xref: /openbmc/qemu/include/hw/virtio/virtio.h (revision 9dbab31d)
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 
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     /* Element has been processed (VIRTIO_F_IN_ORDER) */
73     bool in_order_filled;
74     hwaddr *in_addr;
75     hwaddr *out_addr;
76     struct iovec *in_sg;
77     struct iovec *out_sg;
78 } VirtQueueElement;
79 
80 #define VIRTIO_QUEUE_MAX 1024
81 
82 #define VIRTIO_NO_VECTOR 0xffff
83 
84 /* special index value used internally for config irqs */
85 #define VIRTIO_CONFIG_IRQ_IDX -1
86 
87 #define TYPE_VIRTIO_DEVICE "virtio-device"
88 OBJECT_DECLARE_TYPE(VirtIODevice, VirtioDeviceClass, VIRTIO_DEVICE)
89 
90 typedef struct {
91     int virtio_bit;
92     const char *feature_desc;
93 } qmp_virtio_feature_map_t;
94 
95 enum virtio_device_endian {
96     VIRTIO_DEVICE_ENDIAN_UNKNOWN,
97     VIRTIO_DEVICE_ENDIAN_LITTLE,
98     VIRTIO_DEVICE_ENDIAN_BIG,
99 };
100 
101 /**
102  * struct VirtIODevice - common VirtIO structure
103  * @name: name of the device
104  * @status: VirtIO Device Status field
105  *
106  */
107 struct VirtIODevice
108 {
109     DeviceState parent_obj;
110     const char *name;
111     uint8_t status;
112     uint8_t isr;
113     uint16_t queue_sel;
114     /**
115      * These fields represent a set of VirtIO features at various
116      * levels of the stack. @host_features indicates the complete
117      * feature set the VirtIO device can offer to the driver.
118      * @guest_features indicates which features the VirtIO driver has
119      * selected by writing to the feature register. Finally
120      * @backend_features represents everything supported by the
121      * backend (e.g. vhost) and could potentially be a subset of the
122      * total feature set offered by QEMU.
123      */
124     uint64_t host_features;
125     uint64_t guest_features;
126     uint64_t backend_features;
127 
128     size_t config_len;
129     void *config;
130     uint16_t config_vector;
131     uint32_t generation;
132     int nvectors;
133     VirtQueue *vq;
134     MemoryListener listener;
135     uint16_t device_id;
136     /* @vm_running: current VM running state via virtio_vmstate_change() */
137     bool vm_running;
138     bool broken; /* device in invalid state, needs reset */
139     bool use_disabled_flag; /* allow use of 'disable' flag when needed */
140     bool disabled; /* device in temporarily disabled state */
141     /**
142      * @use_started: true if the @started flag should be used to check the
143      * current state of the VirtIO device. Otherwise status bits
144      * should be checked for a current status of the device.
145      * @use_started is only set via QMP and defaults to true for all
146      * modern machines (since 4.1).
147      */
148     bool use_started;
149     bool started;
150     bool start_on_kick; /* when virtio 1.0 feature has not been negotiated */
151     bool disable_legacy_check;
152     bool vhost_started;
153     VMChangeStateEntry *vmstate;
154     char *bus_name;
155     uint8_t device_endian;
156     /**
157      * @user_guest_notifier_mask: gate usage of ->guest_notifier_mask() callback.
158      * This is used to suppress the masking of guest updates for
159      * vhost-user devices which are asynchronous by design.
160      */
161     bool use_guest_notifier_mask;
162     AddressSpace *dma_as;
163     QLIST_HEAD(, VirtQueue) *vector_queues;
164     QTAILQ_ENTRY(VirtIODevice) next;
165     /**
166      * @config_notifier: the event notifier that handles config events
167      */
168     EventNotifier config_notifier;
169     bool device_iotlb_enabled;
170 };
171 
172 struct VirtioDeviceClass {
173     /*< private >*/
174     DeviceClass parent;
175     /*< public >*/
176 
177     /* This is what a VirtioDevice must implement */
178     DeviceRealize realize;
179     DeviceUnrealize unrealize;
180     uint64_t (*get_features)(VirtIODevice *vdev,
181                              uint64_t requested_features,
182                              Error **errp);
183     uint64_t (*bad_features)(VirtIODevice *vdev);
184     void (*set_features)(VirtIODevice *vdev, uint64_t val);
185     int (*validate_features)(VirtIODevice *vdev);
186     void (*get_config)(VirtIODevice *vdev, uint8_t *config);
187     void (*set_config)(VirtIODevice *vdev, const uint8_t *config);
188     void (*reset)(VirtIODevice *vdev);
189     void (*set_status)(VirtIODevice *vdev, uint8_t val);
190     /* Device must validate queue_index.  */
191     void (*queue_reset)(VirtIODevice *vdev, uint32_t queue_index);
192     /* Device must validate queue_index.  */
193     void (*queue_enable)(VirtIODevice *vdev, uint32_t queue_index);
194     /* For transitional devices, this is a bitmap of features
195      * that are only exposed on the legacy interface but not
196      * the modern one.
