1 #include <linux/virtio_ring.h> 2 #include <linux/types.h> 3 #include <sys/uio.h> 4 #include <stdlib.h> 5 6 #include "kvm/guest_compat.h" 7 #include "kvm/barrier.h" 8 #include "kvm/virtio.h" 9 #include "kvm/virtio-pci.h" 10 #include "kvm/virtio-mmio.h" 11 #include "kvm/util.h" 12 #include "kvm/kvm.h" 13 14 15 const char* virtio_trans_name(enum virtio_trans trans) 16 { 17 if (trans == VIRTIO_PCI) 18 return "pci"; 19 else if (trans == VIRTIO_MMIO) 20 return "mmio"; 21 return "unknown"; 22 } 23 24 void virt_queue__used_idx_advance(struct virt_queue *queue, u16 jump) 25 { 26 u16 idx = virtio_guest_to_host_u16(queue, queue->vring.used->idx); 27 28 /* 29 * Use wmb to assure that used elem was updated with head and len. 30 * We need a wmb here since we can't advance idx unless we're ready 31 * to pass the used element to the guest. 32 */ 33 wmb(); 34 idx += jump; 35 queue->vring.used->idx = virtio_host_to_guest_u16(queue, idx); 36 } 37 38 struct vring_used_elem * 39 virt_queue__set_used_elem_no_update(struct virt_queue *queue, u32 head, 40 u32 len, u16 offset) 41 { 42 struct vring_used_elem *used_elem; 43 u16 idx = virtio_guest_to_host_u16(queue, queue->vring.used->idx); 44 45 idx += offset; 46 used_elem = &queue->vring.used->ring[idx % queue->vring.num]; 47 used_elem->id = virtio_host_to_guest_u32(queue, head); 48 used_elem->len = virtio_host_to_guest_u32(queue, len); 49 50 return used_elem; 51 } 52 53 struct vring_used_elem *virt_queue__set_used_elem(struct virt_queue *queue, u32 head, u32 len) 54 { 55 struct vring_used_elem *used_elem; 56 57 used_elem = virt_queue__set_used_elem_no_update(queue, head, len, 0); 58 virt_queue__used_idx_advance(queue, 1); 59 60 return used_elem; 61 } 62 63 static inline bool virt_desc__test_flag(struct virt_queue *vq, 64 struct vring_desc *desc, u16 flag) 65 { 66 return !!(virtio_guest_to_host_u16(vq, desc->flags) & flag); 67 } 68 69 /* 70 * Each buffer in the virtqueues is actually a chain of descriptors. This 71 * function returns the next descriptor in the chain, or max if we're at the 72 * end. 73 */ 74 static unsigned next_desc(struct virt_queue *vq, struct vring_desc *desc, 75 unsigned int i, unsigned int max) 76 { 77 unsigned int next; 78 79 /* If this descriptor says it doesn't chain, we're done. */ 80 if (!virt_desc__test_flag(vq, &desc[i], VRING_DESC_F_NEXT)) 81 return max; 82 83 next = virtio_guest_to_host_u16(vq, desc[i].next); 84 85 /* Ensure they're not leading us off end of descriptors. */ 86 return min(next, max); 87 } 88 89 u16 virt_queue__get_head_iov(struct virt_queue *vq, struct iovec iov[], u16 *out, u16 *in, u16 head, struct kvm *kvm) 90 { 91 struct vring_desc *desc; 92 u16 idx; 93 u16 max; 94 95 idx = head; 96 *out = *in = 0; 97 max = vq->vring.num; 98 desc = vq->vring.desc; 99 100 if (virt_desc__test_flag(vq, &desc[idx], VRING_DESC_F_INDIRECT)) { 101 max = virtio_guest_to_host_u32(vq, desc[idx].len) / sizeof(struct vring_desc); 102 desc = guest_flat_to_host(kvm, virtio_guest_to_host_u64(vq, desc[idx].addr)); 103 idx = 0; 104 } 105 106 do { 107 /* Grab the first descriptor, and check it's OK. */ 108 iov[*out + *in].iov_len = virtio_guest_to_host_u32(vq, desc[idx].len); 109 iov[*out + *in].iov_base = guest_flat_to_host(kvm, 110 virtio_guest_to_host_u64(vq, desc[idx].addr)); 111 /* If this is