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) | 104*609ab4c3SEugenio Pérez /* VHOST_F_LOG_ALL is exposed by SVQ */ 105*609ab4c3SEugenio Pérez BIT_ULL(VHOST_F_LOG_ALL) | 1061576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_RSC_EXT) | 1071576dbb5SEugenio Pérez BIT_ULL(VIRTIO_NET_F_STANDBY); 1081576dbb5SEugenio Pérez 109c1a10086SEugenio Pérez #define VHOST_VDPA_NET_CVQ_ASID 1 110c1a10086SEugenio Pérez 1111e0a84eaSCindy Lu VHostNetState *vhost_vdpa_get_vhost_net(NetClientState *nc) 1121e0a84eaSCindy Lu { 1131e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 1141e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 1151e0a84eaSCindy Lu return s->vhost_net; 1161e0a84eaSCindy Lu } 1171e0a84eaSCindy Lu 11836e46472SEugenio Pérez static bool vhost_vdpa_net_valid_svq_features(uint64_t features, Error **errp) 11936e46472SEugenio Pérez { 12036e46472SEugenio Pérez uint64_t invalid_dev_features = 12136e46472SEugenio Pérez features & ~vdpa_svq_device_features & 12236e46472SEugenio Pérez /* Transport are all accepted at this point */ 12336e46472SEugenio Pérez ~MAKE_64BIT_MASK(VIRTIO_TRANSPORT_F_START, 12436e46472SEugenio Pérez VIRTIO_TRANSPORT_F_END - VIRTIO_TRANSPORT_F_START); 12536e46472SEugenio Pérez 12636e46472SEugenio Pérez if (invalid_dev_features) { 12736e46472SEugenio Pérez error_setg(errp, "vdpa svq does not work with features 0x%" PRIx64, 12836e46472SEugenio Pérez invalid_dev_features); 129258a0394SEugenio Pérez return false; 13036e46472SEugenio Pérez } 13136e46472SEugenio Pérez 132258a0394SEugenio Pérez return vhost_svq_valid_features(features, errp); 13336e46472SEugenio Pérez } 13436e46472SEugenio Pérez 1351e0a84eaSCindy Lu static int vhost_vdpa_net_check_device_id(struct vhost_net *net) 1361e0a84eaSCindy Lu { 1371e0a84eaSCindy Lu uint32_t device_id; 1381e0a84eaSCindy Lu int ret; 1391e0a84eaSCindy Lu struct vhost_dev *hdev; 1401e0a84eaSCindy Lu 1411e0a84eaSCindy Lu hdev = (struct vhost_dev *)&net->dev; 1421e0a84eaSCindy Lu ret = hdev->vhost_ops->vhost_get_device_id(hdev, &device_id); 1431e0a84eaSCindy Lu if (device_id != VIRTIO_ID_NET) { 1441e0a84eaSCindy Lu return -ENOTSUP; 1451e0a84eaSCindy Lu } 1461e0a84eaSCindy Lu return ret; 1471e0a84eaSCindy Lu } 1481e0a84eaSCindy Lu 14940237840SJason Wang static int vhost_vdpa_add(NetClientState *ncs, void *be, 15040237840SJason Wang int queue_pair_index, int nvqs) 1511e0a84eaSCindy Lu { 1521e0a84eaSCindy Lu VhostNetOptions options; 1531e0a84eaSCindy Lu struct vhost_net *net = NULL; 1541e0a84eaSCindy Lu VhostVDPAState *s; 1551e0a84eaSCindy Lu int ret; 1561e0a84eaSCindy Lu 1571e0a84eaSCindy Lu options.backend_type = VHOST_BACKEND_TYPE_VDPA; 1581e0a84eaSCindy Lu assert(ncs->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 1591e0a84eaSCindy Lu s = DO_UPCAST(VhostVDPAState, nc, ncs); 1601e0a84eaSCindy Lu options.net_backend = ncs; 1611e0a84eaSCindy Lu options.opaque = be; 1621e0a84eaSCindy Lu options.busyloop_timeout = 0; 16340237840SJason Wang options.nvqs = nvqs; 1641e0a84eaSCindy Lu 1651e0a84eaSCindy Lu net = vhost_net_init(&options); 1661e0a84eaSCindy Lu if (!net) { 1671e0a84eaSCindy Lu error_report("failed to init vhost_net for queue"); 168a97ef87aSJason Wang goto err_init; 1691e0a84eaSCindy Lu } 1701e0a84eaSCindy Lu s->vhost_net = net; 1711e0a84eaSCindy Lu ret = vhost_vdpa_net_check_device_id(net); 1721e0a84eaSCindy Lu if (ret) { 173a97ef87aSJason Wang goto err_check; 1741e0a84eaSCindy Lu } 1751e0a84eaSCindy Lu return 0; 176a97ef87aSJason Wang err_check: 1771e0a84eaSCindy Lu vhost_net_cleanup(net); 178ab36edcfSJason Wang g_free(net); 179a97ef87aSJason Wang err_init: 1801e0a84eaSCindy Lu return -1; 1811e0a84eaSCindy Lu } 1821e0a84eaSCindy Lu 1831e0a84eaSCindy Lu static void vhost_vdpa_cleanup(NetClientState *nc) 1841e0a84eaSCindy Lu { 1851e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 1861e0a84eaSCindy Lu 1872df4dd31SEugenio Pérez qemu_vfree(s->cvq_cmd_out_buffer); 18817fb889fSEugenio Pérez qemu_vfree(s->status); 1891e0a84eaSCindy Lu if (s->vhost_net) { 1901e0a84eaSCindy Lu vhost_net_cleanup(s->vhost_net); 1911e0a84eaSCindy Lu g_free(s->vhost_net); 1921e0a84eaSCindy Lu s->vhost_net = NULL; 1931e0a84eaSCindy Lu } 19457b3a7d8SCindy Lu if (s->vhost_vdpa.device_fd >= 0) { 19557b3a7d8SCindy Lu qemu_close(s->vhost_vdpa.device_fd); 19657b3a7d8SCindy Lu s->vhost_vdpa.device_fd = -1; 19757b3a7d8SCindy Lu } 1981e0a84eaSCindy Lu } 1991e0a84eaSCindy Lu 2001e0a84eaSCindy Lu static bool vhost_vdpa_has_vnet_hdr(NetClientState *nc) 2011e0a84eaSCindy Lu { 2021e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 2031e0a84eaSCindy Lu 2041e0a84eaSCindy Lu return true; 2051e0a84eaSCindy Lu } 2061e0a84eaSCindy Lu 2071e0a84eaSCindy Lu static bool vhost_vdpa_has_ufo(NetClientState *nc) 2081e0a84eaSCindy Lu { 2091e0a84eaSCindy Lu assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 2101e0a84eaSCindy Lu VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 2111e0a84eaSCindy Lu uint64_t features = 0; 2121e0a84eaSCindy Lu features |= (1ULL << VIRTIO_NET_F_HOST_UFO); 2131e0a84eaSCindy Lu features = vhost_net_get_features(s->vhost_net, features); 2141e0a84eaSCindy Lu return !!