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>
67 #include <rte_pause.h>
70 #include "eal_private.h"
72 #include "eal_thread.h"
74 #define EAL_INTR_EPOLL_WAIT_FOREVER (-1)
75 #define NB_OTHER_INTR 1
77 static RTE_DEFINE_PER_LCORE(int, _epfd) = -1; /**< epoll fd per thread */
93 * union buffer for reading on different devices
95 union rte_intr_read_buffer {
96 int uio_intr_count; /* for uio device */
98 uint64_t vfio_intr_count; /* for vfio device */
100 uint64_t timerfd_num; /* for timerfd */
101 char charbuf[16]; /* for others */
104 TAILQ_HEAD(rte_intr_cb_list, rte_intr_callback);
105 TAILQ_HEAD(rte_intr_source_list, rte_intr_source);
107 struct rte_intr_callback {
108 TAILQ_ENTRY(rte_intr_callback) next;
109 rte_intr_callback_fn cb_fn; /**< callback address */
110 void *cb_arg; /**< parameter for callback */
113 struct rte_intr_source {
114 TAILQ_ENTRY(rte_intr_source) next;
115 struct rte_intr_handle intr_handle; /**< interrupt handle */
116 struct rte_intr_cb_list callbacks; /**< user callbacks */
120 /* global spinlock for interrupt data operation */
121 static rte_spinlock_t intr_lock = RTE_SPINLOCK_INITIALIZER;
123 /* union buffer for pipe read/write */
124 static union intr_pipefds intr_pipe;
126 /* interrupt sources list */
127 static struct rte_intr_source_list intr_sources;
129 /* interrupt handling thread */
130 static pthread_t intr_thread;
132 /* VFIO interrupts */
135 #define IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + sizeof(int))
136 /* irq set buffer length for queue interrupts and LSC interrupt */
137 #define MSIX_IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + \
138 sizeof(int) * (RTE_MAX_RXTX_INTR_VEC_ID + 1))
140 /* enable legacy (INTx) interrupts */
142 vfio_enable_intx(const struct rte_intr_handle *intr_handle) {
143 struct vfio_irq_set *irq_set;
144 char irq_set_buf[IRQ_SET_BUF_LEN];
148 len = sizeof(irq_set_buf);
151 irq_set = (struct vfio_irq_set *) irq_set_buf;
152 irq_set->argsz = len;
154 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
155 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
157 fd_ptr = (int *) &irq_set->data;
158 *fd_ptr = intr_handle->fd;
160 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
163 RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
168 /* unmask INTx after enabling */
169 memset(irq_set, 0, len);
170 len = sizeof(struct vfio_irq_set);
171 irq_set->argsz = len;
173 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
174 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
177 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
180 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
187 /* disable legacy (INTx) interrupts */
189 vfio_disable_intx(const struct rte_intr_handle *intr_handle) {
190 struct vfio_irq_set *irq_set;
191 char irq_set_buf[IRQ_SET_BUF_LEN];
194 len = sizeof(struct vfio_irq_set);
196 /* mask interrupts before disabling */
197 irq_set = (struct vfio_irq_set *) irq_set_buf;
198 irq_set->argsz = len;
200 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_MASK;
201 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
204 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
207 RTE_LOG(ERR, EAL, "Error masking INTx interrupts for fd %d\n",
213 memset(irq_set, 0, len);
214 irq_set->argsz = len;
216 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
217 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
220 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
224 "Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
230 /* enable MSI interrupts */
232 vfio_enable_msi(const struct rte_intr_handle *intr_handle) {
234 char irq_set_buf[IRQ_SET_BUF_LEN];
235 struct vfio_irq_set *irq_set;
238 len = sizeof(irq_set_buf);
240 irq_set = (struct vfio_irq_set *) irq_set_buf;
241 irq_set->argsz = len;
243 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
244 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
246 fd_ptr = (int *) &irq_set->data;
247 *fd_ptr = intr_handle->fd;
249 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
252 RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
259 /* disable MSI interrupts */
261 vfio_disable_msi(const struct rte_intr_handle *intr_handle) {
262 struct vfio_irq_set *irq_set;
263 char irq_set_buf[IRQ_SET_BUF_LEN];
266 len = sizeof(struct vfio_irq_set);
268 irq_set = (struct vfio_irq_set *) irq_set_buf;
269 irq_set->argsz = len;
271 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
272 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
275 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
279 "Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
284 /* enable MSI-X interrupts */
286 vfio_enable_msix(const struct rte_intr_handle *intr_handle) {
288 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
289 struct vfio_irq_set *irq_set;
292 len = sizeof(irq_set_buf);
294 irq_set = (struct vfio_irq_set *) irq_set_buf;
295 irq_set->argsz = len;
296 /* 0 < irq_set->count < RTE_MAX_RXTX_INTR_VEC_ID + 1 */
297 irq_set->count = intr_handle->max_intr ?