an input descriptor, increment that count. */ 112 if (virt_desc__test_flag(vq, &desc[idx], VRING_DESC_F_WRITE)) 113 (*in)++; 114 else 115 (*out)++; 116 } while ((idx = next_desc(vq, desc, idx, max)) != max); 117 118 return head; 119 } 120 121 u16 virt_queue__get_iov(struct virt_queue *vq, struct iovec iov[], u16 *out, u16 *in, struct kvm *kvm) 122 { 123 u16 head; 124 125 head = virt_queue__pop(vq); 126 127 return virt_queue__get_head_iov(vq, iov, out, in, head, kvm); 128 } 129 130 /* in and out are relative to guest */ 131 u16 virt_queue__get_inout_iov(struct kvm *kvm, struct virt_queue *queue, 132 struct iovec in_iov[], struct iovec out_iov[], 133 u16 *in, u16 *out) 134 { 135 struct vring_desc *desc; 136 u16 head, idx; 137 138 idx = head = virt_queue__pop(queue); 139 *out = *in = 0; 140 do { 141 u64 addr; 142 desc = virt_queue__get_desc(queue, idx); 143 addr = virtio_guest_to_host_u64(queue, desc->addr); 144 if (virt_desc__test_flag(queue, desc, VRING_DESC_F_WRITE)) { 145 in_iov[*in].iov_base = guest_flat_to_host(kvm, addr); 146 in_iov[*in].iov_len = virtio_guest_to_host_u32(queue, desc->len); 147 (*in)++; 148 } else { 149 out_iov[*out].iov_base = guest_flat_to_host(kvm, addr); 150 out_iov[*out].iov_len = virtio_guest_to_host_u32(queue, desc->len); 151 (*out)++; 152 } 153 if (virt_desc__test_flag(queue, desc, VRING_DESC_F_NEXT)) 154 idx = virtio_guest_to_host_u16(queue, desc->next); 155 else 156 break; 157 } while (1); 158 159 return head; 160 } 161 162 void virtio_init_device_vq(struct kvm *kvm, struct virtio_device *vdev, 163 struct virt_queue *vq, size_t nr_descs, 164 u32 page_size, u32 align, u32 pfn) 165 { 166 void *p = guest_flat_to_host(kvm, (u64)pfn * page_size); 167 168 vq->endian = vdev->endian; 169 vq->pfn = pfn; 170 vq->use_event_idx = (vdev->features & VIRTIO_RING_F_EVENT_IDX); 171 vq->enabled = true; 172 173 vring_init(&vq->vring, nr_descs, p, align); 174 } 175 176 void virtio_exit_vq(struct kvm *kvm, struct virtio_device *vdev, 177 void *dev, int num) 178 { 179 struct virt_queue *vq = vdev->ops->get_vq(kvm, dev, num); 180 181 if (vq->enabled && vdev->ops->exit_vq) 182 vdev->ops->exit_vq(kvm, dev, num); 183 memset(vq, 0, sizeof(*vq)); 184 } 185 186 int virtio__get_dev_specific_field(int offset, bool msix, u32 *config_off) 187 { 188 if (msix) { 189 if (offset < 4) 190 return VIRTIO_PCI_O_MSIX; 191 else 192 offset -= 4; 193 } 194 195 *config_off = offset; 196 197 return VIRTIO_PCI_O_CONFIG; 198 } 199 200 bool virtio_queue__should_signal(struct virt_queue *vq) 201 { 202 u16 old_idx, new_idx, event_idx; 203 204 /* 205 * Use mb to assure used idx has been increased before we signal the 206 * guest, and we don't read a stale value for used_event. Without a mb 207 * here we might not send a notification that we need to send, or the 208 * guest may ignore the queue since it won't see an updated idx. 209 */ 210 mb(); 211 212 if (!vq->use_event_idx) { 213 /* 214 * When VIRTIO_RING_F_EVENT_IDX isn't negotiated, interrupt the 215 * guest if it didn't explicitly request to be left alone. 216 */ 217 return !