1 /*
2 * Virtio Support
3 *
4 * Copyright IBM, Corp. 2007
5 *
6 * Authors:
7 * Anthony Liguori <aliguori@us.ibm.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2. See
10 * the COPYING file in the top-level directory.
11 *
12 */
13
14 #ifndef QEMU_VIRTIO_H
15 #define QEMU_VIRTIO_H
16
17 #include "exec/memory.h"
18 #include "hw/qdev-core.h"
19 #include "net/net.h"
20 #include "migration/vmstate.h"
21 #include "qemu/event_notifier.h"
22 #include "standard-headers/linux/virtio_config.h"
23 #include "standard-headers/linux/virtio_ring.h"
24 #include "qom/object.h"
25 #include "block/aio.h"
26
27 /*
28 * A guest should never accept this. It implies negotiation is broken
29 * between the driver frontend and the device. This bit is re-used for
30 * vhost-user to advertise VHOST_USER_F_PROTOCOL_FEATURES between QEMU
31 * and a vhost-user backend.
32 */
33 #define VIRTIO_F_BAD_FEATURE 30
34
35 #define VIRTIO_LEGACY_FEATURES ((0x1ULL << VIRTIO_F_BAD_FEATURE) | \
36 (0x1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) | \
37 (0x1ULL << VIRTIO_F_ANY_LAYOUT))
38
39 struct VirtQueue;
40
vring_align(hwaddr addr,unsigned long align)41 static inline hwaddr vring_align(hwaddr addr,
42 unsigned long align)
43 {
44 return QEMU_ALIGN_UP(addr, align);
45 }
46
47 typedef struct VirtIOFeature {
48 uint64_t flags;
49 size_t end;
50 } VirtIOFeature;
51
52 typedef struct VirtIOConfigSizeParams {
53 size_t min_size;
54 size_t max_size;
55 const VirtIOFeature *feature_sizes;
56 } VirtIOConfigSizeParams;
57
58 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
59 uint64_t host_features);
60
61 typedef struct VirtQueue VirtQueue;
62
63 #define VIRTQUEUE_MAX_SIZE 1024
64
65 typedef struct VirtQueueElement
66 {
67 unsigned int index;
68 unsigned int len;
69 unsigned int ndescs;
70 unsigned int out_num;
71 unsigned int in_num;
72 hwaddr *in_addr;
73 hwaddr *out_addr;
74 struct iovec *in_sg;
75 struct iovec *out_sg;
76 } VirtQueueElement;
77
78 #define VIRTIO_QUEUE_MAX 1024
79
80 #define VIRTIO_NO_VECTOR 0xffff
81
82 /* special index value used internally for config irqs */
83 #define VIRTIO_CONFIG_IRQ_IDX -1
84
85 #define TYPE_VIRTIO_DEVICE "virtio-device"
86 OBJECT_DECLARE_TYPE(VirtIODevice, VirtioDeviceClass, VIRTIO_DEVICE)
87
88 typedef struct {
89 int virtio_bit;
90 const char *feature_desc;
91 } qmp_virtio_feature_map_t;
92
93 enum virtio_device_endian {
94 VIRTIO_DEVICE_ENDIAN_UNKNOWN,
95 VIRTIO_DEVICE_ENDIAN_LITTLE,
96 VIRTIO_DEVICE_ENDIAN_BIG,
97 };
98
99 /**
100 * struct VirtIODevice - common VirtIO structure
101 * @name: name of the device
102 * @status: VirtIO Device Status field
103 *
104 */
105 struct VirtIODevice
106 {
107 DeviceState parent_obj;
108 const char *name;
109 uint8_t status;
110 uint8_t isr;
111 uint16_t queue_sel;
112 /**
113 * These fields represent a set of VirtIO features at various
114 * levels of the stack. @host_features indicates the complete
115 * feature set the VirtIO device can offer to the driver.
