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)
291 src->intr_handle.max_intr = 1;
292 else if (src->intr_handle.max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
293 src->intr_handle.max_intr
294 = RTE_MAX_RXTX_INTR_VEC_ID + 1;
296 return src->intr_handle.max_intr;
302 /* enable MSI-X interrupts */
304 vfio_enable_msix(const struct rte_intr_handle *intr_handle) {
306 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
307 struct vfio_irq_set *irq_set;
310 len = sizeof(irq_set_buf);
312 irq_set = (struct vfio_irq_set *) irq_set_buf;
313 irq_set->argsz = len;
315 ret = get_max_intr(intr_handle);
317 RTE_LOG(ERR, EAL, "Invalid number of MSI-X irqs for fd %d\n",
322 irq_set->count = ret;
323 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
324 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
326 fd_ptr = (int *) &irq_set->data;
327 /* INTR vector offset 0 reserve for non-efds mapping */
328 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
329 memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
330 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
332 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
335 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
343 /* disable MSI-X interrupts */
345 vfio_disable_msix(const struct rte_intr_handle *intr_handle) {
346 struct vfio_irq_set *irq_set;
347 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
350 len = sizeof(struct vfio_irq_set);
352 irq_set = (struct vfio_irq_set *) irq_set_buf;
353 irq_set->argsz = len;
355 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
356 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
359 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
363 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
370 uio_intx_intr_disable(const struct rte_intr_handle *intr_handle)
372 unsigned char command_high;
374 /* use UIO config file descriptor for uio_pci_generic */
375 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
377 "Error reading interrupts status for fd %d\n",
378 intr_handle->uio_cfg_fd);
381 /* disable interrupts */
383 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
385 "Error disabling interrupts for fd %d\n",
386 intr_handle->uio_cfg_fd);
394 uio_intx_intr_enable(const struct rte_intr_handle *intr_handle)
396 unsigned char command_high;
398 /* use UIO config file descriptor for uio_pci_generic */
399 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
401 "Error reading interrupts status for fd %d\n",
402 intr_handle->uio_cfg_fd);
405 /* enable interrupts */
406 command_high &= ~0x4;
407 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
409 "Error enabling interrupts for fd %d\n",
410 intr_handle->uio_cfg_fd);
418 uio_intr_disable(const struct rte_intr_handle *intr_handle)
422 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
424 "Error disabling interrupts for fd %d (%s)\n",
425 intr_handle->fd, strerror(errno));
432 uio_intr_enable(const struct rte_intr_handle *intr_handle)
436 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
438 "Error enabling interrupts for fd %d (%s)\n",
439 intr_handle->fd, strerror(errno));
446 rte_intr_callback_register(const struct rte_intr_handle *intr_handle,
447 rte_intr_callback_fn cb, void *cb_arg)
449 int ret, wake_thread;
450 struct rte_intr_source *src;
451 struct rte_intr_callback *callback;
455 /* first do parameter checking */
456 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
458 "Registering with invalid input parameter\n");
462 /* allocate a new interrupt callback entity */
463 callback = rte_zmalloc("interrupt callback list",
464 sizeof(*callback), 0);
465 if (callback == NULL) {
466 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
469 callback->cb_fn = cb;
470 callback->cb_arg = cb_arg;
472 rte_spinlock_lock(&intr_lock);
474 /* check if there is at least one callback registered for the fd */
475 TAILQ_FOREACH(src, &intr_sources, next) {
476 if (src->intr_handle.fd == intr_handle->fd) {
477 /* we had no interrupts for this */
478 if TAILQ_EMPTY(&src->callbacks)
481 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
487 /* no existing callbacks for this - add new source */
489 if ((src = rte_zmalloc("interrupt source list",
490 sizeof(*src), 0)) == NULL) {
491 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
495 src->intr_handle = *intr_handle;
496 TAILQ_INIT(&src->callbacks);
497 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
498 TAILQ_INSERT_TAIL(&intr_sources, src, next);
504 rte_spinlock_unlock(&intr_lock);
507 * check if need to notify the pipe fd waited by epoll_wait to
508 * rebuild the wait list.
