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.
36 #include <linux/pci_regs.h>
37 #include <sys/eventfd.h>
38 #include <sys/socket.h>
39 #include <sys/ioctl.h>
43 #include <rte_tailq.h>
44 #include <rte_eal_memconfig.h>
45 #include <rte_malloc.h>
47 #include "eal_filesystem.h"
48 #include "eal_pci_init.h"
53 * PCI probing under linux (VFIO version)
55 * This code tries to determine if the PCI device is bound to VFIO driver,
56 * and initialize it (map BARs, set up interrupts) if that's the case.
58 * This file is only compiled if CONFIG_RTE_EAL_VFIO is set to "y".
63 #define VFIO_DIR "/dev/vfio"
64 #define VFIO_CONTAINER_PATH "/dev/vfio/vfio"
65 #define VFIO_GROUP_FMT "/dev/vfio/%u"
66 #define VFIO_GET_REGION_ADDR(x) ((uint64_t) x << 40ULL)
68 /* per-process VFIO config */
69 static struct vfio_config vfio_cfg;
71 /* get PCI BAR number where MSI-X interrupts are */
73 pci_vfio_get_msix_bar(int fd, int *msix_bar)
77 uint8_t cap_id, cap_offset;
79 /* read PCI capability pointer from config space */
80 ret = pread64(fd, ®, sizeof(reg),
81 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
83 if (ret != sizeof(reg)) {
84 RTE_LOG(ERR, EAL, "Cannot read capability pointer from PCI "
89 /* we need first byte */
90 cap_offset = reg & 0xFF;
94 /* read PCI capability ID */
95 ret = pread64(fd, ®, sizeof(reg),
96 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
98 if (ret != sizeof(reg)) {
99 RTE_LOG(ERR, EAL, "Cannot read capability ID from PCI "
104 /* we need first byte */
107 /* if we haven't reached MSI-X, check next capability */
108 if (cap_id != PCI_CAP_ID_MSIX) {
109 ret = pread64(fd, ®, sizeof(reg),
110 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
112 if (ret != sizeof(reg)) {
113 RTE_LOG(ERR, EAL, "Cannot read capability pointer from PCI "
118 /* we need second byte */
119 cap_offset = (reg & 0xFF00) >> 8;
123 /* else, read table offset */
125 /* table offset resides in the next 4 bytes */
126 ret = pread64(fd, ®, sizeof(reg),
127 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
129 if (ret != sizeof(reg)) {
130 RTE_LOG(ERR, EAL, "Cannot read table offset from PCI config "
135 *msix_bar = reg & RTE_PCI_MSIX_TABLE_BIR;
143 /* set PCI bus mastering */
145 pci_vfio_set_bus_master(int dev_fd)
150 ret = pread64(dev_fd, ®, sizeof(reg),
151 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
153 if (ret != sizeof(reg)) {
154 RTE_LOG(ERR, EAL, "Cannot read command from PCI config space!\n");
158 /* set the master bit */
159 reg |= PCI_COMMAND_MASTER;
161 ret = pwrite64(dev_fd, ®, sizeof(reg),
162 VFIO_GET_REGION_ADDR(VFIO_PCI_CONFIG_REGION_INDEX) +
165 if (ret != sizeof(reg)) {
166 RTE_LOG(ERR, EAL, "Cannot write command to PCI config space!\n");
173 /* set up DMA mappings */
175 pci_vfio_setup_dma_maps(int vfio_container_fd)
177 const struct rte_memseg *ms = rte_eal_get_physmem_layout();
180 ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU,
183 RTE_LOG(ERR, EAL, " cannot set IOMMU type, "
184 "error %i (%s)\n", errno, strerror(errno));
188 /* map all DPDK segments for DMA. use 1:1 PA to IOVA mapping */
189 for (i = 0; i < RTE_MAX_MEMSEG; i++) {
190 struct vfio_iommu_type1_dma_map dma_map;
192 if (ms[i].addr == NULL)
195 memset(&dma_map, 0, sizeof(dma_map));
196 dma_map.argsz = sizeof(struct vfio_iommu_type1_dma_map);
197 dma_map.vaddr = ms[i].addr_64;
198 dma_map.size = ms[i].len;
199 dma_map.iova = ms[i].phys_addr;
200 dma_map.flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE;
202 ret = ioctl(vfio_container_fd, VFIO_IOMMU_MAP_DMA, &dma_map);
