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(struct rte_intr_handle *intr_handle) {
140 struct vfio_irq_set *irq_set;
141 char irq_set_buf[IRQ_SET_BUF_LEN];
145 len = sizeof(irq_set_buf);
148 irq_set = (struct vfio_irq_set *) irq_set_buf;
149 irq_set->argsz = len;
151 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
152 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
154 fd_ptr = (int *) &irq_set->data;
155 *fd_ptr = intr_handle->fd;
157 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
160 RTE_LOG(ERR, EAL, "Error enabling INTx interrupts for fd %d\n",
165 /* unmask INTx after enabling */
166 memset(irq_set, 0, len);
167 len = sizeof(struct vfio_irq_set);
168 irq_set->argsz = len;
170 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
171 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
174 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
177 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
184 /* disable legacy (INTx) interrupts */
186 vfio_disable_intx(struct rte_intr_handle *intr_handle) {
187 struct vfio_irq_set *irq_set;
188 char irq_set_buf[IRQ_SET_BUF_LEN];
191 len = sizeof(struct vfio_irq_set);
193 /* mask interrupts before disabling */
194 irq_set = (struct vfio_irq_set *) irq_set_buf;
195 irq_set->argsz = len;
197 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_UNMASK;
198 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
201 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
204 RTE_LOG(ERR, EAL, "Error unmasking INTx interrupts for fd %d\n",
210 memset(irq_set, 0, len);
211 irq_set->argsz = len;
213 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
214 irq_set->index = VFIO_PCI_INTX_IRQ_INDEX;
217 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
221 "Error disabling INTx interrupts for fd %d\n", intr_handle->fd);
227 /* enable MSI interrupts */
229 vfio_enable_msi(struct rte_intr_handle *intr_handle) {
231 char irq_set_buf[IRQ_SET_BUF_LEN];
232 struct vfio_irq_set *irq_set;
235 len = sizeof(irq_set_buf);
237 irq_set = (struct vfio_irq_set *) irq_set_buf;
238 irq_set->argsz = len;
240 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
241 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
243 fd_ptr = (int *) &irq_set->data;
244 *fd_ptr = intr_handle->fd;
246 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
249 RTE_LOG(ERR, EAL, "Error enabling MSI interrupts for fd %d\n",
256 /* disable MSI interrupts */
258 vfio_disable_msi(struct rte_intr_handle *intr_handle) {
259 struct vfio_irq_set *irq_set;
260 char irq_set_buf[IRQ_SET_BUF_LEN];
263 len = sizeof(struct vfio_irq_set);
265 irq_set = (struct vfio_irq_set *) irq_set_buf;
266 irq_set->argsz = len;
268 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
269 irq_set->index = VFIO_PCI_MSI_IRQ_INDEX;
272 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
276 "Error disabling MSI interrupts for fd %d\n", intr_handle->fd);
281 /* enable MSI-X interrupts */
283 vfio_enable_msix(struct rte_intr_handle *intr_handle) {
285 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
286 struct vfio_irq_set *irq_set;
289 len = sizeof(irq_set_buf);
291 irq_set = (struct vfio_irq_set *) irq_set_buf;
292 irq_set->argsz = len;
293 if (!intr_handle->max_intr)
294 intr_handle->max_intr = 1;
295 else if (intr_handle->max_intr > RTE_MAX_RXTX_INTR_VEC_ID)
296 intr_handle->max_intr = RTE_MAX_RXTX_INTR_VEC_ID + 1;
298 irq_set->count = intr_handle->max_intr;
299 irq_set->flags = VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
300 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
