1 /* SPDX-License-Identifier: BSD-3-Clause
3 * Copyright (c) 2015-2016 Freescale Semiconductor, Inc. All rights reserved.
4 * Copyright 2016-2019 NXP
10 #include <sys/types.h>
15 #include <sys/ioctl.h>
21 #include <sys/eventfd.h>
23 #include <eal_filesystem.h>
25 #include <rte_ethdev_driver.h>
26 #include <rte_malloc.h>
27 #include <rte_memcpy.h>
28 #include <rte_string_fns.h>
29 #include <rte_cycles.h>
30 #include <rte_kvargs.h>
33 #include <rte_eal_memconfig.h>
35 #include "rte_fslmc.h"
36 #include "fslmc_vfio.h"
37 #include "fslmc_logs.h"
38 #include <mc/fsl_dpmng.h>
40 #include "portal/dpaa2_hw_pvt.h"
41 #include "portal/dpaa2_hw_dpio.h"
43 /** Pathname of FSL-MC devices directory. */
44 #define SYSFS_FSL_MC_DEVICES "/sys/bus/fsl-mc/devices"
46 #define FSLMC_CONTAINER_MAX_LEN 8 /**< Of the format dprc.XX */
48 /* Number of VFIO containers & groups with in */
49 static struct fslmc_vfio_group vfio_group;
50 static struct fslmc_vfio_container vfio_container;
51 static int container_device_fd;
52 static char *fslmc_container;
53 static int fslmc_iommu_type;
54 static uint32_t *msi_intr_vaddr;
55 void *(*rte_mcp_ptr_list);
57 static struct rte_dpaa2_object_list dpaa2_obj_list =
58 TAILQ_HEAD_INITIALIZER(dpaa2_obj_list);
60 /*register a fslmc bus based dpaa2 driver */
62 rte_fslmc_object_register(struct rte_dpaa2_object *object)
66 TAILQ_INSERT_TAIL(&dpaa2_obj_list, object, next);
70 fslmc_get_container_group(int *groupid)
75 if (!fslmc_container) {
76 container = getenv("DPRC");
77 if (container == NULL) {
78 DPAA2_BUS_DEBUG("DPAA2: DPRC not available");
82 if (strlen(container) >= FSLMC_CONTAINER_MAX_LEN) {
83 DPAA2_BUS_ERR("Invalid container name: %s", container);
87 fslmc_container = strdup(container);
88 if (!fslmc_container) {
89 DPAA2_BUS_ERR("Mem alloc failure; Container name");
94 fslmc_iommu_type = (rte_vfio_noiommu_is_enabled() == 1) ?
95 RTE_VFIO_NOIOMMU : VFIO_TYPE1_IOMMU;
97 /* get group number */
98 ret = rte_vfio_get_group_num(SYSFS_FSL_MC_DEVICES,
99 fslmc_container, groupid);
101 DPAA2_BUS_ERR("Unable to find %s IOMMU group", fslmc_container);
105 DPAA2_BUS_DEBUG("Container: %s has VFIO iommu group id = %d",
106 fslmc_container, *groupid);
112 vfio_connect_container(void)
116 if (vfio_container.used) {
117 DPAA2_BUS_DEBUG("No container available");
121 /* Try connecting to vfio container if already created */
122 if (!ioctl(vfio_group.fd, VFIO_GROUP_SET_CONTAINER,
123 &vfio_container.fd)) {
125 "Container pre-exists with FD[0x%x] for this group",
127 vfio_group.container = &vfio_container;
131 /* Opens main vfio file descriptor which represents the "container" */
132 fd = rte_vfio_get_container_fd();
134 DPAA2_BUS_ERR("Failed to open VFIO container");
138 /* Check whether support for SMMU type IOMMU present or not */
139 if (ioctl(fd, VFIO_CHECK_EXTENSION, fslmc_iommu_type)) {
140 /* Connect group to container */
141 ret = ioctl(vfio_group.fd, VFIO_GROUP_SET_CONTAINER, &fd);
