_intr = 0;
1174 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1176 return !(!intr_handle->nb_efd);
1180 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1182 return !!(intr_handle->max_intr - intr_handle->nb_efd);