302 fd_ptr = (int *) &irq_set->data;
303 /* INTR vector offset 0 reserve for non-efds mapping */
304 fd_ptr[RTE_INTR_VEC_ZERO_OFFSET] = intr_handle->fd;
305 memcpy(&fd_ptr[RTE_INTR_VEC_RXTX_OFFSET], intr_handle->efds,
306 sizeof(*intr_handle->efds) * intr_handle->nb_efd);
308 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
311 RTE_LOG(ERR, EAL, "Error enabling MSI-X interrupts for fd %d\n",
319 /* disable MSI-X interrupts */
321 vfio_disable_msix(struct rte_intr_handle *intr_handle) {
322 struct vfio_irq_set *irq_set;
323 char irq_set_buf[MSIX_IRQ_SET_BUF_LEN];
326 len = sizeof(struct vfio_irq_set);
328 irq_set = (struct vfio_irq_set *) irq_set_buf;
329 irq_set->argsz = len;
331 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
332 irq_set->index = VFIO_PCI_MSIX_IRQ_INDEX;
335 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
339 "Error disabling MSI-X interrupts for fd %d\n", intr_handle->fd);
346 uio_intx_intr_disable(struct rte_intr_handle *intr_handle)
348 unsigned char command_high;
350 /* use UIO config file descriptor for uio_pci_generic */
351 if (pread(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
353 "Error reading interrupts status for fd %d\n",
354 intr_handle->uio_cfg_fd);
357 /* disable interrupts */
359 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
361 "Error disabling interrupts for fd %d\n",
362 intr_handle->uio_cfg_fd);
370 uio_intx_intr_enable(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 /* enable interrupts */
382 command_high &= ~0x4;
383 if (pwrite(intr_handle->uio_cfg_fd, &command_high, 1, 5) != 1) {
385 "Error enabling interrupts for fd %d\n",
386 intr_handle->uio_cfg_fd);
394 uio_intr_disable(struct rte_intr_handle *intr_handle)
398 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
400 "Error disabling interrupts for fd %d (%s)\n",
401 intr_handle->fd, strerror(errno));
408 uio_intr_enable(struct rte_intr_handle *intr_handle)
412 if (write(intr_handle->fd, &value, sizeof(value)) < 0) {
414 "Error enabling interrupts for fd %d (%s)\n",
415 intr_handle->fd, strerror(errno));
422 rte_intr_callback_register(struct rte_intr_handle *intr_handle,
423 rte_intr_callback_fn cb, void *cb_arg)
425 int ret, wake_thread;
426 struct rte_intr_source *src;
427 struct rte_intr_callback *callback;
431 /* first do parameter checking */
432 if (intr_handle == NULL || intr_handle->fd < 0 || cb == NULL) {
434 "Registering with invalid input parameter\n");
438 /* allocate a new interrupt callback entity */
439 callback = rte_zmalloc("interrupt callback list",
440 sizeof(*callback), 0);
441 if (callback == NULL) {
442 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
445 callback->cb_fn = cb;
446 callback->cb_arg = cb_arg;
448 rte_spinlock_lock(&intr_lock);
450 /* check if there is at least one callback registered for the fd */
451 TAILQ_FOREACH(src, &intr_sources, next) {
452 if (src->intr_handle.fd == intr_handle->fd) {
453 /* we had no interrupts for this */
454 if TAILQ_EMPTY(&src->callbacks)
457 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
463 /* no existing callbacks for this - add new source */
465 if ((src = rte_zmalloc("interrupt source list",
466 sizeof(*src), 0)) == NULL) {
467 RTE_LOG(ERR, EAL, "Can not allocate memory\n");
471 src->intr_handle = *intr_handle;
472 TAILQ_INIT(&src->callbacks);
473 TAILQ_INSERT_TAIL(&(src->callbacks), callback, next);
474 TAILQ_INSERT_TAIL(&intr_sources, src, next);
480 rte_spinlock_unlock(&intr_lock);
483 * check if need to notify the pipe fd waited by epoll_wait to
484 * rebuild the wait list.
