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>
50 #include <rte_common.h>
51 #include <rte_interrupts.h>
52 #include <rte_memory.h>
53 #include <rte_memzone.h>
54 #include <rte_launch.h>
56 #include <rte_per_lcore.h>
57 #include <rte_lcore.h>
58 #include <rte_atomic.h>
59 #include <rte_branch_prediction.h>
60 #include <rte_debug.h>
63 #include <rte_malloc.h>
64 #include <rte_errno.h>
65 #include <rte_spinlock.h>
67 #include "eal_private.h"
69 #include "eal_thread.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(const 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(const 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(const 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(const 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);
282 get_max_intr(const struct rte_intr_handle *intr_handle)
284 struct rte_intr_source *src;
286 TAILQ_FOREACH(src, &intr_sources, next) {
287 if (src->intr_handle.fd != intr_handle->fd)
290 if (src->intr_handle.max_intr < intr_handle->max_intr)
291 src->intr_handle.max_intr = intr_handle->max_intr;
292 if (!src->intr_handle.max_intr)
293 src->intr_handle.max_intr = 1;
294 else if (src->intr_handle.max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
295 src->intr_handle.max_intr
296 = RTE_MAX_RXTX_INTR_VEC_ID + 1;
298 return src->intr_handle.max_intr;
304 /* enable MSI-X interrupts */
306 vfio_enable_msix(const struct rte_intr_handle *intr_handle) {
308 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
309 struct vfio_irq_set *irq_set;
312 len = sizeof(irq_set_buf);
314 irq_set = (struct vfio_irq_set *) irq_set_buf;
315 irq_set->argsz = len;
317 ret = get_max_intr(intr_handle);
319 RTE_LOG(ERR, EAL, "Invalid number of MSI-X irqs for fd %d\n",
324 irq_set->count = ret;
325 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
326 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
328 fd_ptr = (int *) &irq_set->data;
329 /* INTR vector offset 0 reserve for non-efds mapping */
330 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
331 memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
332 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
334 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
337 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
345 /* disable MSI-X interrupts */
347 vfio_disable_msix(const struct rte_intr_handle *intr_handle) {
348 struct vfio_irq_set *irq_set;
349 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
352 len = sizeof(struct vfio_irq_set);
354 irq_set = (struct vfio_irq_set *) irq_set_buf;
355 irq_set->argsz = len;
357 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
358 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
361 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
365 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
372 uio_intx_intr_disable(const struct rte_intr_handle *intr_handle)
374 unsigned char command_high;
376 /* use UIO config file descriptor for uio_pci_generic */
377 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
379 "Error reading interrupts status for fd %d\n",
380 intr_handle->uio_cfg_fd);
383 /* disable interrupts */
385 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
387 "Error disabling interrupts for fd %d\n",
388 intr_handle->uio_cfg_fd);
396 uio_intx_intr_enable(const struct rte_intr_handle *intr_handle)
398 unsigned char command_high;
400 /* use UIO config file descriptor for uio_pci_generic */
401 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
403 "Error reading interrupts status for fd %d\n",
404 intr_handle->uio_cfg_fd);
407 /* enable interrupts */
408 command_high &= ~0x4;
409 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
411 "Error enabling interrupts for fd %d\n",
412 intr_handle->uio_cfg_fd);
420 uio_intr_disable(const struct rte_intr_handle *intr_handle)
424 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
426 "Error disabling interrupts for fd %d (%s)\n",
427 intr_handle->fd, strerror(errno));
434 uio_intr_enable(const struct rte_intr_handle *intr_handle)
438 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
440 "Error enabling interrupts for fd %d (%s)\n",
441 intr_handle->fd, strerror(errno));
448 rte_intr_callback_register(const struct rte_intr_handle *intr_handle,
449 rte_intr_callback_fn cb, void *cb_arg)
451 int ret, wake_thread;
452 struct rte_intr_source *src;
453 struct rte_intr_callback *callback;
457 /* first do parameter checking */
458 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
460 "Registering with invalid input parameter\n");
464 /* allocate a new interrupt callback entity */
465 callback = rte_zmalloc("interrupt callback list",
466 sizeof(*callback), 0);
467 if (callback == NULL) {
468 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
471 callback->cb_fn = cb;
472 callback->cb_arg = cb_arg;
474 rte_spinlock_lock(&intr_lock);
476 /* check if there is at least one callback registered for the fd */
477 TAILQ_FOREACH(src, &intr_sources, next) {
478 if (src->intr_handle.fd == intr_handle->fd) {
479 /* we had no interrupts for this */
480 if TAILQ_EMPTY(&src->callbacks)
483 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
489 /* no existing callbacks for this - add new source */
491 if ((src = rte_zmalloc("interrupt source list",
492 sizeof(*src), 0)) == NULL) {
493 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
497 src->intr_handle = *intr_handle;
498 TAILQ_INIT(&src->callbacks);
499 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
500 TAILQ_INSERT_TAIL(&intr_sources, src, next);
506 rte_spinlock_unlock(&intr_lock);
509 * check if need to notify the pipe fd waited by epoll_wait to
510 * rebuild the wait list.