205 RTE_LOG(ERR, EAL, " cannot set up DMA remapping, "
206 "error %i (%s)\n", errno, strerror(errno));
214 /* set up interrupt support (but not enable interrupts) */
216 pci_vfio_setup_interrupts(struct rte_pci_device *dev, int vfio_dev_fd)
218 int i, ret, intr_idx;
220 /* default to invalid index */
221 intr_idx = VFIO_PCI_NUM_IRQS;
223 /* get interrupt type from internal config (MSI-X by default, can be
224 * overriden from the command line
226 switch (internal_config.vfio_intr_mode) {
227 case RTE_INTR_MODE_MSIX:
228 intr_idx = VFIO_PCI_MSIX_IRQ_INDEX;
230 case RTE_INTR_MODE_MSI:
231 intr_idx = VFIO_PCI_MSI_IRQ_INDEX;
233 case RTE_INTR_MODE_LEGACY:
234 intr_idx = VFIO_PCI_INTX_IRQ_INDEX;
236 /* don't do anything if we want to automatically determine interrupt type */
237 case RTE_INTR_MODE_NONE:
240 RTE_LOG(ERR, EAL, " unknown default interrupt type!\n");
244 /* start from MSI-X interrupt type */
245 for (i = VFIO_PCI_MSIX_IRQ_INDEX; i >= 0; i--) {
246 struct vfio_irq_info irq = { .argsz = sizeof(irq) };
249 /* skip interrupt modes we don't want */
250 if (internal_config.vfio_intr_mode != RTE_INTR_MODE_NONE &&
256 ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_IRQ_INFO, &irq);
258 RTE_LOG(ERR, EAL, " cannot get IRQ info, "
259 "error %i (%s)\n", errno, strerror(errno));
263 /* if this vector cannot be used with eventfd, fail if we explicitly
264 * specified interrupt type, otherwise continue */
265 if ((irq.flags & VFIO_IRQ_INFO_EVENTFD) == 0) {
266 if (internal_config.vfio_intr_mode != RTE_INTR_MODE_NONE) {
268 " interrupt vector does not support eventfd!\n");
274 /* set up an eventfd for interrupts */
277 RTE_LOG(ERR, EAL, " cannot set up eventfd, "
278 "error %i (%s)\n", errno, strerror(errno));
282 dev->intr_handle.fd = fd;
283 dev->intr_handle.vfio_dev_fd = vfio_dev_fd;
286 case VFIO_PCI_MSIX_IRQ_INDEX:
287 internal_config.vfio_intr_mode = RTE_INTR_MODE_MSIX;
288 dev->intr_handle.type = RTE_INTR_HANDLE_VFIO_MSIX;
290 case VFIO_PCI_MSI_IRQ_INDEX:
291 internal_config.vfio_intr_mode = RTE_INTR_MODE_MSI;
292 dev->intr_handle.type = RTE_INTR_HANDLE_VFIO_MSI;
294 case VFIO_PCI_INTX_IRQ_INDEX:
295 internal_config.vfio_intr_mode = RTE_INTR_MODE_LEGACY;
296 dev->intr_handle.type = RTE_INTR_HANDLE_VFIO_LEGACY;
299 RTE_LOG(ERR, EAL, " unknown interrupt type!\n");
306 /* if we're here, we haven't found a suitable interrupt vector */
310 /* open container fd or get an existing one */
312 pci_vfio_get_container_fd(void)
314 int ret, vfio_container_fd;
316 /* if we're in a primary process, try to open the container */
317 if (internal_config.process_type == RTE_PROC_PRIMARY) {
318 vfio_container_fd = open(VFIO_CONTAINER_PATH, O_RDWR);
319 if (vfio_container_fd < 0) {
320 RTE_LOG(ERR, EAL, " cannot open VFIO container, "
321 "error %i (%s)\n", errno, strerror(errno));
325 /* check VFIO API version */
326 ret = ioctl(vfio_container_fd, VFIO_GET_API_VERSION);
327 if (ret != VFIO_API_VERSION) {
329 RTE_LOG(ERR, EAL, " could not get VFIO API version, "
330 "error %i (%s)\n", errno, strerror(errno));
332 RTE_LOG(ERR, EAL, " unsupported VFIO API version!\n");
333 close(vfio_container_fd);
337 /* check if we support IOMMU type 1 */
338 ret = ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1_IOMMU);
341 RTE_LOG(ERR, EAL, " could not get IOMMU type, "
342 "error %i (%s)\n", errno, strerror(errno));
344 RTE_LOG(ERR, EAL, " unsupported IOMMU type!\n");
345 close(vfio_container_fd);
349 return vfio_container_fd;
352 * if we're in a secondary process, request container fd from the