487 if (write(intr_pipe.writefd, "1", 1) < 0)
494 rte_intr_callback_unregister(struct rte_intr_handle *intr_handle,
495 rte_intr_callback_fn cb_fn, void *cb_arg)
498 struct rte_intr_source *src;
499 struct rte_intr_callback *cb, *next;
501 /* do parameter checking first */
502 if (intr_handle == NULL || intr_handle->fd < 0) {
504 "Unregistering with invalid input parameter\n");
508 rte_spinlock_lock(&intr_lock);
510 /* check if the insterrupt source for the fd is existent */
511 TAILQ_FOREACH(src, &intr_sources, next)
512 if (src->intr_handle.fd == intr_handle->fd)
515 /* No interrupt source registered for the fd */
519 /* interrupt source has some active callbacks right now. */
520 } else if (src->active != 0) {
527 /*walk through the callbacks and remove all that match. */
528 for (cb = TAILQ_FIRST(&src->callbacks); cb != NULL; cb = next) {
530 next = TAILQ_NEXT(cb, next);
532 if (cb->cb_fn == cb_fn && (cb_arg == (void *)-1 ||
533 cb->cb_arg == cb_arg)) {
534 TAILQ_REMOVE(&src->callbacks, cb, next);
540 /* all callbacks for that source are removed. */
541 if (TAILQ_EMPTY(&src->callbacks)) {
542 TAILQ_REMOVE(&intr_sources, src, next);
547 rte_spinlock_unlock(&intr_lock);
549 /* notify the pipe fd waited by epoll_wait to rebuild the wait list */
550 if (ret >= 0 && write(intr_pipe.writefd, "1", 1) < 0) {
558 rte_intr_enable(struct rte_intr_handle *intr_handle)
560 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
563 switch (intr_handle->type){
564 /* write to the uio fd to enable the interrupt */
565 case RTE_INTR_HANDLE_UIO:
566 if (uio_intr_enable(intr_handle))
569 case RTE_INTR_HANDLE_UIO_INTX:
570 if (uio_intx_intr_enable(intr_handle))
573 /* not used at this moment */
574 case RTE_INTR_HANDLE_ALARM:
577 case RTE_INTR_HANDLE_VFIO_MSIX:
578 if (vfio_enable_msix(intr_handle))
581 case RTE_INTR_HANDLE_VFIO_MSI:
582 if (vfio_enable_msi(intr_handle))
585 case RTE_INTR_HANDLE_VFIO_LEGACY:
586 if (vfio_enable_intx(intr_handle))
590 /* unknown handle type */
593 "Unknown handle type of fd %d\n",
602 rte_intr_disable(struct rte_intr_handle *intr_handle)
604 if (!intr_handle || intr_handle->fd < 0 || intr_handle->uio_cfg_fd < 0)
607 switch (intr_handle->type){
608 /* write to the uio fd to disable the interrupt */
609 case RTE_INTR_HANDLE_UIO:
610 if (uio_intr_disable(intr_handle))
613 case RTE_INTR_HANDLE_UIO_INTX:
614 if (uio_intx_intr_disable(intr_handle))
617 /* not used at this moment */
618 case RTE_INTR_HANDLE_ALARM:
621 case RTE_INTR_HANDLE_VFIO_MSIX:
622 if (vfio_disable_msix(intr_handle))
625 case RTE_INTR_HANDLE_VFIO_MSI:
626 if (vfio_disable_msi(intr_handle))
629 case RTE_INTR_HANDLE_VFIO_LEGACY:
630 if (vfio_disable_intx(intr_handle))
634 /* unknown handle type */
637 "Unknown handle type of fd %d\n",
646 eal_intr_process_interrupts(struct epoll_event *events, int nfds)
649 struct rte_intr_source *src;
650 struct rte_intr_callback *cb;
651 union rte_intr_read_buffer buf;
652 struct rte_intr_callback active_cb;
654 for (n = 0; n < nfds; n++) {
657 * if the pipe fd is ready to read, return out to
658 * rebuild the wait list.
660 if (events[n].data.fd == intr_pipe.readfd){
661 int r = read(intr_pipe.readfd, buf.charbuf,
662 sizeof(buf.charbuf));
666 rte_spinlock_lock(&intr_lock);
667 TAILQ_FOREACH(src, &intr_sources, next)
668 if (src->intr_handle.fd ==
672 rte_spinlock_unlock(&intr_lock);
676 /* mark this interrupt source as active and release the lock. */
678 rte_spinlock_unlock(&intr_lock);
680 /* set the length to be read dor different handle type */
681 switch (src->intr_handle.type) {
682 case RTE_INTR_HANDLE_UIO:
683 case RTE_INTR_HANDLE_UIO_INTX:
684 bytes_read = sizeof(buf.uio_intr_count);
686 case RTE_INTR_HANDLE_ALARM:
687 bytes_read = sizeof(buf.timerfd_num);
690 case RTE_INTR_HANDLE_VFIO_MSIX:
691 case RTE_INTR_HANDLE_VFIO_MSI:
692 case RTE_INTR_HANDLE_VFIO_LEGACY:
693 bytes_read = sizeof(buf.vfio_intr_count);
696 case RTE_INTR_HANDLE_EXT:
702 if (src->intr_handle.type != RTE_INTR_HANDLE_EXT) {
704 * read out to clear the ready-to-be-read flag
707 bytes_read = read(events[n].data.fd, &buf, bytes_read);
708 if (bytes_read < 0) {
709 if (errno == EINTR || errno == EWOULDBLOCK)
712 RTE_LOG(ERR, EAL, "Error reading from file "
713 "descriptor %d: %s\n",
716 } else if (bytes_read == 0)
717 RTE_LOG(ERR, EAL, "Read nothing from file "
718 "descriptor %d\n", events[n].data.fd);
721 /* grab a lock, again to call callbacks and update status. */
722 rte_spinlock_lock(&intr_lock);
724 if (bytes_read > 0) {
726 /* Finally, call all callbacks. */
727 TAILQ_FOREACH(cb, &src->callbacks, next) {
729 /* make a copy and unlock. */
731 rte_spinlock_unlock(&intr_lock);
733 /* call the actual callback */
734 active_cb.cb_fn(&src->intr_handle,
737 /*get the lock back. */
738 rte_spinlock_lock(&intr_lock);
742 /* we done with that interrupt source, release it. */
744 rte_spinlock_unlock(&intr_lock);
751 * It handles all the interrupts.