(features & (1ULL << VIRTIO_NET_F_HOST_UFO)); 2151e0a84eaSCindy Lu 2161e0a84eaSCindy Lu } 2171e0a84eaSCindy Lu 218ee8a1c63SKevin Wolf static bool vhost_vdpa_check_peer_type(NetClientState *nc, ObjectClass *oc, 219ee8a1c63SKevin Wolf Error **errp) 220ee8a1c63SKevin Wolf { 221ee8a1c63SKevin Wolf const char *driver = object_class_get_name(oc); 222ee8a1c63SKevin Wolf 223ee8a1c63SKevin Wolf if (!g_str_has_prefix(driver, "virtio-net-")) { 224ee8a1c63SKevin Wolf error_setg(errp, "vhost-vdpa requires frontend driver virtio-net-*"); 225ee8a1c63SKevin Wolf return false; 226ee8a1c63SKevin Wolf } 227ee8a1c63SKevin Wolf 228ee8a1c63SKevin Wolf return true; 229ee8a1c63SKevin Wolf } 230ee8a1c63SKevin Wolf 231846a1e85SEugenio Pérez /** Dummy receive in case qemu falls back to userland tap networking */ 232846a1e85SEugenio Pérez static ssize_t vhost_vdpa_receive(NetClientState *nc, const uint8_t *buf, 233846a1e85SEugenio Pérez size_t size) 234846a1e85SEugenio Pérez { 235bc5add1dSSi-Wei Liu return size; 236846a1e85SEugenio Pérez } 237846a1e85SEugenio Pérez 23800ef422eSEugenio Pérez /** From any vdpa net client, get the netclient of the first queue pair */ 23900ef422eSEugenio Pérez static VhostVDPAState *vhost_vdpa_net_first_nc_vdpa(VhostVDPAState *s) 24000ef422eSEugenio Pérez { 24100ef422eSEugenio Pérez NICState *nic = qemu_get_nic(s->nc.peer); 24200ef422eSEugenio Pérez NetClientState *nc0 = qemu_get_peer(nic->ncs, 0); 24300ef422eSEugenio Pérez 24400ef422eSEugenio Pérez return DO_UPCAST(VhostVDPAState, nc, nc0); 24500ef422eSEugenio Pérez } 24600ef422eSEugenio Pérez 24769498430SEugenio Pérez static void vhost_vdpa_net_log_global_enable(VhostVDPAState *s, bool enable) 24869498430SEugenio Pérez { 24969498430SEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 25069498430SEugenio Pérez VirtIONet *n; 25169498430SEugenio Pérez VirtIODevice *vdev; 25269498430SEugenio Pérez int data_queue_pairs, cvq, r; 25369498430SEugenio Pérez 25469498430SEugenio Pérez /* We are only called on the first data vqs and only if x-svq is not set */ 25569498430SEugenio Pérez if (s->vhost_vdpa.shadow_vqs_enabled == enable) { 25669498430SEugenio Pérez return; 25769498430SEugenio Pérez } 25869498430SEugenio Pérez 25969498430SEugenio Pérez vdev = v->dev->vdev; 26069498430SEugenio Pérez n = VIRTIO_NET(vdev); 26169498430SEugenio Pérez if (!n->vhost_started) { 26269498430SEugenio Pérez return; 26369498430SEugenio Pérez } 26469498430SEugenio Pérez 26569498430SEugenio Pérez data_queue_pairs = n->multiqueue ? n->max_queue_pairs : 1; 26669498430SEugenio Pérez cvq = virtio_vdev_has_feature(vdev, VIRTIO_NET_F_CTRL_VQ) ? 26769498430SEugenio Pérez n->max_ncs - n->max_queue_pairs : 0; 26869498430SEugenio Pérez /* 26969498430SEugenio Pérez * TODO: vhost_net_stop does suspend, get_base and reset. We can be smarter 27069498430SEugenio Pérez * in the future and resume the device if read-only operations between 27169498430SEugenio Pérez * suspend and reset goes wrong. 27269498430SEugenio Pérez */ 27369498430SEugenio Pérez vhost_net_stop(vdev, n->nic->ncs, data_queue_pairs, cvq); 27469498430SEugenio Pérez 27569498430SEugenio Pérez /* Start will check migration setup_or_active to configure or not SVQ */ 27669498430SEugenio Pérez r = vhost_net_start(vdev, n->nic->ncs, data_queue_pairs, cvq); 27769498430SEugenio Pérez if (unlikely(r < 0)) { 27869498430SEugenio Pérez error_report("unable to start vhost net: %s(%d)", g_strerror(-r), -r); 27969498430SEugenio Pérez } 28069498430SEugenio Pérez } 28169498430SEugenio Pérez 28269498430SEugenio Pérez static void vdpa_net_migration_state_notifier(Notifier *notifier, void *data) 28369498430SEugenio Pérez { 28469498430SEugenio Pérez MigrationState *migration = data; 28569498430SEugenio Pérez VhostVDPAState *s = container_of(notifier, VhostVDPAState, 28669498430SEugenio Pérez migration_state); 28769498430SEugenio Pérez 28869498430SEugenio Pérez if (migration_in_setup(migration)) { 28969498430SEugenio Pérez vhost_vdpa_net_log_global_enable(s, true); 29069498430SEugenio Pérez } else if (migration_has_failed(migration)) { 29169498430SEugenio Pérez vhost_vdpa_net_log_global_enable(s, false); 29269498430SEugenio Pérez } 29369498430SEugenio Pérez } 29469498430SEugenio Pérez 29500ef422eSEugenio Pérez static void vhost_vdpa_net_data_start_first(VhostVDPAState *s) 29600ef422eSEugenio Pérez { 29700ef422eSEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 29800ef422eSEugenio Pérez 29969498430SEugenio Pérez add_migration_state_change_notifier(&s->migration_state); 30000ef422eSEugenio Pérez if (v->shadow_vqs_enabled) { 30100ef422eSEugenio Pérez v->iova_tree = vhost_iova_tree_new(v->iova_range.first, 30200ef422eSEugenio Pérez v->iova_range.last); 30300ef422eSEugenio Pérez } 30400ef422eSEugenio Pérez } 30500ef422eSEugenio Pérez 30600ef422eSEugenio Pérez static int vhost_vdpa_net_data_start(NetClientState *nc) 30700ef422eSEugenio Pérez { 30800ef422eSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 30900ef422eSEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 31000ef422eSEugenio Pérez 31100ef422eSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 31200ef422eSEugenio Pérez 31369498430SEugenio Pérez if (s->always_svq || 31469498430SEugenio Pérez migration_is_setup_or_active(migrate_get_current()->state)) { 31569498430SEugenio Pérez v->shadow_vqs_enabled = true; 31669498430SEugenio Pérez v->shadow_data = true; 31769498430SEugenio Pérez } else { 31869498430SEugenio Pérez v->shadow_vqs_enabled = false; 31969498430SEugenio Pérez v->shadow_data = false; 32069498430SEugenio Pérez } 32169498430SEugenio Pérez 32200ef422eSEugenio Pérez if (v->index == 0) { 32300ef422eSEugenio Pérez vhost_vdpa_net_data_start_first(s); 32400ef422eSEugenio Pérez return 0; 32500ef422eSEugenio Pérez } 32600ef422eSEugenio Pérez 32700ef422eSEugenio Pérez if (v->shadow_vqs_enabled) { 32800ef422eSEugenio Pérez VhostVDPAState *s0 = vhost_vdpa_net_first_nc_vdpa(s); 