config: add pkgconfig for arm64
[dpdk.git] / drivers / bus / fslmc / fslmc_vfio.c
index 96ba568..970969d 100644 (file)
@@ -1,7 +1,7 @@
 /* SPDX-License-Identifier: BSD-3-Clause
  *
  *   Copyright (c) 2015-2016 Freescale Semiconductor, Inc. All rights reserved.
- *   Copyright 2016 NXP
+ *   Copyright 2016-2019 NXP
  *
  */
 
 #include "portal/dpaa2_hw_pvt.h"
 #include "portal/dpaa2_hw_dpio.h"
 
-/** Pathname of FSL-MC devices directory. */
-#define SYSFS_FSL_MC_DEVICES "/sys/bus/fsl-mc/devices"
-
 #define FSLMC_CONTAINER_MAX_LEN 8 /**< Of the format dprc.XX */
 
 /* Number of VFIO containers & groups with in */
 static struct fslmc_vfio_group vfio_group;
 static struct fslmc_vfio_container vfio_container;
 static int container_device_fd;
-static char *g_container;
+char *fslmc_container;
+static int fslmc_iommu_type;
 static uint32_t *msi_intr_vaddr;
 void *(*rte_mcp_ptr_list);
 
@@ -71,7 +69,7 @@ fslmc_get_container_group(int *groupid)
        int ret;
        char *container;
 
-       if (!g_container) {
+       if (!fslmc_container) {
                container = getenv("DPRC");
                if (container == NULL) {
                        DPAA2_BUS_DEBUG("DPAA2: DPRC not available");
@@ -83,23 +81,26 @@ fslmc_get_container_group(int *groupid)
                        return -1;
                }
 
-               g_container = strdup(container);
-               if (!g_container) {
+               fslmc_container = strdup(container);
+               if (!fslmc_container) {
                        DPAA2_BUS_ERR("Mem alloc failure; Container name");
                        return -ENOMEM;
                }
        }
 
+       fslmc_iommu_type = (rte_vfio_noiommu_is_enabled() == 1) ?
+               RTE_VFIO_NOIOMMU : VFIO_TYPE1_IOMMU;
+
        /* get group number */
        ret = rte_vfio_get_group_num(SYSFS_FSL_MC_DEVICES,
-                                    g_container, groupid);
+                                    fslmc_container, groupid);
        if (ret <= 0) {
-               DPAA2_BUS_ERR("Unable to find %s IOMMU group", g_container);
+               DPAA2_BUS_ERR("Unable to find %s IOMMU group", fslmc_container);
                return -1;
        }
 
        DPAA2_BUS_DEBUG("Container: %s has VFIO iommu group id = %d",
-                       g_container, *groupid);
+                       fslmc_container, *groupid);
 
        return 0;
 }
@@ -132,7 +133,7 @@ vfio_connect_container(void)
        }
 
        /* Check whether support for SMMU type IOMMU present or not */
-       if (ioctl(fd, VFIO_CHECK_EXTENSION, VFIO_TYPE1_IOMMU)) {
+       if (ioctl(fd, VFIO_CHECK_EXTENSION, fslmc_iommu_type)) {
                /* Connect group to container */
                ret = ioctl(vfio_group.fd, VFIO_GROUP_SET_CONTAINER, &fd);
                if (ret) {
@@ -141,7 +142,7 @@ vfio_connect_container(void)
                        return -errno;
                }
 
-               ret = ioctl(fd, VFIO_SET_IOMMU, VFIO_TYPE1_IOMMU);
+               ret = ioctl(fd, VFIO_SET_IOMMU, fslmc_iommu_type);
                if (ret) {
                        DPAA2_BUS_ERR("Failed to setup VFIO iommu");
                        close(fd);
@@ -176,7 +177,7 @@ static int vfio_map_irq_region(struct fslmc_vfio_group *group)
        vaddr = (unsigned long *)mmap(NULL, 0x1000, PROT_WRITE |
                PROT_READ, MAP_SHARED, container_device_fd, 0x6030000);
        if (vaddr == MAP_FAILED) {
-               DPAA2_BUS_ERR("Unable to map region (errno = %d)", errno);
+               DPAA2_BUS_INFO("Unable to map region (errno = %d)", errno);
                return -errno;
        }
 