197      */
198     uint64_t legacy_features;
199     /* Test and clear event pending status.
200      * Should be called after unmask to avoid losing events.
201      * If backend does not support masking,
202      * must check in frontend instead.
203      */
204     bool (*guest_notifier_pending)(VirtIODevice *vdev, int n);
205     /* Mask/unmask events from this vq. Any events reported
206      * while masked will become pending.
207      * If backend does not support masking,
208      * must mask in frontend instead.
209      */
210     void (*guest_notifier_mask)(VirtIODevice *vdev, int n, bool mask);
211     int (*start_ioeventfd)(VirtIODevice *vdev);
212     void (*stop_ioeventfd)(VirtIODevice *vdev);
213     /* Saving and loading of a device; trying to deprecate save/load
214      * use vmsd for new devices.
215      */
216     void (*save)(VirtIODevice *vdev, QEMUFile *f);
217     int (*load)(VirtIODevice *vdev, QEMUFile *f, int version_id);
218     /* Post load hook in vmsd is called early while device is processed, and
219      * when VirtIODevice isn't fully initialized.  Devices should use this instead,
220      * unless they specifically want to verify the migration stream as it's
221      * processed, e.g. for bounds checking.
222      */
223     int (*post_load)(VirtIODevice *vdev);
224     const VMStateDescription *vmsd;
225     bool (*primary_unplug_pending)(void *opaque);
226     struct vhost_dev *(*get_vhost)(VirtIODevice *vdev);
227     void (*toggle_device_iotlb)(VirtIODevice *vdev);
228 };
229 
230 void virtio_instance_init_common(Object *proxy_obj, void *data,
231                                  size_t vdev_size, const char *vdev_name);
232 
233 /**
234  * virtio_init() - initialise the common VirtIODevice structure
235  * @vdev: pointer to VirtIODevice
236  * @device_id: the VirtIO device ID (see virtio_ids.h)
237  * @config_size: size of the config space
238  */
239 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size);
240 
241 void virtio_cleanup(VirtIODevice *vdev);
242 
243 void virtio_error(VirtIODevice *vdev, const char *fmt, ...) G_GNUC_PRINTF(2, 3);
244 
245 /* Set the child bus name. */
246 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name);
247 
248 typedef void (*VirtIOHandleOutput)(VirtIODevice *, VirtQueue *);
249 
250 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
251                             VirtIOHandleOutput handle_output);
252 
253 void virtio_del_queue(VirtIODevice *vdev, int n);
254 
255 void virtio_delete_queue(VirtQueue *vq);
256 
257 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
258                     unsigned int len);
259 void virtqueue_flush(VirtQueue *vq, unsigned int count);
260 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
261                               unsigned int len);
262 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
263                      unsigned int len);
264 bool virtqueue_rewind(VirtQueue *vq, unsigned int num);
265 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
266                     unsigned int len, unsigned int idx);
267 
268 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem);
269 void *virtqueue_pop(VirtQueue *vq, size_t sz);
270 unsigned int virtqueue_drop_all(VirtQueue *vq);
271 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz);
272 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
273                                 VirtQueueElement *elem);
274 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
275                           unsigned int out_bytes);
276 /**
277  * Return <0 on error or an opaque >=0 to pass to
278  * virtio_queue_enable_notification_and_check on success.
279  */
280 int virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
281                               unsigned int *out_bytes, unsigned max_in_bytes,
282                               unsigned max_out_bytes);
283 
284 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq);
285 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq);
286 
287 int virtio_save(VirtIODevice *vdev, QEMUFile *f);
288 
289 extern const VMStateInfo virtio_vmstate_info;
290 
291 #define VMSTATE_VIRTIO_DEVICE \
292     {                                         \
293         .name = "virtio",                     \
294         .info = &virtio_vmstate_info,         \
295         .flags = VMS_SINGLE,                  \
296     }
297 
298 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id);
299 
300 /**
301  * virtio_notify_config() - signal a change to device config
302  * @vdev: the virtio device
303  *
304  * Assuming the virtio device is up (VIRTIO_CONFIG_S_DRIVER_OK) this
305  * will trigger a guest interrupt and update the config version.