32900ef422eSEugenio Pérez v->iova_tree = s0->vhost_vdpa.iova_tree; 33000ef422eSEugenio Pérez } 33100ef422eSEugenio Pérez 33200ef422eSEugenio Pérez return 0; 33300ef422eSEugenio Pérez } 33400ef422eSEugenio Pérez 33500ef422eSEugenio Pérez static void vhost_vdpa_net_client_stop(NetClientState *nc) 33600ef422eSEugenio Pérez { 33700ef422eSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 33800ef422eSEugenio Pérez struct vhost_dev *dev; 33900ef422eSEugenio Pérez 34000ef422eSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 34100ef422eSEugenio Pérez 34269498430SEugenio Pérez if (s->vhost_vdpa.index == 0) { 34369498430SEugenio Pérez remove_migration_state_change_notifier(&s->migration_state); 34469498430SEugenio Pérez } 34569498430SEugenio Pérez 34600ef422eSEugenio Pérez dev = s->vhost_vdpa.dev; 34700ef422eSEugenio Pérez if (dev->vq_index + dev->nvqs == dev->vq_index_end) { 34800ef422eSEugenio Pérez g_clear_pointer(&s->vhost_vdpa.iova_tree, vhost_iova_tree_delete); 34900ef422eSEugenio Pérez } 35000ef422eSEugenio Pérez } 35100ef422eSEugenio Pérez 3521e0a84eaSCindy Lu static NetClientInfo net_vhost_vdpa_info = { 3531e0a84eaSCindy Lu .type = NET_CLIENT_DRIVER_VHOST_VDPA, 3541e0a84eaSCindy Lu .size = sizeof(VhostVDPAState), 355846a1e85SEugenio Pérez .receive = vhost_vdpa_receive, 35600ef422eSEugenio Pérez .start = vhost_vdpa_net_data_start, 35700ef422eSEugenio Pérez .stop = vhost_vdpa_net_client_stop, 3581e0a84eaSCindy Lu .cleanup = vhost_vdpa_cleanup, 3591e0a84eaSCindy Lu .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, 3601e0a84eaSCindy Lu .has_ufo = vhost_vdpa_has_ufo, 361ee8a1c63SKevin Wolf .check_peer_type = vhost_vdpa_check_peer_type, 3621e0a84eaSCindy Lu }; 3631e0a84eaSCindy Lu 364c1a10086SEugenio Pérez static int64_t vhost_vdpa_get_vring_group(int device_fd, unsigned vq_index) 365c1a10086SEugenio Pérez { 366c1a10086SEugenio Pérez struct vhost_vring_state state = { 367c1a10086SEugenio Pérez .index = vq_index, 368c1a10086SEugenio Pérez }; 369c1a10086SEugenio Pérez int r = ioctl(device_fd, VHOST_VDPA_GET_VRING_GROUP, &state); 370c1a10086SEugenio Pérez 371c1a10086SEugenio Pérez if (unlikely(r < 0)) { 372c1a10086SEugenio Pérez error_report("Cannot get VQ %u group: %s", vq_index, 373c1a10086SEugenio Pérez g_strerror(errno)); 374c1a10086SEugenio Pérez return r; 375c1a10086SEugenio Pérez } 376c1a10086SEugenio Pérez 377c1a10086SEugenio Pérez return state.num; 378c1a10086SEugenio Pérez } 379c1a10086SEugenio Pérez 380c1a10086SEugenio Pérez static int vhost_vdpa_set_address_space_id(struct vhost_vdpa *v, 381c1a10086SEugenio Pérez unsigned vq_group, 382c1a10086SEugenio Pérez unsigned asid_num) 383c1a10086SEugenio Pérez { 384c1a10086SEugenio Pérez struct vhost_vring_state asid = { 385c1a10086SEugenio Pérez .index = vq_group, 386c1a10086SEugenio Pérez .num = asid_num, 387c1a10086SEugenio Pérez }; 388c1a10086SEugenio Pérez int r; 389c1a10086SEugenio Pérez 390c1a10086SEugenio Pérez r = ioctl(v->device_fd, VHOST_VDPA_SET_GROUP_ASID, &asid); 391c1a10086SEugenio Pérez if (unlikely(r < 0)) { 392c1a10086SEugenio Pérez error_report("Can't set vq group %u asid %u, errno=%d (%s)", 393c1a10086SEugenio Pérez asid.index, asid.num, errno, g_strerror(errno)); 394c1a10086SEugenio Pérez } 395c1a10086SEugenio Pérez return r; 396c1a10086SEugenio Pérez } 397c1a10086SEugenio Pérez 3982df4dd31SEugenio Pérez static void vhost_vdpa_cvq_unmap_buf(struct vhost_vdpa *v, void *addr) 3992df4dd31SEugenio Pérez { 4002df4dd31SEugenio Pérez VhostIOVATree *tree = v->iova_tree; 4012df4dd31SEugenio Pérez DMAMap needle = { 4022df4dd31SEugenio Pérez /* 4032df4dd31SEugenio Pérez * No need to specify size or to look for more translations since 4042df4dd31SEugenio Pérez * this contiguous chunk was allocated by us. 4052df4dd31SEugenio Pérez */ 4062df4dd31SEugenio Pérez .translated_addr = (hwaddr)(uintptr_t)addr, 4072df4dd31SEugenio Pérez }; 4082df4dd31SEugenio Pérez const DMAMap *map = vhost_iova_tree_find_iova(tree, &needle); 4092df4dd31SEugenio Pérez int r; 4102df4dd31SEugenio Pérez 4112df4dd31SEugenio Pérez if (unlikely(!map)) { 4122df4dd31SEugenio Pérez error_report("Cannot locate expected map"); 4132df4dd31SEugenio Pérez return; 4142df4dd31SEugenio Pérez } 4152df4dd31SEugenio Pérez 416cd831ed5SEugenio Pérez r = vhost_vdpa_dma_unmap(v, v->address_space_id, map->iova, map->size + 1); 4172df4dd31SEugenio Pérez if (unlikely(r != 0)) { 4182df4dd31SEugenio Pérez error_report("Device cannot unmap: %s(%d)", g_strerror(r), r); 4192df4dd31SEugenio Pérez } 4202df4dd31SEugenio Pérez 42169292a8eSEugenio Pérez vhost_iova_tree_remove(tree, *map); 4222df4dd31SEugenio Pérez } 4232df4dd31SEugenio Pérez 4242df4dd31SEugenio Pérez static size_t vhost_vdpa_net_cvq_cmd_len(void) 4252df4dd31SEugenio Pérez { 4262df4dd31SEugenio Pérez /* 4272df4dd31SEugenio Pérez * MAC_TABLE_SET is the ctrl command that produces the longer out buffer. 