(virtio_guest_to_host_u16(vq, vq->vring.avail->flags) & 218 VRING_AVAIL_F_NO_INTERRUPT); 219 } 220 221 old_idx = vq->last_used_signalled; 222 new_idx = virtio_guest_to_host_u16(vq, vq->vring.used->idx); 223 event_idx = virtio_guest_to_host_u16(vq, vring_used_event(&vq->vring)); 224 225 if (vring_need_event(event_idx, new_idx, old_idx)) { 226 vq->last_used_signalled = new_idx; 227 return true; 228 } 229 230 return false; 231 } 232 233 void virtio_set_guest_features(struct kvm *kvm, struct virtio_device *vdev, 234 void *dev, u32 features) 235 { 236 /* TODO: fail negotiation if features & ~host_features */ 237 238 vdev->features = features; 239 vdev->ops->set_guest_features(kvm, dev, features); 240 } 241 242 void virtio_notify_status(struct kvm *kvm, struct virtio_device *vdev, 243 void *dev, u8 status) 244 { 245 u32 ext_status = status; 246 247 vdev->status &= ~VIRTIO_CONFIG_S_MASK; 248 vdev->status |= status; 249 250 /* Add a few hints to help devices */ 251 if ((status & VIRTIO_CONFIG_S_DRIVER_OK) && 252 !(vdev->status & VIRTIO__STATUS_START)) { 253 vdev->status |= VIRTIO__STATUS_START; 254 ext_status |= VIRTIO__STATUS_START; 255 256 } else if (!status && (vdev->status & VIRTIO__STATUS_START)) { 257 vdev->status &= ~VIRTIO__STATUS_START; 258 ext_status |= VIRTIO__STATUS_STOP; 259 260 /* 261 * Reset virtqueues and stop all traffic now, so that the device 262 * can safely reset the backend in notify_status(). 263 */ 264 vdev->ops->reset(kvm, vdev); 265 } 266 267 if (vdev->ops->notify_status) 268 vdev->ops->notify_status(kvm, dev, ext_status); 269 } 270 271 int virtio_init(struct kvm *kvm, void *dev, struct virtio_device *vdev, 272 struct virtio_ops *ops, enum virtio_trans trans, 273 int device_id, int subsys_id, int class) 274 { 275 void *virtio; 276 int r; 277 278 switch (trans) { 279 case VIRTIO_PCI: 280 virtio = calloc(sizeof(struct virtio_pci), 1); 281 if (!virtio) 282 return -ENOMEM; 283 vdev->virtio = virtio; 284 vdev->ops = ops; 285 vdev->ops->signal_vq = virtio_pci__signal_vq; 286 vdev->ops->signal_config = virtio_pci__signal_config; 287 vdev->ops->init = virtio_pci__init; 288 vdev->ops->exit = virtio_pci__exit; 289 vdev->ops->reset = virtio_pci__reset; 290 r = vdev->ops->init(kvm, dev, vdev, device_id, subsys_id, class); 291 break; 292 case VIRTIO_MMIO: 293 virtio = calloc(sizeof(struct virtio_mmio), 1); 294 if (!virtio) 295 return -ENOMEM; 296 vdev->virtio = virtio; 297 vdev->ops = ops; 298 vdev->ops->signal_vq = virtio_mmio_signal_vq; 299 vdev->ops->signal_config = virtio_mmio_signal_config; 300 vdev->ops->init = virtio_mmio_init; 301 vdev->ops->exit = virtio_mmio_exit; 302 vdev->ops->reset = virtio_mmio_reset; 303 r = vdev->ops->init(kvm, dev, vdev, device_id, subsys_id, class); 304 break; 305 default: 306 r = -1; 307 }; 308 309 return r; 310 } 311 312 int virtio_compat_add_message(const char *device, const char *config) 313 { 314 int len = 1024; 315 int compat_id; 316 char *title; 317 char *desc; 318 319 title = malloc(len); 320 if (!title) 321 return -ENOMEM; 322 323 desc = malloc(len); 324 if (!desc) { 325 free(title); 326 return -ENOMEM; 327 } 328 329 snprintf(title, len, "%s device was not detected.", device); 330 snprintf(desc, len, "While you have requested a %s device, " 331 "the guest kernel did not initialize it.\n" 332 "\tPlease make sure that the guest kernel was " 333 "compiled with %s=y enabled in .config.", 334 device, config); 335 336 compat_id = compat__add_message(title, desc); 337 338 free(desc); 339 free(title); 340 341 return compat_id; 342 } 343