306  */
307 void virtio_notify_config(VirtIODevice *vdev);
308 
309 bool virtio_queue_get_notification(VirtQueue *vq);
310 void virtio_queue_set_notification(VirtQueue *vq, int enable);
311 
312 int virtio_queue_ready(VirtQueue *vq);
313 
314 int virtio_queue_empty(VirtQueue *vq);
315 
316 /**
317  * Enable notification and check whether guest has added some
318  * buffers since last call to virtqueue_get_avail_bytes.
319  *
320  * @opaque: value returned from virtqueue_get_avail_bytes
321  */
322 bool virtio_queue_enable_notification_and_check(VirtQueue *vq,
323                                                 int opaque);
324 
325 void virtio_queue_set_shadow_avail_idx(VirtQueue *vq, uint16_t idx);
326 
327 /* Host binding interface.  */
328 
329 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr);
330 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr);
331 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr);
332 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data);
333 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data);
334 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data);
335 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr);
336 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr);
337 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr);
338 void virtio_config_modern_writeb(VirtIODevice *vdev,
339                                  uint32_t addr, uint32_t data);
340 void virtio_config_modern_writew(VirtIODevice *vdev,
341                                  uint32_t addr, uint32_t data);
342 void virtio_config_modern_writel(VirtIODevice *vdev,
343                                  uint32_t addr, uint32_t data);
344 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr);
345 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n);
346 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num);
347 int virtio_queue_get_num(VirtIODevice *vdev, int n);
348 int virtio_queue_get_max_num(VirtIODevice *vdev, int n);
349 int virtio_get_num_queues(VirtIODevice *vdev);
350 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
351                             hwaddr avail, hwaddr used);
352 void virtio_queue_update_rings(VirtIODevice *vdev, int n);
353 void virtio_init_region_cache(VirtIODevice *vdev, int n);
354 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align);
355 void virtio_queue_notify(VirtIODevice *vdev, int n);
356 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n);
357 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector);
358 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
359                                       MemoryRegion *mr, bool assign);
360 int virtio_set_status(VirtIODevice *vdev, uint8_t val);
361 void virtio_reset(void *opaque);
362 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index);
363 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index);
364 void virtio_update_irq(VirtIODevice *vdev);
365 int virtio_set_features(VirtIODevice *vdev, uint64_t val);
366 
367 /* Base devices.  */
368 typedef struct VirtIOBlkConf VirtIOBlkConf;
369 struct virtio_net_conf;
370 typedef struct virtio_serial_conf virtio_serial_conf;
371 typedef struct virtio_input_conf virtio_input_conf;
372 typedef struct VirtIOSCSIConf VirtIOSCSIConf;
373 typedef struct VirtIORNGConf VirtIORNGConf;
374 
375 #define DEFINE_VIRTIO_COMMON_FEATURES(_state, _field) \
376     DEFINE_PROP_BIT64("indirect_desc", _state, _field,    \
377                       VIRTIO_RING_F_INDIRECT_DESC, true), \
378     DEFINE_PROP_BIT64("event_idx", _state, _field,        \
379                       VIRTIO_RING_F_EVENT_IDX, true),     \
380     DEFINE_PROP_BIT64("notify_on_empty", _state, _field,  \
381                       VIRTIO_F_NOTIFY_ON_EMPTY, true), \
382     DEFINE_PROP_BIT64("any_layout", _state, _field, \
383                       VIRTIO_F_ANY_LAYOUT, true), \
384     DEFINE_PROP_BIT64("iommu_platform", _state, _field, \
385                       VIRTIO_F_IOMMU_PLATFORM, false), \
386     DEFINE_PROP_BIT64("packed", _state, _field, \
387                       VIRTIO_F_RING_PACKED, false), \
388     DEFINE_PROP_BIT64("queue_reset", _state, _field, \
389                       VIRTIO_F_RING_RESET, true), \
390     DEFINE_PROP_BIT64("in_order", _state, _field, \
391                       VIRTIO_F_IN_ORDER, false)
392 
393 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n);
394 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n);
395 bool virtio_queue_enabled(VirtIODevice *vdev, int n);
396 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n);
397 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n);
398 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n);
399 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n);
400 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n);
401 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n);
402 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
403                                      unsigned int idx);