4282df4dd31SEugenio Pérez * In buffer is always 1 byte, so it should fit here 4292df4dd31SEugenio Pérez */ 4302df4dd31SEugenio Pérez return sizeof(struct virtio_net_ctrl_hdr) + 4312df4dd31SEugenio Pérez 2 * sizeof(struct virtio_net_ctrl_mac) + 4322df4dd31SEugenio Pérez MAC_TABLE_ENTRIES * ETH_ALEN; 4332df4dd31SEugenio Pérez } 4342df4dd31SEugenio Pérez 4352df4dd31SEugenio Pérez static size_t vhost_vdpa_net_cvq_cmd_page_len(void) 4362df4dd31SEugenio Pérez { 4372df4dd31SEugenio Pérez return ROUND_UP(vhost_vdpa_net_cvq_cmd_len(), qemu_real_host_page_size()); 4382df4dd31SEugenio Pérez } 4392df4dd31SEugenio Pérez 4407a7f87e9SEugenio Pérez /** Map CVQ buffer. */ 4417a7f87e9SEugenio Pérez static int vhost_vdpa_cvq_map_buf(struct vhost_vdpa *v, void *buf, size_t size, 4427a7f87e9SEugenio Pérez bool write) 4432df4dd31SEugenio Pérez { 4442df4dd31SEugenio Pérez DMAMap map = {}; 4452df4dd31SEugenio Pérez int r; 4462df4dd31SEugenio Pérez 4472df4dd31SEugenio Pérez map.translated_addr = (hwaddr)(uintptr_t)buf; 4487a7f87e9SEugenio Pérez map.size = size - 1; 4492df4dd31SEugenio Pérez map.perm = write ? IOMMU_RW : IOMMU_RO, 4502df4dd31SEugenio Pérez r = vhost_iova_tree_map_alloc(v->iova_tree, &map); 4512df4dd31SEugenio Pérez if (unlikely(r != IOVA_OK)) { 4522df4dd31SEugenio Pérez error_report("Cannot map injected element"); 4537a7f87e9SEugenio Pérez return r; 4542df4dd31SEugenio Pérez } 4552df4dd31SEugenio Pérez 456cd831ed5SEugenio Pérez r = vhost_vdpa_dma_map(v, v->address_space_id, map.iova, 457cd831ed5SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), buf, !write); 4582df4dd31SEugenio Pérez if (unlikely(r < 0)) { 4592df4dd31SEugenio Pérez goto dma_map_err; 4602df4dd31SEugenio Pérez } 4612df4dd31SEugenio Pérez 4627a7f87e9SEugenio Pérez return 0; 4632df4dd31SEugenio Pérez 4642df4dd31SEugenio Pérez dma_map_err: 46569292a8eSEugenio Pérez vhost_iova_tree_remove(v->iova_tree, map); 4667a7f87e9SEugenio Pérez return r; 4672df4dd31SEugenio Pérez } 4682df4dd31SEugenio Pérez 4697a7f87e9SEugenio Pérez static int vhost_vdpa_net_cvq_start(NetClientState *nc) 4702df4dd31SEugenio Pérez { 47100ef422eSEugenio Pérez VhostVDPAState *s, *s0; 472c1a10086SEugenio Pérez struct vhost_vdpa *v; 473c1a10086SEugenio Pérez uint64_t backend_features; 474c1a10086SEugenio Pérez int64_t cvq_group; 475c1a10086SEugenio Pérez int cvq_index, r; 4762df4dd31SEugenio Pérez 4777a7f87e9SEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 4787a7f87e9SEugenio Pérez 4797a7f87e9SEugenio Pérez s = DO_UPCAST(VhostVDPAState, nc, nc); 480c1a10086SEugenio Pérez v = &s->vhost_vdpa; 481c1a10086SEugenio Pérez 48269498430SEugenio Pérez s0 = vhost_vdpa_net_first_nc_vdpa(s); 48369498430SEugenio Pérez v->shadow_data = s0->vhost_vdpa.shadow_vqs_enabled; 484c1a10086SEugenio Pérez v->shadow_vqs_enabled = s->always_svq; 485c1a10086SEugenio Pérez s->vhost_vdpa.address_space_id = VHOST_VDPA_GUEST_PA_ASID; 486c1a10086SEugenio Pérez 48769498430SEugenio Pérez if (s->vhost_vdpa.shadow_data) { 488c1a10086SEugenio Pérez /* SVQ is already configured for all virtqueues */ 489c1a10086SEugenio Pérez goto out; 490c1a10086SEugenio Pérez } 491c1a10086SEugenio Pérez 492c1a10086SEugenio Pérez /* 493c1a10086SEugenio Pérez * If we early return in these cases SVQ will not be enabled. The migration 494c1a10086SEugenio Pérez * will be blocked as long as vhost-vdpa backends will not offer _F_LOG. 495c1a10086SEugenio Pérez * 496c1a10086SEugenio Pérez * Calling VHOST_GET_BACKEND_FEATURES as they are not available in v->dev 497c1a10086SEugenio Pérez * yet. 498c1a10086SEugenio Pérez */ 499c1a10086SEugenio Pérez r = ioctl(v->device_fd, VHOST_GET_BACKEND_FEATURES, &backend_features); 500c1a10086SEugenio Pérez if (unlikely(r < 0)) { 501c1a10086SEugenio Pérez error_report("Cannot get vdpa backend_features: %s(%d)", 502c1a10086SEugenio Pérez g_strerror(errno), errno); 503c1a10086SEugenio Pérez return -1; 504c1a10086SEugenio Pérez } 505525ae115SEugenio Pérez if (!(backend_features & BIT_ULL(VHOST_BACKEND_F_IOTLB_ASID)) || 506c1a10086SEugenio Pérez !vhost_vdpa_net_valid_svq_features(v->dev->features, NULL)) { 507c1a10086SEugenio Pérez return 0; 508c1a10086SEugenio Pérez } 509c1a10086SEugenio Pérez 510c1a10086SEugenio Pérez /* 511c1a10086SEugenio Pérez * Check if all the virtqueues of the virtio device are in a different vq 512c1a10086SEugenio Pérez * than the last vq. VQ group of last group passed in cvq_group. 513c1a10086SEugenio Pérez */ 514c1a10086SEugenio Pérez cvq_index = v->dev->vq_index_end - 1; 515c1a10086SEugenio Pérez cvq_group = vhost_vdpa_get_vring_group(v->device_fd, cvq_index); 516c1a10086SEugenio Pérez if (unlikely(cvq_group < 0)) { 517c1a10086SEugenio Pérez return cvq_group; 518c1a10086SEugenio Pérez } 519c1a10086SEugenio Pérez for (int i = 0; i < cvq_index; ++i) { 520c1a10086SEugenio Pérez int64_t group = vhost_vdpa_get_vring_group(v->device_fd, i); 521c1a10086SEugenio Pérez 522c1a10086SEugenio Pérez if (unlikely(group < 0)) { 523c1a10086SEugenio Pérez return group; 524c1a10086SEugenio Pérez } 525c1a10086SEugenio Pérez 526c1a10086SEugenio Pérez if (group == cvq_group) { 527c1a10086SEugenio Pérez return 0; 528c1a10086SEugenio Pérez } 529c1a10086SEugenio Pérez } 530c1a10086SEugenio Pérez 531c1a10086SEugenio Pérez r = vhost_vdpa_set_address_space_id(v, cvq_group, VHOST_VDPA_NET_CVQ_ASID); 532c1a10086SEugenio Pérez if (unlikely(r < 0)) { 533c1a10086SEugenio Pérez return r; 534c1a10086SEugenio Pérez } 535c1a10086SEugenio Pérez 536c1a10086SEugenio Pérez v->shadow_vqs_enabled = true; 537c1a10086SEugenio Pérez s->vhost_vdpa.address_space_id = VHOST_VDPA_NET_CVQ_ASID; 538c1a10086SEugenio Pérez 539c1a10086SEugenio Pérez out: 5407a7f87e9SEugenio Pérez if (!s->vhost_vdpa.shadow_vqs_enabled) { 5417a7f87e9SEugenio Pérez return 0; 5422df4dd31SEugenio Pérez } 5432df4dd31SEugenio Pérez 54400ef422eSEugenio Pérez if (s0->vhost_vdpa.iova_tree) { 54500ef422eSEugenio Pérez /* 54600ef422eSEugenio Pérez * SVQ is already configured for all virtqueues. Reuse IOVA tree for 54700ef422eSEugenio Pérez * simplicity, whether CVQ shares ASID with guest or not, because: 54800ef422eSEugenio Pérez * - Memory listener need access to guest's memory addresses allocated 54900ef422eSEugenio Pérez * in the IOVA tree. 55000ef422eSEugenio Pérez * - There should be plenty of IOVA address space for both ASID not to 55100ef422eSEugenio Pérez * worry about collisions between them. Guest's translations are 55200ef422eSEugenio Pérez * still validated with virtio virtqueue_pop so there is no risk for 55300ef422eSEugenio Pérez * the guest to access memory that it shouldn't. 55400ef422eSEugenio Pérez * 55500ef422eSEugenio Pérez * To allocate a iova tree per ASID is doable but it complicates the 55600ef422eSEugenio Pérez * code and it is not worth it for the moment. 