298 (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID + 1 ?
299 RTE_MAX_RXTX_INTR_VEC_ID + 1 : intr_handle->max_intr) : 1;
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(const 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(const 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(const 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(const 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(const 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(const 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(const 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(const struct rte_intr_handle *intr_handle)
561 if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
564 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
567 switch (intr_handle->type){
568 /* write to the uio fd to enable the interrupt */
569 case RTE_INTR_HANDLE_UIO:
570 if (uio_intr_enable(intr_handle))
573 case RTE_INTR_HANDLE_UIO_INTX:
574 if (uio_intx_intr_enable(intr_handle))
577 /* not used at this moment */
578 case RTE_INTR_HANDLE_ALARM:
581 case RTE_INTR_HANDLE_VFIO_MSIX:
582 if (vfio_enable_msix(intr_handle))
585 case RTE_INTR_HANDLE_VFIO_MSI:
586 if (vfio_enable_msi(intr_handle))
589 case RTE_INTR_HANDLE_VFIO_LEGACY:
590 if (vfio_enable_intx(intr_handle))
594 /* unknown handle type */
597 "Unknown handle type of fd %d\n",
606 rte_intr_disable(const struct rte_intr_handle *intr_handle)
608 if (intr_handle && intr_handle->type == RTE_INTR_HANDLE_VDEV)
611 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
614 switch (intr_handle->type){
615 /* write to the uio fd to disable the interrupt */
616 case RTE_INTR_HANDLE_UIO:
617 if (uio_intr_disable(intr_handle))
620 case RTE_INTR_HANDLE_UIO_INTX:
621 if (uio_intx_intr_disable(intr_handle))
624 /* not used at this moment */
625 case RTE_INTR_HANDLE_ALARM:
628 case RTE_INTR_HANDLE_VFIO_MSIX:
629 if (vfio_disable_msix(intr_handle))
632 case RTE_INTR_HANDLE_VFIO_MSI:
633 if (vfio_disable_msi(intr_handle))
636 case RTE_INTR_HANDLE_VFIO_LEGACY:
637 if (vfio_disable_intx(intr_handle))
641 /* unknown handle type */
644 "Unknown handle type of fd %d\n",
653 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
657 struct rte_intr_source *src;
658 struct rte_intr_callback *cb;
659 union rte_intr_read_buffer buf;
660 struct rte_intr_callback active_cb;
662 for (n = 0; n < nfds; n++) {
665 * if the pipe fd is ready to read, return out to
666 * rebuild the wait list.
