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>
51 #include <rte_common.h>
52 #include <rte_interrupts.h>
53 #include <rte_memory.h>
54 #include <rte_memzone.h>
55 #include <rte_launch.h>
57 #include <rte_per_lcore.h>
58 #include <rte_lcore.h>
59 #include <rte_atomic.h>
60 #include <rte_branch_prediction.h>
61 #include <rte_debug.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(const 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(const 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_MASK;
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 masking 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(const 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(const 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(const 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 /* 0 < irq_set->count < RTE_MAX_RXTX_INTR_VEC_ID + 1 */
295 irq_set->count = intr_handle->max_intr ?
296 (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID + 1 ?
297 RTE_MAX_RXTX_INTR_VEC_ID + 1 : intr_handle->max_intr) : 1;
298 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
299 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
301 fd_ptr = (int *) &irq_set->data;
302 /* INTR vector offset 0 reserve for non-efds mapping */
303 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
304 memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
305 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
307 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
310 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
318 /* disable MSI-X interrupts */
320 vfio_disable_msix(const struct rte_intr_handle *intr_handle) {
321 struct vfio_irq_set *irq_set;
322 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
325 len = sizeof(struct vfio_irq_set);
327 irq_set = (struct vfio_irq_set *) irq_set_buf;
328 irq_set->argsz = len;
330 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
331 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
334 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
338 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
345 uio_intx_intr_disable(const struct rte_intr_handle *intr_handle)
347 unsigned char command_high;
349 /* use UIO config file descriptor for uio_pci_generic */
350 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
352 "Error reading interrupts status for fd %d\n",
353 intr_handle->uio_cfg_fd);
356 /* disable interrupts */
358 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
360 "Error disabling interrupts for fd %d\n",
361 intr_handle->uio_cfg_fd);
369 uio_intx_intr_enable(const struct rte_intr_handle *intr_handle)
371 unsigned char command_high;
373 /* use UIO config file descriptor for uio_pci_generic */
374 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
376 "Error reading interrupts status for fd %d\n",
377 intr_handle->uio_cfg_fd);
380 /* enable interrupts */
381 command_high &= ~0x4;
382 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
384 "Error enabling interrupts for fd %d\n",
385 intr_handle->uio_cfg_fd);
393 uio_intr_disable(const struct rte_intr_handle *intr_handle)
397 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
399 "Error disabling interrupts for fd %d (%s)\n",
400 intr_handle->fd, strerror(errno));
407 uio_intr_enable(const struct rte_intr_handle *intr_handle)
411 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
413 "Error enabling interrupts for fd %d (%s)\n",
414 intr_handle->fd, strerror(errno));
421 rte_intr_callback_register(const struct rte_intr_handle *intr_handle,
422 rte_intr_callback_fn cb, void *cb_arg)
424 int ret, wake_thread;
425 struct rte_intr_source *src;
426 struct rte_intr_callback *callback;
430 /* first do parameter checking */
431 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
433 "Registering with invalid input parameter\n");
437 /* allocate a new interrupt callback entity */
438 callback = rte_zmalloc("interrupt callback list",
439 sizeof(*callback), 0);
440 if (callback == NULL) {
441 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
444 callback->cb_fn = cb;
445 callback->cb_arg = cb_arg;
447 rte_spinlock_lock(&intr_lock);
449 /* check if there is at least one callback registered for the fd */
450 TAILQ_FOREACH(src, &intr_sources, next) {
451 if (src->intr_handle.fd == intr_handle->fd) {
452 /* we had no interrupts for this */
453 if TAILQ_EMPTY(&src->callbacks)
456 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
462 /* no existing callbacks for this - add new source */
464 if ((src = rte_zmalloc("interrupt source list",
465 sizeof(*src), 0)) == NULL) {
466 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
470 src->intr_handle = *intr_handle;
471 TAILQ_INIT(&src->callbacks);
472 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
473 TAILQ_INSERT_TAIL(&intr_sources, src, next);
479 rte_spinlock_unlock(&intr_lock);
482 * check if need to notify the pipe fd waited by epoll_wait to
483 * rebuild the wait list.