@@ -221,6 +222,13 @@ fslmc_memevent_cb(enum rte_mem_event type, const void *addr, size_t len,
                                        "alloc" : "dealloc",
                                va, virt_addr, iova_addr, map_len);
 
+               /* iova_addr may be set to RTE_BAD_IOVA */
+               if (iova_addr == RTE_BAD_IOVA) {
+                       DPAA2_BUS_DEBUG("Segment has invalid iova, skipping\n");
+                       cur_len += map_len;
+                       continue;
+               }
+
                if (type == RTE_MEM_EVENT_ALLOC)
                        ret = fslmc_map_dma(virt_addr, iova_addr, map_len);
                else
@@ -254,6 +262,11 @@ fslmc_map_dma(uint64_t vaddr, rte_iova_t iovaddr __rte_unused, size_t len)
        };
        int ret;
 
+       if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
+               DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
+               return 0;
+       }
+
        dma_map.size = len;
        dma_map.vaddr = vaddr;
 
@@ -293,6 +306,11 @@ fslmc_unmap_dma(uint64_t vaddr, uint64_t iovaddr __rte_unused, size_t len)
        };
        int ret;
 
+       if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
+               DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
+               return 0;
+       }
+
        dma_unmap.size = len;
        dma_unmap.iova = vaddr;
 
@@ -318,11 +336,15 @@ fslmc_unmap_dma(uint64_t vaddr, uint64_t iovaddr __rte_unused, size_t len)
 
 static int
 fslmc_dmamap_seg(const struct rte_memseg_list *msl __rte_unused,
-                const struct rte_memseg *ms, void *arg)
+               const struct rte_memseg *ms, void *arg)
 {
        int *n_segs = arg;
        int ret;
 
+       /* if IOVA address is invalid, skip */
+       if (ms->iova == RTE_BAD_IOVA)
+               return 0;
+
        ret = fslmc_map_dma(ms->addr_64, ms->iova, ms->len);
        if (ret)
                DPAA2_BUS_ERR("Unable to VFIO map (addr=%p, len=%zu)",
@@ -333,17 +355,55 @@ fslmc_dmamap_seg(const struct rte_memseg_list *msl __rte_unused,
        return ret;
 }
 
+int
+rte_fslmc_vfio_mem_dmamap(uint64_t vaddr, uint64_t iova, uint64_t size)
+{
+       int ret;
+       struct fslmc_vfio_group *group;
+       struct vfio_iommu_type1_dma_map dma_map = {
+               .argsz = sizeof(struct vfio_iommu_type1_dma_map),
+               .flags = VFIO_DMA_MAP_FLAG_READ | VFIO_DMA_MAP_FLAG_WRITE,
+       };
+
+       if (fslmc_iommu_type == RTE_VFIO_NOIOMMU) {
+               DPAA2_BUS_DEBUG("Running in NOIOMMU mode");
+               return 0;
+       }
+
+       /* SET DMA MAP for IOMMU */
+       group = &vfio_group;
+       if (!group->container) {
+               DPAA2_BUS_ERR("Container is not connected");
+               return -1;
+       }
+
+       dma_map.size = size;
+       dma_map.vaddr = vaddr;
+       dma_map.iova = iova;
+
+       DPAA2_BUS_DEBUG("VFIOdmamap 0x%"PRIx64":0x%"PRIx64",size 0x%"PRIx64"\n",
+                       (uint64_t)dma_map.vaddr, (uint64_t)dma_map.iova,
+                       (uint64_t)dma_map.size);
+       ret = ioctl(group->container->fd, VFIO_IOMMU_MAP_DMA,
+                   &dma_map);
+       if (ret) {
+               printf("Unable to map DMA address (errno = %d)\n",
+                       errno);
+               return ret;
+       }
+
+       return 0;
+}
+
 int rte_fslmc_vfio_dmamap(void)
 {
        int i = 0, ret;
-       struct rte_mem_config *mcfg = rte_eal_get_configuration()->mem_config;
-       rte_rwlock_t *mem_lock = &mcfg->memory_hotplug_lock;
 
