ceph/src/seastar/dpdk/drivers/net/virtio/virtio_user/vhost_kernel.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2016 Intel Corporation
   3  */
   4
   5 #include <sys/types.h>
   6 #include <sys/stat.h>
   7 #include <fcntl.h>
   8 #include <unistd.h>
   9
  10 #include <rte_memory.h>
  11 #include <rte_eal_memconfig.h>
  12
  13 #include "vhost.h"
  14 #include "virtio_user_dev.h"
  15 #include "vhost_kernel_tap.h"
  16
  17 struct vhost_memory_kernel {
  18         uint32_t nregions;
  19         uint32_t padding;
  20         struct vhost_memory_region regions[0];
  21 };
  22
  23 /* vhost kernel ioctls */
  24 #define VHOST_VIRTIO 0xAF
  25 #define VHOST_GET_FEATURES _IOR(VHOST_VIRTIO, 0x00, __u64)
  26 #define VHOST_SET_FEATURES _IOW(VHOST_VIRTIO, 0x00, __u64)
  27 #define VHOST_SET_OWNER _IO(VHOST_VIRTIO, 0x01)
  28 #define VHOST_RESET_OWNER _IO(VHOST_VIRTIO, 0x02)
  29 #define VHOST_SET_MEM_TABLE _IOW(VHOST_VIRTIO, 0x03, struct vhost_memory_kernel)
  30 #define VHOST_SET_LOG_BASE _IOW(VHOST_VIRTIO, 0x04, __u64)
  31 #define VHOST_SET_LOG_FD _IOW(VHOST_VIRTIO, 0x07, int)
  32 #define VHOST_SET_VRING_NUM _IOW(VHOST_VIRTIO, 0x10, struct vhost_vring_state)
  33 #define VHOST_SET_VRING_ADDR _IOW(VHOST_VIRTIO, 0x11, struct vhost_vring_addr)
  34 #define VHOST_SET_VRING_BASE _IOW(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  35 #define VHOST_GET_VRING_BASE _IOWR(VHOST_VIRTIO, 0x12, struct vhost_vring_state)
  36 #define VHOST_SET_VRING_KICK _IOW(VHOST_VIRTIO, 0x20, struct vhost_vring_file)
  37 #define VHOST_SET_VRING_CALL _IOW(VHOST_VIRTIO, 0x21, struct vhost_vring_file)
  38 #define VHOST_SET_VRING_ERR _IOW(VHOST_VIRTIO, 0x22, struct vhost_vring_file)
  39 #define VHOST_NET_SET_BACKEND _IOW(VHOST_VIRTIO, 0x30, struct vhost_vring_file)
  40
  41 static uint64_t max_regions = 64;
  42
  43 static void
  44 get_vhost_kernel_max_regions(void)
  45 {
  46         int fd;
  47         char buf[20] = {'\0'};
  48
  49         fd = open("/sys/module/vhost/parameters/max_mem_regions", O_RDONLY);
  50         if (fd < 0)
  51                 return;
  52
  53         if (read(fd, buf, sizeof(buf) - 1) > 0)
  54                 max_regions = strtoull(buf, NULL, 10);
  55
  56         close(fd);
  57 }
  58
  59 static uint64_t vhost_req_user_to_kernel[] = {
  60         [VHOST_USER_SET_OWNER] = VHOST_SET_OWNER,
  61         [VHOST_USER_RESET_OWNER] = VHOST_RESET_OWNER,
  62         [VHOST_USER_SET_FEATURES] = VHOST_SET_FEATURES,
  63         [VHOST_USER_GET_FEATURES] = VHOST_GET_FEATURES,
  64         [VHOST_USER_SET_VRING_CALL] = VHOST_SET_VRING_CALL,
  65         [VHOST_USER_SET_VRING_NUM] = VHOST_SET_VRING_NUM,
  66         [VHOST_USER_SET_VRING_BASE] = VHOST_SET_VRING_BASE,
  67         [VHOST_USER_GET_VRING_BASE] = VHOST_GET_VRING_BASE,
  68         [VHOST_USER_SET_VRING_ADDR] = VHOST_SET_VRING_ADDR,
  69         [VHOST_USER_SET_VRING_KICK] = VHOST_SET_VRING_KICK,