486 if (write(intr_pipe.writefd, "1", 1) < 0)
493 rte_intr_callback_unregister(const struct rte_intr_handle *intr_handle,
494 rte_intr_callback_fn cb_fn, void *cb_arg)
497 struct rte_intr_source *src;
498 struct rte_intr_callback *cb, *next;
500 /* do parameter checking first */
501 if (intr_handle == NULL || intr_handle->fd < 0) {
503 "Unregistering with invalid input parameter\n");
507 rte_spinlock_lock(&intr_lock);
509 /* check if the insterrupt source for the fd is existent */
510 TAILQ_FOREACH(src, &intr_sources, next)
511 if (src->intr_handle.fd == intr_handle->fd)
514 /* No interrupt source registered for the fd */
518 /* interrupt source has some active callbacks right now. */
519 } else if (src->active != 0) {
526 /*walk through the callbacks and remove all that match. */
527 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
529 next = TAILQ_NEXT(cb, next);
531 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
532 cb->cb_arg == cb_arg)) {
533 TAILQ_REMOVE(&src->callbacks, cb, next);
539 /* all callbacks for that source are removed. */
540 if (TAILQ_EMPTY(&src->callbacks)) {
541 TAILQ_REMOVE(&intr_sources, src, next);
546 rte_spinlock_unlock(&intr_lock);
548 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
549 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
557 rte_intr_enable(const struct rte_intr_handle *intr_handle)
559 if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
562 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
565 switch (intr_handle->type){
566 /* write to the uio fd to enable the interrupt */
567 case RTE_INTR_HANDLE_UIO:
568 if (uio_intr_enable(intr_handle))
571 case RTE_INTR_HANDLE_UIO_INTX:
572 if (uio_intx_intr_enable(intr_handle))
575 /* not used at this moment */
576 case RTE_INTR_HANDLE_ALARM:
579 case RTE_INTR_HANDLE_VFIO_MSIX:
580 if (vfio_enable_msix(intr_handle))
583 case RTE_INTR_HANDLE_VFIO_MSI:
584 if (vfio_enable_msi(intr_handle))
587 case RTE_INTR_HANDLE_VFIO_LEGACY:
588 if (vfio_enable_intx(intr_handle))
592 /* unknown handle type */
595 "Unknown handle type of fd %d\n",
604 rte_intr_disable(const struct rte_intr_handle *intr_handle)
606 if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
609 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
612 switch (intr_handle->type){
613 /* write to the uio fd to disable the interrupt */
614 case RTE_INTR_HANDLE_UIO:
615 if (uio_intr_disable(intr_handle))
618 case RTE_INTR_HANDLE_UIO_INTX:
619 if (uio_intx_intr_disable(intr_handle))
622 /* not used at this moment */
623 case RTE_INTR_HANDLE_ALARM:
626 case RTE_INTR_HANDLE_VFIO_MSIX:
627 if (vfio_disable_msix(intr_handle))
630 case RTE_INTR_HANDLE_VFIO_MSI:
631 if (vfio_disable_msi(intr_handle))
634 case RTE_INTR_HANDLE_VFIO_LEGACY:
635 if (vfio_disable_intx(intr_handle))
639 /* unknown handle type */
642 "Unknown handle type of fd %d\n",
651 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
655 struct rte_intr_source *src;
656 struct rte_intr_callback *cb;
657 union rte_intr_read_buffer buf;
658 struct rte_intr_callback active_cb;
660 for (n = 0; n < nfds; n++) {
663 * if the pipe fd is ready to read, return out to
664 * rebuild the wait list.