116 * @guest_features indicates which features the VirtIO driver has
117 * selected by writing to the feature register. Finally
118 * @backend_features represents everything supported by the
119 * backend (e.g. vhost) and could potentially be a subset of the
120 * total feature set offered by QEMU.
121 */
122 uint64_t host_features;
123 uint64_t guest_features;
124 uint64_t backend_features;
125
126 size_t config_len;
127 void *config;
128 uint16_t config_vector;
129 uint32_t generation;
130 int nvectors;
131 VirtQueue *vq;
132 MemoryListener listener;
133 uint16_t device_id;
134 /* @vm_running: current VM running state via virtio_vmstate_change() */
135 bool vm_running;
136 bool broken; /* device in invalid state, needs reset */
137 bool use_disabled_flag; /* allow use of 'disable' flag when needed */
138 bool disabled; /* device in temporarily disabled state */
139 /**
140 * @use_started: true if the @started flag should be used to check the
141 * current state of the VirtIO device. Otherwise status bits
142 * should be checked for a current status of the device.
143 * @use_started is only set via QMP and defaults to true for all
144 * modern machines (since 4.1).
145 */
146 bool use_started;
147 bool started;
148 bool start_on_kick; /* when virtio 1.0 feature has not been negotiated */
149 bool disable_legacy_check;
150 bool vhost_started;
151 VMChangeStateEntry *vmstate;
152 char *bus_name;
153 uint8_t device_endian;
154 /**
155 * @user_guest_notifier_mask: gate usage of ->guest_notifier_mask() callback.
156 * This is used to suppress the masking of guest updates for
157 * vhost-user devices which are asynchronous by design.
158 */
159 bool use_guest_notifier_mask;
160 AddressSpace *dma_as;
161 QLIST_HEAD(, VirtQueue) *vector_queues;
162 QTAILQ_ENTRY(VirtIODevice) next;
163 /**
164 * @config_notifier: the event notifier that handles config events
165 */
166 EventNotifier config_notifier;
167 bool device_iotlb_enabled;
168 };
169
170 struct VirtioDeviceClass {
171 /*< private >*/
172 DeviceClass parent;
173 /*< public >*/
174
175 /* This is what a VirtioDevice must implement */
176 DeviceRealize realize;
177 DeviceUnrealize unrealize;
178 uint64_t (*get_features)(VirtIODevice *vdev,
179 uint64_t requested_features,
180 Error **errp);
181 uint64_t (*bad_features)(VirtIODevice *vdev);
182 void (*set_features)(VirtIODevice *vdev, uint64_t val);
183 int (*validate_features)(VirtIODevice *vdev);
184 void (*get_config)(VirtIODevice *vdev, uint8_t *config);
185 void (*set_config)(VirtIODevice *vdev, const uint8_t *config);
186 void (*reset)(VirtIODevice *vdev);
187 void (*set_status)(VirtIODevice *vdev, uint8_t val);
188 /* Device must validate queue_index. */
189 void (*queue_reset)(VirtIODevice *vdev, uint32_t queue_index);
190 /* Device must validate queue_index. */
191 void (*queue_enable)(VirtIODevice *vdev, uint32_t queue_index);
192 /* For transitional devices, this is a bitmap of features
193 * that are only exposed on the legacy interface but not
194 * the modern one.
195 */
196 uint64_t legacy_features;
197 /* Test and clear event pending status.
198 * Should be called after unmask to avoid losing events.
199 * If backend does not support masking,
200 * must check in frontend instead.
201 */
202 bool (*guest_notifier_pending)(VirtIODevice *vdev, int n);
203 /* Mask/unmask events from this vq. Any events reported
204 * while masked will become pending.
205 * If backend does not support masking,
206 * must mask in frontend instead.
207 */
208 void (*guest_notifier_mask)(VirtIODevice *vdev, int n, bool mask);
209 int (*start_ioeventfd)(VirtIODevice *vdev);
210 void (*stop_ioeventfd)(VirtIODevice *vdev);
211 /* Saving and loading of a device; trying to deprecate save/load
212 * use vmsd for new devices.
