11e0a84eaSCindy Lu /* 21e0a84eaSCindy Lu * vhost-vdpa.c 31e0a84eaSCindy Lu * 41e0a84eaSCindy Lu * Copyright(c) 2017-2018 Intel Corporation. 51e0a84eaSCindy Lu * Copyright(c) 2020 Red Hat, Inc. 61e0a84eaSCindy Lu * 71e0a84eaSCindy Lu * This work is licensed under the terms of the GNU GPL, version 2 or later. 81e0a84eaSCindy Lu * See the COPYING file in the top-level directory. 91e0a84eaSCindy Lu * 101e0a84eaSCindy Lu */ 111e0a84eaSCindy Lu 121e0a84eaSCindy Lu #include "qemu/osdep.h" 131e0a84eaSCindy Lu #include "clients.h" 14bd907ae4SEugenio Pérez #include "hw/virtio/virtio-net.h" 151e0a84eaSCindy Lu #include "net/vhost_net.h" 161e0a84eaSCindy Lu #include "net/vhost-vdpa.h" 171e0a84eaSCindy Lu #include "hw/virtio/vhost-vdpa.h" 181e0a84eaSCindy Lu #include "qemu/config-file.h" 191e0a84eaSCindy Lu #include "qemu/error-report.h" 20bd907ae4SEugenio Pérez #include "qemu/log.h" 21bd907ae4SEugenio Pérez #include "qemu/memalign.h" 221e0a84eaSCindy Lu #include "qemu/option.h" 231e0a84eaSCindy Lu #include "qapi/error.h" 2440237840SJason Wang #include <linux/vhost.h> 251e0a84eaSCindy Lu #include <sys/ioctl.h> 261e0a84eaSCindy Lu #include <err.h> 271e0a84eaSCindy Lu #include "standard-headers/linux/virtio_net.h" 281e0a84eaSCindy Lu #include "monitor/monitor.h" 2969498430SEugenio Pérez #include "migration/migration.h" 3069498430SEugenio Pérez #include "migration/misc.h" 311e0a84eaSCindy Lu #include "hw/virtio/vhost.h" 321e0a84eaSCindy Lu 331e0a84eaSCindy Lu /* Todo:need to add the multiqueue support here */ 341e0a84eaSCindy Lu typedef struct VhostVDPAState { 351e0a84eaSCindy Lu NetClientState nc; 361e0a84eaSCindy Lu struct vhost_vdpa vhost_vdpa; 3769498430SEugenio Pérez Notifier migration_state; 381e0a84eaSCindy Lu VHostNetState *vhost_net; 392df4dd31SEugenio Pérez 402df4dd31SEugenio Pérez /* Control commands shadow buffers */ 4117fb889fSEugenio Pérez void *cvq_cmd_out_buffer; 4217fb889fSEugenio Pérez virtio_net_ctrl_ack *status; 4317fb889fSEugenio Pérez 447f211a28SEugenio Pérez /* The device always have SVQ enabled */ 457f211a28SEugenio Pérez bool always_svq; 461e0a84eaSCindy Lu bool started; 471e0a84eaSCindy Lu } VhostVDPAState; 481e0a84eaSCindy Lu 491e0a84eaSCindy Lu const int vdpa_feature_bits[] = { 501e0a84eaSCindy Lu VIRTIO_F_NOTIFY_ON_EMPTY, 511e0a84eaSCindy Lu VIRTIO_RING_F_INDIRECT_DESC, 521e0a84eaSCindy Lu VIRTIO_RING_F_EVENT_IDX, 531e0a84eaSCindy Lu VIRTIO_F_ANY_LAYOUT, 541e0a84eaSCindy Lu VIRTIO_F_VERSION_1, 551e0a84eaSCindy Lu VIRTIO_NET_F_CSUM, 561e0a84eaSCindy Lu VIRTIO_NET_F_GUEST_CSUM, 571e0a84eaSCindy Lu VIRTIO_NET_F_GSO, 581e0a84eaSCindy Lu VIRTIO_NET_F_GUEST_TSO4, 591e0a84eaSCindy Lu VIRTIO_NET_F_GUEST_TSO6, 601e0a84eaSCindy Lu VIRTIO_NET_F_GUEST_ECN, 611e0a84eaSCindy Lu VIRTIO_NET_F_GUEST_UFO, 621e0a84eaSCindy Lu VIRTIO_NET_F_HOST_TSO4, 631e0a84eaSCindy Lu VIRTIO_NET_F_HOST_TSO6, 641e0a84eaSCindy Lu VIRTIO_NET_F_HOST_ECN, 651e0a84eaSCindy Lu VIRTIO_NET_F_HOST_UFO, 661e0a84eaSCindy Lu VIRTIO_NET_F_MRG_RXBUF, 671e0a84eaSCindy Lu VIRTIO_NET_F_MTU, 6840237840SJason Wang VIRTIO_NET_F_CTRL_RX, 6940237840SJason Wang VIRTIO_NET_F_CTRL_RX_EXTRA, 7040237840SJason Wang VIRTIO_NET_F_CTRL_VLAN, 7140237840SJason Wang VIRTIO_NET_F_CTRL_MAC_ADDR, 7240237840SJason Wang VIRTIO_NET_F_RSS, 7340237840SJason Wang VIRTIO_NET_F_MQ, 7440237840SJason Wang VIRTIO_NET_F_CTRL_VQ, 751e0a84eaSCindy Lu VIRTIO_F_IOMMU_PLATFORM, 761e0a84eaSCindy Lu VIRTIO_F_RING_PACKED, 77562a7d23SStefano Garzarella VIRTIO_F_RING_RESET, 780145c393SAndrew Melnychenko VIRTIO_NET_F_RSS, 790145c393SAndrew Melnychenko VIRTIO_NET_F_HASH_REPORT, 809aa47eddSSi-Wei Liu VIRTIO_NET_F_STATUS, 811e0a84eaSCindy Lu VHOST_INVALID_FEATURE_BIT 821e0a84eaSCindy Lu }; 831e0a84eaSCindy Lu 841576dbb5SEugenio Pérez /** Supported device specific feature bits with SVQ */ 851576dbb5SEugenio Pérez static const uint64_t vdpa_svq_device_features = 861576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_CSUM) | 871576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_GUEST_CSUM) | 881576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_MTU) | 891576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_MAC) | 901576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_GUEST_TSO4) | 911576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_GUEST_TSO6) | 921576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_GUEST_ECN) | 931576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_GUEST_UFO) | 941576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_HOST_TSO4) | 951576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_HOST_TSO6) | 961576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_HOST_ECN) | 971576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_HOST_UFO) | 981576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_MRG_RXBUF) | 991576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_STATUS) | 1001576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_CTRL_VQ) | 10172b99a87SEugenio Pérez BIT_ULL(VIRTIO_NET_F_MQ) | 1021576dbb5SEugenio Pérez BIT_ULL(VIRTIO_F_ANY_LAYOUT) | 1031576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR) | 104609ab4c3SEugenio Pérez /* VHOST_F_LOG_ALL is exposed by SVQ */ 105609ab4c3SEugenio Pérez BIT_ULL(VHOST_F_LOG_ALL) | 1061576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_RSC_EXT) | 1070d74e2b7SEugenio Pérez BIT_ULL(VIRTIO_NET_F_STANDBY) | 1080d74e2b7SEugenio Pérez BIT_ULL(VIRTIO_NET_F_SPEED_DUPLEX); 1091576dbb5SEugenio Pérez 110c1a10086SEugenio Pérez #define VHOST_VDPA_NET_CVQ_ASID 1 111c1a10086SEugenio Pérez 1121e0a84eaSCindy Lu VHostNetState *vhost_vdpa_get_vhost_net(NetClientState *nc) 1131e0a84eaSCindy Lu { 1141e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 1151e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 1161e0a84eaSCindy Lu return s->vhost_net; 1171e0a84eaSCindy Lu } 1181e0a84eaSCindy Lu 11936e46472SEugenio Pérez static bool vhost_vdpa_net_valid_svq_features(uint64_t features, Error **errp) 12036e46472SEugenio Pérez { 12136e46472SEugenio Pérez uint64_t invalid_dev_features = 12236e46472SEugenio Pérez features & ~vdpa_svq_device_features & 12336e46472SEugenio Pérez /* Transport are all accepted at this point */ 12436e46472SEugenio Pérez ~MAKE_64BIT_MASK(VIRTIO_TRANSPORT_F_START, 12536e46472SEugenio Pérez VIRTIO_TRANSPORT_F_END - VIRTIO_TRANSPORT_F_START); 12636e46472SEugenio Pérez 12736e46472SEugenio Pérez if (invalid_dev_features) { 12836e46472SEugenio Pérez error_setg(errp, "vdpa svq does not work with features 0x%" PRIx64, 12936e46472SEugenio Pérez invalid_dev_features); 130258a0394SEugenio Pérez return false; 13136e46472SEugenio Pérez } 13236e46472SEugenio Pérez 133258a0394SEugenio Pérez return vhost_svq_valid_features(features, errp); 13436e46472SEugenio Pérez } 13536e46472SEugenio Pérez 1361e0a84eaSCindy Lu static int vhost_vdpa_net_check_device_id(struct vhost_net *net) 1371e0a84eaSCindy Lu { 1381e0a84eaSCindy Lu uint32_t device_id; 1391e0a84eaSCindy Lu int ret; 1401e0a84eaSCindy Lu struct vhost_dev *hdev; 1411e0a84eaSCindy Lu 1421e0a84eaSCindy Lu hdev = (struct vhost_dev *)&net->dev; 1431e0a84eaSCindy Lu ret = hdev->vhost_ops->vhost_get_device_id(hdev, &device_id); 1441e0a84eaSCindy Lu if (device_id != VIRTIO_ID_NET) { 1451e0a84eaSCindy Lu return -ENOTSUP; 1461e0a84eaSCindy Lu } 1471e0a84eaSCindy Lu return ret; 1481e0a84eaSCindy Lu } 1491e0a84eaSCindy Lu 15040237840SJason Wang static int vhost_vdpa_add(NetClientState *ncs, void *be, 15140237840SJason Wang int queue_pair_index, int nvqs) 1521e0a84eaSCindy Lu { 1531e0a84eaSCindy Lu VhostNetOptions options; 1541e0a84eaSCindy Lu struct vhost_net *net = NULL; 1551e0a84eaSCindy Lu VhostVDPAState *s; 1561e0a84eaSCindy Lu int ret; 1571e0a84eaSCindy Lu 1581e0a84eaSCindy Lu options.backend_type = VHOST_BACKEND_TYPE_VDPA; 1591e0a84eaSCindy Lu assert(ncs->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 1601e0a84eaSCindy Lu s = DO_UPCAST(VhostVDPAState, nc, ncs); 1611e0a84eaSCindy Lu options.net_backend = ncs; 1621e0a84eaSCindy Lu options.opaque = be; 1631e0a84eaSCindy Lu options.busyloop_timeout = 0; 16440237840SJason Wang options.nvqs = nvqs; 1651e0a84eaSCindy Lu 1661e0a84eaSCindy Lu net = vhost_net_init(&options); 1671e0a84eaSCindy Lu if (!net) { 1681e0a84eaSCindy Lu error_report("failed to init vhost_net for queue"); 169a97ef87aSJason Wang goto err_init; 1701e0a84eaSCindy Lu } 1711e0a84eaSCindy Lu s->vhost_net = net; 1721e0a84eaSCindy Lu ret = vhost_vdpa_net_check_device_id(net); 1731e0a84eaSCindy Lu if (ret) { 174a97ef87aSJason Wang goto err_check; 1751e0a84eaSCindy Lu } 1761e0a84eaSCindy Lu return 0; 177a97ef87aSJason Wang err_check: 1781e0a84eaSCindy Lu vhost_net_cleanup(net); 179ab36edcfSJason Wang g_free(net); 180a97ef87aSJason Wang err_init: 1811e0a84eaSCindy Lu return -1; 1821e0a84eaSCindy Lu } 1831e0a84eaSCindy Lu 1841e0a84eaSCindy Lu static void vhost_vdpa_cleanup(NetClientState *nc) 1851e0a84eaSCindy Lu { 1861e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 1871e0a84eaSCindy Lu 1882df4dd31SEugenio Pérez qemu_vfree(s->cvq_cmd_out_buffer); 18917fb889fSEugenio Pérez qemu_vfree(s->status); 1901e0a84eaSCindy Lu if (s->vhost_net) { 1911e0a84eaSCindy Lu vhost_net_cleanup(s->vhost_net); 1921e0a84eaSCindy Lu g_free(s->vhost_net); 1931e0a84eaSCindy Lu s->vhost_net = NULL; 1941e0a84eaSCindy Lu } 19557b3a7d8SCindy Lu if (s->vhost_vdpa.device_fd >= 0) { 19657b3a7d8SCindy Lu qemu_close(s->vhost_vdpa.device_fd); 19757b3a7d8SCindy Lu s->vhost_vdpa.device_fd = -1; 19857b3a7d8SCindy Lu } 1991e0a84eaSCindy Lu } 2001e0a84eaSCindy Lu 2011e0a84eaSCindy Lu static bool vhost_vdpa_has_vnet_hdr(NetClientState *nc) 2021e0a84eaSCindy Lu { 2031e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 2041e0a84eaSCindy Lu 2051e0a84eaSCindy Lu return true; 2061e0a84eaSCindy Lu } 2071e0a84eaSCindy Lu 2081e0a84eaSCindy Lu static bool vhost_vdpa_has_ufo(NetClientState *nc) 2091e0a84eaSCindy Lu { 2101e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 2111e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 2121e0a84eaSCindy Lu uint64_t features = 0; 2131e0a84eaSCindy Lu features |= (1ULL << VIRTIO_NET_F_HOST_UFO); 2141e0a84eaSCindy Lu features = vhost_net_get_features(s->vhost_net, features); 2151e0a84eaSCindy Lu return !!(features & (1ULL << VIRTIO_NET_F_HOST_UFO)); 2161e0a84eaSCindy Lu 2171e0a84eaSCindy Lu } 2181e0a84eaSCindy Lu 219ee8a1c63SKevin Wolf static bool vhost_vdpa_check_peer_type(NetClientState *nc, ObjectClass *oc, 220ee8a1c63SKevin Wolf Error **errp) 221ee8a1c63SKevin Wolf { 222ee8a1c63SKevin Wolf const char *driver = object_class_get_name(oc); 223ee8a1c63SKevin Wolf 224ee8a1c63SKevin Wolf if (!g_str_has_prefix(driver, "virtio-net-")) { 225ee8a1c63SKevin Wolf error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*"); 226ee8a1c63SKevin Wolf return false; 227ee8a1c63SKevin Wolf } 228ee8a1c63SKevin Wolf 229ee8a1c63SKevin Wolf return true; 230ee8a1c63SKevin Wolf } 231ee8a1c63SKevin Wolf 232846a1e85SEugenio Pérez /** Dummy receive in case qemu falls back to userland tap networking */ 233846a1e85SEugenio Pérez static ssize_t vhost_vdpa_receive(NetClientState *nc, const uint8_t *buf, 234846a1e85SEugenio Pérez size_t size) 235846a1e85SEugenio Pérez { 236bc5add1dSSi-Wei Liu return size; 237846a1e85SEugenio Pérez } 238846a1e85SEugenio Pérez 23900ef422eSEugenio Pérez /** From any vdpa net client, get the netclient of the first queue pair */ 24000ef422eSEugenio Pérez static VhostVDPAState *vhost_vdpa_net_first_nc_vdpa(VhostVDPAState *s) 24100ef422eSEugenio Pérez { 24200ef422eSEugenio Pérez NICState *nic = qemu_get_nic(s->nc.peer); 24300ef422eSEugenio Pérez NetClientState *nc0 = qemu_get_peer(nic->ncs, 0); 24400ef422eSEugenio Pérez 24500ef422eSEugenio Pérez return DO_UPCAST(VhostVDPAState, nc, nc0); 24600ef422eSEugenio Pérez } 24700ef422eSEugenio Pérez 24869498430SEugenio Pérez static void vhost_vdpa_net_log_global_enable(VhostVDPAState *s, bool enable) 24969498430SEugenio Pérez { 25069498430SEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 25169498430SEugenio Pérez VirtIONet *n; 25269498430SEugenio Pérez VirtIODevice *vdev; 25369498430SEugenio Pérez int data_queue_pairs, cvq, r; 25469498430SEugenio Pérez 25569498430SEugenio Pérez /* We are only called on the first data vqs and only if x-svq is not set */ 25669498430SEugenio Pérez if (s->vhost_vdpa.shadow_vqs_enabled == enable) { 25769498430SEugenio Pérez return; 25869498430SEugenio Pérez } 25969498430SEugenio Pérez 26069498430SEugenio Pérez vdev = v->dev->vdev; 26169498430SEugenio Pérez n = VIRTIO_NET(vdev); 26269498430SEugenio Pérez if (!n->vhost_started) { 26369498430SEugenio Pérez return; 26469498430SEugenio Pérez } 26569498430SEugenio Pérez 26669498430SEugenio Pérez data_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 26769498430SEugenio Pérez cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ? 