55700ef422eSEugenio Pérez */ 55800ef422eSEugenio Pérez v->iova_tree = s0->vhost_vdpa.iova_tree; 55900ef422eSEugenio Pérez } else { 56000ef422eSEugenio Pérez v->iova_tree = vhost_iova_tree_new(v->iova_range.first, 56100ef422eSEugenio Pérez v->iova_range.last); 56200ef422eSEugenio Pérez } 56300ef422eSEugenio Pérez 5647a7f87e9SEugenio Pérez r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer, 5657a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), false); 5667a7f87e9SEugenio Pérez if (unlikely(r < 0)) { 5677a7f87e9SEugenio Pérez return r; 5687a7f87e9SEugenio Pérez } 5697a7f87e9SEugenio Pérez 57017fb889fSEugenio Pérez r = vhost_vdpa_cvq_map_buf(&s->vhost_vdpa, s->status, 5717a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len(), true); 5727a7f87e9SEugenio Pérez if (unlikely(r < 0)) { 5732df4dd31SEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); 5742df4dd31SEugenio Pérez } 5752df4dd31SEugenio Pérez 5767a7f87e9SEugenio Pérez return r; 5777a7f87e9SEugenio Pérez } 5787a7f87e9SEugenio Pérez 5797a7f87e9SEugenio Pérez static void vhost_vdpa_net_cvq_stop(NetClientState *nc) 5807a7f87e9SEugenio Pérez { 5817a7f87e9SEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 5827a7f87e9SEugenio Pérez 5837a7f87e9SEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 5847a7f87e9SEugenio Pérez 5857a7f87e9SEugenio Pérez if (s->vhost_vdpa.shadow_vqs_enabled) { 5867a7f87e9SEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->cvq_cmd_out_buffer); 58717fb889fSEugenio Pérez vhost_vdpa_cvq_unmap_buf(&s->vhost_vdpa, s->status); 588c1a10086SEugenio Pérez } 58900ef422eSEugenio Pérez 59000ef422eSEugenio Pérez vhost_vdpa_net_client_stop(nc); 5912df4dd31SEugenio Pérez } 5922df4dd31SEugenio Pérez 593be4278b6SEugenio Pérez static ssize_t vhost_vdpa_net_cvq_add(VhostVDPAState *s, size_t out_len, 594be4278b6SEugenio Pérez size_t in_len) 595be4278b6SEugenio Pérez { 596be4278b6SEugenio Pérez /* Buffers for the device */ 597be4278b6SEugenio Pérez const struct iovec out = { 598be4278b6SEugenio Pérez .iov_base = s->cvq_cmd_out_buffer, 599be4278b6SEugenio Pérez .iov_len = out_len, 600be4278b6SEugenio Pérez }; 601be4278b6SEugenio Pérez const struct iovec in = { 60217fb889fSEugenio Pérez .iov_base = s->status, 603be4278b6SEugenio Pérez .iov_len = sizeof(virtio_net_ctrl_ack), 604be4278b6SEugenio Pérez }; 605be4278b6SEugenio Pérez VhostShadowVirtqueue *svq = g_ptr_array_index(s->vhost_vdpa.shadow_vqs, 0); 606be4278b6SEugenio Pérez int r; 607be4278b6SEugenio Pérez 608be4278b6SEugenio Pérez r = vhost_svq_add(svq, &out, 1, &in, 1, NULL); 609be4278b6SEugenio Pérez if (unlikely(r != 0)) { 610be4278b6SEugenio Pérez if (unlikely(r == -ENOSPC)) { 611be4278b6SEugenio Pérez qemu_log_mask(LOG_GUEST_ERROR, "%s: No space on device queue\n", 612be4278b6SEugenio Pérez __func__); 613be4278b6SEugenio Pérez } 614be4278b6SEugenio Pérez return r; 615be4278b6SEugenio Pérez } 616be4278b6SEugenio Pérez 617be4278b6SEugenio Pérez /* 618be4278b6SEugenio Pérez * We can poll here since we've had BQL from the time we sent the 619be4278b6SEugenio Pérez * descriptor. Also, we need to take the answer before SVQ pulls by itself, 620be4278b6SEugenio Pérez * when BQL is released 621be4278b6SEugenio Pérez */ 622be4278b6SEugenio Pérez return vhost_svq_poll(svq); 623be4278b6SEugenio Pérez } 624be4278b6SEugenio Pérez 625f73c0c43SEugenio Pérez static ssize_t vhost_vdpa_net_load_cmd(VhostVDPAState *s, uint8_t class, 626f73c0c43SEugenio Pérez uint8_t cmd, const void *data, 627f73c0c43SEugenio Pérez size_t data_size) 628f73c0c43SEugenio Pérez { 629f73c0c43SEugenio Pérez const struct virtio_net_ctrl_hdr ctrl = { 630f73c0c43SEugenio Pérez .class = class, 631f73c0c43SEugenio Pérez .cmd = cmd, 632f73c0c43SEugenio Pérez }; 633f73c0c43SEugenio Pérez 634f73c0c43SEugenio Pérez assert(data_size < vhost_vdpa_net_cvq_cmd_page_len() - sizeof(ctrl)); 635f73c0c43SEugenio Pérez 636f73c0c43SEugenio Pérez memcpy(s->cvq_cmd_out_buffer, &ctrl, sizeof(ctrl)); 637f73c0c43SEugenio Pérez memcpy(s->cvq_cmd_out_buffer + sizeof(ctrl), data, data_size); 638f73c0c43SEugenio Pérez 639f73c0c43SEugenio Pérez return vhost_vdpa_net_cvq_add(s, sizeof(ctrl) + data_size, 640f73c0c43SEugenio Pérez sizeof(virtio_net_ctrl_ack)); 641f73c0c43SEugenio Pérez } 642f73c0c43SEugenio Pérez 643f73c0c43SEugenio Pérez static int vhost_vdpa_net_load_mac(VhostVDPAState *s, const VirtIONet *n) 644f73c0c43SEugenio Pérez { 645f73c0c43SEugenio Pérez uint64_t features = n->parent_obj.guest_features; 646f73c0c43SEugenio Pérez if (features & BIT_ULL(VIRTIO_NET_F_CTRL_MAC_ADDR)) { 647f73c0c43SEugenio Pérez ssize_t dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MAC, 648f73c0c43SEugenio Pérez VIRTIO_NET_CTRL_MAC_ADDR_SET, 649f73c0c43SEugenio Pérez n->mac, sizeof(n->mac)); 650f73c0c43SEugenio Pérez if (unlikely(dev_written < 0)) { 651f73c0c43SEugenio Pérez return dev_written; 652f73c0c43SEugenio Pérez } 653f73c0c43SEugenio Pérez 654f73c0c43SEugenio Pérez return *s->status != VIRTIO_NET_OK; 655f73c0c43SEugenio Pérez } 656f73c0c43SEugenio Pérez 657f73c0c43SEugenio Pérez return 0; 658f73c0c43SEugenio Pérez } 659f73c0c43SEugenio Pérez 660f64c7cdaSEugenio Pérez static int vhost_vdpa_net_load_mq(VhostVDPAState *s, 661f64c7cdaSEugenio Pérez const VirtIONet *n) 662f64c7cdaSEugenio Pérez { 663f64c7cdaSEugenio Pérez struct virtio_net_ctrl_mq mq; 664f64c7cdaSEugenio Pérez uint64_t features = n->parent_obj.guest_features; 665f64c7cdaSEugenio Pérez ssize_t dev_written; 666f64c7cdaSEugenio Pérez 667f64c7cdaSEugenio Pérez if (!