213 */
214 void (*save)(VirtIODevice *vdev, QEMUFile *f);
215 int (*load)(VirtIODevice *vdev, QEMUFile *f, int version_id);
216 /* Post load hook in vmsd is called early while device is processed, and
217 * when VirtIODevice isn't fully initialized. Devices should use this instead,
218 * unless they specifically want to verify the migration stream as it's
219 * processed, e.g. for bounds checking.
220 */
221 int (*post_load)(VirtIODevice *vdev);
222 const VMStateDescription *vmsd;
223 bool (*primary_unplug_pending)(void *opaque);
224 struct vhost_dev *(*get_vhost)(VirtIODevice *vdev);
225 void (*toggle_device_iotlb)(VirtIODevice *vdev);
226 };
227
228 void virtio_instance_init_common(Object *proxy_obj, void *data,
229 size_t vdev_size, const char *vdev_name);
230
231 /**
232 * virtio_init() - initialise the common VirtIODevice structure
233 * @vdev: pointer to VirtIODevice
234 * @device_id: the VirtIO device ID (see virtio_ids.h)
235 * @config_size: size of the config space
236 */
237 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size);
238
239 void virtio_cleanup(VirtIODevice *vdev);
240
241 void virtio_error(VirtIODevice *vdev, const char *fmt, ...) G_GNUC_PRINTF(2, 3);
242
243 /* Set the child bus name. */
244 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name);
245
246 typedef void (*VirtIOHandleOutput)(VirtIODevice *, VirtQueue *);
247
248 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
249 VirtIOHandleOutput handle_output);
250
251 void virtio_del_queue(VirtIODevice *vdev, int n);
252
253 void virtio_delete_queue(VirtQueue *vq);
254
255 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
256 unsigned int len);
257 void virtqueue_flush(VirtQueue *vq, unsigned int count);
258 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
259 unsigned int len);
260 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
261 unsigned int len);
262 bool virtqueue_rewind(VirtQueue *vq, unsigned int num);
263 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
264 unsigned int len, unsigned int idx);
265
266 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem);
267 void *virtqueue_pop(VirtQueue *vq, size_t sz);
268 unsigned int virtqueue_drop_all(VirtQueue *vq);
269 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz);
270 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
271 VirtQueueElement *elem);
272 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
273 unsigned int out_bytes);
274 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
275 unsigned int *out_bytes,
276 unsigned max_in_bytes, unsigned max_out_bytes);
277
278 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq);
279 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq);
280
281 int virtio_save(VirtIODevice *vdev, QEMUFile *f);
282
283 extern const VMStateInfo virtio_vmstate_info;
284
285 #define VMSTATE_VIRTIO_DEVICE \
286 { \
287 .name = "virtio", \
288 .info = &virtio_vmstate_info, \
289 .flags = VMS_SINGLE, \
290 }
291
292 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id);
293
294 /**
295 * virtio_notify_config() - signal a change to device config
296 * @vdev: the virtio device
297 *
298 * Assuming the virtio device is up (VIRTIO_CONFIG_S_DRIVER_OK) this
299 * will trigger a guest interrupt and update the config version.