511 if (write(intr_pipe.writefd, "1", 1) < 0)
518 rte_intr_callback_unregister(const struct rte_intr_handle *intr_handle,
519 rte_intr_callback_fn cb_fn, void *cb_arg)
522 struct rte_intr_source *src;
523 struct rte_intr_callback *cb, *next;
525 /* do parameter checking first */
526 if (intr_handle == NULL || intr_handle->fd < 0) {
528 "Unregistering with invalid input parameter\n");
532 rte_spinlock_lock(&intr_lock);
534 /* check if the insterrupt source for the fd is existent */
535 TAILQ_FOREACH(src, &intr_sources, next)
536 if (src->intr_handle.fd == intr_handle->fd)
539 /* No interrupt source registered for the fd */
543 /* interrupt source has some active callbacks right now. */
544 } else if (src->active != 0) {
551 /*walk through the callbacks and remove all that match. */
552 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
554 next = TAILQ_NEXT(cb, next);
556 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
557 cb->cb_arg == cb_arg)) {
558 TAILQ_REMOVE(&src->callbacks, cb, next);
564 /* all callbacks for that source are removed. */
565 if (TAILQ_EMPTY(&src->callbacks)) {
566 TAILQ_REMOVE(&intr_sources, src, next);
571 rte_spinlock_unlock(&intr_lock);
573 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
574 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
582 rte_intr_enable(const struct rte_intr_handle *intr_handle)
584 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
587 switch (intr_handle->type){
588 /* write to the uio fd to enable the interrupt */
589 case RTE_INTR_HANDLE_UIO:
590 if (uio_intr_enable(intr_handle))
593 case RTE_INTR_HANDLE_UIO_INTX:
594 if (uio_intx_intr_enable(intr_handle))
597 /* not used at this moment */
598 case RTE_INTR_HANDLE_ALARM:
601 case RTE_INTR_HANDLE_VFIO_MSIX:
602 if (vfio_enable_msix(intr_handle))
605 case RTE_INTR_HANDLE_VFIO_MSI:
606 if (vfio_enable_msi(intr_handle))
609 case RTE_INTR_HANDLE_VFIO_LEGACY:
610 if (vfio_enable_intx(intr_handle))
614 /* unknown handle type */
617 "Unknown handle type of fd %d\n",
626 rte_intr_disable(const struct rte_intr_handle *intr_handle)
628 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
631 switch (intr_handle->type){
632 /* write to the uio fd to disable the interrupt */
633 case RTE_INTR_HANDLE_UIO:
634 if (uio_intr_disable(intr_handle))
637 case RTE_INTR_HANDLE_UIO_INTX:
638 if (uio_intx_intr_disable(intr_handle))
641 /* not used at this moment */
642 case RTE_INTR_HANDLE_ALARM:
645 case RTE_INTR_HANDLE_VFIO_MSIX:
646 if (vfio_disable_msix(intr_handle))
649 case RTE_INTR_HANDLE_VFIO_MSI:
650 if (vfio_disable_msi(intr_handle))
653 case RTE_INTR_HANDLE_VFIO_LEGACY:
654 if (vfio_disable_intx(intr_handle))
658 /* unknown handle type */
661 "Unknown handle type of fd %d\n",
670 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
673 struct rte_intr_source *src;
674 struct rte_intr_callback *cb;
675 union rte_intr_read_buffer buf;
676 struct rte_intr_callback active_cb;
678 for (n = 0; n < nfds; n++) {
681 * if the pipe fd is ready to read, return out to
682 * rebuild the wait list.