353 * primary process via our socket
357 socket_fd = vfio_mp_sync_connect_to_primary();
359 RTE_LOG(ERR, EAL, " cannot connect to primary process!\n");
362 if (vfio_mp_sync_send_request(socket_fd, SOCKET_REQ_CONTAINER) < 0) {
363 RTE_LOG(ERR, EAL, " cannot request container fd!\n");
367 vfio_container_fd = vfio_mp_sync_receive_fd(socket_fd);
368 if (vfio_container_fd < 0) {
369 RTE_LOG(ERR, EAL, " cannot get container fd!\n");
374 return vfio_container_fd;
380 /* open group fd or get an existing one */
382 pci_vfio_get_group_fd(int iommu_group_no)
386 char filename[PATH_MAX];
388 /* check if we already have the group descriptor open */
389 for (i = 0; i < vfio_cfg.vfio_group_idx; i++)
390 if (vfio_cfg.vfio_groups[i].group_no == iommu_group_no)
391 return vfio_cfg.vfio_groups[i].fd;
393 /* if primary, try to open the group */
394 if (internal_config.process_type == RTE_PROC_PRIMARY) {
395 snprintf(filename, sizeof(filename),
396 VFIO_GROUP_FMT, iommu_group_no);
397 vfio_group_fd = open(filename, O_RDWR);
398 if (vfio_group_fd < 0) {
399 /* if file not found, it's not an error */
400 if (errno != ENOENT) {
401 RTE_LOG(ERR, EAL, "Cannot open %s: %s\n", filename,
408 /* if the fd is valid, create a new group for it */
409 if (vfio_cfg.vfio_group_idx == VFIO_MAX_GROUPS) {
410 RTE_LOG(ERR, EAL, "Maximum number of VFIO groups reached!\n");
413 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = iommu_group_no;
414 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = vfio_group_fd;
415 return vfio_group_fd;
417 /* if we're in a secondary process, request group fd from the primary
418 * process via our socket
423 socket_fd = vfio_mp_sync_connect_to_primary();
426 RTE_LOG(ERR, EAL, " cannot connect to primary process!\n");
429 if (vfio_mp_sync_send_request(socket_fd, SOCKET_REQ_GROUP) < 0) {
430 RTE_LOG(ERR, EAL, " cannot request container fd!\n");
434 if (vfio_mp_sync_send_request(socket_fd, iommu_group_no) < 0) {
435 RTE_LOG(ERR, EAL, " cannot send group number!\n");
439 ret = vfio_mp_sync_receive_request(socket_fd);
445 vfio_group_fd = vfio_mp_sync_receive_fd(socket_fd);
446 /* if we got the fd, return it */
447 if (vfio_group_fd > 0) {
449 return vfio_group_fd;
451 /* fall-through on error */
453 RTE_LOG(ERR, EAL, " cannot get container fd!\n");
461 /* parse IOMMU group number for a PCI device
462 * returns -1 for errors, 0 for non-existent group */
464 pci_vfio_get_group_no(const char *pci_addr)
466 char linkname[PATH_MAX];
467 char filename[PATH_MAX];
468 char *tok[16], *group_tok, *end;
469 int ret, iommu_group_no;
471 memset(linkname, 0, sizeof(linkname));
472 memset(filename, 0, sizeof(filename));
474 /* try to find out IOMMU group for this device */
475 snprintf(linkname, sizeof(linkname),
476 SYSFS_PCI_DEVICES "/%s/iommu_group", pci_addr);
478 ret = readlink(linkname, filename, sizeof(filename));
480 /* if the link doesn't exist, no VFIO for us */
484 ret = rte_strsplit(filename, sizeof(filename),
485 tok, RTE_DIM(tok), '/');
488 RTE_LOG(ERR, EAL, " %s cannot get IOMMU group\n", pci_addr);
492 /* IOMMU group is always the last token */
494 group_tok = tok[ret - 1];
496 iommu_group_no = strtol(group_tok, &end, 10);
497 if ((end != group_tok && *end != '\0') || errno != 0) {
498 RTE_LOG(ERR, EAL, " %s error parsing IOMMU number!\n", pci_addr);
502 return iommu_group_no;
506 clear_current_group(void)
508 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = 0;
509 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = -1;
514 * map the PCI resources of a PCI device in virtual memory (VFIO version).