668 if (events[n].data.fd == intr_pipe.readfd){
669 int r = read(intr_pipe.readfd, buf.charbuf,
670 sizeof(buf.charbuf));
674 rte_spinlock_lock(&intr_lock);
675 TAILQ_FOREACH(src, &intr_sources, next)
676 if (src->intr_handle.fd ==
680 rte_spinlock_unlock(&intr_lock);
684 /* mark this interrupt source as active and release the lock. */
686 rte_spinlock_unlock(&intr_lock);
688 /* set the length to be read dor different handle type */
689 switch (src->intr_handle.type) {
690 case RTE_INTR_HANDLE_UIO:
691 case RTE_INTR_HANDLE_UIO_INTX:
692 bytes_read = sizeof(buf.uio_intr_count);
694 case RTE_INTR_HANDLE_ALARM:
695 bytes_read = sizeof(buf.timerfd_num);
698 case RTE_INTR_HANDLE_VFIO_MSIX:
699 case RTE_INTR_HANDLE_VFIO_MSI:
700 case RTE_INTR_HANDLE_VFIO_LEGACY:
701 bytes_read = sizeof(buf.vfio_intr_count);
704 case RTE_INTR_HANDLE_VDEV:
705 case RTE_INTR_HANDLE_EXT:
715 if (bytes_read > 0) {
717 * read out to clear the ready-to-be-read flag
720 bytes_read = read(events[n].data.fd, &buf, bytes_read);
721 if (bytes_read < 0) {
722 if (errno == EINTR || errno == EWOULDBLOCK)
725 RTE_LOG(ERR, EAL, "Error reading from file "
726 "descriptor %d: %s\n",
729 } else if (bytes_read == 0)
730 RTE_LOG(ERR, EAL, "Read nothing from file "
731 "descriptor %d\n", events[n].data.fd);
736 /* grab a lock, again to call callbacks and update status. */
737 rte_spinlock_lock(&intr_lock);
741 /* Finally, call all callbacks. */
742 TAILQ_FOREACH(cb, &src->callbacks, next) {
744 /* make a copy and unlock. */
746 rte_spinlock_unlock(&intr_lock);
748 /* call the actual callback */
749 active_cb.cb_fn(active_cb.cb_arg);
751 /*get the lock back. */
752 rte_spinlock_lock(&intr_lock);
756 /* we done with that interrupt source, release it. */
758 rte_spinlock_unlock(&intr_lock);
765 * It handles all the interrupts.
768 * epoll file descriptor.
770 * The number of file descriptors added in epoll.
776 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
778 struct epoll_event events[totalfds];
782 nfds = epoll_wait(pfd, events, totalfds,
783 EAL_INTR_EPOLL_WAIT_FOREVER);
784 /* epoll_wait fail */
789 "epoll_wait returns with fail\n");
792 /* epoll_wait timeout, will never happens here */
795 /* epoll_wait has at least one fd ready to read */
796 if (eal_intr_process_interrupts(events, nfds) < 0)
802 * It builds/rebuilds up the epoll file descriptor with all the
803 * file descriptors being waited on. Then handles the interrupts.
811 static __attribute__((noreturn)) void *
812 eal_intr_thread_main(__rte_unused void *arg)
814 struct epoll_event ev;
816 /* host thread, never break out */
818 /* build up the epoll fd with all descriptors we are to
819 * wait on then pass it to the handle_interrupts function
821 static struct epoll_event pipe_event = {
822 .events = EPOLLIN | EPOLLPRI,
824 struct rte_intr_source *src;
827 /* create epoll fd */
828 int pfd = epoll_create(1);
830 rte_panic("Cannot create epoll instance\n");
832 pipe_event.data.fd = intr_pipe.readfd;
834 * add pipe fd into wait list, this pipe is used to
835 * rebuild the wait list.
837 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
839 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
840 intr_pipe.readfd, strerror(errno));
844 rte_spinlock_lock(&intr_lock);
846 TAILQ_FOREACH(src, &intr_sources, next) {
847 if (src->callbacks.tqh_first == NULL)
848 continue; /* skip those with no callbacks */
849 ev.events = EPOLLIN | EPOLLPRI | EPOLLRDHUP | EPOLLHUP;
850 ev.data.fd = src->intr_handle.fd;
853 * add all the uio device file descriptor
856 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
857 src->intr_handle.fd, &ev) < 0){
858 rte_panic("Error adding fd %d epoll_ctl, %s\n",
859 src->intr_handle.fd, strerror(errno));
864 rte_spinlock_unlock(&intr_lock);
865 /* serve the interrupt */
866 eal_intr_handle_interrupts(pfd, numfds);
869 * when we return, we need to rebuild the
870 * list of fds to monitor.