  70         [VHOST_USER_SET_MEM_TABLE] = VHOST_SET_MEM_TABLE,
  71 };
  72
  73 static int
  74 add_memseg_list(const struct rte_memseg_list *msl, void *arg)
  75 {
  76         struct vhost_memory_kernel *vm = arg;
  77         struct vhost_memory_region *mr;
  78         void *start_addr;
  79         uint64_t len;
  80
  81         if (msl->external)
  82                 return 0;
  83
  84         if (vm->nregions >= max_regions)
  85                 return -1;
  86
  87         start_addr = msl->base_va;
  88         len = msl->page_sz * msl->memseg_arr.len;
  89
  90         mr = &vm->regions[vm->nregions++];
  91
  92         mr->guest_phys_addr = (uint64_t)(uintptr_t)start_addr;
  93         mr->userspace_addr = (uint64_t)(uintptr_t)start_addr;
  94         mr->memory_size = len;
  95         mr->mmap_offset = 0; /* flags_padding */
  96
  97         PMD_DRV_LOG(DEBUG, "index=%u addr=%p len=%" PRIu64,
  98                         vm->nregions - 1, start_addr, len);
  99
 100         return 0;
 101 }
 102
 103 /* By default, vhost kernel module allows 64 regions, but DPDK may
 104  * have much more memory regions. Below function will treat each
 105  * contiguous memory space reserved by DPDK as one region.
 106  */
 107 static struct vhost_memory_kernel *
 108 prepare_vhost_memory_kernel(void)
 109 {
 110         struct vhost_memory_kernel *vm;
 111
 112         vm = malloc(sizeof(struct vhost_memory_kernel) +
 113                         max_regions *
 114                         sizeof(struct vhost_memory_region));
 115         if (!vm)
 116                 return NULL;
 117
 118         vm->nregions = 0;
 119         vm->padding = 0;
 120
 121         /*
 122          * The memory lock has already been taken by memory subsystem
 123          * or virtio_user_start_device().
 124          */
 125         if (rte_memseg_list_walk_thread_unsafe(add_memseg_list, vm) < 0) {
 126                 free(vm);
 127                 return NULL;
 128         }
 129
 130         return vm;
 131 }
 132
 133 /* with below features, vhost kernel does not need to do the checksum and TSO,
 134  * these info will be passed to virtio_user through virtio net header.
 135  */
 136 #define VHOST_KERNEL_GUEST_OFFLOADS_MASK        \
 137         ((1ULL << VIRTIO_NET_F_GUEST_CSUM) |    \
 138          (1ULL << VIRTIO_NET_F_GUEST_TSO4) |    \
 139          (1ULL << VIRTIO_NET_F_GUEST_TSO6) |    \
 140          (1ULL << VIRTIO_NET_F_GUEST_ECN)  |    \
 141          (1ULL << VIRTIO_NET_F_GUEST_UFO))
 142
 143 /* with below features, when flows from virtio_user to vhost kernel
 144  * (1) if flows goes up through the kernel networking stack, it does not need
 145  * to verify checksum, which can save CPU cycles;
 146  * (2) if flows goes through a Linux bridge and outside from an interface
 147  * (kernel driver), checksum and TSO will be done by GSO in kernel or even
 148  * offloaded into real physical device.