        /* Lock before parsing and registering callback to memory subsystem */
-       rte_rwlock_read_lock(mem_lock);
+       rte_mcfg_mem_read_lock();
 
        if (rte_memseg_walk(fslmc_dmamap_seg, &i) < 0) {
-               rte_rwlock_read_unlock(mem_lock);
+               rte_mcfg_mem_read_unlock();
                return -1;
        }
 
@@ -367,33 +427,141 @@ int rte_fslmc_vfio_dmamap(void)
        /* Existing segments have been mapped and memory callback for hotplug
         * has been installed.
         */
-       rte_rwlock_read_unlock(mem_lock);
+       rte_mcfg_mem_read_unlock();
+
+       return 0;
+}
+
+static int
+fslmc_vfio_setup_device(const char *sysfs_base, const char *dev_addr,
+               int *vfio_dev_fd, struct vfio_device_info *device_info)
+{
+       struct vfio_group_status group_status = {
+                       .argsz = sizeof(group_status)
+       };
+       int vfio_group_fd, vfio_container_fd, iommu_group_no, ret;
+
+       /* get group number */
+       ret = rte_vfio_get_group_num(sysfs_base, dev_addr, &iommu_group_no);
+       if (ret < 0)
+               return -1;
+
+       /* get the actual group fd */
+       vfio_group_fd = rte_vfio_get_group_fd(iommu_group_no);
+       if (vfio_group_fd < 0)
+               return -1;
+
+       /* if group_fd == 0, that means the device isn't managed by VFIO */
+       if (vfio_group_fd == 0) {
+               RTE_LOG(WARNING, EAL, " %s not managed by VFIO driver, skipping\n",
+                               dev_addr);
+               return 1;
+       }
+
+       /* Opens main vfio file descriptor which represents the "container" */
+       vfio_container_fd = rte_vfio_get_container_fd();
+       if (vfio_container_fd < 0) {
+               DPAA2_BUS_ERR("Failed to open VFIO container");
+               return -errno;
+       }
+
+       /* check if the group is viable */
+       ret = ioctl(vfio_group_fd, VFIO_GROUP_GET_STATUS, &group_status);
+       if (ret) {
+               DPAA2_BUS_ERR("  %s cannot get group status, "
+                               "error %i (%s)\n", dev_addr,
+                               errno, strerror(errno));
+               close(vfio_group_fd);
+               rte_vfio_clear_group(vfio_group_fd);
+               return -1;
+       } else if (!(group_status.flags & VFIO_GROUP_FLAGS_VIABLE)) {
+               DPAA2_BUS_ERR("  %s VFIO group is not viable!\n", dev_addr);
+               close(vfio_group_fd);
+               rte_vfio_clear_group(vfio_group_fd);
+               return -1;
+       }
+       /* At this point, we know that this group is viable (meaning,
+        * all devices are either bound to VFIO or not bound to anything)
+        */
+
+       /* check if group does not have a container yet */
+       if (!(group_status.flags & VFIO_GROUP_FLAGS_CONTAINER_SET)) {
+
+               /* add group to a container */
+               ret = ioctl(vfio_group_fd, VFIO_GROUP_SET_CONTAINER,
+                               &vfio_container_fd);
+               if (ret) {
+                       DPAA2_BUS_ERR("  %s cannot add VFIO group to container, "
+                                       "error %i (%s)\n", dev_addr,
+                                       errno, strerror(errno));
+                       close(vfio_group_fd);
+                       close(vfio_container_fd);
+                       rte_vfio_clear_group(vfio_group_fd);
+                       return -1;
+               }
+
+               /*
+                * set an IOMMU type for container
+                *
+                */
+               if (ioctl(vfio_container_fd, VFIO_CHECK_EXTENSION,
+                         fslmc_iommu_type)) {
+                       ret = ioctl(vfio_container_fd, VFIO_SET_IOMMU,
+                                   fslmc_iommu_type);
+                       if (ret) {
+                               DPAA2_BUS_ERR("Failed to setup VFIO iommu");
+                               close(vfio_group_fd);
+                               close(vfio_container_fd);
+                               return -errno;
+                       }
+               } else {
+                       DPAA2_BUS_ERR("No supported IOMMU available");
+                       close(vfio_group_fd);
+                       close(vfio_container_fd);
+                       return -EINVAL;
+               }
+       }
+
+       /* get a file descriptor for the device */
+       *vfio_dev_fd = ioctl(vfio_group_fd, VFIO_GROUP_GET_DEVICE_FD, dev_addr);
+       if (*vfio_dev_fd < 0) {
+               /* if we cannot get a device fd, this implies a problem with
+                * the VFIO group or the container not having IOMMU configured.
+                */
+
+               DPAA2_BUS_WARN("Getting a vfio_dev_fd for %s failed", dev_addr);
+               close(vfio_group_fd);
+               close(vfio_container_fd);
+               rte_vfio_clear_group(vfio_group_fd);
+               return -1;
+       }
+
+       /* test and setup the device */
+       ret = ioctl(*vfio_dev_fd, VFIO_DEVICE_GET_INFO, device_info);
+       if (ret) {
+               DPAA2_BUS_ERR("  %s cannot get device info, error %i (%s)",
+                               dev_addr, errno, strerror(errno));
+               close(*vfio_dev_fd);
+               close(vfio_group_fd);
+               close(vfio_container_fd);
+               rte_vfio_clear_group(vfio_group_fd);
+               return -1;
+       }
 