513 if (write(intr_pipe.writefd, "1", 1) < 0)
520 rte_intr_callback_unregister(const struct rte_intr_handle *intr_handle,
521 rte_intr_callback_fn cb_fn, void *cb_arg)
524 struct rte_intr_source *src;
525 struct rte_intr_callback *cb, *next;
527 /* do parameter checking first */
528 if (intr_handle == NULL || intr_handle->fd < 0) {
530 "Unregistering with invalid input parameter\n");
534 rte_spinlock_lock(&intr_lock);
536 /* check if the insterrupt source for the fd is existent */
537 TAILQ_FOREACH(src, &intr_sources, next)
538 if (src->intr_handle.fd == intr_handle->fd)
541 /* No interrupt source registered for the fd */
545 /* interrupt source has some active callbacks right now. */
546 } else if (src->active != 0) {
553 /*walk through the callbacks and remove all that match. */
554 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
556 next = TAILQ_NEXT(cb, next);
558 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
559 cb->cb_arg == cb_arg)) {
560 TAILQ_REMOVE(&src->callbacks, cb, next);
566 /* all callbacks for that source are removed. */
567 if (TAILQ_EMPTY(&src->callbacks)) {
568 TAILQ_REMOVE(&intr_sources, src, next);
573 rte_spinlock_unlock(&intr_lock);
575 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
576 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
584 rte_intr_enable(const struct rte_intr_handle *intr_handle)
586 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
589 switch (intr_handle->type){
590 /* write to the uio fd to enable the interrupt */
591 case RTE_INTR_HANDLE_UIO:
592 if (uio_intr_enable(intr_handle))
595 case RTE_INTR_HANDLE_UIO_INTX:
596 if (uio_intx_intr_enable(intr_handle))
599 /* not used at this moment */
600 case RTE_INTR_HANDLE_ALARM:
603 case RTE_INTR_HANDLE_VFIO_MSIX:
604 if (vfio_enable_msix(intr_handle))
607 case RTE_INTR_HANDLE_VFIO_MSI:
608 if (vfio_enable_msi(intr_handle))
611 case RTE_INTR_HANDLE_VFIO_LEGACY:
612 if (vfio_enable_intx(intr_handle))
616 /* unknown handle type */
619 "Unknown handle type of fd %d\n",
628 rte_intr_disable(const struct rte_intr_handle *intr_handle)
630 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
633 switch (intr_handle->type){
634 /* write to the uio fd to disable the interrupt */
635 case RTE_INTR_HANDLE_UIO:
636 if (uio_intr_disable(intr_handle))
639 case RTE_INTR_HANDLE_UIO_INTX:
640 if (uio_intx_intr_disable(intr_handle))
643 /* not used at this moment */
644 case RTE_INTR_HANDLE_ALARM:
647 case RTE_INTR_HANDLE_VFIO_MSIX:
648 if (vfio_disable_msix(intr_handle))
651 case RTE_INTR_HANDLE_VFIO_MSI:
652 if (vfio_disable_msi(intr_handle))
655 case RTE_INTR_HANDLE_VFIO_LEGACY:
656 if (vfio_disable_intx(intr_handle))
660 /* unknown handle type */
663 "Unknown handle type of fd %d\n",
672 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
675 struct rte_intr_source *src;
676 struct rte_intr_callback *cb;
677 union rte_intr_read_buffer buf;
678 struct rte_intr_callback active_cb;
680 for (n = 0; n < nfds; n++) {
683 * if the pipe fd is ready to read, return out to
684 * rebuild the wait list.