666 if (events[n].data.fd == intr_pipe.readfd){
667 int r = read(intr_pipe.readfd, buf.charbuf,
668 sizeof(buf.charbuf));
672 rte_spinlock_lock(&intr_lock);
673 TAILQ_FOREACH(src, &intr_sources, next)
674 if (src->intr_handle.fd ==
678 rte_spinlock_unlock(&intr_lock);
682 /* mark this interrupt source as active and release the lock. */
684 rte_spinlock_unlock(&intr_lock);
686 /* set the length to be read dor different handle type */
687 switch (src->intr_handle.type) {
688 case RTE_INTR_HANDLE_UIO:
689 case RTE_INTR_HANDLE_UIO_INTX:
690 bytes_read = sizeof(buf.uio_intr_count);
692 case RTE_INTR_HANDLE_ALARM:
693 bytes_read = sizeof(buf.timerfd_num);
696 case RTE_INTR_HANDLE_VFIO_MSIX:
697 case RTE_INTR_HANDLE_VFIO_MSI:
698 case RTE_INTR_HANDLE_VFIO_LEGACY:
699 bytes_read = sizeof(buf.vfio_intr_count);
702 case RTE_INTR_HANDLE_VDEV:
703 case RTE_INTR_HANDLE_EXT:
713 if (bytes_read > 0) {
715 * read out to clear the ready-to-be-read flag
718 bytes_read = read(events[n].data.fd, &buf, bytes_read);
719 if (bytes_read < 0) {
720 if (errno == EINTR || errno == EWOULDBLOCK)
723 RTE_LOG(ERR, EAL, "Error reading from file "
724 "descriptor %d: %s\n",
727 } else if (bytes_read == 0)
728 RTE_LOG(ERR, EAL, "Read nothing from file "
729 "descriptor %d\n", events[n].data.fd);
734 /* grab a lock, again to call callbacks and update status. */
735 rte_spinlock_lock(&intr_lock);
739 /* Finally, call all callbacks. */
740 TAILQ_FOREACH(cb, &src->callbacks, next) {
742 /* make a copy and unlock. */
744 rte_spinlock_unlock(&intr_lock);
746 /* call the actual callback */
747 active_cb.cb_fn(active_cb.cb_arg);
749 /*get the lock back. */
750 rte_spinlock_lock(&intr_lock);
754 /* we done with that interrupt source, release it. */
756 rte_spinlock_unlock(&intr_lock);
763 * It handles all the interrupts.
766 * epoll file descriptor.
768 * The number of file descriptors added in epoll.
774 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
776 struct epoll_event events[totalfds];
780 nfds = epoll_wait(pfd, events, totalfds,
781 EAL_INTR_EPOLL_WAIT_FOREVER);
782 /* epoll_wait fail */
787 "epoll_wait returns with fail\n");
790 /* epoll_wait timeout, will never happens here */
793 /* epoll_wait has at least one fd ready to read */
794 if (eal_intr_process_interrupts(events, nfds) < 0)
800 * It builds/rebuilds up the epoll file descriptor with all the
801 * file descriptors being waited on. Then handles the interrupts.
809 static __attribute__((noreturn)) void *
810 eal_intr_thread_main(__rte_unused void *arg)
812 struct epoll_event ev;
814 /* host thread, never break out */
816 /* build up the epoll fd with all descriptors we are to
817 * wait on then pass it to the handle_interrupts function
819 static struct epoll_event pipe_event = {
820 .events = EPOLLIN | EPOLLPRI,
822 struct rte_intr_source *src;
825 /* create epoll fd */
826 int pfd = epoll_create(1);
828 rte_panic("Cannot create epoll instance\n");
830 pipe_event.data.fd = intr_pipe.readfd;
832 * add pipe fd into wait list, this pipe is used to
833 * rebuild the wait list.
835 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
837 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
838 intr_pipe.readfd, strerror(errno));
842 rte_spinlock_lock(&intr_lock);
844 TAILQ_FOREACH(src, &intr_sources, next) {
845 if (src->callbacks.tqh_first == NULL)
846 continue; /* skip those with no callbacks */
847 ev.events = EPOLLIN | EPOLLPRI | EPOLLRDHUP | EPOLLHUP;
848 ev.data.fd = src->intr_handle.fd;
851 * add all the uio device file descriptor
854 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
855 src->intr_handle.fd, &ev) < 0){
856 rte_panic("Error adding fd %d epoll_ctl, %s\n",
857 src->intr_handle.fd, strerror(errno));
862 rte_spinlock_unlock(&intr_lock);
863 /* serve the interrupt */
864 eal_intr_handle_interrupts(pfd, numfds);
867 * when we return, we need to rebuild the
868 * list of fds to monitor.