        return 0;
 }
 
-static int64_t vfio_map_mcp_obj(struct fslmc_vfio_group *group, char *mcp_obj)
+static intptr_t vfio_map_mcp_obj(const char *mcp_obj)
 {
        intptr_t v_addr = (intptr_t)MAP_FAILED;
        int32_t ret, mc_fd;
+       struct vfio_group_status status = { .argsz = sizeof(status) };
 
        struct vfio_device_info d_info = { .argsz = sizeof(d_info) };
        struct vfio_region_info reg_info = { .argsz = sizeof(reg_info) };
 
-       /* getting the mcp object's fd*/
-       mc_fd = ioctl(group->fd, VFIO_GROUP_GET_DEVICE_FD, mcp_obj);
-       if (mc_fd < 0) {
-               DPAA2_BUS_ERR("Error in VFIO get dev %s fd from group %d",
-                             mcp_obj, group->fd);
-               return v_addr;
-       }
-
-       /* getting device info*/
-       ret = ioctl(mc_fd, VFIO_DEVICE_GET_INFO, &d_info);
-       if (ret < 0) {
-               DPAA2_BUS_ERR("Error in VFIO getting DEVICE_INFO");
-               goto MC_FAILURE;
-       }
+       fslmc_vfio_setup_device(SYSFS_FSL_MC_DEVICES, mcp_obj,
+                       &mc_fd, &d_info);
 
        /* getting device region info*/
        ret = ioctl(mc_fd, VFIO_DEVICE_GET_REGION_INFO, &reg_info);
@@ -526,19 +694,8 @@ fslmc_process_iodevices(struct rte_dpaa2_device *dev)
        struct vfio_device_info device_info = { .argsz = sizeof(device_info) };
        struct rte_dpaa2_object *object = NULL;
 