143 DPAA2_BUS_ERR("Failed to setup group container");
148 ret = ioctl(fd, VFIO_SET_IOMMU, fslmc_iommu_type);
150 DPAA2_BUS_ERR("Failed to setup VFIO iommu");
155 DPAA2_BUS_ERR("No supported IOMMU available");
160 vfio_container.used = 1;
161 vfio_container.fd = fd;
162 vfio_container.group = &vfio_group;
163 vfio_group.container = &vfio_container;
168 static int vfio_map_irq_region(struct fslmc_vfio_group *group)
171 unsigned long *vaddr = NULL;
172 struct vfio_iommu_type1_dma_map map = {
173 .argsz = sizeof(map),
174 .flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
180 vaddr = (unsigned long *)mmap(NULL, 0x1000, PROT_WRITE |
181 PROT_READ, MAP_SHARED, container_device_fd, 0x6030000);
182 if (vaddr == MAP_FAILED) {
183 DPAA2_BUS_INFO("Unable to map region (errno = %d)", errno);
187 msi_intr_vaddr = (uint32_t *)((char *)(vaddr) + 64);
188 map.vaddr = (unsigned long)vaddr;
189 ret = ioctl(group->container->fd, VFIO_IOMMU_MAP_DMA, &map);
193 DPAA2_BUS_ERR("Unable to map DMA address (errno = %d)", errno);
197 static int fslmc_map_dma(uint64_t vaddr, rte_iova_t iovaddr, size_t len);
198 static int fslmc_unmap_dma(uint64_t vaddr, rte_iova_t iovaddr, size_t len);
201 fslmc_memevent_cb(enum rte_mem_event type, const void *addr, size_t len,
202 void *arg __rte_unused)
204 struct rte_memseg_list *msl;
205 struct rte_memseg *ms;
206 size_t cur_len = 0, map_len = 0;
208 rte_iova_t iova_addr;
211 msl = rte_mem_virt2memseg_list(addr);
213 while (cur_len < len) {
214 const void *va = RTE_PTR_ADD(addr, cur_len);
216 ms = rte_mem_virt2memseg(va, msl);
217 iova_addr = ms->iova;
218 virt_addr = ms->addr_64;
221 DPAA2_BUS_DEBUG("Request for %s, va=%p, "
222 "virt_addr=0x%" PRIx64 ", "
223 "iova=0x%" PRIx64 ", map_len=%zu",
224 type == RTE_MEM_EVENT_ALLOC ?
226 va, virt_addr, iova_addr, map_len);
228 /* iova_addr may be set to RTE_BAD_IOVA */
229 if (iova_addr == RTE_BAD_IOVA) {
230 DPAA2_BUS_DEBUG("Segment has invalid iova, skipping\n");
235 if (type == RTE_MEM_EVENT_ALLOC)
236 ret = fslmc_map_dma(virt_addr, iova_addr, map_len);
238 ret = fslmc_unmap_dma(virt_addr, iova_addr, map_len);
241 DPAA2_BUS_ERR("DMA Mapping/Unmapping failed. "
242 "Map=%d, addr=%p, len=%zu, err:(%d)",
243 type, va, map_len, ret);
250 if (type == RTE_MEM_EVENT_ALLOC)
251 DPAA2_BUS_DEBUG("Total Mapped: addr=%p, len=%zu",
254 DPAA2_BUS_DEBUG("Total Unmapped: addr=%p, len=%zu",
259 fslmc_map_dma(uint64_t vaddr, rte_iova_t iovaddr __rte_unused, size_t len)
261 struct fslmc_vfio_group *group;
262 struct vfio_iommu_type1_dma_map dma_map = {
263 .argsz = sizeof(struct vfio_iommu_type1_dma_map),
264 .flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
268 if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
269 DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
274 dma_map.vaddr = vaddr;
276 #ifdef RTE_LIBRTE_DPAA2_USE_PHYS_IOVA
277 dma_map.iova = iovaddr;
279 dma_map.iova = dma_map.vaddr;
282 /* SET DMA MAP for IOMMU */
285 if (!group->container) {