754 * epoll file descriptor.
756 * The number of file descriptors added in epoll.
762 eal_intr_handle_interrupts(int pfd, unsigned totalfds)
764 struct epoll_event events[totalfds];
768 nfds = epoll_wait(pfd, events, totalfds,
769 EAL_INTR_EPOLL_WAIT_FOREVER);
770 /* epoll_wait fail */
775 "epoll_wait returns with fail\n");
778 /* epoll_wait timeout, will never happens here */
781 /* epoll_wait has at least one fd ready to read */
782 if (eal_intr_process_interrupts(events, nfds) < 0)
788 * It builds/rebuilds up the epoll file descriptor with all the
789 * file descriptors being waited on. Then handles the interrupts.
797 static __attribute__((noreturn)) void *
798 eal_intr_thread_main(__rte_unused void *arg)
800 struct epoll_event ev;
802 /* host thread, never break out */
804 /* build up the epoll fd with all descriptors we are to
805 * wait on then pass it to the handle_interrupts function
807 static struct epoll_event pipe_event = {
808 .events = EPOLLIN | EPOLLPRI,
810 struct rte_intr_source *src;
813 /* create epoll fd */
814 int pfd = epoll_create(1);
816 rte_panic("Cannot create epoll instance\n");
818 pipe_event.data.fd = intr_pipe.readfd;
820 * add pipe fd into wait list, this pipe is used to
821 * rebuild the wait list.
823 if (epoll_ctl(pfd, EPOLL_CTL_ADD, intr_pipe.readfd,
825 rte_panic("Error adding fd to %d epoll_ctl, %s\n",
826 intr_pipe.readfd, strerror(errno));
830 rte_spinlock_lock(&intr_lock);
832 TAILQ_FOREACH(src, &intr_sources, next) {
833 if (src->callbacks.tqh_first == NULL)
834 continue; /* skip those with no callbacks */
835 ev.events = EPOLLIN | EPOLLPRI;
836 ev.data.fd = src->intr_handle.fd;
839 * add all the uio device file descriptor
842 if (epoll_ctl(pfd, EPOLL_CTL_ADD,
843 src->intr_handle.fd, &ev) < 0){
844 rte_panic("Error adding fd %d epoll_ctl, %s\n",
845 src->intr_handle.fd, strerror(errno));
850 rte_spinlock_unlock(&intr_lock);
851 /* serve the interrupt */
852 eal_intr_handle_interrupts(pfd, numfds);
855 * when we return, we need to rebuild the
856 * list of fds to monitor.
863 rte_eal_intr_init(void)
865 int ret = 0, ret_1 = 0;
866 char thread_name[RTE_MAX_THREAD_NAME_LEN];
868 /* init the global interrupt source head */
869 TAILQ_INIT(&intr_sources);
872 * create a pipe which will be waited by epoll and notified to
873 * rebuild the wait list of epoll.