 149  */
 150 #define VHOST_KERNEL_HOST_OFFLOADS_MASK         \
 151         ((1ULL << VIRTIO_NET_F_HOST_TSO4) |     \
 152          (1ULL << VIRTIO_NET_F_HOST_TSO6) |     \
 153          (1ULL << VIRTIO_NET_F_CSUM))
 154
 155 static unsigned int
 156 tap_support_features(void)
 157 {
 158         int tapfd;
 159         unsigned int tap_features;
 160
 161         tapfd = open(PATH_NET_TUN, O_RDWR);
 162         if (tapfd < 0) {
 163                 PMD_DRV_LOG(ERR, "fail to open %s: %s",
 164                             PATH_NET_TUN, strerror(errno));
 165                 return -1;
 166         }
 167
 168         if (ioctl(tapfd, TUNGETFEATURES, &tap_features) == -1) {
 169                 PMD_DRV_LOG(ERR, "TUNGETFEATURES failed: %s", strerror(errno));
 170                 close(tapfd);
 171                 return -1;
 172         }
 173
 174         close(tapfd);
 175         return tap_features;
 176 }
 177
 178 static int
 179 vhost_kernel_ioctl(struct virtio_user_dev *dev,
 180                    enum vhost_user_request req,
 181                    void *arg)
 182 {
 183         int ret = -1;
 184         unsigned int i;
 185         uint64_t req_kernel;
 186         struct vhost_memory_kernel *vm = NULL;
 187         int vhostfd;
 188         unsigned int queue_sel;
 189         unsigned int features;
 190
 191         PMD_DRV_LOG(INFO, "%s", vhost_msg_strings[req]);
 192
 193         req_kernel = vhost_req_user_to_kernel[req];
 194
 195         if (req_kernel == VHOST_SET_MEM_TABLE) {
 196                 vm = prepare_vhost_memory_kernel();
 197                 if (!vm)
 198                         return -1;
 199                 arg = (void *)vm;
 200         }
 201
 202         if (req_kernel == VHOST_SET_FEATURES) {
 203                 /* We don't need memory protection here */
 204                 *(uint64_t *)arg &= ~(1ULL << VIRTIO_F_IOMMU_PLATFORM);
 205
 206                 /* VHOST kernel does not know about below flags */
 207                 *(uint64_t *)arg &= ~VHOST_KERNEL_GUEST_OFFLOADS_MASK;
 208                 *(uint64_t *)arg &= ~VHOST_KERNEL_HOST_OFFLOADS_MASK;
 209
 210                 *(uint64_t *)arg &= ~(1ULL << VIRTIO_NET_F_MQ);
 211         }
 212
 213         switch (req_kernel) {
 214         case VHOST_SET_VRING_NUM:
 215         case VHOST_SET_VRING_ADDR:
 216         case VHOST_SET_VRING_BASE:
 217         case VHOST_GET_VRING_BASE:
 218         case VHOST_SET_VRING_KICK:
 219         case VHOST_SET_VRING_CALL:
 220                 queue_sel = *(unsigned int *)arg;
 221                 vhostfd = dev->vhostfds[queue_sel / 2];
 222                 *(unsigned int *)arg = queue_sel % 2;
 223                 PMD_DRV_LOG(DEBUG, "vhostfd=%d, index=%u",
 224                             vhostfd, *(unsigned int *)arg);
 225                 break;
 226         default:
 227                 vhostfd = -1;
 228         }
 229         if (vhostfd == -1) {
 230                 for (i = 0; i < dev->max_queue_pairs; ++i) {
 231                         if (dev->vhostfds[i] < 0)
 232                                 continue;
 233
 234                         ret = ioctl(dev->vhostfds[i], req_kernel, arg);
 235                         if (ret < 0)
 236                                 break;
 237                 }
 238         } else {
 239                 ret = ioctl(vhostfd, req_kernel, arg);
 240         }
 241
 242         if (!ret && req_kernel == VHOST_GET_FEATURES) {
 243                 features = tap_support_features();
 244                 /* with tap as the backend, all these features are supported
 245                  * but not claimed by vhost-net, so we add them back when
 246                  * reporting to upper layer.
 247                  */
 248                 if (features & IFF_VNET_HDR) {
 249                         *((uint64_t *)arg) |= VHOST_KERNEL_GUEST_OFFLOADS_MASK;
 250                         *((uint64_t *)arg) |= VHOST_KERNEL_HOST_OFFLOADS_MASK;
 251                 }
 252
 253                 /* vhost_kernel will not declare this feature, but it does
 254                  * support multi-queue.