684 if (events[n].data.fd == intr_pipe.readfd){
685 int r = read(intr_pipe.readfd, buf.charbuf,
686 sizeof(buf.charbuf));
690 rte_spinlock_lock(&intr_lock);
691 TAILQ_FOREACH(src, &intr_sources, next)
692 if (src->intr_handle.fd ==
696 rte_spinlock_unlock(&intr_lock);
700 /* mark this interrupt source as active and release the lock. */
702 rte_spinlock_unlock(&intr_lock);
704 /* set the length to be read dor different handle type */
705 switch (src->intr_handle.type) {
706 case RTE_INTR_HANDLE_UIO:
707 case RTE_INTR_HANDLE_UIO_INTX:
708 bytes_read = sizeof(buf.uio_intr_count);
710 case RTE_INTR_HANDLE_ALARM:
711 bytes_read = sizeof(buf.timerfd_num);
714 case RTE_INTR_HANDLE_VFIO_MSIX:
715 case RTE_INTR_HANDLE_VFIO_MSI:
716 case RTE_INTR_HANDLE_VFIO_LEGACY:
717 bytes_read = sizeof(buf.vfio_intr_count);
720 case RTE_INTR_HANDLE_EXT:
726 if (src->intr_handle.type != RTE_INTR_HANDLE_EXT) {
728 * read out to clear the ready-to-be-read flag
731 bytes_read = read(events[n].data.fd, &buf, bytes_read);
732 if (bytes_read < 0) {
733 if (errno == EINTR || errno == EWOULDBLOCK)
736 RTE_LOG(ERR, EAL, "Error reading from file "
737 "descriptor %d: %s\n",
740 } else if (bytes_read == 0)
741 RTE_LOG(ERR, EAL, "Read nothing from file "
742 "descriptor %d\n", events[n].data.fd);
745 /* grab a lock, again to call callbacks and update status. */
746 rte_spinlock_lock(&intr_lock);
748 if (bytes_read > 0) {
750 /* Finally, call all callbacks. */
751 TAILQ_FOREACH(cb, &src->callbacks, next) {
753 /* make a copy and unlock. */
755 rte_spinlock_unlock(&intr_lock);
757 /* call the actual callback */
758 active_cb.cb_fn(&src->intr_handle,
761 /*get the lock back. */
762 rte_spinlock_lock(&intr_lock);
766 /* we done with that interrupt source, release it. */
768 rte_spinlock_unlock(&intr_lock);
775 * It handles all the interrupts.
778 * epoll file descriptor.
780 * The number of file descriptors added in epoll.
786 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
788 struct epoll_event events[totalfds];
792 nfds = epoll_wait(pfd, events, totalfds,
793 EAL_INTR_EPOLL_WAIT_FOREVER);
794 /* epoll_wait fail */
799 "epoll_wait returns with fail\n");
802 /* epoll_wait timeout, will never happens here */
805 /* epoll_wait has at least one fd ready to read */
806 if (eal_intr_process_interrupts(events, nfds) < 0)
812 * It builds/rebuilds up the epoll file descriptor with all the
813 * file descriptors being waited on. Then handles the interrupts.
821 static __attribute__((noreturn)) void *
822 eal_intr_thread_main(__rte_unused void *arg)
824 struct epoll_event ev;
826 /* host thread, never break out */
828 /* build up the epoll fd with all descriptors we are to
829 * wait on then pass it to the handle_interrupts function
831 static struct epoll_event pipe_event = {
832 .events = EPOLLIN | EPOLLPRI,
834 struct rte_intr_source *src;
837 /* create epoll fd */
838 int pfd = epoll_create(1);
840 rte_panic("Cannot create epoll instance\n");
842 pipe_event.data.fd = intr_pipe.readfd;
844 * add pipe fd into wait list, this pipe is used to
845 * rebuild the wait list.