875 if (pipe(intr_pipe.pipefd) < 0)
878 /* create the host thread to wait/handle the interrupt */
879 ret = pthread_create(&intr_thread, NULL,
880 eal_intr_thread_main, NULL);
883 "Failed to create thread for interrupt handling\n");
885 /* Set thread_name for aid in debugging. */
886 snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
888 ret_1 = rte_thread_setname(intr_thread, thread_name);
891 "Failed to set thread name for interrupt handling\n");
898 eal_intr_proc_rxtx_intr(int fd, const struct rte_intr_handle *intr_handle)
900 union rte_intr_read_buffer buf;
904 switch (intr_handle->type) {
905 case RTE_INTR_HANDLE_UIO:
906 case RTE_INTR_HANDLE_UIO_INTX:
907 bytes_read = sizeof(buf.uio_intr_count);
910 case RTE_INTR_HANDLE_VFIO_MSIX:
911 case RTE_INTR_HANDLE_VFIO_MSI:
912 case RTE_INTR_HANDLE_VFIO_LEGACY:
913 bytes_read = sizeof(buf.vfio_intr_count);
918 RTE_LOG(INFO, EAL, "unexpected intr type\n");
923 * read out to clear the ready-to-be-read flag
927 nbytes = read(fd, &buf, bytes_read);
929 if (errno == EINTR || errno == EWOULDBLOCK ||
933 "Error reading from fd %d: %s\n",
934 fd, strerror(errno));
935 } else if (nbytes == 0)
936 RTE_LOG(ERR, EAL, "Read nothing from fd %d\n", fd);
942 eal_epoll_process_event(struct epoll_event *evs, unsigned int n,
943 struct rte_epoll_event *events)
945 unsigned int i, count = 0;
946 struct rte_epoll_event *rev;
948 for (i = 0; i < n; i++) {
949 rev = evs[i].data.ptr;
950 if (!rev || !rte_atomic32_cmpset(&rev->status, RTE_EPOLL_VALID,
954 events[count].status = RTE_EPOLL_VALID;
955 events[count].fd = rev->fd;
956 events[count].epfd = rev->epfd;
957 events[count].epdata.event = rev->epdata.event;
958 events[count].epdata.data = rev->epdata.data;
959 if (rev->epdata.cb_fun)
960 rev->epdata.cb_fun(rev->fd,
963 rte_compiler_barrier();
964 rev->status = RTE_EPOLL_VALID;
971 eal_init_tls_epfd(void)
973 int pfd = epoll_create(255);
977 "Cannot create epoll instance\n");
984 rte_intr_tls_epfd(void)
986 if (RTE_PER_LCORE(_epfd) == -1)
987 RTE_PER_LCORE(_epfd) = eal_init_tls_epfd();
989 return RTE_PER_LCORE(_epfd);
993 rte_epoll_wait(int epfd, struct rte_epoll_event *events,
994 int maxevents, int timeout)
996 struct epoll_event evs[maxevents];
1000 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1004 /* using per thread epoll fd */
1005 if (epfd == RTE_EPOLL_PER_THREAD)
1006 epfd = rte_intr_tls_epfd();
1009 rc = epoll_wait(epfd, evs, maxevents, timeout);
1010 if (likely(rc > 0)) {
1011 /* epoll_wait has at least one fd ready to read */
1012 rc = eal_epoll_process_event(evs, rc, events);
1014 } else if (rc < 0) {
1017 /* epoll_wait fail */
1018 RTE_LOG(ERR, EAL, "epoll_wait returns with fail %s\n",
1023 /* rc == 0, epoll_wait timed out */
1032 eal_epoll_data_safe_free(struct rte_epoll_event *ev)
1034 while (!rte_atomic32_cmpset(&ev->status, RTE_EPOLL_VALID,
1036 while (ev->status != RTE_EPOLL_VALID)
1038 memset(&ev->epdata, 0, sizeof(ev->epdata));
1044 rte_epoll_ctl(int epfd, int op, int fd,
1045 struct rte_epoll_event *event)
1047 struct epoll_event ev;
1050 RTE_LOG(ERR, EAL, "rte_epoll_event can't be NULL\n");
1054 /* using per thread epoll fd */
1055 if (epfd == RTE_EPOLL_PER_THREAD)
1056 epfd = rte_intr_tls_epfd();
1058 if (op == EPOLL_CTL_ADD) {
1059 event->status = RTE_EPOLL_VALID;
1060 event->fd = fd; /* ignore fd in event */
1062 ev.data.ptr = (void *)event;
1065 ev.events = event->epdata.event;
1066 if (epoll_ctl(epfd, op, fd, &ev) < 0) {
1067 RTE_LOG(ERR, EAL, "Error op %d fd %d epoll_ctl, %s\n",
1068 op, fd, strerror(errno));
1069 if (op == EPOLL_CTL_ADD)
1070 /* rollback status when CTL_ADD fail */
1071 event->status = RTE_EPOLL_INVALID;
1075 if (op == EPOLL_CTL_DEL && event->status != RTE_EPOLL_INVALID)
1076 eal_epoll_data_safe_free(event);
1082 rte_intr_rx_ctl(struct rte_intr_handle *intr_handle, int epfd,
1083 int op, unsigned int vec, void *data)
1085 struct rte_epoll_event *rev;
1086 struct rte_epoll_data *epdata;
1088 unsigned int efd_idx;
1091 efd_idx = (vec >= RTE_INTR_VEC_RXTX_OFFSET) ?