(features & BIT_ULL(VIRTIO_NET_F_MQ))) { 668f64c7cdaSEugenio Pérez return 0; 669f64c7cdaSEugenio Pérez } 670f64c7cdaSEugenio Pérez 671f64c7cdaSEugenio Pérez mq.virtqueue_pairs = cpu_to_le16(n->curr_queue_pairs); 672f64c7cdaSEugenio Pérez dev_written = vhost_vdpa_net_load_cmd(s, VIRTIO_NET_CTRL_MQ, 673f64c7cdaSEugenio Pérez VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, &mq, 674f64c7cdaSEugenio Pérez sizeof(mq)); 675f64c7cdaSEugenio Pérez if (unlikely(dev_written < 0)) { 676f64c7cdaSEugenio Pérez return dev_written; 677f64c7cdaSEugenio Pérez } 678f64c7cdaSEugenio Pérez 679f64c7cdaSEugenio Pérez return *s->status != VIRTIO_NET_OK; 680f64c7cdaSEugenio Pérez } 681f64c7cdaSEugenio Pérez 682dd036d8dSEugenio Pérez static int vhost_vdpa_net_load(NetClientState *nc) 683dd036d8dSEugenio Pérez { 684dd036d8dSEugenio Pérez VhostVDPAState *s = DO_UPCAST(VhostVDPAState, nc, nc); 685f73c0c43SEugenio Pérez struct vhost_vdpa *v = &s->vhost_vdpa; 686dd036d8dSEugenio Pérez const VirtIONet *n; 687f73c0c43SEugenio Pérez int r; 688dd036d8dSEugenio Pérez 689dd036d8dSEugenio Pérez assert(nc->info->type == NET_CLIENT_DRIVER_VHOST_VDPA); 690dd036d8dSEugenio Pérez 691dd036d8dSEugenio Pérez if (!v->shadow_vqs_enabled) { 692dd036d8dSEugenio Pérez return 0; 693dd036d8dSEugenio Pérez } 694dd036d8dSEugenio Pérez 695dd036d8dSEugenio Pérez n = VIRTIO_NET(v->dev->vdev); 696f73c0c43SEugenio Pérez r = vhost_vdpa_net_load_mac(s, n); 697f73c0c43SEugenio Pérez if (unlikely(r < 0)) { 698f73c0c43SEugenio Pérez return r; 699dd036d8dSEugenio Pérez } 700f64c7cdaSEugenio Pérez r = vhost_vdpa_net_load_mq(s, n); 701f64c7cdaSEugenio Pérez if (unlikely(r)) { 702f64c7cdaSEugenio Pérez return r; 703f64c7cdaSEugenio Pérez } 704dd036d8dSEugenio Pérez 705dd036d8dSEugenio Pérez return 0; 706dd036d8dSEugenio Pérez } 707dd036d8dSEugenio Pérez 708f8972b56SEugenio Pérez static NetClientInfo net_vhost_vdpa_cvq_info = { 709f8972b56SEugenio Pérez .type = NET_CLIENT_DRIVER_VHOST_VDPA, 710f8972b56SEugenio Pérez .size = sizeof(VhostVDPAState), 711f8972b56SEugenio Pérez .receive = vhost_vdpa_receive, 7127a7f87e9SEugenio Pérez .start = vhost_vdpa_net_cvq_start, 713dd036d8dSEugenio Pérez .load = vhost_vdpa_net_load, 7147a7f87e9SEugenio Pérez .stop = vhost_vdpa_net_cvq_stop, 715f8972b56SEugenio Pérez .cleanup = vhost_vdpa_cleanup, 716f8972b56SEugenio Pérez .has_vnet_hdr = vhost_vdpa_has_vnet_hdr, 717f8972b56SEugenio Pérez .has_ufo = vhost_vdpa_has_ufo, 718f8972b56SEugenio Pérez .check_peer_type = vhost_vdpa_check_peer_type, 719f8972b56SEugenio Pérez }; 720f8972b56SEugenio Pérez 7212df4dd31SEugenio Pérez /** 7222df4dd31SEugenio Pérez * Validate and copy control virtqueue commands. 7232df4dd31SEugenio Pérez * 7242df4dd31SEugenio Pérez * Following QEMU guidelines, we offer a copy of the buffers to the device to 7252df4dd31SEugenio Pérez * prevent TOCTOU bugs. 726bd907ae4SEugenio Pérez */ 727bd907ae4SEugenio Pérez static int vhost_vdpa_net_handle_ctrl_avail(VhostShadowVirtqueue *svq, 728bd907ae4SEugenio Pérez VirtQueueElement *elem, 729bd907ae4SEugenio Pérez void *opaque) 730bd907ae4SEugenio Pérez { 7312df4dd31SEugenio Pérez VhostVDPAState *s = opaque; 732be4278b6SEugenio Pérez size_t in_len; 733bd907ae4SEugenio Pérez virtio_net_ctrl_ack status = VIRTIO_NET_ERR; 7347a7f87e9SEugenio Pérez /* Out buffer sent to both the vdpa device and the device model */ 7357a7f87e9SEugenio Pérez struct iovec out = { 7367a7f87e9SEugenio Pérez .iov_base = s->cvq_cmd_out_buffer, 7377a7f87e9SEugenio Pérez }; 7382df4dd31SEugenio Pérez /* in buffer used for device model */ 7392df4dd31SEugenio Pérez const struct iovec in = { 7402df4dd31SEugenio Pérez .iov_base = &status, 7412df4dd31SEugenio Pérez .iov_len = sizeof(status), 7422df4dd31SEugenio Pérez }; 743be4278b6SEugenio Pérez ssize_t dev_written = -EINVAL; 744bd907ae4SEugenio Pérez 7457a7f87e9SEugenio Pérez out.iov_len = iov_to_buf(elem->out_sg, elem->out_num, 0, 7467a7f87e9SEugenio Pérez s->cvq_cmd_out_buffer, 7477a7f87e9SEugenio Pérez vhost_vdpa_net_cvq_cmd_len()); 7483f9a3eebSEugenio Pérez if (*(uint8_t *)s->cvq_cmd_out_buffer == VIRTIO_NET_CTRL_ANNOUNCE) { 7493f9a3eebSEugenio Pérez /* 7503f9a3eebSEugenio Pérez * Guest announce capability is emulated by qemu, so don't forward to 7513f9a3eebSEugenio Pérez * the device. 