515 * primary and secondary processes follow almost exactly the same path
518 pci_vfio_map_resource(struct rte_pci_device *dev)
520 struct vfio_group_status group_status = {
521 .argsz = sizeof(group_status)
523 struct vfio_device_info device_info = { .argsz = sizeof(device_info) };
524 int vfio_group_fd, vfio_dev_fd;
526 char pci_addr[PATH_MAX] = {0};
527 struct rte_pci_addr *loc = &dev->addr;
528 int i, ret, msix_bar;
529 struct mapped_pci_resource *vfio_res = NULL;
530 struct pci_map *maps;
532 dev->intr_handle.fd = -1;
533 dev->intr_handle.type = RTE_INTR_HANDLE_UNKNOWN;
535 /* store PCI address string */
536 snprintf(pci_addr, sizeof(pci_addr), PCI_PRI_FMT,
537 loc->domain, loc->bus, loc->devid, loc->function);
539 /* get group number */
540 iommu_group_no = pci_vfio_get_group_no(pci_addr);
542 /* if 0, group doesn't exist */
543 if (iommu_group_no == 0) {
544 RTE_LOG(WARNING, EAL, " %s not managed by VFIO driver, skipping\n",
548 /* if negative, something failed */
549 else if (iommu_group_no < 0)
552 /* get the actual group fd */
553 vfio_group_fd = pci_vfio_get_group_fd(iommu_group_no);
554 if (vfio_group_fd < 0)
558 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].group_no = iommu_group_no;
559 vfio_cfg.vfio_groups[vfio_cfg.vfio_group_idx].fd = vfio_group_fd;
561 /* if group_fd == 0, that means the device isn't managed by VFIO */
562 if (vfio_group_fd == 0) {
563 RTE_LOG(WARNING, EAL, " %s not managed by VFIO driver, skipping\n",
565 /* we store 0 as group fd to distinguish between existing but
566 * unbound VFIO groups, and groups that don't exist at all.
568 vfio_cfg.vfio_group_idx++;
573 * at this point, we know at least one port on this device is bound to VFIO,
574 * so we can proceed to try and set this particular port up
577 /* check if the group is viable */
578 ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status);
580 RTE_LOG(ERR, EAL, " %s cannot get group status, "
581 "error %i (%s)\n", pci_addr, errno, strerror(errno));
582 close(vfio_group_fd);
583 clear_current_group();
585 } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) {
586 RTE_LOG(ERR, EAL, " %s VFIO group is not viable!\n", pci_addr);
587 close(vfio_group_fd);
588 clear_current_group();
593 * at this point, we know that this group is viable (meaning, all devices
594 * are either bound to VFIO or not bound to anything)
597 /* check if group does not have a container yet */
598 if (!(group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) {
600 /* add group to a container */
601 ret = ioctl(vfio_group_fd, VFIO_GROUP_SET_CONTAINER,
602 &vfio_cfg.vfio_container_fd);
604 RTE_LOG(ERR, EAL, " %s cannot add VFIO group to container, "
605 "error %i (%s)\n", pci_addr, errno, strerror(errno));
606 close(vfio_group_fd);
607 clear_current_group();
611 * at this point we know that this group has been successfully
612 * initialized, so we increment vfio_group_idx to indicate that we can
615 vfio_cfg.vfio_group_idx++;
619 * set up DMA mappings for container
621 * needs to be done only once, only when at least one group is assigned to
622 * a container and only in primary process
624 if (internal_config.process_type == RTE_PROC_PRIMARY &&
625 vfio_cfg.vfio_container_has_dma == 0) {
626 ret = pci_vfio_setup_dma_maps(vfio_cfg.vfio_container_fd);
628 RTE_LOG(ERR, EAL, " %s DMA remapping failed, "
629 "error %i (%s)\n", pci_addr, errno, strerror(errno));
632 vfio_cfg.vfio_container_has_dma = 1;
635 /* get a file descriptor for the device */