875 rte_eal_intr_init(void)
877 int ret = 0, ret_1 = 0;
878 char thread_name[RTE_MAX_THREAD_NAME_LEN];
880 /* init the global interrupt source head */
881 TAILQ_INIT(&intr_sources);
884 * create a pipe which will be waited by epoll and notified to
885 * rebuild the wait list of epoll.
887 if (pipe(intr_pipe.pipefd) < 0) {
892 /* create the host thread to wait/handle the interrupt */
893 ret = pthread_create(&intr_thread, NULL,
894 eal_intr_thread_main, NULL);
898 "Failed to create thread for interrupt handling\n");
900 /* Set thread_name for aid in debugging. */
901 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
903 ret_1 = rte_thread_setname(intr_thread, thread_name);
906 "Failed to set thread name for interrupt handling\n");
913 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
915 union rte_intr_read_buffer buf;
919 switch (intr_handle->type) {
920 case RTE_INTR_HANDLE_UIO:
921 case RTE_INTR_HANDLE_UIO_INTX:
922 bytes_read = sizeof(buf.uio_intr_count);
925 case RTE_INTR_HANDLE_VFIO_MSIX:
926 case RTE_INTR_HANDLE_VFIO_MSI:
927 case RTE_INTR_HANDLE_VFIO_LEGACY:
928 bytes_read = sizeof(buf.vfio_intr_count);
931 case RTE_INTR_HANDLE_VDEV:
932 /* for vdev, fd points to:
933 * a. eventfd which does not need to read out;
934 * b. datapath fd which needs PMD to read out.
937 case RTE_INTR_HANDLE_EXT:
941 RTE_LOG(INFO, EAL, "unexpected intr type\n");
946 * read out to clear the ready-to-be-read flag
950 nbytes = read(fd, &buf, bytes_read);
952 if (errno == EINTR || errno == EWOULDBLOCK ||
956 "Error reading from fd %d: %s\n",
957 fd, strerror(errno));
958 } else if (nbytes == 0)
959 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
965 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
966 struct rte_epoll_event *events)
968 unsigned int i, count = 0;
969 struct rte_epoll_event *rev;
971 for (i = 0; i < n; i++) {
972 rev = evs[i].data.ptr;
973 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
977 events[count].status = RTE_EPOLL_VALID;
978 events[count].fd = rev->fd;
979 events[count].epfd = rev->epfd;
980 events[count].epdata.event = rev->epdata.event;
981 events[count].epdata.data = rev->epdata.data;
982 if (rev->epdata.cb_fun)
983 rev->epdata.cb_fun(rev->fd,
986 rte_compiler_barrier();
987 rev->status = RTE_EPOLL_VALID;
994 eal_init_tls_epfd(void)
996 int pfd = epoll_create(255);
1000 "Cannot create epoll instance\n");
1007 rte_intr_tls_epfd(void)
1009 if (RTE_PER_LCORE(_epfd) == -1)
1010 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
1012 return RTE_PER_LCORE(_epfd);
1016 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
1017 int maxevents, int timeout)
1019 struct epoll_event evs[maxevents];
1023 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1027 /* using per thread epoll fd */
1028 if (epfd == RTE_EPOLL_PER_THREAD)
1029 epfd = rte_intr_tls_epfd();
1032 rc = epoll_wait(epfd, evs, maxevents, timeout);
1033 if (likely(rc > 0)) {
1034 /* epoll_wait has at least one fd ready to read */
1035 rc = eal_epoll_process_event(evs, rc, events);
1037 } else if (rc < 0) {
1040 /* epoll_wait fail */
1041 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1046 /* rc == 0, epoll_wait timed out */
1055 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1057 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1059 while (ev->status != RTE_EPOLL_VALID)
1061 memset(&ev->epdata, 0, sizeof(ev->epdata));
1067 rte_epoll_ctl(int epfd, int op, int fd,
1068 struct rte_epoll_event *event)
1070 struct epoll_event ev;
1073 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1077 /* using per thread epoll fd */
1078 if (epfd == RTE_EPOLL_PER_THREAD)
1079 epfd = rte_intr_tls_epfd();
1081 if (op == EPOLL_CTL_ADD) {
1082 event->status = RTE_EPOLL_VALID;