286 DPAA2_BUS_ERR("Container is not connected ");
290 DPAA2_BUS_DEBUG("--> Map address: 0x%"PRIx64", size: %"PRIu64"",
291 (uint64_t)dma_map.vaddr, (uint64_t)dma_map.size);
292 ret = ioctl(group->container->fd, VFIO_IOMMU_MAP_DMA, &dma_map);
294 DPAA2_BUS_ERR("VFIO_IOMMU_MAP_DMA API(errno = %d)",
303 fslmc_unmap_dma(uint64_t vaddr, uint64_t iovaddr __rte_unused, size_t len)
305 struct fslmc_vfio_group *group;
306 struct vfio_iommu_type1_dma_unmap dma_unmap = {
307 .argsz = sizeof(struct vfio_iommu_type1_dma_unmap),
312 if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
313 DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
317 dma_unmap.size = len;
318 dma_unmap.iova = vaddr;
320 /* SET DMA MAP for IOMMU */
323 if (!group->container) {
324 DPAA2_BUS_ERR("Container is not connected ");
328 DPAA2_BUS_DEBUG("--> Unmap address: 0x%"PRIx64", size: %"PRIu64"",
329 (uint64_t)dma_unmap.iova, (uint64_t)dma_unmap.size);
330 ret = ioctl(group->container->fd, VFIO_IOMMU_UNMAP_DMA, &dma_unmap);
332 DPAA2_BUS_ERR("VFIO_IOMMU_UNMAP_DMA API(errno = %d)",
341 fslmc_dmamap_seg(const struct rte_memseg_list *msl __rte_unused,
342 const struct rte_memseg *ms, void *arg)
347 /* if IOVA address is invalid, skip */
348 if (ms->iova == RTE_BAD_IOVA)
351 ret = fslmc_map_dma(ms->addr_64, ms->iova, ms->len);
353 DPAA2_BUS_ERR("Unable to VFIO map (addr=%p, len=%zu)",
362 rte_fslmc_vfio_mem_dmamap(uint64_t vaddr, uint64_t iova, uint64_t size)
365 struct fslmc_vfio_group *group;
366 struct vfio_iommu_type1_dma_map dma_map = {
367 .argsz = sizeof(struct vfio_iommu_type1_dma_map),
368 .flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
371 if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
372 DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
376 /* SET DMA MAP for IOMMU */
378 if (!group->container) {
379 DPAA2_BUS_ERR("Container is not connected");
384 dma_map.vaddr = vaddr;
387 DPAA2_BUS_DEBUG("VFIOdmamap 0x%"PRIx64":0x%"PRIx64",size 0x%"PRIx64"\n",
388 (uint64_t)dma_map.vaddr, (uint64_t)dma_map.iova,
389 (uint64_t)dma_map.size);
390 ret = ioctl(group->container->fd, VFIO_IOMMU_MAP_DMA,
393 printf("Unable to map DMA address (errno = %d)\n",
401 int rte_fslmc_vfio_dmamap(void)
405 /* Lock before parsing and registering callback to memory subsystem */
406 rte_mcfg_mem_read_lock();
408 if (rte_memseg_walk(fslmc_dmamap_seg, &i) < 0) {
409 rte_mcfg_mem_read_unlock();
413 ret = rte_mem_event_callback_register("fslmc_memevent_clb",
414 fslmc_memevent_cb, NULL);
415 if (ret && rte_errno == ENOTSUP)
416 DPAA2_BUS_DEBUG("Memory event callbacks not supported");
418 DPAA2_BUS_DEBUG("Unable to install memory handler");
420 DPAA2_BUS_DEBUG("Installed memory callback handler");
422 DPAA2_BUS_DEBUG("Total %d segments found.", i);
424 /* TODO - This is a W.A. as VFIO currently does not add the mapping of
425 * the interrupt region to SMMU. This should be removed once the
426 * support is added in the Kernel.