26869498430SEugenio Pérez n->max_ncs - n->max_queue_pairs : 0; 26969498430SEugenio Pérez /* 27069498430SEugenio Pérez * TODO: vhost_net_stop does suspend, get_base and reset. We can be smarter 27169498430SEugenio Pérez * in the future and resume the device if read-only operations between 27269498430SEugenio Pérez * suspend and reset goes wrong. 27369498430SEugenio Pérez */ 27469498430SEugenio Pérez vhost_net_stop(vdev, n->nic->ncs, data_queue_pairs, cvq); 27569498430SEugenio Pérez 27669498430SEugenio Pérez /* Start will check migration setup_or_active to configure or not SVQ */ 27769498430SEugenio Pérez r = vhost_net_start(vdev, n->nic->ncs, data_queue_pairs, cvq); 27869498430SEugenio Pérez if (unlikely(r < 0)) { 27969498430SEugenio Pérez error_report("unable to start vhost net: %s(%d)", g_strerror(-r), -r); 28069498430SEugenio Pérez } 28169498430SEugenio Pérez } 28269498430SEugenio Pérez 28369498430SEugenio Pérez static void vdpa_net_migration_state_notifier(Notifier *notifier, void *data) 28469498430SEugenio Pérez { 28569498430SEugenio Pérez MigrationState *migration = data; 28669498430SEugenio Pérez VhostVDPAState *s = container_of(notifier, VhostVDPAState, 28769498430SEugenio Pérez migration_state); 28869498430SEugenio Pérez 28969498430SEugenio Pérez if (migration_in_setup(migration)) { 29069498430SEugenio Pérez vhost_vdpa_net_log_global_enable(s, true); 29169498430SEugenio Pérez } else if (migration_has_failed(migration)) { 29269498430SEugenio Pérez vhost_vdpa_net_log_global_enable(s, false); 29369498430SEugenio Pérez } 29469498430SEugenio Pérez } 29569498430SEugenio Pérez 29600ef422eSEugenio Pérez static void vhost_vdpa_net_data_start_first(VhostVDPAState *s) 29700ef422eSEugenio Pérez { 29800ef422eSEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 29900ef422eSEugenio Pérez 30069498430SEugenio Pérez add_migration_state_change_notifier(&s->migration_state); 30100ef422eSEugenio Pérez if (v->shadow_vqs_enabled) { 30200ef422eSEugenio Pérez v->iova_tree = vhost_iova_tree_new(v->iova_range.first, 30300ef422eSEugenio Pérez v->iova_range.last); 30400ef422eSEugenio Pérez } 30500ef422eSEugenio Pérez } 30600ef422eSEugenio Pérez 30700ef422eSEugenio Pérez static int vhost_vdpa_net_data_start(NetClientState *nc) 30800ef422eSEugenio Pérez { 30900ef422eSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 31000ef422eSEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 31100ef422eSEugenio Pérez 31200ef422eSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 31300ef422eSEugenio Pérez 31469498430SEugenio Pérez if (s->always_svq || 31569498430SEugenio Pérez migration_is_setup_or_active(migrate_get_current()->state)) { 31669498430SEugenio Pérez v->shadow_vqs_enabled = true; 31769498430SEugenio Pérez v->shadow_data = true; 31869498430SEugenio Pérez } else { 31969498430SEugenio Pérez v->shadow_vqs_enabled = false; 32069498430SEugenio Pérez v->shadow_data = false; 32169498430SEugenio Pérez } 32269498430SEugenio Pérez 32300ef422eSEugenio Pérez if (v->index == 0) { 32400ef422eSEugenio Pérez vhost_vdpa_net_data_start_first(s); 32500ef422eSEugenio Pérez return 0; 32600ef422eSEugenio Pérez } 32700ef422eSEugenio Pérez 32800ef422eSEugenio Pérez if (v->shadow_vqs_enabled) { 32900ef422eSEugenio Pérez VhostVDPAState *s0 = vhost_vdpa_net_first_nc_vdpa(s); 33000ef422eSEugenio Pérez v->iova_tree = s0->vhost_vdpa.iova_tree; 33100ef422eSEugenio Pérez } 33200ef422eSEugenio Pérez 33300ef422eSEugenio Pérez return 0; 33400ef422eSEugenio Pérez } 33500ef422eSEugenio Pérez 33600ef422eSEugenio Pérez static void vhost_vdpa_net_client_stop(NetClientState *nc) 33700ef422eSEugenio Pérez { 33800ef422eSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 33900ef422eSEugenio Pérez struct vhost_dev *dev; 34000ef422eSEugenio Pérez 34100ef422eSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 34200ef422eSEugenio Pérez 34369498430SEugenio Pérez if (s->vhost_vdpa.index == 0) { 34469498430SEugenio Pérez remove_migration_state_change_notifier(&s->migration_state); 34569498430SEugenio Pérez } 34669498430SEugenio Pérez 34700ef422eSEugenio Pérez dev = s->vhost_vdpa.dev; 34800ef422eSEugenio Pérez if (dev->vq_index + dev->nvqs == dev->vq_index_end) { 34900ef422eSEugenio Pérez g_clear_pointer(&s->vhost_vdpa.iova_tree, vhost_iova_tree_delete); 35000ef422eSEugenio Pérez } 35100ef422eSEugenio Pérez } 35200ef422eSEugenio Pérez 3531e0a84eaSCindy Lu static NetClientInfo net_vhost_vdpa_info = { 3541e0a84eaSCindy Lu .type = NET_CLIENT_DRIVER_VHOST_VDPA, 3551e0a84eaSCindy Lu .size = sizeof(VhostVDPAState), 356846a1e85SEugenio Pérez .receive = vhost_vdpa_receive, 35700ef422eSEugenio Pérez .start = vhost_vdpa_net_data_start, 35800ef422eSEugenio Pérez .stop = vhost_vdpa_net_client_stop, 3591e0a84eaSCindy Lu .cleanup = vhost_vdpa_cleanup, 3601e0a84eaSCindy Lu .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, 3611e0a84eaSCindy Lu .has_ufo = vhost_vdpa_has_ufo, 362ee8a1c63SKevin Wolf .check_peer_type = vhost_vdpa_check_peer_type, 3631e0a84eaSCindy Lu }; 3641e0a84eaSCindy Lu 365*0f2bb0bfSEugenio Pérez /** 366*0f2bb0bfSEugenio Pérez * Get vring virtqueue group 367*0f2bb0bfSEugenio Pérez * 368*0f2bb0bfSEugenio Pérez * @device_fd vdpa device fd 369*0f2bb0bfSEugenio Pérez * @vq_index Virtqueue index 370*0f2bb0bfSEugenio Pérez * 371*0f2bb0bfSEugenio Pérez * Return -errno in case of error, or vq group if success. 