877 rte_eal_intr_init(void)
879 int ret = 0, ret_1 = 0;
880 char thread_name[RTE_MAX_THREAD_NAME_LEN];
882 /* init the global interrupt source head */
883 TAILQ_INIT(&intr_sources);
886 * create a pipe which will be waited by epoll and notified to
887 * rebuild the wait list of epoll.
889 if (pipe(intr_pipe.pipefd) < 0) {
894 /* create the host thread to wait/handle the interrupt */
895 ret = pthread_create(&intr_thread, NULL,
896 eal_intr_thread_main, NULL);
900 "Failed to create thread for interrupt handling\n");
902 /* Set thread_name for aid in debugging. */
903 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
905 ret_1 = rte_thread_setname(intr_thread, thread_name);
908 "Failed to set thread name for interrupt handling\n");
915 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
917 union rte_intr_read_buffer buf;
921 switch (intr_handle->type) {
922 case RTE_INTR_HANDLE_UIO:
923 case RTE_INTR_HANDLE_UIO_INTX:
924 bytes_read = sizeof(buf.uio_intr_count);
927 case RTE_INTR_HANDLE_VFIO_MSIX:
928 case RTE_INTR_HANDLE_VFIO_MSI:
929 case RTE_INTR_HANDLE_VFIO_LEGACY:
930 bytes_read = sizeof(buf.vfio_intr_count);
933 case RTE_INTR_HANDLE_VDEV:
934 /* for vdev, fd points to:
935 * a. eventfd which does not need to read out;
936 * b. datapath fd which needs PMD to read out.
939 case RTE_INTR_HANDLE_EXT:
943 RTE_LOG(INFO, EAL, "unexpected intr type\n");
948 * read out to clear the ready-to-be-read flag
952 nbytes = read(fd, &buf, bytes_read);
954 if (errno == EINTR || errno == EWOULDBLOCK ||
958 "Error reading from fd %d: %s\n",
959 fd, strerror(errno));
960 } else if (nbytes == 0)
961 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
967 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
968 struct rte_epoll_event *events)
970 unsigned int i, count = 0;
971 struct rte_epoll_event *rev;
973 for (i = 0; i < n; i++) {
974 rev = evs[i].data.ptr;
975 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
979 events[count].status = RTE_EPOLL_VALID;
980 events[count].fd = rev->fd;
981 events[count].epfd = rev->epfd;
982 events[count].epdata.event = rev->epdata.event;
983 events[count].epdata.data = rev->epdata.data;
984 if (rev->epdata.cb_fun)
985 rev->epdata.cb_fun(rev->fd,
988 rte_compiler_barrier();
989 rev->status = RTE_EPOLL_VALID;
996 eal_init_tls_epfd(void)
998 int pfd = epoll_create(255);
1002 "Cannot create epoll instance\n");
1009 rte_intr_tls_epfd(void)
1011 if (RTE_PER_LCORE(_epfd) == -1)
1012 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
1014 return RTE_PER_LCORE(_epfd);
1018 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
1019 int maxevents, int timeout)
1021 struct epoll_event evs[maxevents];
1025 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1029 /* using per thread epoll fd */
1030 if (epfd == RTE_EPOLL_PER_THREAD)
1031 epfd = rte_intr_tls_epfd();
1034 rc = epoll_wait(epfd, evs, maxevents, timeout);
1035 if (likely(rc > 0)) {
1036 /* epoll_wait has at least one fd ready to read */
1037 rc = eal_epoll_process_event(evs, rc, events);
1039 } else if (rc < 0) {
1042 /* epoll_wait fail */
1043 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1048 /* rc == 0, epoll_wait timed out */
1057 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1059 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1061 while (ev->status != RTE_EPOLL_VALID)
1063 memset(&ev->epdata, 0, sizeof(ev->epdata));
1069 rte_epoll_ctl(int epfd, int op, int fd,
1070 struct rte_epoll_event *event)
1072 struct epoll_event ev;
1075 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1079 /* using per thread epoll fd */
1080 if (epfd == RTE_EPOLL_PER_THREAD)
1081 epfd = rte_intr_tls_epfd();
1083 if (op == EPOLL_CTL_ADD) {
1084 event->status = RTE_EPOLL_VALID;