1083 event->fd = fd; /* ignore fd in event */
1085 ev.data.ptr = (void *)event;
1088 ev.events = event->epdata.event;
1089 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1090 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1091 op, fd, strerror(errno));
1092 if (op == EPOLL_CTL_ADD)
1093 /* rollback status when CTL_ADD fail */
1094 event->status = RTE_EPOLL_INVALID;
1098 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1099 eal_epoll_data_safe_free(event);
1105 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1106 int op, unsigned int vec, void *data)
1108 struct rte_epoll_event *rev;
1109 struct rte_epoll_data *epdata;
1111 unsigned int efd_idx;
1114 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1115 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1117 if (!intr_handle || intr_handle->nb_efd == 0 ||
1118 efd_idx >= intr_handle->nb_efd) {
1119 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1124 case RTE_INTR_EVENT_ADD:
1125 epfd_op = EPOLL_CTL_ADD;
1126 rev = &intr_handle->elist[efd_idx];
1127 if (rev->status != RTE_EPOLL_INVALID) {
1128 RTE_LOG(INFO, EAL, "Event already been added.\n");
1132 /* attach to intr vector fd */
1133 epdata = &rev->epdata;
1134 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1135 epdata->data = data;
1136 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1137 epdata->cb_arg = (void *)intr_handle;
1138 rc = rte_epoll_ctl(epfd, epfd_op,
1139 intr_handle->efds[efd_idx], rev);
1142 "efd %d associated with vec %d added on epfd %d"
1143 "\n", rev->fd, vec, epfd);
1147 case RTE_INTR_EVENT_DEL:
1148 epfd_op = EPOLL_CTL_DEL;
1149 rev = &intr_handle->elist[efd_idx];
1150 if (rev->status == RTE_EPOLL_INVALID) {
1151 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1155 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1160 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1168 rte_intr_free_epoll_fd(struct rte_intr_handle *intr_handle)
1171 struct rte_epoll_event *rev;
1173 for (i = 0; i < intr_handle->nb_efd; i++) {
1174 rev = &intr_handle->elist[i];
1175 if (rev->status == RTE_EPOLL_INVALID)
1177 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1178 /* force free if the entry valid */
1179 eal_epoll_data_safe_free(rev);
1180 rev->status = RTE_EPOLL_INVALID;
1186 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1190 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1192 assert(nb_efd != 0);
1194 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1195 for (i = 0; i < n; i++) {
1196 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1199 "can't setup eventfd, error %i (%s)\n",
1200 errno, strerror(errno));
1203 intr_handle->efds[i] = fd;
1205 intr_handle->nb_efd = n;
1206 intr_handle->max_intr = NB_OTHER_INTR + n;
1207 } else if (intr_handle->type == RTE_INTR_HANDLE_VDEV) {
1208 /* do nothing, and let vdev driver to initialize this struct */
1210 intr_handle->efds[0] = intr_handle->fd;
1211 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1212 intr_handle->max_intr = NB_OTHER_INTR;
1219 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1223 rte_intr_free_epoll_fd(intr_handle);
1224 if (intr_handle->max_intr > intr_handle->nb_efd) {
1225 for (i = 0; i < intr_handle->nb_efd; i++)
1226 close(intr_handle->efds[i]);
1228 intr_handle->nb_efd = 0;
1229 intr_handle->max_intr = 0;
1233 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1235 return !(!intr_handle->nb_efd);
1239 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1241 if (!rte_intr_dp_is_en(intr_handle))
1244 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1248 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1250 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)
1253 if (intr_handle->type == RTE_INTR_HANDLE_VDEV)