372*0f2bb0bfSEugenio Pérez */ 373c1a10086SEugenio Pérez static int64_t vhost_vdpa_get_vring_group(int device_fd, unsigned vq_index) 374c1a10086SEugenio Pérez { 375c1a10086SEugenio Pérez struct vhost_vring_state state = { 376c1a10086SEugenio Pérez .index = vq_index, 377c1a10086SEugenio Pérez }; 378c1a10086SEugenio Pérez int r = ioctl(device_fd, VHOST_VDPA_GET_VRING_GROUP, &state); 379c1a10086SEugenio Pérez 380c1a10086SEugenio Pérez if (unlikely(r < 0)) { 381*0f2bb0bfSEugenio Pérez r = -errno; 382c1a10086SEugenio Pérez error_report("Cannot get VQ %u group: %s", vq_index, 383c1a10086SEugenio Pérez g_strerror(errno)); 384c1a10086SEugenio Pérez return r; 385c1a10086SEugenio Pérez } 386c1a10086SEugenio Pérez 387c1a10086SEugenio Pérez return state.num; 388c1a10086SEugenio Pérez } 389c1a10086SEugenio Pérez 390c1a10086SEugenio Pérez static int vhost_vdpa_set_address_space_id(struct vhost_vdpa *v, 391c1a10086SEugenio Pérez unsigned vq_group, 392c1a10086SEugenio Pérez unsigned asid_num) 393c1a10086SEugenio Pérez { 394c1a10086SEugenio Pérez struct vhost_vring_state asid = { 395c1a10086SEugenio Pérez .index = vq_group, 396c1a10086SEugenio Pérez .num = asid_num, 397c1a10086SEugenio Pérez }; 398c1a10086SEugenio Pérez int r; 399c1a10086SEugenio Pérez 400c1a10086SEugenio Pérez r = ioctl(v->device_fd, VHOST_VDPA_SET_GROUP_ASID, &asid); 401c1a10086SEugenio Pérez if (unlikely(r < 0)) { 402c1a10086SEugenio Pérez error_report("Can't set vq group %u asid %u, errno=%d (%s)", 403c1a10086SEugenio Pérez asid.index, asid.num, errno, g_strerror(errno)); 404c1a10086SEugenio Pérez } 405c1a10086SEugenio Pérez return r; 406c1a10086SEugenio Pérez } 407c1a10086SEugenio Pérez 4082df4dd31SEugenio Pérez static void vhost_vdpa_cvq_unmap_buf(struct vhost_vdpa *v, void *addr) 4092df4dd31SEugenio Pérez { 4102df4dd31SEugenio Pérez VhostIOVATree *tree = v->iova_tree; 4112df4dd31SEugenio Pérez DMAMap needle = { 4122df4dd31SEugenio Pérez /* 4132df4dd31SEugenio Pérez * No need to specify size or to look for more translations since 4142df4dd31SEugenio Pérez * this contiguous chunk was allocated by us. 4152df4dd31SEugenio Pérez */ 4162df4dd31SEugenio Pérez .translated_addr = (hwaddr)(uintptr_t)addr, 4172df4dd31SEugenio Pérez }; 4182df4dd31SEugenio Pérez const DMAMap *map = vhost_iova_tree_find_iova(tree, &needle); 4192df4dd31SEugenio Pérez int r; 4202df4dd31SEugenio Pérez 4212df4dd31SEugenio Pérez if (unlikely(!map)) { 4222df4dd31SEugenio Pérez error_report("Cannot locate expected map"); 4232df4dd31SEugenio Pérez return; 4242df4dd31SEugenio Pérez } 4252df4dd31SEugenio Pérez 426cd831ed5SEugenio Pérez r = vhost_vdpa_dma_unmap(v, v->address_space_id, map->iova, map->size + 1); 4272df4dd31SEugenio Pérez if (unlikely(r != 0)) { 4282df4dd31SEugenio Pérez error_report("Device cannot unmap: %s(%d)", g_strerror(r), r); 4292df4dd31SEugenio Pérez } 4302df4dd31SEugenio Pérez 43169292a8eSEugenio Pérez vhost_iova_tree_remove(tree, *map); 4322df4dd31SEugenio Pérez } 4332df4dd31SEugenio Pérez 4342df4dd31SEugenio Pérez static size_t vhost_vdpa_net_cvq_cmd_len(void) 4352df4dd31SEugenio Pérez { 4362df4dd31SEugenio Pérez /* 4372df4dd31SEugenio Pérez * MAC_TABLE_SET is the ctrl command that produces the longer out buffer. 4382df4dd31SEugenio Pérez * In buffer is always 1 byte, so it should fit here 4392df4dd31SEugenio Pérez */ 4402df4dd31SEugenio Pérez return sizeof(struct virtio_net_ctrl_hdr) + 4412df4dd31SEugenio Pérez 2 * sizeof(struct virtio_net_ctrl_mac) + 4422df4dd31SEugenio Pérez MAC_TABLE_ENTRIES * ETH_ALEN; 4432df4dd31SEugenio Pérez } 4442df4dd31SEugenio Pérez 4452df4dd31SEugenio Pérez static size_t vhost_vdpa_net_cvq_cmd_page_len(void) 4462df4dd31SEugenio Pérez { 4472df4dd31SEugenio Pérez return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size()); 4482df4dd31SEugenio Pérez } 4492df4dd31SEugenio Pérez 4507a7f87e9SEugenio Pérez /** Map CVQ buffer. */ 4517a7f87e9SEugenio Pérez static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size, 4527a7f87e9SEugenio Pérez bool write) 4532df4dd31SEugenio Pérez { 4542df4dd31SEugenio Pérez DMAMap map = {}; 4552df4dd31SEugenio Pérez int r; 4562df4dd31SEugenio Pérez 4572df4dd31SEugenio Pérez map.translated_addr = (hwaddr)(uintptr_t)buf; 4587a7f87e9SEugenio Pérez map.size = size - 1; 4592df4dd31SEugenio Pérez map.perm = write ? IOMMU_RW : IOMMU_RO, 4602df4dd31SEugenio Pérez r = vhost_iova_tree_map_alloc(v->iova_tree, &map); 4612df4dd31SEugenio Pérez if (unlikely(r != IOVA_OK)) { 4622df4dd31SEugenio Pérez error_report("Cannot map injected element"); 4637a7f87e9SEugenio Pérez return r; 4642df4dd31SEugenio Pérez } 4652df4dd31SEugenio Pérez 466cd831ed5SEugenio Pérez r = vhost_vdpa_dma_map(v, v->address_space_id, map.iova, 467cd831ed5SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), buf, !write); 4682df4dd31SEugenio Pérez if (unlikely(r < 0)) { 4692df4dd31SEugenio Pérez goto dma_map_err; 4702df4dd31SEugenio Pérez } 4712df4dd31SEugenio Pérez 4727a7f87e9SEugenio Pérez return 0; 4732df4dd31SEugenio Pérez 4742df4dd31SEugenio Pérez dma_map_err: 47569292a8eSEugenio Pérez vhost_iova_tree_remove(v->iova_tree, map); 4767a7f87e9SEugenio Pérez return r; 4772df4dd31SEugenio Pérez } 4782df4dd31SEugenio Pérez 4797a7f87e9SEugenio Pérez static int vhost_vdpa_net_cvq_start(NetClientState *nc) 4802df4dd31SEugenio Pérez { 48100ef422eSEugenio Pérez VhostVDPAState *s, *s0; 482c1a10086SEugenio Pérez struct vhost_vdpa *v; 483c1a10086SEugenio Pérez uint64_t backend_features; 484c1a10086SEugenio Pérez int64_t cvq_group; 485c1a10086SEugenio Pérez int cvq_index, r; 4862df4dd31SEugenio Pérez 4877a7f87e9SEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 4887a7f87e9SEugenio Pérez 4897a7f87e9SEugenio Pérez s = DO_UPCAST(VhostVDPAState, nc, nc); 490c1a10086SEugenio Pérez v = &s->vhost_vdpa; 491c1a10086SEugenio Pérez 49269498430SEugenio Pérez s0 = vhost_vdpa_net_first_nc_vdpa(s); 49369498430SEugenio Pérez v->shadow_data = s0->vhost_vdpa.shadow_vqs_enabled; 494c1a10086SEugenio Pérez v->shadow_vqs_enabled = s->always_svq; 495c1a10086SEugenio Pérez s->vhost_vdpa.address_space_id = VHOST_VDPA_GUEST_PA_ASID; 496c1a10086SEugenio Pérez 49769498430SEugenio Pérez if (s->vhost_vdpa.shadow_data) { 498c1a10086SEugenio Pérez /* SVQ is already configured for all virtqueues */ 499c1a10086SEugenio Pérez goto out; 500c1a10086SEugenio Pérez } 501c1a10086SEugenio Pérez 502c1a10086SEugenio Pérez /* 503c1a10086SEugenio Pérez * If we early return in these cases SVQ will not be enabled. The migration 504c1a10086SEugenio Pérez * will be blocked as long as vhost-vdpa backends will not offer _F_LOG. 505c1a10086SEugenio Pérez * 506c1a10086SEugenio Pérez * Calling VHOST_GET_BACKEND_FEATURES as they are not available in v->dev 507c1a10086SEugenio Pérez * yet. 508c1a10086SEugenio Pérez */ 509c1a10086SEugenio Pérez r = ioctl(v->device_fd, VHOST_GET_BACKEND_FEATURES, &backend_features); 510c1a10086SEugenio Pérez if (unlikely(r < 0)) { 511c1a10086SEugenio Pérez error_report("Cannot get vdpa backend_features: %s(%d)", 512c1a10086SEugenio Pérez g_strerror(errno), errno); 513c1a10086SEugenio Pérez return -1; 514c1a10086SEugenio Pérez } 515525ae115SEugenio Pérez if (!