686 if (events[n].data.fd == intr_pipe.readfd){
687 int r = read(intr_pipe.readfd, buf.charbuf,
688 sizeof(buf.charbuf));
692 rte_spinlock_lock(&intr_lock);
693 TAILQ_FOREACH(src, &intr_sources, next)
694 if (src->intr_handle.fd ==
698 rte_spinlock_unlock(&intr_lock);
702 /* mark this interrupt source as active and release the lock. */
704 rte_spinlock_unlock(&intr_lock);
706 /* set the length to be read dor different handle type */
707 switch (src->intr_handle.type) {
708 case RTE_INTR_HANDLE_UIO:
709 case RTE_INTR_HANDLE_UIO_INTX:
710 bytes_read = sizeof(buf.uio_intr_count);
712 case RTE_INTR_HANDLE_ALARM:
713 bytes_read = sizeof(buf.timerfd_num);
716 case RTE_INTR_HANDLE_VFIO_MSIX:
717 case RTE_INTR_HANDLE_VFIO_MSI:
718 case RTE_INTR_HANDLE_VFIO_LEGACY:
719 bytes_read = sizeof(buf.vfio_intr_count);
722 case RTE_INTR_HANDLE_EXT:
728 if (src->intr_handle.type != RTE_INTR_HANDLE_EXT) {
730 * read out to clear the ready-to-be-read flag
733 bytes_read = read(events[n].data.fd, &buf, bytes_read);
734 if (bytes_read < 0) {
735 if (errno == EINTR || errno == EWOULDBLOCK)
738 RTE_LOG(ERR, EAL, "Error reading from file "
739 "descriptor %d: %s\n",
742 } else if (bytes_read == 0)
743 RTE_LOG(ERR, EAL, "Read nothing from file "
744 "descriptor %d\n", events[n].data.fd);
747 /* grab a lock, again to call callbacks and update status. */
748 rte_spinlock_lock(&intr_lock);
750 if (bytes_read > 0) {
752 /* Finally, call all callbacks. */
753 TAILQ_FOREACH(cb, &src->callbacks, next) {
755 /* make a copy and unlock. */
757 rte_spinlock_unlock(&intr_lock);
759 /* call the actual callback */
760 active_cb.cb_fn(&src->intr_handle,
763 /*get the lock back. */
764 rte_spinlock_lock(&intr_lock);
768 /* we done with that interrupt source, release it. */
770 rte_spinlock_unlock(&intr_lock);
777 * It handles all the interrupts.
780 * epoll file descriptor.
782 * The number of file descriptors added in epoll.
788 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
790 struct epoll_event events[totalfds];
794 nfds = epoll_wait(pfd, events, totalfds,
795 EAL_INTR_EPOLL_WAIT_FOREVER);
796 /* epoll_wait fail */
801 "epoll_wait returns with fail\n");
804 /* epoll_wait timeout, will never happens here */
807 /* epoll_wait has at least one fd ready to read */
808 if (eal_intr_process_interrupts(events, nfds) < 0)
814 * It builds/rebuilds up the epoll file descriptor with all the
815 * file descriptors being waited on. Then handles the interrupts.
823 static __attribute__((noreturn)) void *
824 eal_intr_thread_main(__rte_unused void *arg)
826 struct epoll_event ev;
828 /* host thread, never break out */
830 /* build up the epoll fd with all descriptors we are to
831 * wait on then pass it to the handle_interrupts function
833 static struct epoll_event pipe_event = {
834 .events = EPOLLIN | EPOLLPRI,
836 struct rte_intr_source *src;
839 /* create epoll fd */
840 int pfd = epoll_create(1);
842 rte_panic("Cannot create epoll instance\n");
844 pipe_event.data.fd = intr_pipe.readfd;
846 * add pipe fd into wait list, this pipe is used to
847 * rebuild the wait list.