428 vfio_map_irq_region(&vfio_group);
430 /* Existing segments have been mapped and memory callback for hotplug
431 * has been installed.
433 rte_mcfg_mem_read_unlock();
438 static int64_t vfio_map_mcp_obj(struct fslmc_vfio_group *group, char *mcp_obj)
440 intptr_t v_addr = (intptr_t)MAP_FAILED;
443 struct vfio_device_info d_info = { .argsz = sizeof(d_info) };
444 struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) };
446 /* getting the mcp object's fd*/
447 mc_fd = ioctl(group->fd, VFIO_GROUP_GET_DEVICE_FD, mcp_obj);
449 DPAA2_BUS_ERR("Error in VFIO get dev %s fd from group %d",
454 /* getting device info*/
455 ret = ioctl(mc_fd, VFIO_DEVICE_GET_INFO, &d_info);
457 DPAA2_BUS_ERR("Error in VFIO getting DEVICE_INFO");
461 /* getting device region info*/
462 ret = ioctl(mc_fd, VFIO_DEVICE_GET_REGION_INFO, ®_info);
464 DPAA2_BUS_ERR("Error in VFIO getting REGION_INFO");
468 v_addr = (size_t)mmap(NULL, reg_info.size,
469 PROT_WRITE | PROT_READ, MAP_SHARED,
470 mc_fd, reg_info.offset);
478 #define IRQ_SET_BUF_LEN (sizeof(struct vfio_irq_set) + sizeof(int))
480 int rte_dpaa2_intr_enable(struct rte_intr_handle *intr_handle, int index)
483 char irq_set_buf[IRQ_SET_BUF_LEN];
484 struct vfio_irq_set *irq_set;
487 len = sizeof(irq_set_buf);
489 irq_set = (struct vfio_irq_set *)irq_set_buf;
490 irq_set->argsz = len;
493 VFIO_IRQ_SET_DATA_EVENTFD | VFIO_IRQ_SET_ACTION_TRIGGER;
494 irq_set->index = index;
496 fd_ptr = (int *)&irq_set->data;
497 *fd_ptr = intr_handle->fd;
499 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
501 DPAA2_BUS_ERR("Error:dpaa2 SET IRQs fd=%d, err = %d(%s)",
502 intr_handle->fd, errno, strerror(errno));
509 int rte_dpaa2_intr_disable(struct rte_intr_handle *intr_handle, int index)
511 struct vfio_irq_set *irq_set;
512 char irq_set_buf[IRQ_SET_BUF_LEN];
515 len = sizeof(struct vfio_irq_set);
517 irq_set = (struct vfio_irq_set *)irq_set_buf;
518 irq_set->argsz = len;
519 irq_set->flags = VFIO_IRQ_SET_DATA_NONE | VFIO_IRQ_SET_ACTION_TRIGGER;
520 irq_set->index = index;
524 ret = ioctl(intr_handle->vfio_dev_fd, VFIO_DEVICE_SET_IRQS, irq_set);
527 "Error disabling dpaa2 interrupts for fd %d",
533 /* set up interrupt support (but not enable interrupts) */
535 rte_dpaa2_vfio_setup_intr(struct rte_intr_handle *intr_handle,
541 /* start from MSI-X interrupt type */
542 for (i = 0; i < num_irqs; i++) {
543 struct vfio_irq_info irq_info = { .argsz = sizeof(irq_info) };
548 ret = ioctl(vfio_dev_fd, VFIO_DEVICE_GET_IRQ_INFO, &irq_info);
550 DPAA2_BUS_ERR("Cannot get IRQ(%d) info, error %i (%s)",
551 i, errno, strerror(errno));
555 /* if this vector cannot be used with eventfd,
556 * fail if we explicitly
557 * specified interrupt type, otherwise continue
559 if ((irq_info.flags & VFIO_IRQ_INFO_EVENTFD) == 0)
562 /* set up an eventfd for interrupts */
563 fd = eventfd(0, EFD_NONBLOCK | EFD_CLOEXEC);
565 DPAA2_BUS_ERR("Cannot set up eventfd, error %i (%s)",
566 errno, strerror(errno));
570 intr_handle->fd = fd;
571 intr_handle->type = RTE_INTR_HANDLE_VFIO_MSI;