-       dev_fd = ioctl(vfio_group.fd, VFIO_GROUP_GET_DEVICE_FD,
-                      dev->device.name);
-       if (dev_fd <= 0) {
-               DPAA2_BUS_ERR("Unable to obtain device FD for device:%s",
-                             dev->device.name);
-               return -1;
-       }
-
-       if (ioctl(dev_fd, VFIO_DEVICE_GET_INFO, &device_info)) {
-               DPAA2_BUS_ERR("Unable to obtain information for device:%s",
-                             dev->device.name);
-               return -1;
-       }
+       fslmc_vfio_setup_device(SYSFS_FSL_MC_DEVICES, dev->device.name,
+                       &dev_fd, &device_info);
 
        switch (dev->dev_type) {
        case DPAA2_ETH:
@@ -549,6 +706,8 @@ fslmc_process_iodevices(struct rte_dpaa2_device *dev)
        case DPAA2_IO:
        case DPAA2_CI:
        case DPAA2_BPOOL:
+       case DPAA2_DPRTC:
+       case DPAA2_MUX:
                TAILQ_FOREACH(object, &dpaa2_obj_list, next) {
                        if (dev->dev_type == object->dev_type)
                                object->create(dev_fd, &device_info,
@@ -569,38 +728,48 @@ fslmc_process_iodevices(struct rte_dpaa2_device *dev)
 static int
 fslmc_process_mcp(struct rte_dpaa2_device *dev)
 {
+       int ret;
        intptr_t v_addr;
-       char *dev_name;
+       char *dev_name = NULL;
        struct fsl_mc_io dpmng  = {0};
        struct mc_version mc_ver_info = {0};
 
        rte_mcp_ptr_list = malloc(sizeof(void *) * 1);
        if (!rte_mcp_ptr_list) {
                DPAA2_BUS_ERR("Unable to allocate MC portal memory");
-               return -ENOMEM;
+               ret = -ENOMEM;
+               goto cleanup;
        }
 
        dev_name = strdup(dev->device.name);
        if (!dev_name) {
                DPAA2_BUS_ERR("Unable to allocate MC device name memory");
-               free(rte_mcp_ptr_list);
-               rte_mcp_ptr_list = NULL;
-               return -ENOMEM;
+               ret = -ENOMEM;
+               goto cleanup;
        }
 
-       v_addr = vfio_map_mcp_obj(&vfio_group, dev_name);
+       v_addr = vfio_map_mcp_obj(dev->device.name);
        if (v_addr == (intptr_t)MAP_FAILED) {
                DPAA2_BUS_ERR("Error mapping region (errno = %d)", errno);
-               free(rte_mcp_ptr_list);
-               rte_mcp_ptr_list = NULL;
-               return -1;
+               ret = -1;
+               goto cleanup;
        }
 
        /* check the MC version compatibility */
        dpmng.regs = (void *)v_addr;
+
+       /* In case of secondary processes, MC version check is no longer
+        * required.
+        */
+       if (rte_eal_process_type() == RTE_PROC_SECONDARY) {
+               rte_mcp_ptr_list[0] = (void *)v_addr;
+               return 0;
+       }
+
        if (mc_get_version(&dpmng, CMD_PRI_LOW, &mc_ver_info)) {
                DPAA2_BUS_ERR("Unable to obtain MC version");
-               return -1;
+               ret = -1;
+               goto cleanup;
        }
 
        if ((mc_ver_info.major != MC_VER_MAJOR) ||
@@ -610,13 +779,24 @@ fslmc_process_mcp(struct rte_dpaa2_device *dev)
                              MC_VER_MAJOR, MC_VER_MINOR,
                              mc_ver_info.major, mc_ver_info.minor,
                              mc_ver_info.revision);
-               free(rte_mcp_ptr_list);
-               rte_mcp_ptr_list = NULL;
-               return -1;
+               ret = -1;
+               goto cleanup;
        }
        rte_mcp_ptr_list[0] = (void *)v_addr;
 