1085 event->fd = fd; /* ignore fd in event */
1087 ev.data.ptr = (void *)event;
1090 ev.events = event->epdata.event;
1091 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1092 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1093 op, fd, strerror(errno));
1094 if (op == EPOLL_CTL_ADD)
1095 /* rollback status when CTL_ADD fail */
1096 event->status = RTE_EPOLL_INVALID;
1100 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1101 eal_epoll_data_safe_free(event);
1107 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1108 int op, unsigned int vec, void *data)
1110 struct rte_epoll_event *rev;
1111 struct rte_epoll_data *epdata;
1113 unsigned int efd_idx;
1116 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1117 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1119 if (!intr_handle || intr_handle->nb_efd == 0 ||
1120 efd_idx >= intr_handle->nb_efd) {
1121 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1126 case RTE_INTR_EVENT_ADD:
1127 epfd_op = EPOLL_CTL_ADD;
1128 rev = &intr_handle->elist[efd_idx];
1129 if (rev->status != RTE_EPOLL_INVALID) {
1130 RTE_LOG(INFO, EAL, "Event already been added.\n");
1134 /* attach to intr vector fd */
1135 epdata = &rev->epdata;
1136 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1137 epdata->data = data;
1138 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1139 epdata->cb_arg = (void *)intr_handle;
1140 rc = rte_epoll_ctl(epfd, epfd_op,
1141 intr_handle->efds[efd_idx], rev);
1144 "efd %d associated with vec %d added on epfd %d"
1145 "\n", rev->fd, vec, epfd);
1149 case RTE_INTR_EVENT_DEL:
1150 epfd_op = EPOLL_CTL_DEL;
1151 rev = &intr_handle->elist[efd_idx];
1152 if (rev->status == RTE_EPOLL_INVALID) {
1153 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1157 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1162 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1170 rte_intr_free_epoll_fd(struct rte_intr_handle *intr_handle)
1173 struct rte_epoll_event *rev;
1175 for (i = 0; i < intr_handle->nb_efd; i++) {
1176 rev = &intr_handle->elist[i];
1177 if (rev->status == RTE_EPOLL_INVALID)
1179 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1180 /* force free if the entry valid */
1181 eal_epoll_data_safe_free(rev);
1182 rev->status = RTE_EPOLL_INVALID;
1188 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1192 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1194 assert(nb_efd != 0);
1196 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1197 for (i = 0; i < n; i++) {
1198 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1201 "can't setup eventfd, error %i (%s)\n",
1202 errno, strerror(errno));
1205 intr_handle->efds[i] = fd;
1207 intr_handle->nb_efd = n;
1208 intr_handle->max_intr = NB_OTHER_INTR + n;
1209 } else if (intr_handle->type == RTE_INTR_HANDLE_VDEV) {
1210 /* do nothing, and let vdev driver to initialize this struct */
1212 intr_handle->efds[0] = intr_handle->fd;
1213 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1214 intr_handle->max_intr = NB_OTHER_INTR;
1221 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1225 rte_intr_free_epoll_fd(intr_handle);
1226 if (intr_handle->max_intr > intr_handle->nb_efd) {
1227 for (i = 0; i < intr_handle->nb_efd; i++)
1228 close(intr_handle->efds[i]);
1230 intr_handle->nb_efd = 0;
1231 intr_handle->max_intr = 0;
1235 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1237 return !(!intr_handle->nb_efd);
1241 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1243 if (!rte_intr_dp_is_en(intr_handle))
1246 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1250 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1252 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)
1255 if (intr_handle->type == RTE_INTR_HANDLE_VDEV)