(backend_features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID)) || 516c1a10086SEugenio Pérez !vhost_vdpa_net_valid_svq_features(v->dev->features, NULL)) { 517c1a10086SEugenio Pérez return 0; 518c1a10086SEugenio Pérez } 519c1a10086SEugenio Pérez 520c1a10086SEugenio Pérez /* 521c1a10086SEugenio Pérez * Check if all the virtqueues of the virtio device are in a different vq 522c1a10086SEugenio Pérez * than the last vq. VQ group of last group passed in cvq_group. 523c1a10086SEugenio Pérez */ 524c1a10086SEugenio Pérez cvq_index = v->dev->vq_index_end - 1; 525c1a10086SEugenio Pérez cvq_group = vhost_vdpa_get_vring_group(v->device_fd, cvq_index); 526c1a10086SEugenio Pérez if (unlikely(cvq_group < 0)) { 527c1a10086SEugenio Pérez return cvq_group; 528c1a10086SEugenio Pérez } 529c1a10086SEugenio Pérez for (int i = 0; i < cvq_index; ++i) { 530c1a10086SEugenio Pérez int64_t group = vhost_vdpa_get_vring_group(v->device_fd, i); 531c1a10086SEugenio Pérez 532c1a10086SEugenio Pérez if (unlikely(group < 0)) { 533c1a10086SEugenio Pérez return group; 534c1a10086SEugenio Pérez } 535c1a10086SEugenio Pérez 536c1a10086SEugenio Pérez if (group == cvq_group) { 537c1a10086SEugenio Pérez return 0; 538c1a10086SEugenio Pérez } 539c1a10086SEugenio Pérez } 540c1a10086SEugenio Pérez 541c1a10086SEugenio Pérez r = vhost_vdpa_set_address_space_id(v, cvq_group, VHOST_VDPA_NET_CVQ_ASID); 542c1a10086SEugenio Pérez if (unlikely(r < 0)) { 543c1a10086SEugenio Pérez return r; 544c1a10086SEugenio Pérez } 545c1a10086SEugenio Pérez 546c1a10086SEugenio Pérez v->shadow_vqs_enabled = true; 547c1a10086SEugenio Pérez s->vhost_vdpa.address_space_id = VHOST_VDPA_NET_CVQ_ASID; 548c1a10086SEugenio Pérez 549c1a10086SEugenio Pérez out: 5507a7f87e9SEugenio Pérez if (!s->vhost_vdpa.shadow_vqs_enabled) { 5517a7f87e9SEugenio Pérez return 0; 5522df4dd31SEugenio Pérez } 5532df4dd31SEugenio Pérez 55400ef422eSEugenio Pérez if (s0->vhost_vdpa.iova_tree) { 55500ef422eSEugenio Pérez /* 55600ef422eSEugenio Pérez * SVQ is already configured for all virtqueues. Reuse IOVA tree for 55700ef422eSEugenio Pérez * simplicity, whether CVQ shares ASID with guest or not, because: 55800ef422eSEugenio Pérez * - Memory listener need access to guest's memory addresses allocated 55900ef422eSEugenio Pérez * in the IOVA tree. 56000ef422eSEugenio Pérez * - There should be plenty of IOVA address space for both ASID not to 56100ef422eSEugenio Pérez * worry about collisions between them. Guest's translations are 56200ef422eSEugenio Pérez * still validated with virtio virtqueue_pop so there is no risk for 56300ef422eSEugenio Pérez * the guest to access memory that it shouldn't. 56400ef422eSEugenio Pérez * 56500ef422eSEugenio Pérez * To allocate a iova tree per ASID is doable but it complicates the 56600ef422eSEugenio Pérez * code and it is not worth it for the moment. 56700ef422eSEugenio Pérez */ 56800ef422eSEugenio Pérez v->iova_tree = s0->vhost_vdpa.iova_tree; 56900ef422eSEugenio Pérez } else { 57000ef422eSEugenio Pérez v->iova_tree = vhost_iova_tree_new(v->iova_range.first, 57100ef422eSEugenio Pérez v->iova_range.last); 57200ef422eSEugenio Pérez } 57300ef422eSEugenio Pérez 5747a7f87e9SEugenio Pérez r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer, 5757a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), false); 5767a7f87e9SEugenio Pérez if (unlikely(r < 0)) { 5777a7f87e9SEugenio Pérez return r; 5787a7f87e9SEugenio Pérez } 5797a7f87e9SEugenio Pérez 58017fb889fSEugenio Pérez r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->status, 5817a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), true); 5827a7f87e9SEugenio Pérez if (unlikely(r < 0)) { 5832df4dd31SEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); 5842df4dd31SEugenio Pérez } 5852df4dd31SEugenio Pérez 5867a7f87e9SEugenio Pérez return r; 5877a7f87e9SEugenio Pérez } 5887a7f87e9SEugenio Pérez 5897a7f87e9SEugenio Pérez static void vhost_vdpa_net_cvq_stop(NetClientState *nc) 5907a7f87e9SEugenio Pérez { 5917a7f87e9SEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 5927a7f87e9SEugenio Pérez 5937a7f87e9SEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 5947a7f87e9SEugenio Pérez 5957a7f87e9SEugenio Pérez if (s->vhost_vdpa.shadow_vqs_enabled) { 5967a7f87e9SEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); 59717fb889fSEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->status); 598c1a10086SEugenio Pérez } 59900ef422eSEugenio Pérez 60000ef422eSEugenio Pérez vhost_vdpa_net_client_stop(nc); 6012df4dd31SEugenio Pérez } 6022df4dd31SEugenio Pérez 603be4278b6SEugenio Pérez static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, 604be4278b6SEugenio Pérez size_t in_len) 605be4278b6SEugenio Pérez { 606be4278b6SEugenio Pérez /* Buffers for the device */ 607be4278b6SEugenio Pérez const struct iovec out = { 608be4278b6SEugenio Pérez .iov_base = s->cvq_cmd_out_buffer, 609be4278b6SEugenio Pérez .iov_len = out_len, 610be4278b6SEugenio Pérez }; 611be4278b6SEugenio Pérez const struct iovec in = { 61217fb889fSEugenio Pérez .iov_base = s->status, 613be4278b6SEugenio Pérez .iov_len = sizeof(virtio_net_ctrl_ack), 614be4278b6SEugenio Pérez }; 615be4278b6SEugenio Pérez VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0); 616be4278b6SEugenio Pérez int r; 617be4278b6SEugenio Pérez 618be4278b6SEugenio Pérez r = vhost_svq_add(svq, &out, 1, &in, 1, NULL); 619be4278b6SEugenio Pérez if (unlikely(r != 0)) { 620be4278b6SEugenio Pérez if (unlikely(r == -ENOSPC)) { 621be4278b6SEugenio Pérez qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n", 622be4278b6SEugenio Pérez __func__); 623be4278b6SEugenio Pérez } 624be4278b6SEugenio Pérez return r; 625be4278b6SEugenio