7523f9a3eebSEugenio Pérez */ 7533f9a3eebSEugenio Pérez dev_written = sizeof(status); 7543f9a3eebSEugenio Pérez *s->status = VIRTIO_NET_OK; 7553f9a3eebSEugenio Pérez } else { 756be4278b6SEugenio Pérez dev_written = vhost_vdpa_net_cvq_add(s, out.iov_len, sizeof(status)); 757be4278b6SEugenio Pérez if (unlikely(dev_written < 0)) { 758bd907ae4SEugenio Pérez goto out; 759bd907ae4SEugenio Pérez } 7603f9a3eebSEugenio Pérez } 761bd907ae4SEugenio Pérez 762bd907ae4SEugenio Pérez if (unlikely(dev_written < sizeof(status))) { 763bd907ae4SEugenio Pérez error_report("Insufficient written data (%zu)", dev_written); 7642df4dd31SEugenio Pérez goto out; 7652df4dd31SEugenio Pérez } 7662df4dd31SEugenio Pérez 76717fb889fSEugenio Pérez if (*s->status != VIRTIO_NET_OK) { 768be4278b6SEugenio Pérez return VIRTIO_NET_ERR; 7692df4dd31SEugenio Pérez } 7702df4dd31SEugenio Pérez 7712df4dd31SEugenio Pérez status = VIRTIO_NET_ERR; 7727a7f87e9SEugenio Pérez virtio_net_handle_ctrl_iov(svq->vdev, &in, 1, &out, 1); 7732df4dd31SEugenio Pérez if (status != VIRTIO_NET_OK) { 7742df4dd31SEugenio Pérez error_report("Bad CVQ processing in model"); 775bd907ae4SEugenio Pérez } 776bd907ae4SEugenio Pérez 777bd907ae4SEugenio Pérez out: 778bd907ae4SEugenio Pérez in_len = iov_from_buf(elem->in_sg, elem->in_num, 0, &status, 779bd907ae4SEugenio Pérez sizeof(status)); 780bd907ae4SEugenio Pérez if (unlikely(in_len < sizeof(status))) { 781bd907ae4SEugenio Pérez error_report("Bad device CVQ written length"); 782bd907ae4SEugenio Pérez } 783bd907ae4SEugenio Pérez vhost_svq_push_elem(svq, elem, MIN(in_len, sizeof(status))); 784bd907ae4SEugenio Pérez g_free(elem); 785be4278b6SEugenio Pérez return dev_written < 0 ? dev_written : 0; 786bd907ae4SEugenio Pérez } 787bd907ae4SEugenio Pérez 788bd907ae4SEugenio Pérez static const VhostShadowVirtqueueOps vhost_vdpa_net_svq_ops = { 789bd907ae4SEugenio Pérez .avail_handler = vhost_vdpa_net_handle_ctrl_avail, 790bd907ae4SEugenio Pérez }; 791bd907ae4SEugenio Pérez 792654790b6SJason Wang static NetClientState *net_vhost_vdpa_init(NetClientState *peer, 793654790b6SJason Wang const char *device, 794654790b6SJason Wang const char *name, 79540237840SJason Wang int vdpa_device_fd, 79640237840SJason Wang int queue_pair_index, 79740237840SJason Wang int nvqs, 7981576dbb5SEugenio Pérez bool is_datapath, 7991576dbb5SEugenio Pérez bool svq, 8005c1ebd4cSEugenio Pérez struct vhost_vdpa_iova_range iova_range, 8015c1ebd4cSEugenio Pérez uint64_t features) 8021e0a84eaSCindy Lu { 8031e0a84eaSCindy Lu NetClientState *nc = NULL; 8041e0a84eaSCindy Lu VhostVDPAState *s; 8051e0a84eaSCindy Lu int ret = 0; 8061e0a84eaSCindy Lu assert(name); 80740237840SJason Wang if (is_datapath) { 80840237840SJason Wang nc = qemu_new_net_client(&net_vhost_vdpa_info, peer, device, 80940237840SJason Wang name); 81040237840SJason Wang } else { 811f8972b56SEugenio Pérez nc = qemu_new_net_control_client(&net_vhost_vdpa_cvq_info, peer, 81240237840SJason Wang device, name); 81340237840SJason Wang } 81453b85d95SLaurent Vivier qemu_set_info_str(nc, TYPE_VHOST_VDPA); 8151e0a84eaSCindy Lu s = DO_UPCAST(VhostVDPAState, nc, nc); 8167327813dSJason Wang 8171e0a84eaSCindy Lu s->vhost_vdpa.device_fd = vdpa_device_fd; 81840237840SJason Wang s->vhost_vdpa.index = queue_pair_index; 8197f211a28SEugenio Pérez s->always_svq = svq; 82069498430SEugenio Pérez s->migration_state.notify = vdpa_net_migration_state_notifier; 8211576dbb5SEugenio Pérez s->vhost_vdpa.shadow_vqs_enabled = svq; 822a585fad2SEugenio Pérez s->vhost_vdpa.iova_range = iova_range; 8236188d78aSEugenio Pérez s->vhost_vdpa.shadow_data = svq; 8245c1ebd4cSEugenio Pérez if (queue_pair_index == 0) { 8255c1ebd4cSEugenio Pérez vhost_vdpa_net_valid_svq_features(features, 8265c1ebd4cSEugenio Pérez &s->vhost_vdpa.migration_blocker); 8275c1ebd4cSEugenio Pérez } else if (!is_datapath) { 8282df4dd31SEugenio Pérez s->cvq_cmd_out_buffer = qemu_memalign(qemu_real_host_page_size(), 8292df4dd31SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len()); 8302df4dd31SEugenio Pérez memset(s->cvq_cmd_out_buffer, 0, vhost_vdpa_net_cvq_cmd_page_len()); 83117fb889fSEugenio Pérez s->status = qemu_memalign(qemu_real_host_page_size(), 8322df4dd31SEugenio Pérez vhost_vdpa_net_cvq_cmd_page_len()); 83317fb889fSEugenio Pérez memset(s->status, 0, vhost_vdpa_net_cvq_cmd_page_len()); 8342df4dd31SEugenio Pérez 835bd907ae4SEugenio Pérez s->vhost_vdpa.shadow_vq_ops = &vhost_vdpa_net_svq_ops; 836bd907ae4SEugenio Pérez s->vhost_vdpa.shadow_vq_ops_opaque = s; 8379c363cf6SEugenio Pérez 8389c363cf6SEugenio Pérez /* 8399c363cf6SEugenio Pérez * TODO: We cannot migrate devices with CVQ as there is no way to set 8409c363cf6SEugenio Pérez * the device state (MAC, MQ, etc) before starting the datapath. 8419c363cf6SEugenio Pérez * 8429c363cf6SEugenio Pérez * Migration blocker ownership now belongs to s->vhost_vdpa. 8439c363cf6SEugenio Pérez */ 8449c363cf6SEugenio Pérez error_setg(&s->vhost_vdpa.migration_blocker, 8459c363cf6SEugenio Pérez "net vdpa cannot migrate with CVQ feature"); 846bd907ae4SEugenio Pérez } 84740237840SJason Wang ret = vhost_vdpa_add(nc, (void *)&s->vhost_vdpa, queue_pair_index, nvqs); 84874af5eecSJason Wang if (ret) { 84974af5eecSJason Wang qemu_del_net_client(nc); 850654790b6SJason Wang return NULL; 85174af5eecSJason Wang } 852654790b6SJason Wang return nc; 8531e0a84eaSCindy Lu } 8541e0a84eaSCindy Lu 8558170ab3fSEugenio Pérez static int vhost_vdpa_get_features(int fd, uint64_t *features, Error **errp) 8568170ab3fSEugenio Pérez { 8578170ab3fSEugenio Pérez int ret = ioctl(fd, VHOST_GET_FEATURES, features); 8588170ab3fSEugenio Pérez if (unlikely(ret < 0)) { 8598170ab3fSEugenio Pérez error_setg_errno(errp, errno, 8608170ab3fSEugenio Pérez "Fail to query features from vhost-vDPA device"); 8618170ab3fSEugenio Pérez } 8628170ab3fSEugenio Pérez return ret; 8638170ab3fSEugenio Pérez } 8648170ab3fSEugenio Pérez 8658170ab3fSEugenio Pérez static int vhost_vdpa_get_max_queue_pairs(int fd, uint64_t features, 8668170ab3fSEugenio Pérez int *has_cvq, Error **errp) 86740237840SJason Wang { 86840237840SJason Wang unsigned long config_size = offsetof(struct vhost_vdpa_config, buf); 869cd523a41SStefano