849 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
851 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
852 intr_pipe.readfd, strerror(errno));
856 rte_spinlock_lock(&intr_lock);
858 TAILQ_FOREACH(src, &intr_sources, next) {
859 if (src->callbacks.tqh_first == NULL)
860 continue; /* skip those with no callbacks */
861 ev.events = EPOLLIN | EPOLLPRI;
862 ev.data.fd = src->intr_handle.fd;
865 * add all the uio device file descriptor
868 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
869 src->intr_handle.fd, &ev) < 0){
870 rte_panic("Error adding fd %d epoll_ctl, %s\n",
871 src->intr_handle.fd, strerror(errno));
876 rte_spinlock_unlock(&intr_lock);
877 /* serve the interrupt */
878 eal_intr_handle_interrupts(pfd, numfds);
881 * when we return, we need to rebuild the
882 * list of fds to monitor.
889 rte_eal_intr_init(void)
891 int ret = 0, ret_1 = 0;
892 char thread_name[RTE_MAX_THREAD_NAME_LEN];
894 /* init the global interrupt source head */
895 TAILQ_INIT(&intr_sources);
898 * create a pipe which will be waited by epoll and notified to
899 * rebuild the wait list of epoll.
901 if (pipe(intr_pipe.pipefd) < 0) {
906 /* create the host thread to wait/handle the interrupt */
907 ret = pthread_create(&intr_thread, NULL,
908 eal_intr_thread_main, NULL);
912 "Failed to create thread for interrupt handling\n");
914 /* Set thread_name for aid in debugging. */
915 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
917 ret_1 = rte_thread_setname(intr_thread, thread_name);
920 "Failed to set thread name for interrupt handling\n");
927 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
929 union rte_intr_read_buffer buf;
933 switch (intr_handle->type) {
934 case RTE_INTR_HANDLE_UIO:
935 case RTE_INTR_HANDLE_UIO_INTX:
936 bytes_read = sizeof(buf.uio_intr_count);
939 case RTE_INTR_HANDLE_VFIO_MSIX:
940 case RTE_INTR_HANDLE_VFIO_MSI:
941 case RTE_INTR_HANDLE_VFIO_LEGACY:
942 bytes_read = sizeof(buf.vfio_intr_count);
947 RTE_LOG(INFO, EAL, "unexpected intr type\n");
952 * read out to clear the ready-to-be-read flag
956 nbytes = read(fd, &buf, bytes_read);
958 if (errno == EINTR || errno == EWOULDBLOCK ||
962 "Error reading from fd %d: %s\n",
963 fd, strerror(errno));
964 } else if (nbytes == 0)
965 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
971 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
972 struct rte_epoll_event *events)
974 unsigned int i, count = 0;
975 struct rte_epoll_event *rev;
977 for (i = 0; i < n; i++) {
978 rev = evs[i].data.ptr;
979 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
983 events[count].status = RTE_EPOLL_VALID;
984 events[count].fd = rev->fd;
985 events[count].epfd = rev->epfd;
986 events[count].epdata.event = rev->epdata.event;
987 events[count].epdata.data = rev->epdata.data;
988 if (rev->epdata.cb_fun)
989 rev->epdata.cb_fun(rev->fd,
992 rte_compiler_barrier();
993 rev->status = RTE_EPOLL_VALID;
1000 eal_init_tls_epfd(void)
1002 int pfd = epoll_create(255);
1006 "Cannot create epoll instance\n");
1013 rte_intr_tls_epfd(void)
1015 if (RTE_PER_LCORE(_epfd) == -1)
1016 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
1018 return RTE_PER_LCORE(_epfd);
1022 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
1023 int maxevents, int timeout)
1025 struct epoll_event evs[maxevents];
1029 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1033 /* using per thread epoll fd */
1034 if (epfd == RTE_EPOLL_PER_THREAD)
1035 epfd = rte_intr_tls_epfd();
1038 rc = epoll_wait(epfd, evs, maxevents, timeout);
1039 if (likely(rc > 0)) {
1040 /* epoll_wait has at least one fd ready to read */
1041 rc = eal_epoll_process_event(evs, rc, events);
1043 } else if (rc < 0) {
1046 /* epoll_wait fail */
1047 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1052 /* rc == 0, epoll_wait timed out */
1061 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1063 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1065 while (ev->status != RTE_EPOLL_VALID)