+       free(dev_name);
        return 0;
+
+cleanup:
+       if (dev_name)
+               free(dev_name);
+
+       if (rte_mcp_ptr_list) {
+               free(rte_mcp_ptr_list);
+               rte_mcp_ptr_list = NULL;
+       }
+
+       return ret;
 }
 
 int
@@ -629,6 +809,15 @@ fslmc_vfio_process_group(void)
        /* Search the MCP as that should be initialized first. */
        TAILQ_FOREACH_SAFE(dev, &rte_fslmc_bus.device_list, next, dev_temp) {
                if (dev->dev_type == DPAA2_MPORTAL) {
+                       if (dev->device.devargs &&
+                           dev->device.devargs->policy == RTE_DEV_BLACKLISTED) {
+                               DPAA2_BUS_LOG(DEBUG, "%s Blacklisted, skipping",
+                                             dev->device.name);
+                               TAILQ_REMOVE(&rte_fslmc_bus.device_list,
+                                               dev, next);
+                               continue;
+                       }
+
                        ret = fslmc_process_mcp(dev);
                        if (ret) {
                                DPAA2_BUS_ERR("Unable to map MC Portal");
@@ -653,9 +842,13 @@ fslmc_vfio_process_group(void)
        }
 
        TAILQ_FOREACH_SAFE(dev, &rte_fslmc_bus.device_list, next, dev_temp) {
-               if (!dev)
-                       break;
-
+               if (dev->device.devargs &&
+                   dev->device.devargs->policy == RTE_DEV_BLACKLISTED) {
+                       DPAA2_BUS_LOG(DEBUG, "%s Blacklisted, skipping",
+                                     dev->device.name);
+                       TAILQ_REMOVE(&rte_fslmc_bus.device_list, dev, next);
+                       continue;
+               }
                switch (dev->dev_type) {
                case DPAA2_ETH:
                case DPAA2_CRYPTO:
@@ -668,9 +861,18 @@ fslmc_vfio_process_group(void)
                        }
                        break;
                case DPAA2_CON:
-               case DPAA2_IO:
                case DPAA2_CI:
                case DPAA2_BPOOL:
+               case DPAA2_DPRTC:
+               case DPAA2_MUX:
+                       /* IN case of secondary processes, all control objects
+                        * like dpbp, dpcon, dpci are not initialized/required
+                        * - all of these are assumed to be initialized and made
+                        *   available by primary.
+                        */
+                       if (rte_eal_process_type() == RTE_PROC_SECONDARY)
+                               continue;
+
                        /* Call the object creation routine and remove the
                         * device entry from device list
                         */
@@ -681,12 +883,15 @@ fslmc_vfio_process_group(void)
                                return -1;
                        }
 
-                       /* This device is not required to be in the DPDK
-                        * exposed device list.
-                        */
-                       TAILQ_REMOVE(&rte_fslmc_bus.device_list, dev, next);
-                       free(dev);
-                       dev = NULL;
+                       break;
+               case DPAA2_IO:
+                       ret = fslmc_process_iodevices(dev);
+                       if (ret) {
+                               DPAA2_BUS_DEBUG("Dev (%s) init failed",
+                                               dev->device.name);
+                               return -1;
+                       }
+
                        break;
                case DPAA2_UNKNOWN:
                default:
@@ -764,10 +969,10 @@ fslmc_vfio_setup_group(void)
        }
 
        /* Get Device information */
-       ret = ioctl(vfio_group.fd, VFIO_GROUP_GET_DEVICE_FD, g_container);
+       ret = ioctl(vfio_group.fd, VFIO_GROUP_GET_DEVICE_FD, fslmc_container);
        if (ret < 0) {
                DPAA2_BUS_ERR("Error getting device %s fd from group %d",
-                             g_container, vfio_group.groupid);
+                             fslmc_container, vfio_group.groupid);
                close(vfio_group.fd);
                rte_vfio_clear_group(vfio_group.fd);
                return ret;