Garzarella g_autofree struct vhost_vdpa_config *config = NULL; 87040237840SJason Wang __virtio16 *max_queue_pairs; 87140237840SJason Wang int ret; 87240237840SJason Wang 87340237840SJason Wang if (features & (1 << VIRTIO_NET_F_CTRL_VQ)) { 87440237840SJason Wang *has_cvq = 1; 87540237840SJason Wang } else { 87640237840SJason Wang *has_cvq = 0; 87740237840SJason Wang } 87840237840SJason Wang 87940237840SJason Wang if (features & (1 << VIRTIO_NET_F_MQ)) { 88040237840SJason Wang config = g_malloc0(config_size + sizeof(*max_queue_pairs)); 88140237840SJason Wang config->off = offsetof(struct virtio_net_config, max_virtqueue_pairs); 88240237840SJason Wang config->len = sizeof(*max_queue_pairs); 88340237840SJason Wang 88440237840SJason Wang ret = ioctl(fd, VHOST_VDPA_GET_CONFIG, config); 88540237840SJason Wang if (ret) { 88640237840SJason Wang error_setg(errp, "Fail to get config from vhost-vDPA device"); 88740237840SJason Wang return -ret; 88840237840SJason Wang } 88940237840SJason Wang 89040237840SJason Wang max_queue_pairs = (__virtio16 *)&config->buf; 89140237840SJason Wang 89240237840SJason Wang return lduw_le_p(max_queue_pairs); 89340237840SJason Wang } 89440237840SJason Wang 89540237840SJason Wang return 1; 89640237840SJason Wang } 89740237840SJason Wang 8981e0a84eaSCindy Lu int net_init_vhost_vdpa(const Netdev *netdev, const char *name, 8991e0a84eaSCindy Lu NetClientState *peer, Error **errp) 9001e0a84eaSCindy Lu { 9011e0a84eaSCindy Lu const NetdevVhostVDPAOptions *opts; 9028170ab3fSEugenio Pérez uint64_t features; 903654790b6SJason Wang int vdpa_device_fd; 904eb3cb751SEugenio Pérez g_autofree NetClientState **ncs = NULL; 905a585fad2SEugenio Pérez struct vhost_vdpa_iova_range iova_range; 906eb3cb751SEugenio Pérez NetClientState *nc; 907aed5da45SEugenio Pérez int queue_pairs, r, i = 0, has_cvq = 0; 9081e0a84eaSCindy Lu 9091e0a84eaSCindy Lu assert(netdev->type == NET_CLIENT_DRIVER_VHOST_VDPA); 9101e0a84eaSCindy Lu opts = &netdev->u.vhost_vdpa; 9117480874aSMarkus Armbruster if (!opts->vhostdev && !opts->vhostfd) { 9128801ccd0SSi-Wei Liu error_setg(errp, 9138801ccd0SSi-Wei Liu "vhost-vdpa: neither vhostdev= nor vhostfd= was specified"); 914c8295404SEugenio Pérez return -1; 915c8295404SEugenio Pérez } 9167327813dSJason Wang 9177480874aSMarkus Armbruster if (opts->vhostdev && opts->vhostfd) { 9188801ccd0SSi-Wei Liu error_setg(errp, 9198801ccd0SSi-Wei Liu "vhost-vdpa: vhostdev= and vhostfd= are mutually exclusive"); 9208801ccd0SSi-Wei Liu return -1; 9218801ccd0SSi-Wei Liu } 9228801ccd0SSi-Wei Liu 9237480874aSMarkus Armbruster if (opts->vhostdev) { 9240351152bSEugenio Pérez vdpa_device_fd = qemu_open(opts->vhostdev, O_RDWR, errp); 9257327813dSJason Wang if (vdpa_device_fd == -1) { 9267327813dSJason Wang return -errno; 9277327813dSJason Wang } 9285107fd3eSPeter Maydell } else { 9295107fd3eSPeter Maydell /* has_vhostfd */ 9308801ccd0SSi-Wei Liu vdpa_device_fd = monitor_fd_param(monitor_cur(), opts->vhostfd, errp); 9318801ccd0SSi-Wei Liu if (vdpa_device_fd == -1) { 9328801ccd0SSi-Wei Liu error_prepend(errp, "vhost-vdpa: unable to parse vhostfd: "); 9338801ccd0SSi-Wei Liu return -1; 9348801ccd0SSi-Wei Liu } 9358801ccd0SSi-Wei Liu } 9367327813dSJason Wang 9378170ab3fSEugenio Pérez r = vhost_vdpa_get_features(vdpa_device_fd, &features, errp); 9388170ab3fSEugenio Pérez if (unlikely(r < 0)) { 939aed5da45SEugenio Pérez goto err; 9408170ab3fSEugenio Pérez } 9418170ab3fSEugenio Pérez 9428170ab3fSEugenio Pérez queue_pairs = vhost_vdpa_get_max_queue_pairs(vdpa_device_fd, features, 94340237840SJason Wang &has_cvq, errp); 94440237840SJason Wang if (queue_pairs < 0) { 9457327813dSJason Wang qemu_close(vdpa_device_fd); 94640237840SJason Wang return queue_pairs; 9477327813dSJason Wang } 9487327813dSJason Wang 949bf7a2ad8SLongpeng r = vhost_vdpa_get_iova_range(vdpa_device_fd, &iova_range); 950bf7a2ad8SLongpeng if (unlikely(r < 0)) { 951bf7a2ad8SLongpeng error_setg(errp, "vhost-vdpa: get iova range failed: %s", 952bf7a2ad8SLongpeng strerror(-r)); 953bf7a2ad8SLongpeng goto err; 954bf7a2ad8SLongpeng } 955bf7a2ad8SLongpeng 95600ef422eSEugenio Pérez if (opts->x_svq && !vhost_vdpa_net_valid_svq_features(features, errp)) { 95700ef422eSEugenio Pérez goto err; 9581576dbb5SEugenio Pérez } 9591576dbb5SEugenio Pérez 96040237840SJason Wang ncs = g_malloc0(sizeof(*ncs) * queue_pairs); 96140237840SJason Wang 96240237840SJason Wang for (i = 0; i < queue_pairs; i++) { 96340237840SJason Wang ncs[i] = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, 9641576dbb5SEugenio Pérez vdpa_device_fd, i, 2, true, opts->x_svq, 9655c1ebd4cSEugenio Pérez iova_range, features); 96640237840SJason Wang if (!ncs[i]) 96740237840SJason Wang goto err; 96840237840SJason Wang } 96940237840SJason Wang 97040237840SJason Wang if (has_cvq) { 97140237840SJason Wang nc = net_vhost_vdpa_init(peer, TYPE_VHOST_VDPA, name, 9721576dbb5SEugenio Pérez vdpa_device_fd, i, 1, false, 9735c1ebd4cSEugenio Pérez opts->x_svq, iova_range, features); 97440237840SJason Wang if (!nc) 97540237840SJason Wang goto err; 97640237840SJason Wang } 97740237840SJason Wang 978654790b6SJason Wang return 0; 97940237840SJason Wang 98040237840SJason Wang err: 98140237840SJason Wang if (i) { 9829bd05507SSi-Wei Liu for (i--; i >= 0; i--) { 9839bd05507SSi-Wei Liu qemu_del_net_client(ncs[i]); 9849bd05507SSi-Wei Liu } 98540237840SJason Wang } 9861576dbb5SEugenio Pérez 98740237840SJason Wang qemu_close(vdpa_device_fd); 98840237840SJason Wang 98940237840SJason Wang return -1; 9901e0a84eaSCindy Lu } 991