572 intr_handle->vfio_dev_fd = vfio_dev_fd;
577 /* if we're here, we haven't found a suitable interrupt vector */
582 * fslmc_process_iodevices for processing only IO (ETH, CRYPTO, and possibly
586 fslmc_process_iodevices(struct rte_dpaa2_device *dev)
589 struct vfio_device_info device_info = { .argsz = sizeof(device_info) };
590 struct rte_dpaa2_object *object = NULL;
592 dev_fd = ioctl(vfio_group.fd, VFIO_GROUP_GET_DEVICE_FD,
595 DPAA2_BUS_ERR("Unable to obtain device FD for device:%s",
600 if (ioctl(dev_fd, VFIO_DEVICE_GET_INFO, &device_info)) {
601 DPAA2_BUS_ERR("Unable to obtain information for device:%s",
606 switch (dev->dev_type) {
608 rte_dpaa2_vfio_setup_intr(&dev->intr_handle, dev_fd,
609 device_info.num_irqs);
616 TAILQ_FOREACH(object, &dpaa2_obj_list, next) {
617 if (dev->dev_type == object->dev_type)
618 object->create(dev_fd, &device_info,
628 DPAA2_BUS_LOG(DEBUG, "Device (%s) abstracted from VFIO",
634 fslmc_process_mcp(struct rte_dpaa2_device *dev)
638 char *dev_name = NULL;
639 struct fsl_mc_io dpmng = {0};
640 struct mc_version mc_ver_info = {0};
642 rte_mcp_ptr_list = malloc(sizeof(void *) * 1);
643 if (!rte_mcp_ptr_list) {
644 DPAA2_BUS_ERR("Unable to allocate MC portal memory");
649 dev_name = strdup(dev->device.name);
651 DPAA2_BUS_ERR("Unable to allocate MC device name memory");
656 v_addr = vfio_map_mcp_obj(&vfio_group, dev_name);
657 if (v_addr == (intptr_t)MAP_FAILED) {
658 DPAA2_BUS_ERR("Error mapping region (errno = %d)", errno);
663 /* check the MC version compatibility */
664 dpmng.regs = (void *)v_addr;
666 /* In case of secondary processes, MC version check is no longer
669 if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
670 rte_mcp_ptr_list[0] = (void *)v_addr;
674 if (mc_get_version(&dpmng, CMD_PRI_LOW, &mc_ver_info)) {
675 DPAA2_BUS_ERR("Unable to obtain MC version");
680 if ((mc_ver_info.major != MC_VER_MAJOR) ||
681 (mc_ver_info.minor < MC_VER_MINOR)) {
682 DPAA2_BUS_ERR("DPAA2 MC version not compatible!"
683 " Expected %d.%d.x, Detected %d.%d.%d",
684 MC_VER_MAJOR, MC_VER_MINOR,
685 mc_ver_info.major, mc_ver_info.minor,
686 mc_ver_info.revision);
690 rte_mcp_ptr_list[0] = (void *)v_addr;
699 if (rte_mcp_ptr_list) {
700 free(rte_mcp_ptr_list);
701 rte_mcp_ptr_list = NULL;
708 fslmc_vfio_process_group(void)
711 int found_mportal = 0;
712 struct rte_dpaa2_device *dev, *dev_temp;
714 /* Search the MCP as that should be initialized first. */
715 TAILQ_FOREACH_SAFE(dev, &rte_fslmc_bus.device_list, next, dev_temp) {
716 if (dev->dev_type == DPAA2_MPORTAL) {
717 if (dev->device.devargs &&
718 dev->device.devargs->policy == RTE_DEV_BLACKLISTED) {
719 DPAA2_BUS_LOG(DEBUG, "%s Blacklisted, skipping",
721 TAILQ_REMOVE(&rte_fslmc_bus.device_list,
726 ret = fslmc_process_mcp(dev);
728 DPAA2_BUS_ERR("Unable to map MC Portal");
734 TAILQ_REMOVE(&rte_fslmc_bus.device_list, dev, next);
737 /* Ideally there is only a single dpmcp, but in case
738 * multiple exists, looping on remaining devices.