300 */
301 void virtio_notify_config(VirtIODevice *vdev);
302
303 bool virtio_queue_get_notification(VirtQueue *vq);
304 void virtio_queue_set_notification(VirtQueue *vq, int enable);
305
306 int virtio_queue_ready(VirtQueue *vq);
307
308 int virtio_queue_empty(VirtQueue *vq);
309
310 /* Host binding interface. */
311
312 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr);
313 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr);
314 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr);
315 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data);
316 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data);
317 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data);
318 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr);
319 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr);
320 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr);
321 void virtio_config_modern_writeb(VirtIODevice *vdev,
322 uint32_t addr, uint32_t data);
323 void virtio_config_modern_writew(VirtIODevice *vdev,
324 uint32_t addr, uint32_t data);
325 void virtio_config_modern_writel(VirtIODevice *vdev,
326 uint32_t addr, uint32_t data);
327 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr);
328 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n);
329 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num);
330 int virtio_queue_get_num(VirtIODevice *vdev, int n);
331 int virtio_queue_get_max_num(VirtIODevice *vdev, int n);
332 int virtio_get_num_queues(VirtIODevice *vdev);
333 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
334 hwaddr avail, hwaddr used);
335 void virtio_queue_update_rings(VirtIODevice *vdev, int n);
336 void virtio_init_region_cache(VirtIODevice *vdev, int n);
337 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align);
338 void virtio_queue_notify(VirtIODevice *vdev, int n);
339 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n);
340 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector);
341 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
342 MemoryRegion *mr, bool assign);
343 int virtio_set_status(VirtIODevice *vdev, uint8_t val);
344 void virtio_reset(void *opaque);
345 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index);
346 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index);
347 void virtio_update_irq(VirtIODevice *vdev);
348 int virtio_set_features(VirtIODevice *vdev, uint64_t val);
349
350 /* Base devices. */
351 typedef struct VirtIOBlkConf VirtIOBlkConf;
352 struct virtio_net_conf;
353 typedef struct virtio_serial_conf virtio_serial_conf;
354 typedef struct virtio_input_conf virtio_input_conf;
355 typedef struct VirtIOSCSIConf VirtIOSCSIConf;
356 typedef struct VirtIORNGConf VirtIORNGConf;
357
358 #define DEFINE_VIRTIO_COMMON_FEATURES(_state, _field) \
359 DEFINE_PROP_BIT64("indirect_desc", _state, _field, \
360 VIRTIO_RING_F_INDIRECT_DESC, true), \
361 DEFINE_PROP_BIT64("event_idx", _state, _field, \
362 VIRTIO_RING_F_EVENT_IDX, true), \
363 DEFINE_PROP_BIT64("notify_on_empty", _state, _field, \
364 VIRTIO_F_NOTIFY_ON_EMPTY, true), \
365 DEFINE_PROP_BIT64("any_layout", _state, _field, \
366 VIRTIO_F_ANY_LAYOUT, true), \
367 DEFINE_PROP_BIT64("iommu_platform", _state, _field, \
368 VIRTIO_F_IOMMU_PLATFORM, false), \
369 DEFINE_PROP_BIT64("packed", _state, _field, \
370 VIRTIO_F_RING_PACKED, false), \
371 DEFINE_PROP_BIT64("queue_reset", _state, _field, \
372 VIRTIO_F_RING_RESET, true)
373
374 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n);
375 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n);
376 bool virtio_queue_enabled(VirtIODevice *vdev, int n);
377 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n);
378 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n);
379 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n);
380 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n);
381 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n);
382 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n);
383 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
384 unsigned int idx);
385 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n);
386 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n);
387 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n);
388 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n);
389 uint16_t virtio_get_queue_index(VirtQueue *vq);
390 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq);
391 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
392 bool with_irqfd);
393 int virtio_device_start_ioeventfd(VirtIODevice *vdev);