847 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
849 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
850 intr_pipe.readfd, strerror(errno));
854 rte_spinlock_lock(&intr_lock);
856 TAILQ_FOREACH(src, &intr_sources, next) {
857 if (src->callbacks.tqh_first == NULL)
858 continue; /* skip those with no callbacks */
859 ev.events = EPOLLIN | EPOLLPRI;
860 ev.data.fd = src->intr_handle.fd;
863 * add all the uio device file descriptor
866 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
867 src->intr_handle.fd, &ev) < 0){
868 rte_panic("Error adding fd %d epoll_ctl, %s\n",
869 src->intr_handle.fd, strerror(errno));
874 rte_spinlock_unlock(&intr_lock);
875 /* serve the interrupt */
876 eal_intr_handle_interrupts(pfd, numfds);
879 * when we return, we need to rebuild the
880 * list of fds to monitor.
887 rte_eal_intr_init(void)
889 int ret = 0, ret_1 = 0;
890 char thread_name[RTE_MAX_THREAD_NAME_LEN];
892 /* init the global interrupt source head */
893 TAILQ_INIT(&intr_sources);
896 * create a pipe which will be waited by epoll and notified to
897 * rebuild the wait list of epoll.
899 if (pipe(intr_pipe.pipefd) < 0)
902 /* create the host thread to wait/handle the interrupt */
903 ret = pthread_create(&intr_thread, NULL,
904 eal_intr_thread_main, NULL);
907 "Failed to create thread for interrupt handling\n");
909 /* Set thread_name for aid in debugging. */
910 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
912 ret_1 = rte_thread_setname(intr_thread, thread_name);
915 "Failed to set thread name for interrupt handling\n");
922 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
924 union rte_intr_read_buffer buf;
928 switch (intr_handle->type) {
929 case RTE_INTR_HANDLE_UIO:
930 case RTE_INTR_HANDLE_UIO_INTX:
931 bytes_read = sizeof(buf.uio_intr_count);
934 case RTE_INTR_HANDLE_VFIO_MSIX:
935 case RTE_INTR_HANDLE_VFIO_MSI:
936 case RTE_INTR_HANDLE_VFIO_LEGACY:
937 bytes_read = sizeof(buf.vfio_intr_count);
942 RTE_LOG(INFO, EAL, "unexpected intr type\n");
947 * read out to clear the ready-to-be-read flag
951 nbytes = read(fd, &buf, bytes_read);
953 if (errno == EINTR || errno == EWOULDBLOCK ||
957 "Error reading from fd %d: %s\n",
958 fd, strerror(errno));
959 } else if (nbytes == 0)
960 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
966 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
967 struct rte_epoll_event *events)
969 unsigned int i, count = 0;
970 struct rte_epoll_event *rev;
972 for (i = 0; i < n; i++) {
973 rev = evs[i].data.ptr;
974 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
978 events[count].status = RTE_EPOLL_VALID;
979 events[count].fd = rev->fd;
980 events[count].epfd = rev->epfd;
981 events[count].epdata.event = rev->epdata.event;
982 events[count].epdata.data = rev->epdata.data;
983 if (rev->epdata.cb_fun)
984 rev->epdata.cb_fun(rev->fd,
987 rte_compiler_barrier();
988 rev->status = RTE_EPOLL_VALID;
995 eal_init_tls_epfd(void)
997 int pfd = epoll_create(255);
1001 "Cannot create epoll instance\n");
1008 rte_intr_tls_epfd(void)
1010 if (RTE_PER_LCORE(_epfd) == -1)
1011 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
1013 return RTE_PER_LCORE(_epfd);
1017 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
1018 int maxevents, int timeout)
1020 struct epoll_event evs[maxevents];
1024 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1028 /* using per thread epoll fd */
1029 if (epfd == RTE_EPOLL_PER_THREAD)
1030 epfd = rte_intr_tls_epfd();
1033 rc = epoll_wait(epfd, evs, maxevents, timeout);
1034 if (likely(rc > 0)) {
1035 /* epoll_wait has at least one fd ready to read */
1036 rc = eal_epoll_process_event(evs, rc, events);
1038 } else if (rc < 0) {
1041 /* epoll_wait fail */
1042 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1047 /* rc == 0, epoll_wait timed out */
1056 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1058 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1060 while (ev->status != RTE_EPOLL_VALID)