Pérez } 626be4278b6SEugenio Pérez 627be4278b6SEugenio Pérez /* 628be4278b6SEugenio Pérez * We can poll here since we've had BQL from the time we sent the 629be4278b6SEugenio Pérez * descriptor. Also, we need to take the answer before SVQ pulls by itself, 630be4278b6SEugenio Pérez * when BQL is released 631be4278b6SEugenio Pérez */ 632be4278b6SEugenio Pérez return vhost_svq_poll(svq); 633be4278b6SEugenio Pérez } 634be4278b6SEugenio Pérez 635f73c0c43SEugenio Pérez static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class, 636f73c0c43SEugenio Pérez uint8_t cmd, const void *data, 637f73c0c43SEugenio Pérez size_t data_size) 638f73c0c43SEugenio Pérez { 639f73c0c43SEugenio Pérez const struct virtio_net_ctrl_hdr ctrl = { 640f73c0c43SEugenio Pérez .class = class, 641f73c0c43SEugenio Pérez .cmd = cmd, 642f73c0c43SEugenio Pérez }; 643f73c0c43SEugenio Pérez 644f73c0c43SEugenio Pérez assert(data_size < vhost_vdpa_net_cvq_cmd_page_len() - sizeof(ctrl)); 645f73c0c43SEugenio Pérez 646f73c0c43SEugenio Pérez memcpy(s->cvq_cmd_out_buffer, &ctrl, sizeof(ctrl)); 647f73c0c43SEugenio Pérez memcpy(s->cvq_cmd_out_buffer + sizeof(ctrl), data, data_size); 648f73c0c43SEugenio Pérez 649f73c0c43SEugenio Pérez return vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + data_size, 650f73c0c43SEugenio Pérez sizeof(virtio_net_ctrl_ack)); 651f73c0c43SEugenio Pérez } 652f73c0c43SEugenio Pérez 653f73c0c43SEugenio Pérez static int vhost_vdpa_net_load_mac(VhostVDPAState *s, const VirtIONet *n) 654f73c0c43SEugenio Pérez { 655f73c0c43SEugenio Pérez uint64_t features = n->parent_obj.guest_features; 656f73c0c43SEugenio Pérez if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) { 657f73c0c43SEugenio Pérez ssize_t dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MAC, 658f73c0c43SEugenio Pérez VIRTIO_NET_CTRL_MAC_ADDR_SET, 659f73c0c43SEugenio Pérez n->mac, sizeof(n->mac)); 660f73c0c43SEugenio Pérez if (unlikely(dev_written < 0)) { 661f73c0c43SEugenio Pérez return dev_written; 662f73c0c43SEugenio Pérez } 663f73c0c43SEugenio Pérez 664f73c0c43SEugenio Pérez return *s->status != VIRTIO_NET_OK; 665f73c0c43SEugenio Pérez } 666f73c0c43SEugenio Pérez 667f73c0c43SEugenio Pérez return 0; 668f73c0c43SEugenio Pérez } 669f73c0c43SEugenio Pérez 670f64c7cdaSEugenio Pérez static int vhost_vdpa_net_load_mq(VhostVDPAState *s, 671f64c7cdaSEugenio Pérez const VirtIONet *n) 672f64c7cdaSEugenio Pérez { 673f64c7cdaSEugenio Pérez struct virtio_net_ctrl_mq mq; 674f64c7cdaSEugenio Pérez uint64_t features = n->parent_obj.guest_features; 675f64c7cdaSEugenio Pérez ssize_t dev_written; 676f64c7cdaSEugenio Pérez 677f64c7cdaSEugenio Pérez if (!(features & BIT_ULL(VIRTIO_NET_F_MQ))) { 678f64c7cdaSEugenio Pérez return 0; 679f64c7cdaSEugenio Pérez } 680f64c7cdaSEugenio Pérez 681f64c7cdaSEugenio Pérez mq.virtqueue_pairs = cpu_to_le16(n->curr_queue_pairs); 682f64c7cdaSEugenio Pérez dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MQ, 683f64c7cdaSEugenio Pérez VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &mq, 684f64c7cdaSEugenio Pérez sizeof(mq)); 685f64c7cdaSEugenio Pérez if (unlikely(dev_written < 0)) { 686f64c7cdaSEugenio Pérez return dev_written; 687f64c7cdaSEugenio Pérez } 688f64c7cdaSEugenio Pérez 689f64c7cdaSEugenio Pérez return *s->status != VIRTIO_NET_OK; 690f64c7cdaSEugenio Pérez } 691f64c7cdaSEugenio Pérez 692dd036d8dSEugenio Pérez static int vhost_vdpa_net_load(NetClientState *nc) 693dd036d8dSEugenio Pérez { 694dd036d8dSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 695f73c0c43SEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 696dd036d8dSEugenio Pérez const VirtIONet *n; 697f73c0c43SEugenio Pérez int r; 698dd036d8dSEugenio Pérez 699dd036d8dSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 700dd036d8dSEugenio Pérez 701dd036d8dSEugenio Pérez if (!v->shadow_vqs_enabled) { 702dd036d8dSEugenio Pérez return 0; 703dd036d8dSEugenio Pérez } 704dd036d8dSEugenio Pérez 705dd036d8dSEugenio Pérez n = VIRTIO_NET(v->dev->vdev); 706f73c0c43SEugenio Pérez r = vhost_vdpa_net_load_mac(s, n); 707f73c0c43SEugenio Pérez if (unlikely(r < 0)) { 708f73c0c43SEugenio Pérez return r; 709dd036d8dSEugenio Pérez } 710f64c7cdaSEugenio Pérez r = vhost_vdpa_net_load_mq(s, n); 711f64c7cdaSEugenio Pérez if (unlikely(r)) { 712f64c7cdaSEugenio Pérez return r; 713f64c7cdaSEugenio Pérez } 714dd036d8dSEugenio Pérez 715dd036d8dSEugenio Pérez return 0; 716dd036d8dSEugenio Pérez } 717dd036d8dSEugenio Pérez 718f8972b56SEugenio Pérez static NetClientInfo net_vhost_vdpa_cvq_info = { 719f8972b56SEugenio Pérez .type = NET_CLIENT_DRIVER_VHOST_VDPA, 720f8972b56SEugenio Pérez .size = sizeof(VhostVDPAState), 721f8972b56SEugenio Pérez .receive = vhost_vdpa_receive, 7227a7f87e9SEugenio Pérez .start = vhost_vdpa_net_cvq_start, 723dd036d8dSEugenio Pérez .load = vhost_vdpa_net_load, 7247a7f87e9SEugenio Pérez .stop = vhost_vdpa_net_cvq_stop, 725f8972b56SEugenio Pérez .cleanup = vhost_vdpa_cleanup, 726f8972b56SEugenio Pérez .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, 727f8972b56SEugenio Pérez .has_ufo = vhost_vdpa_has_ufo, 728f8972b56SEugenio Pérez .check_peer_type = vhost_vdpa_check_peer_type, 729f8972b56SEugenio Pérez }; 730f8972b56SEugenio Pérez 7312df4dd31SEugenio Pérez /** 7322df4dd31SEugenio Pérez * Validate and copy control virtqueue commands. 7332df4dd31SEugenio Pérez * 7342df4dd31SEugenio Pérez * Following QEMU guidelines, we offer a copy of the buffers to the device to 7352df4dd31SEugenio Pérez * prevent TOCTOU bugs. 736bd907ae4SEugenio Pérez */ 737bd907ae4SEugenio Pérez static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, 738bd907ae4SEugenio Pérez VirtQueueElement *elem, 739bd907ae4SEugenio Pérez void *opaque) 740bd907ae4SEugenio Pérez { 7412df4dd31SEugenio Pérez VhostVDPAState *s = opaque; 742be4278b6SEugenio Pérez size_t in_len; 743bd907ae4SEugenio Pérez virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 7447a7f87e9SEugenio Pérez /* Out buffer sent to both the vdpa device and the device model */ 7457a7f87e9SEugenio Pérez struct iovec out = { 7467a7f87e9SEugenio Pérez .iov_base = s->cvq_cmd_out_buffer, 7477a7f87e9SEugenio Pérez }; 7482df4dd31SEugenio Pérez /* in buffer used for device model */ 7492df4dd31SEugenio Pérez const struct iovec in = { 7502df4dd31SEugenio Pérez .iov_base = &status, 7512df4dd31SEugenio Pérez .iov_len = sizeof(status), 7522df4dd31SEugenio Pérez }; 753be4278b6SEugenio Pérez ssize_t dev_written = -EINVAL; 754bd907ae4SEugenio Pérez 7557a7f87e9SEugenio Pérez out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0, 7567a7f87e9SEugenio Pérez s->cvq_cmd_out_buffer, 7577a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_len()); 7583f9a3eebSEugenio Pérez if (*(uint8_t *)s->cvq_cmd_out_buffer == VIRTIO_NET_CTRL_ANNOUNCE) { 7593f9a3eebSEugenio Pérez /* 7603f9a3eebSEugenio Pérez * Guest announce capability is emulated by qemu, so don't forward to 7613f9a3eebSEugenio Pérez * the device. 