743 /* Cannot continue if there is not even a single mportal */
744 if (!found_mportal) {
745 DPAA2_BUS_ERR("No MC Portal device found. Not continuing");
749 TAILQ_FOREACH_SAFE(dev, &rte_fslmc_bus.device_list, next, dev_temp) {
750 if (dev->device.devargs &&
751 dev->device.devargs->policy == RTE_DEV_BLACKLISTED) {
752 DPAA2_BUS_LOG(DEBUG, "%s Blacklisted, skipping",
754 TAILQ_REMOVE(&rte_fslmc_bus.device_list, dev, next);
757 switch (dev->dev_type) {
761 ret = fslmc_process_iodevices(dev);
763 DPAA2_BUS_DEBUG("Dev (%s) init failed",
772 /* IN case of secondary processes, all control objects
773 * like dpbp, dpcon, dpci are not initialized/required
774 * - all of these are assumed to be initialized and made
775 * available by primary.
777 if (rte_eal_process_type() == RTE_PROC_SECONDARY)
780 /* Call the object creation routine and remove the
781 * device entry from device list
783 ret = fslmc_process_iodevices(dev);
785 DPAA2_BUS_DEBUG("Dev (%s) init failed",
792 ret = fslmc_process_iodevices(dev);
794 DPAA2_BUS_DEBUG("Dev (%s) init failed",
802 /* Unknown - ignore */
803 DPAA2_BUS_DEBUG("Found unknown device (%s)",
805 TAILQ_REMOVE(&rte_fslmc_bus.device_list, dev, next);
815 fslmc_vfio_setup_group(void)
819 struct vfio_group_status status = { .argsz = sizeof(status) };
821 /* if already done once */
822 if (container_device_fd)
825 ret = fslmc_get_container_group(&groupid);
829 /* In case this group was already opened, continue without any
832 if (vfio_group.groupid == groupid) {
833 DPAA2_BUS_ERR("groupid already exists %d", groupid);
837 /* Get the actual group fd */
838 ret = rte_vfio_get_group_fd(groupid);
843 /* Check group viability */
844 ret = ioctl(vfio_group.fd, VFIO_GROUP_GET_STATUS, &status);
846 DPAA2_BUS_ERR("VFIO error getting group status");
847 close(vfio_group.fd);
848 rte_vfio_clear_group(vfio_group.fd);
852 if (!(status.flags & VFIO_GROUP_FLAGS_VIABLE)) {
853 DPAA2_BUS_ERR("VFIO group not viable");
854 close(vfio_group.fd);
855 rte_vfio_clear_group(vfio_group.fd);
858 /* Since Group is VIABLE, Store the groupid */
859 vfio_group.groupid = groupid;
861 /* check if group does not have a container yet */
862 if (!(status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) {
863 /* Now connect this IOMMU group to given container */
864 ret = vfio_connect_container();
867 "Error connecting container with groupid %d",
869 close(vfio_group.fd);
870 rte_vfio_clear_group(vfio_group.fd);
875 /* Get Device information */
876 ret = ioctl(vfio_group.fd, VFIO_GROUP_GET_DEVICE_FD, fslmc_container);
878 DPAA2_BUS_ERR("Error getting device %s fd from group %d",
879 fslmc_container, vfio_group.groupid);
880 close(vfio_group.fd);
881 rte_vfio_clear_group(vfio_group.fd);
884 container_device_fd = ret;
885 DPAA2_BUS_DEBUG("VFIO Container FD is [0x%X]",
886 container_device_fd);