1062 memset(&ev->epdata, 0, sizeof(ev->epdata));
1068 rte_epoll_ctl(int epfd, int op, int fd,
1069 struct rte_epoll_event *event)
1071 struct epoll_event ev;
1074 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1078 /* using per thread epoll fd */
1079 if (epfd == RTE_EPOLL_PER_THREAD)
1080 epfd = rte_intr_tls_epfd();
1082 if (op == EPOLL_CTL_ADD) {
1083 event->status = RTE_EPOLL_VALID;
1084 event->fd = fd; /* ignore fd in event */
1086 ev.data.ptr = (void *)event;
1089 ev.events = event->epdata.event;
1090 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1091 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1092 op, fd, strerror(errno));
1093 if (op == EPOLL_CTL_ADD)
1094 /* rollback status when CTL_ADD fail */
1095 event->status = RTE_EPOLL_INVALID;
1099 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1100 eal_epoll_data_safe_free(event);
1106 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1107 int op, unsigned int vec, void *data)
1109 struct rte_epoll_event *rev;
1110 struct rte_epoll_data *epdata;
1112 unsigned int efd_idx;
1115 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1116 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1118 if (!intr_handle || intr_handle->nb_efd == 0 ||
1119 efd_idx >= intr_handle->nb_efd) {
1120 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1125 case RTE_INTR_EVENT_ADD:
1126 epfd_op = EPOLL_CTL_ADD;
1127 rev = &intr_handle->elist[efd_idx];
1128 if (rev->status != RTE_EPOLL_INVALID) {
1129 RTE_LOG(INFO, EAL, "Event already been added.\n");
1133 /* attach to intr vector fd */
1134 epdata = &rev->epdata;
1135 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1136 epdata->data = data;
1137 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1138 epdata->cb_arg = (void *)intr_handle;
1139 rc = rte_epoll_ctl(epfd, epfd_op,
1140 intr_handle->efds[efd_idx], rev);
1143 "efd %d associated with vec %d added on epfd %d"
1144 "\n", rev->fd, vec, epfd);
1148 case RTE_INTR_EVENT_DEL:
1149 epfd_op = EPOLL_CTL_DEL;
1150 rev = &intr_handle->elist[efd_idx];
1151 if (rev->status == RTE_EPOLL_INVALID) {
1152 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1156 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1161 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1169 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1173 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1175 assert(nb_efd != 0);
1177 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1178 for (i = 0; i < n; i++) {
1179 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1182 "can't setup eventfd, error %i (%s)\n",
1183 errno, strerror(errno));
1186 intr_handle->efds[i] = fd;
1188 intr_handle->nb_efd = n;
1189 intr_handle->max_intr = NB_OTHER_INTR + n;
1191 intr_handle->efds[0] = intr_handle->fd;
1192 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1193 intr_handle->max_intr = NB_OTHER_INTR;
1200 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1203 struct rte_epoll_event *rev;
1205 for (i = 0; i < intr_handle->nb_efd; i++) {
1206 rev = &intr_handle->elist[i];
1207 if (rev->status == RTE_EPOLL_INVALID)
1209 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1210 /* force free if the entry valid */
1211 eal_epoll_data_safe_free(rev);
1212 rev->status = RTE_EPOLL_INVALID;
1216 if (intr_handle->max_intr > intr_handle->nb_efd) {
1217 for (i = 0; i < intr_handle->nb_efd; i++)
1218 close(intr_handle->efds[i]);
1220 intr_handle->nb_efd = 0;
1221 intr_handle->max_intr = 0;
1225 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1227 return !(!intr_handle->nb_efd);
1231 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1233 if (!rte_intr_dp_is_en(intr_handle))
1236 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1240 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1242 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)