7623f9a3eebSEugenio Pérez */ 7633f9a3eebSEugenio Pérez dev_written = sizeof(status); 7643f9a3eebSEugenio Pérez *s->status = VIRTIO_NET_OK; 7653f9a3eebSEugenio Pérez } else { 766be4278b6SEugenio Pérez dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status)); 767be4278b6SEugenio Pérez if (unlikely(dev_written < 0)) { 768bd907ae4SEugenio Pérez goto out; 769bd907ae4SEugenio Pérez } 7703f9a3eebSEugenio Pérez } 771bd907ae4SEugenio Pérez 772bd907ae4SEugenio Pérez if (unlikely(dev_written < sizeof(status))) { 773bd907ae4SEugenio Pérez error_report("Insufficient written data (%zu)", dev_written); 7742df4dd31SEugenio Pérez goto out; 7752df4dd31SEugenio Pérez } 7762df4dd31SEugenio Pérez 77717fb889fSEugenio Pérez if (*s->status != VIRTIO_NET_OK) { 778be4278b6SEugenio Pérez return VIRTIO_NET_ERR; 7792df4dd31SEugenio Pérez } 7802df4dd31SEugenio Pérez 7812df4dd31SEugenio Pérez status = VIRTIO_NET_ERR; 7827a7f87e9SEugenio Pérez virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1); 7832df4dd31SEugenio Pérez if (status != VIRTIO_NET_OK) { 7842df4dd31SEugenio Pérez error_report("Bad CVQ processing in model"); 785bd907ae4SEugenio Pérez } 786bd907ae4SEugenio Pérez 787bd907ae4SEugenio Pérez out: 788bd907ae4SEugenio Pérez in_len = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, 789bd907ae4SEugenio Pérez sizeof(status)); 790bd907ae4SEugenio Pérez if (unlikely(in_len < sizeof(status))) { 791bd907ae4SEugenio Pérez error_report("Bad device CVQ written length"); 792bd907ae4SEugenio Pérez } 793bd907ae4SEugenio Pérez vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status))); 794bd907ae4SEugenio Pérez g_free(elem); 795be4278b6SEugenio Pérez return dev_written < 0 ? dev_written : 0; 796bd907ae4SEugenio Pérez } 797bd907ae4SEugenio Pérez 798bd907ae4SEugenio Pérez static const VhostShadowVirtqueueOps vhost_vdpa_net_svq_ops = { 799bd907ae4SEugenio Pérez .avail_handler = vhost_vdpa_net_handle_ctrl_avail, 800bd907ae4SEugenio Pérez }; 801bd907ae4SEugenio Pérez 802654790b6SJason Wang static NetClientState *net_vhost_vdpa_init(NetClientState *peer, 803654790b6SJason Wang const char *device, 804654790b6SJason Wang const char *name, 80540237840SJason Wang int vdpa_device_fd, 80640237840SJason Wang int queue_pair_index, 80740237840SJason Wang int nvqs, 8081576dbb5SEugenio Pérez bool is_datapath, 8091576dbb5SEugenio Pérez bool svq, 8105c1ebd4cSEugenio Pérez struct vhost_vdpa_iova_range iova_range, 8115c1ebd4cSEugenio Pérez uint64_t features) 8121e0a84eaSCindy Lu { 8131e0a84eaSCindy Lu NetClientState *nc = NULL; 8141e0a84eaSCindy Lu VhostVDPAState *s; 8151e0a84eaSCindy Lu int ret = 0; 8161e0a84eaSCindy Lu assert(name); 81740237840SJason Wang if (is_datapath) { 81840237840SJason Wang nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, 81940237840SJason Wang name); 82040237840SJason Wang } else { 821f8972b56SEugenio Pérez nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer, 82240237840SJason Wang device, name); 82340237840SJason Wang } 82453b85d95SLaurent Vivier qemu_set_info_str(nc, TYPE_VHOST_VDPA); 8251e0a84eaSCindy Lu s = DO_UPCAST(VhostVDPAState, nc, nc); 8267327813dSJason Wang 8271e0a84eaSCindy Lu s->vhost_vdpa.device_fd = vdpa_device_fd; 82840237840SJason Wang s->vhost_vdpa.index = queue_pair_index; 8297f211a28SEugenio Pérez s->always_svq = svq; 83069498430SEugenio Pérez s->migration_state.notify = vdpa_net_migration_state_notifier; 8311576dbb5SEugenio Pérez s->vhost_vdpa.shadow_vqs_enabled = svq; 832a585fad2SEugenio Pérez s->vhost_vdpa.iova_range = iova_range; 8336188d78aSEugenio Pérez s->vhost_vdpa.shadow_data = svq; 8345c1ebd4cSEugenio Pérez if (queue_pair_index == 0) { 8355c1ebd4cSEugenio Pérez vhost_vdpa_net_valid_svq_features(features, 8365c1ebd4cSEugenio Pérez &s->vhost_vdpa.migration_blocker); 8375c1ebd4cSEugenio Pérez } else if (!is_datapath) { 8382df4dd31SEugenio Pérez s->cvq_cmd_out_buffer = qemu_memalign(qemu_real_host_page_size(), 8392df4dd31SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len()); 8402df4dd31SEugenio Pérez memset(s->cvq_cmd_out_buffer, 0, vhost_vdpa_net_cvq_cmd_page_len()); 84117fb889fSEugenio Pérez s->status = qemu_memalign(qemu_real_host_page_size(), 8422df4dd31SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len()); 84317fb889fSEugenio Pérez memset(s->status, 0, vhost_vdpa_net_cvq_cmd_page_len()); 8442df4dd31SEugenio Pérez 845bd907ae4SEugenio Pérez s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops; 846bd907ae4SEugenio Pérez s->vhost_vdpa.shadow_vq_ops_opaque = s; 8479c363cf6SEugenio Pérez 8489c363cf6SEugenio Pérez /* 8499c363cf6SEugenio Pérez * TODO: We cannot migrate devices with CVQ as there is no way to set 8509c363cf6SEugenio Pérez * the device state (MAC, MQ, etc) before starting the datapath. 