 255                  */
 256                 if (features & IFF_MULTI_QUEUE)
 257                         *(uint64_t *)arg |= (1ull << VIRTIO_NET_F_MQ);
 258         }
 259
 260         if (vm)
 261                 free(vm);
 262
 263         if (ret < 0)
 264                 PMD_DRV_LOG(ERR, "%s failed: %s",
 265                             vhost_msg_strings[req], strerror(errno));
 266
 267         return ret;
 268 }
 269
 270 /**
 271  * Set up environment to talk with a vhost kernel backend.
 272  *
 273  * @return
 274  *   - (-1) if fail to set up;
 275  *   - (>=0) if successful.
 276  */
 277 static int
 278 vhost_kernel_setup(struct virtio_user_dev *dev)
 279 {
 280         int vhostfd;
 281         uint32_t i;
 282
 283         get_vhost_kernel_max_regions();
 284
 285         for (i = 0; i < dev->max_queue_pairs; ++i) {
 286                 vhostfd = open(dev->path, O_RDWR);
 287                 if (vhostfd < 0) {
 288                         PMD_DRV_LOG(ERR, "fail to open %s, %s",
 289                                     dev->path, strerror(errno));
 290                         return -1;
 291                 }
 292
 293                 dev->vhostfds[i] = vhostfd;
 294         }
 295
 296         return 0;
 297 }
 298
 299 static int
 300 vhost_kernel_set_backend(int vhostfd, int tapfd)
 301 {
 302         struct vhost_vring_file f;
 303
 304         f.fd = tapfd;
 305         f.index = 0;
 306         if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
 307                 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
 308                                 strerror(errno));
 309                 return -1;
 310         }
 311
 312         f.index = 1;
 313         if (ioctl(vhostfd, VHOST_NET_SET_BACKEND, &f) < 0) {
 314                 PMD_DRV_LOG(ERR, "VHOST_NET_SET_BACKEND fails, %s",
 315                                 strerror(errno));
 316                 return -1;
 317         }
 318
 319         return 0;
 320 }
 321
 322 static int
 323 vhost_kernel_enable_queue_pair(struct virtio_user_dev *dev,
 324                                uint16_t pair_idx,
 325                                int enable)
 326 {
 327         int hdr_size;
 328         int vhostfd;
 329         int tapfd;
 330         int req_mq = (dev->max_queue_pairs > 1);
 331
 332         vhostfd = dev->vhostfds[pair_idx];
 333
 334         if (!enable) {
 335                 if (dev->tapfds[pair_idx] >= 0) {
 336                         close(dev->tapfds[pair_idx]);
 337                         dev->tapfds[pair_idx] = -1;
 338                 }
 339                 return vhost_kernel_set_backend(vhostfd, -1);
 340         } else if (dev->tapfds[pair_idx] >= 0) {
 341                 return 0;
 342         }
 343
 344         if ((dev->features & (1ULL << VIRTIO_NET_F_MRG_RXBUF)) ||
 345             (dev->features & (1ULL << VIRTIO_F_VERSION_1)))
 346                 hdr_size = sizeof(struct virtio_net_hdr_mrg_rxbuf);
 347         else
 348                 hdr_size = sizeof(struct virtio_net_hdr);
 349
 350         tapfd = vhost_kernel_open_tap(&dev->ifname, hdr_size, req_mq,
 351                          (char *)dev->mac_addr, dev->features);
 352         if (tapfd < 0) {
 353                 PMD_DRV_LOG(ERR, "fail to open tap for vhost kernel");
 354                 return -1;
 355         }
 356
 357         if (vhost_kernel_set_backend(vhostfd, tapfd) < 0) {
 358                 PMD_DRV_LOG(ERR, "fail to set backend for vhost kernel");
 359                 close(tapfd);
 360                 return -1;
 361         }
 362
 363         dev->tapfds[pair_idx] = tapfd;
 364         return 0;
 365 }
 366
 367 struct virtio_user_backend_ops virtio_ops_kernel = {
 368         .setup = vhost_kernel_setup,
 369         .send_request = vhost_kernel_ioctl,
 370         .enable_qp = vhost_kernel_enable_queue_pair
 371 };