1067 memset(&ev->epdata, 0, sizeof(ev->epdata));
1073 rte_epoll_ctl(int epfd, int op, int fd,
1074 struct rte_epoll_event *event)
1076 struct epoll_event ev;
1079 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1083 /* using per thread epoll fd */
1084 if (epfd == RTE_EPOLL_PER_THREAD)
1085 epfd = rte_intr_tls_epfd();
1087 if (op == EPOLL_CTL_ADD) {
1088 event->status = RTE_EPOLL_VALID;
1089 event->fd = fd; /* ignore fd in event */
1091 ev.data.ptr = (void *)event;
1094 ev.events = event->epdata.event;
1095 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1096 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1097 op, fd, strerror(errno));
1098 if (op == EPOLL_CTL_ADD)
1099 /* rollback status when CTL_ADD fail */
1100 event->status = RTE_EPOLL_INVALID;
1104 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1105 eal_epoll_data_safe_free(event);
1111 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1112 int op, unsigned int vec, void *data)
1114 struct rte_epoll_event *rev;
1115 struct rte_epoll_data *epdata;
1117 unsigned int efd_idx;
1120 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1121 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1123 if (!intr_handle || intr_handle->nb_efd == 0 ||
1124 efd_idx >= intr_handle->nb_efd) {
1125 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1130 case RTE_INTR_EVENT_ADD:
1131 epfd_op = EPOLL_CTL_ADD;
1132 rev = &intr_handle->elist[efd_idx];
1133 if (rev->status != RTE_EPOLL_INVALID) {
1134 RTE_LOG(INFO, EAL, "Event already been added.\n");
1138 /* attach to intr vector fd */
1139 epdata = &rev->epdata;
1140 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1141 epdata->data = data;
1142 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1143 epdata->cb_arg = (void *)intr_handle;
1144 rc = rte_epoll_ctl(epfd, epfd_op,
1145 intr_handle->efds[efd_idx], rev);
1148 "efd %d associated with vec %d added on epfd %d"
1149 "\n", rev->fd, vec, epfd);
1153 case RTE_INTR_EVENT_DEL:
1154 epfd_op = EPOLL_CTL_DEL;
1155 rev = &intr_handle->elist[efd_idx];
1156 if (rev->status == RTE_EPOLL_INVALID) {
1157 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1161 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1166 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1174 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1178 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1180 assert(nb_efd != 0);
1182 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1183 for (i = 0; i < n; i++) {
1184 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1187 "can't setup eventfd, error %i (%s)\n",
1188 errno, strerror(errno));
1191 intr_handle->efds[i] = fd;
1193 intr_handle->nb_efd = n;
1194 intr_handle->max_intr = NB_OTHER_INTR + n;
1196 intr_handle->efds[0] = intr_handle->fd;
1197 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1198 intr_handle->max_intr = NB_OTHER_INTR;
1205 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1208 struct rte_epoll_event *rev;
1210 for (i = 0; i < intr_handle->nb_efd; i++) {
1211 rev = &intr_handle->elist[i];
1212 if (rev->status == RTE_EPOLL_INVALID)
1214 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1215 /* force free if the entry valid */
1216 eal_epoll_data_safe_free(rev);
1217 rev->status = RTE_EPOLL_INVALID;
1221 if (intr_handle->max_intr > intr_handle->nb_efd) {
1222 for (i = 0; i < intr_handle->nb_efd; i++)
1223 close(intr_handle->efds[i]);
1225 intr_handle->nb_efd = 0;
1226 intr_handle->max_intr = 0;
1230 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1232 return !(!intr_handle->nb_efd);
1236 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1238 if (!rte_intr_dp_is_en(intr_handle))
1241 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1245 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1247 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)