394 int virtio_device_grab_ioeventfd(VirtIODevice *vdev);
395 void virtio_device_release_ioeventfd(VirtIODevice *vdev);
396 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev);
397 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq);
398 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled);
399 void virtio_queue_host_notifier_read(EventNotifier *n);
400 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx);
401 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx);
402 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx);
403 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector);
404 VirtQueue *virtio_vector_next_queue(VirtQueue *vq);
405 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev);
406 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
407 bool assign, bool with_irqfd);
408
virtio_add_feature(uint64_t * features,unsigned int fbit)409 static inline void virtio_add_feature(uint64_t *features, unsigned int fbit)
410 {
411 assert(fbit < 64);
412 *features |= (1ULL << fbit);
413 }
414
virtio_clear_feature(uint64_t * features,unsigned int fbit)415 static inline void virtio_clear_feature(uint64_t *features, unsigned int fbit)
416 {
417 assert(fbit < 64);
418 *features &= ~(1ULL << fbit);
419 }
420
virtio_has_feature(uint64_t features,unsigned int fbit)421 static inline bool virtio_has_feature(uint64_t features, unsigned int fbit)
422 {
423 assert(fbit < 64);
424 return !!(features & (1ULL << fbit));
425 }
426
virtio_vdev_has_feature(const VirtIODevice * vdev,unsigned int fbit)427 static inline bool virtio_vdev_has_feature(const VirtIODevice *vdev,
428 unsigned int fbit)
429 {
430 return virtio_has_feature(vdev->guest_features, fbit);
431 }
432
virtio_host_has_feature(VirtIODevice * vdev,unsigned int fbit)433 static inline bool virtio_host_has_feature(VirtIODevice *vdev,
434 unsigned int fbit)
435 {
436 return virtio_has_feature(vdev->host_features, fbit);
437 }
438
virtio_is_big_endian(VirtIODevice * vdev)439 static inline bool virtio_is_big_endian(VirtIODevice *vdev)
440 {
441 if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
442 assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
443 return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_BIG;
444 }
445 /* Devices conforming to VIRTIO 1.0 or later are always LE. */
446 return false;
447 }
448
449 /**
450 * virtio_device_started() - check if device started
451 * @vdev - the VirtIO device
452 * @status - the devices status bits
453 *
454 * Check if the device is started. For most modern machines this is
455 * tracked via the @vdev->started field (to support migration),
456 * otherwise we check for the final negotiated status bit that
457 * indicates everything is ready.
458 */
virtio_device_started(VirtIODevice * vdev,uint8_t status)459 static inline bool virtio_device_started(VirtIODevice *vdev, uint8_t status)
460 {
461 if (vdev->use_started) {
462 return vdev->started;
463 }
464
465 return status & VIRTIO_CONFIG_S_DRIVER_OK;
466 }
467
468 /**
469 * virtio_device_should_start() - check if device startable
470 * @vdev - the VirtIO device
471 * @status - the devices status bits
472 *
473 * This is similar to virtio_device_started() but also encapsulates a
474 * check on the VM status which would prevent a device starting
475 * anyway.
476 */
virtio_device_should_start(VirtIODevice * vdev,uint8_t status)477 static inline bool virtio_device_should_start(VirtIODevice *vdev, uint8_t status)
478 {
479 if (!vdev->vm_running) {
480 return false;
481 }
482
483 return virtio_device_started(vdev, status);
484 }
485
virtio_set_started(VirtIODevice * vdev,bool started)486 static inline void virtio_set_started(VirtIODevice *vdev, bool started)
487 {
488 if (started) {
489 vdev->start_on_kick = false;
490 }
491
492 if (vdev->use_started) {
493 vdev->started = started;
494 }
495 }
496
virtio_set_disabled(VirtIODevice * vdev,bool disable)497 static inline void virtio_set_disabled(VirtIODevice *vdev, bool disable)
498 {
499 if (vdev->use_disabled_flag) {
500 vdev->disabled = disable;
501 }
502 }
503
virtio_device_disabled(VirtIODevice * vdev)504 static inline bool virtio_device_disabled(VirtIODevice *vdev)
505 {
506 return unlikely(vdev->disabled || vdev->broken);
507 }
508
509 bool virtio_legacy_allowed(VirtIODevice *vdev);
510 bool virtio_legacy_check_disabled(VirtIODevice *vdev);
511
512 QEMUBH *virtio_bh_new_guarded_full(DeviceState *dev,
513 QEMUBHFunc *cb, void *opaque,
514 const char *name);
515 #define virtio_bh_new_guarded(dev, cb, opaque) \
516 virtio_bh_new_guarded_full((dev), (cb), (opaque), (stringify(cb)))
517
518 #endif
519