1092 (vec - RTE_INTR_VEC_RXTX_OFFSET) : vec;
1094 if (!intr_handle || intr_handle->nb_efd == 0 ||
1095 efd_idx >= intr_handle->nb_efd) {
1096 RTE_LOG(ERR, EAL, "Wrong intr vector number.\n");
1101 case RTE_INTR_EVENT_ADD:
1102 epfd_op = EPOLL_CTL_ADD;
1103 rev = &intr_handle->elist[efd_idx];
1104 if (rev->status != RTE_EPOLL_INVALID) {
1105 RTE_LOG(INFO, EAL, "Event already been added.\n");
1109 /* attach to intr vector fd */
1110 epdata = &rev->epdata;
1111 epdata->event = EPOLLIN | EPOLLPRI | EPOLLET;
1112 epdata->data = data;
1113 epdata->cb_fun = (rte_intr_event_cb_t)eal_intr_proc_rxtx_intr;
1114 epdata->cb_arg = (void *)intr_handle;
1115 rc = rte_epoll_ctl(epfd, epfd_op,
1116 intr_handle->efds[efd_idx], rev);
1119 "efd %d associated with vec %d added on epfd %d"
1120 "\n", rev->fd, vec, epfd);
1124 case RTE_INTR_EVENT_DEL:
1125 epfd_op = EPOLL_CTL_DEL;
1126 rev = &intr_handle->elist[efd_idx];
1127 if (rev->status == RTE_EPOLL_INVALID) {
1128 RTE_LOG(INFO, EAL, "Event does not exist.\n");
1132 rc = rte_epoll_ctl(rev->epfd, epfd_op, rev->fd, rev);
1137 RTE_LOG(ERR, EAL, "event op type mismatch\n");
1145 rte_intr_efd_enable(struct rte_intr_handle *intr_handle, uint32_t nb_efd)
1149 uint32_t n = RTE_MIN(nb_efd, (uint32_t)RTE_MAX_RXTX_INTR_VEC_ID);
1151 assert(nb_efd != 0);
1153 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX) {
1154 for (i = 0; i < n; i++) {
1155 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
1158 "can't setup eventfd, error %i (%s)\n",
1159 errno, strerror(errno));
1162 intr_handle->efds[i] = fd;
1164 intr_handle->nb_efd = n;
1165 intr_handle->max_intr = NB_OTHER_INTR + n;
1167 intr_handle->efds[0] = intr_handle->fd;
1168 intr_handle->nb_efd = RTE_MIN(nb_efd, 1U);
1169 intr_handle->max_intr = NB_OTHER_INTR;
1176 rte_intr_efd_disable(struct rte_intr_handle *intr_handle)
1179 struct rte_epoll_event *rev;
1181 for (i = 0; i < intr_handle->nb_efd; i++) {
1182 rev = &intr_handle->elist[i];
1183 if (rev->status == RTE_EPOLL_INVALID)
1185 if (rte_epoll_ctl(rev->epfd, EPOLL_CTL_DEL, rev->fd, rev)) {
1186 /* force free if the entry valid */
1187 eal_epoll_data_safe_free(rev);
1188 rev->status = RTE_EPOLL_INVALID;
1192 if (intr_handle->max_intr > intr_handle->nb_efd) {
1193 for (i = 0; i < intr_handle->nb_efd; i++)
1194 close(intr_handle->efds[i]);
1196 intr_handle->nb_efd = 0;
1197 intr_handle->max_intr = 0;
1201 rte_intr_dp_is_en(struct rte_intr_handle *intr_handle)
1203 return !(!intr_handle->nb_efd);
1207 rte_intr_allow_others(struct rte_intr_handle *intr_handle)
1209 if (!rte_intr_dp_is_en(intr_handle))
1212 return !!(intr_handle->max_intr - intr_handle->nb_efd);
1216 rte_intr_cap_multiple(struct rte_intr_handle *intr_handle)
1218 if (intr_handle->type == RTE_INTR_HANDLE_VFIO_MSIX)