636 vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, pci_addr);
637 if (vfio_dev_fd < 0) {
638 /* if we cannot get a device fd, this simply means that this
639 * particular port is not bound to VFIO
641 RTE_LOG(WARNING, EAL, " %s not managed by VFIO driver, skipping\n",
646 /* test and setup the device */
647 ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_INFO, &device_info);
649 RTE_LOG(ERR, EAL, " %s cannot get device info, "
650 "error %i (%s)\n", pci_addr, errno, strerror(errno));
655 /* get MSI-X BAR, if any (we have to know where it is because we can't
656 * mmap it when using VFIO) */
658 ret = pci_vfio_get_msix_bar(vfio_dev_fd, &msix_bar);
660 RTE_LOG(ERR, EAL, " %s cannot get MSI-X BAR number!\n", pci_addr);
665 /* if we're in a primary process, allocate vfio_res and get region info */
666 if (internal_config.process_type == RTE_PROC_PRIMARY) {
667 vfio_res = rte_zmalloc("VFIO_RES", sizeof(*vfio_res), 0);
668 if (vfio_res == NULL) {
670 "%s(): cannot store uio mmap details\n", __func__);
674 memcpy(&vfio_res->pci_addr, &dev->addr, sizeof(vfio_res->pci_addr));
676 /* get number of registers (up to BAR5) */
677 vfio_res->nb_maps = RTE_MIN((int) device_info.num_regions,
678 VFIO_PCI_BAR5_REGION_INDEX + 1);
680 /* if we're in a secondary process, just find our tailq entry */
681 TAILQ_FOREACH(vfio_res, pci_res_list, next) {
682 if (memcmp(&vfio_res->pci_addr, &dev->addr, sizeof(dev->addr)))
686 /* if we haven't found our tailq entry, something's wrong */
687 if (vfio_res == NULL) {
688 RTE_LOG(ERR, EAL, " %s cannot find TAILQ entry for PCI device!\n",
696 maps = vfio_res->maps;
698 for (i = 0; i < (int) vfio_res->nb_maps; i++) {
699 struct vfio_region_info reg = { .argsz = sizeof(reg) };
704 ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_REGION_INFO, ®);
707 RTE_LOG(ERR, EAL, " %s cannot get device region info "
708 "error %i (%s)\n", pci_addr, errno, strerror(errno));
710 if (internal_config.process_type == RTE_PROC_PRIMARY)
715 /* skip non-mmapable BARs */
716 if ((reg.flags & VFIO_REGION_INFO_FLAG_MMAP) == 0)
723 bar_addr = pci_map_resource(maps[i].addr, vfio_dev_fd, reg.offset,
726 if (bar_addr == NULL) {
727 RTE_LOG(ERR, EAL, " %s mapping BAR%i failed: %s\n", pci_addr, i,
730 if (internal_config.process_type == RTE_PROC_PRIMARY)
735 maps[i].addr = bar_addr;
736 maps[i].offset = reg.offset;
737 maps[i].size = reg.size;
738 dev->mem_resource[i].addr = bar_addr;
741 /* if secondary process, do not set up interrupts */
742 if (internal_config.process_type == RTE_PROC_PRIMARY) {
743 if (pci_vfio_setup_interrupts(dev, vfio_dev_fd) != 0) {
744 RTE_LOG(ERR, EAL, " %s error setting up interrupts!\n", pci_addr);
750 /* set bus mastering for the device */
751 if (pci_vfio_set_bus_master(vfio_dev_fd)) {
752 RTE_LOG(ERR, EAL, " %s cannot set up bus mastering!\n", pci_addr);
758 /* Reset the device */
759 ioctl(vfio_dev_fd, VFIO_DEVICE_RESET);
762 if (internal_config.process_type == RTE_PROC_PRIMARY)
763 TAILQ_INSERT_TAIL(pci_res_list, vfio_res, next);
769 pci_vfio_enable(void)
771 /* initialize group list */
774 for (i = 0; i < VFIO_MAX_GROUPS; i++) {
775 vfio_cfg.vfio_groups[i].fd = -1;
776 vfio_cfg.vfio_groups[i].group_no = -1;
778 vfio_cfg.vfio_container_fd = pci_vfio_get_container_fd();
780 /* check if we have VFIO driver enabled */
781 if (vfio_cfg.vfio_container_fd != -1)
782 vfio_cfg.vfio_enabled = 1;
784 RTE_LOG(INFO, EAL, "VFIO support could not be initialized\n");
790 pci_vfio_is_enabled(void)
792 return vfio_cfg.vfio_enabled;