8519c363cf6SEugenio Pérez * 8529c363cf6SEugenio Pérez * Migration blocker ownership now belongs to s->vhost_vdpa. 8539c363cf6SEugenio Pérez */ 8549c363cf6SEugenio Pérez error_setg(&s->vhost_vdpa.migration_blocker, 8559c363cf6SEugenio Pérez "net vdpa cannot migrate with CVQ feature"); 856bd907ae4SEugenio Pérez } 85740237840SJason Wang ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs); 85874af5eecSJason Wang if (ret) { 85974af5eecSJason Wang qemu_del_net_client(nc); 860654790b6SJason Wang return NULL; 86174af5eecSJason Wang } 862654790b6SJason Wang return nc; 8631e0a84eaSCindy Lu } 8641e0a84eaSCindy Lu 8658170ab3fSEugenio Pérez static int vhost_vdpa_get_features(int fd, uint64_t *features, Error **errp) 8668170ab3fSEugenio Pérez { 8678170ab3fSEugenio Pérez int ret = ioctl(fd, VHOST_GET_FEATURES, features); 8688170ab3fSEugenio Pérez if (unlikely(ret < 0)) { 8698170ab3fSEugenio Pérez error_setg_errno(errp, errno, 8708170ab3fSEugenio Pérez "Fail to query features from vhost-vDPA device"); 8718170ab3fSEugenio Pérez } 8728170ab3fSEugenio Pérez return ret; 8738170ab3fSEugenio Pérez } 8748170ab3fSEugenio Pérez 8758170ab3fSEugenio Pérez static int vhost_vdpa_get_max_queue_pairs(int fd, uint64_t features, 8768170ab3fSEugenio Pérez int *has_cvq, Error **errp) 87740237840SJason Wang { 87840237840SJason Wang unsigned long config_size = offsetof(struct vhost_vdpa_config, buf); 879cd523a41SStefano Garzarella g_autofree struct vhost_vdpa_config *config = NULL; 88040237840SJason Wang __virtio16 *max_queue_pairs; 88140237840SJason Wang int ret; 88240237840SJason Wang 88340237840SJason Wang if (features & (1 << VIRTIO_NET_F_CTRL_VQ)) { 88440237840SJason Wang *has_cvq = 1; 88540237840SJason Wang } else { 88640237840SJason Wang *has_cvq = 0; 88740237840SJason Wang } 88840237840SJason Wang 88940237840SJason Wang if (features & (1 << VIRTIO_NET_F_MQ)) { 89040237840SJason Wang config = g_malloc0(config_size + sizeof(*max_queue_pairs)); 89140237840SJason Wang config->off = offsetof(struct virtio_net_config, max_virtqueue_pairs); 89240237840SJason Wang config->len = sizeof(*max_queue_pairs); 89340237840SJason Wang 89440237840SJason Wang ret = ioctl(fd, VHOST_VDPA_GET_CONFIG, config); 89540237840SJason Wang if (ret) { 89640237840SJason Wang error_setg(errp, "Fail to get config from vhost-vDPA device"); 89740237840SJason Wang return -ret; 89840237840SJason Wang } 89940237840SJason Wang 90040237840SJason Wang max_queue_pairs = (__virtio16 *)&config->buf; 90140237840SJason Wang 90240237840SJason Wang return lduw_le_p(max_queue_pairs); 90340237840SJason Wang } 90440237840SJason Wang 90540237840SJason Wang return 1; 90640237840SJason Wang } 90740237840SJason Wang 9081e0a84eaSCindy Lu int net_init_vhost_vdpa(const Netdev *netdev, const char *name, 9091e0a84eaSCindy Lu NetClientState *peer, Error **errp) 9101e0a84eaSCindy Lu { 9111e0a84eaSCindy Lu const NetdevVhostVDPAOptions *opts; 9128170ab3fSEugenio Pérez uint64_t features; 913654790b6SJason Wang int vdpa_device_fd; 914eb3cb751SEugenio Pérez g_autofree NetClientState **ncs = NULL; 915a585fad2SEugenio Pérez struct vhost_vdpa_iova_range iova_range; 916eb3cb751SEugenio Pérez NetClientState *nc; 917aed5da45SEugenio Pérez int queue_pairs, r, i = 0, has_cvq = 0; 9181e0a84eaSCindy Lu 9191e0a84eaSCindy Lu assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA); 9201e0a84eaSCindy Lu opts = &netdev->u.vhost_vdpa; 9217480874aSMarkus Armbruster if (!opts->vhostdev && !opts->vhostfd) { 9228801ccd0SSi-Wei Liu error_setg(errp, 9238801ccd0SSi-Wei Liu "vhost-vdpa: neither vhostdev= nor vhostfd= was specified"); 924c8295404SEugenio Pérez return -1; 925c8295404SEugenio Pérez } 9267327813dSJason Wang 9277480874aSMarkus Armbruster if (opts->vhostdev && opts->vhostfd) { 9288801ccd0SSi-Wei Liu error_setg(errp, 9298801ccd0SSi-Wei Liu "vhost-vdpa: vhostdev= and vhostfd= are mutually exclusive"); 9308801ccd0SSi-Wei Liu return -1; 9318801ccd0SSi-Wei Liu } 9328801ccd0SSi-Wei Liu 9337480874aSMarkus Armbruster if (opts->vhostdev) { 9340351152bSEugenio Pérez vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR, errp); 9357327813dSJason Wang if (vdpa_device_fd == -1) { 9367327813dSJason Wang return -errno; 9377327813dSJason Wang } 9385107fd3eSPeter Maydell } else { 9395107fd3eSPeter Maydell /* has_vhostfd */ 9408801ccd0SSi-Wei Liu vdpa_device_fd = monitor_fd_param(monitor_cur(), opts->vhostfd, errp); 9418801ccd0SSi-Wei Liu if (vdpa_device_fd == -1) { 9428801ccd0SSi-Wei Liu error_prepend(errp, "vhost-vdpa: unable to parse vhostfd: "); 9438801ccd0SSi-Wei Liu return -1; 9448801ccd0SSi-Wei Liu } 9458801ccd0SSi-Wei Liu } 9467327813dSJason Wang 9478170ab3fSEugenio Pérez r = vhost_vdpa_get_features(vdpa_device_fd, &features, errp); 9488170ab3fSEugenio Pérez if (unlikely(r < 0)) { 949aed5da45SEugenio Pérez goto err; 9508170ab3fSEugenio Pérez } 9518170ab3fSEugenio Pérez 9528170ab3fSEugenio Pérez queue_pairs = vhost_vdpa_get_max_queue_pairs(vdpa_device_fd, features, 95340237840SJason Wang &has_cvq, errp); 95440237840SJason Wang if (queue_pairs < 0) { 9557327813dSJason Wang qemu_close(vdpa_device_fd); 95640237840SJason Wang return queue_pairs; 9577327813dSJason Wang } 9587327813dSJason Wang 959bf7a2ad8SLongpeng r = vhost_vdpa_get_iova_range(vdpa_device_fd, &iova_range); 960bf7a2ad8SLongpeng if (unlikely(r < 0)) { 961bf7a2ad8SLongpeng error_setg(errp, "vhost-vdpa: get iova range failed: %s", 962bf7a2ad8SLongpeng strerror(-r)); 963bf7a2ad8SLongpeng goto err; 964bf7a2ad8SLongpeng } 965bf7a2ad8SLongpeng 96600ef422eSEugenio Pérez if (opts->x_svq && !vhost_vdpa_net_valid_svq_features(features, errp)) { 96700ef422eSEugenio Pérez goto err; 9681576dbb5SEugenio Pérez } 9691576dbb5SEugenio Pérez 97040237840SJason Wang ncs = g_malloc0(sizeof(*ncs) * queue_pairs); 97140237840SJason Wang 97240237840SJason Wang for (i = 0; i < queue_pairs; i++) { 97340237840SJason Wang ncs[i] = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, 9741576dbb5SEugenio Pérez vdpa_device_fd, i, 2, true, opts->x_svq, 9755c1ebd4cSEugenio Pérez iova_range, features); 97640237840SJason Wang if (!ncs[i]) 97740237840SJason Wang goto err; 97840237840SJason Wang } 97940237840SJason Wang 98040237840SJason Wang if (has_cvq) { 98140237840SJason Wang nc = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, 9821576dbb5SEugenio Pérez vdpa_device_fd, i, 1, false, 9835c1ebd4cSEugenio Pérez opts->x_svq, iova_range, features); 98440237840SJason Wang if (!nc) 98540237840SJason Wang goto err; 98640237840SJason Wang } 98740237840SJason Wang 988654790b6SJason Wang return 0; 98940237840SJason Wang 99040237840SJason Wang err: 99140237840SJason Wang if (i) { 9929bd05507SSi-Wei Liu for (i--; i >= 0; i--) { 9939bd05507SSi-Wei Liu qemu_del_net_client(ncs[i]); 9949bd05507SSi-Wei Liu } 99540237840SJason Wang } 9961576dbb5SEugenio Pérez 99740237840SJason Wang qemu_close(vdpa_device_fd); 99840237840SJason Wang 99940237840SJason Wang return -1; 10001e0a84eaSCindy Lu } 1001