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