404 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n);
405 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n);
406 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n);
407 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n);
408 uint16_t virtio_get_queue_index(VirtQueue *vq);
409 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq);
410 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
411                                                 bool with_irqfd);
412 int virtio_device_start_ioeventfd(VirtIODevice *vdev);
413 int virtio_device_grab_ioeventfd(VirtIODevice *vdev);
414 void virtio_device_release_ioeventfd(VirtIODevice *vdev);
415 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev);
416 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq);
417 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled);
418 void virtio_queue_host_notifier_read(EventNotifier *n);
419 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx);
420 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx);
421 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx);
422 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector);
423 VirtQueue *virtio_vector_next_queue(VirtQueue *vq);
424 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev);
425 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
426                                                  bool assign, bool with_irqfd);
427 
428 static inline void virtio_add_feature(uint64_t *features, unsigned int fbit)
429 {
430     assert(fbit < 64);
431     *features |= (1ULL << fbit);
432 }
433 
434 static inline void virtio_clear_feature(uint64_t *features, unsigned int fbit)
435 {
436     assert(fbit < 64);
437     *features &= ~(1ULL << fbit);
438 }
439 
440 static inline bool virtio_has_feature(uint64_t features, unsigned int fbit)
441 {
442     assert(fbit < 64);
443     return !!(features & (1ULL << fbit));
444 }
445 
446 static inline bool virtio_vdev_has_feature(const VirtIODevice *vdev,
447                                            unsigned int fbit)
448 {
449     return virtio_has_feature(vdev->guest_features, fbit);
450 }
451 
452 static inline bool virtio_host_has_feature(VirtIODevice *vdev,
453                                            unsigned int fbit)
454 {
455     return virtio_has_feature(vdev->host_features, fbit);
456 }
457 
458 static inline bool virtio_is_big_endian(VirtIODevice *vdev)
459 {
460     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
461         assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
462         return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_BIG;
463     }
464     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
465     return false;
466 }
467 
468 /**
469  * virtio_device_started() - check if device started
470  * @vdev - the VirtIO device
471  * @status - the devices status bits
472  *
473  * Check if the device is started. For most modern machines this is
474  * tracked via the @vdev->started field (to support migration),
475  * otherwise we check for the final negotiated status bit that
476  * indicates everything is ready.
477  */
478 static inline bool virtio_device_started(VirtIODevice *vdev, uint8_t status)
479 {
480     if (vdev->use_started) {
481         return vdev->started;
482     }
483 
484     return status & VIRTIO_CONFIG_S_DRIVER_OK;
485 }
486 
487 /**
488  * virtio_device_should_start() - check if device startable
489  * @vdev - the VirtIO device
490  * @status - the devices status bits
491  *
492  * This is similar to virtio_device_started() but ignores vdev->started
493  * and also encapsulates a check on the VM status which would prevent a
494  * device from starting anyway.
495  */
496 static inline bool virtio_device_should_start(VirtIODevice *vdev, uint8_t status)
497 {
498     if (!vdev->vm_running) {
499         return false;
500     }
501 
502     return status & VIRTIO_CONFIG_S_DRIVER_OK;
503 }
504 
505 static inline void virtio_set_started(VirtIODevice *vdev, bool started)
506 {
507     if (started) {
508         vdev->start_on_kick = false;
509     }
510 
511     if (vdev->use_started) {
512         vdev->started = started;
513     }
514 }
515 
516 static inline void virtio_set_disabled(VirtIODevice *vdev, bool disable)
517 {
518     if (vdev->use_disabled_flag) {
519         vdev->disabled = disable;
520     }
521 }
522 
523 static inline bool virtio_device_disabled(VirtIODevice *vdev)
524 {
525     return unlikely(vdev->disabled || vdev->broken);
526 }
527 
528 bool virtio_legacy_allowed(VirtIODevice *vdev);
529 bool virtio_legacy_check_disabled(VirtIODevice *vdev);
530 
531 QEMUBH *virtio_bh_new_guarded_full(DeviceState *dev,
532                                    QEMUBHFunc *cb, void *opaque,
533                                    const char *name);
534 #define virtio_bh_new_guarded(dev, cb, opaque) \
535     virtio_bh_new_guarded_full((dev), (cb), (opaque), (stringify(cb)))
536 
537 #endif
538