xref: /openbmc/qemu/include/hw/virtio/virtio.h (revision bac4711b)
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 
26 /*
27  * A guest should never accept this. It implies negotiation is broken
28  * between the driver frontend and the device. This bit is re-used for
29  * vhost-user to advertise VHOST_USER_F_PROTOCOL_FEATURES between QEMU
30  * and a vhost-user backend.
31  */
32 #define VIRTIO_F_BAD_FEATURE 30
33 
34 #define VIRTIO_LEGACY_FEATURES ((0x1ULL << VIRTIO_F_BAD_FEATURE) | \
35                                 (0x1ULL << VIRTIO_F_NOTIFY_ON_EMPTY) | \
36                                 (0x1ULL << VIRTIO_F_ANY_LAYOUT))
37 
38 struct VirtQueue;
39 
40 static inline hwaddr vring_align(hwaddr addr,
41                                              unsigned long align)
42 {
43     return QEMU_ALIGN_UP(addr, align);
44 }
45 
46 typedef struct VirtIOFeature {
47     uint64_t flags;
48     size_t end;
49 } VirtIOFeature;
50 
51 typedef struct VirtIOConfigSizeParams {
52     size_t min_size;
53     size_t max_size;
54     const VirtIOFeature *feature_sizes;
55 } VirtIOConfigSizeParams;
56 
57 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
58                               uint64_t host_features);
59 
60 typedef struct VirtQueue VirtQueue;
61 
62 #define VIRTQUEUE_MAX_SIZE 1024
63 
64 typedef struct VirtQueueElement
65 {
66     unsigned int index;
67     unsigned int len;
68     unsigned int ndescs;
69     unsigned int out_num;
70     unsigned int in_num;
71     hwaddr *in_addr;
72     hwaddr *out_addr;
73     struct iovec *in_sg;
74     struct iovec *out_sg;
75 } VirtQueueElement;
76 
77 #define VIRTIO_QUEUE_MAX 1024
78 
79 #define VIRTIO_NO_VECTOR 0xffff
80 
81 /* special index value used internally for config irqs */
82 #define VIRTIO_CONFIG_IRQ_IDX -1
83 
84 #define TYPE_VIRTIO_DEVICE "virtio-device"
85 OBJECT_DECLARE_TYPE(VirtIODevice, VirtioDeviceClass, VIRTIO_DEVICE)
86 
87 typedef struct {
88     int virtio_bit;
89     const char *feature_desc;
90 } qmp_virtio_feature_map_t;
91 
92 enum virtio_device_endian {
93     VIRTIO_DEVICE_ENDIAN_UNKNOWN,
94     VIRTIO_DEVICE_ENDIAN_LITTLE,
95     VIRTIO_DEVICE_ENDIAN_BIG,
96 };
97 
98 /**
99  * struct VirtIODevice - common VirtIO structure
100  * @name: name of the device
101  * @status: VirtIO Device Status field
102  *
103  */
104 struct VirtIODevice
105 {
106     DeviceState parent_obj;
107     const char *name;
108     uint8_t status;
109     uint8_t isr;
110     uint16_t queue_sel;
111     /**
112      * These fields represent a set of VirtIO features at various
113      * levels of the stack. @host_features indicates the complete
114      * feature set the VirtIO device can offer to the driver.
115      * @guest_features indicates which features the VirtIO driver has
116      * selected by writing to the feature register. Finally
117      * @backend_features represents everything supported by the
118      * backend (e.g. vhost) and could potentially be a subset of the
119      * total feature set offered by QEMU.
120      */
121     uint64_t host_features;
122     uint64_t guest_features;
123     uint64_t backend_features;
124 
125     size_t config_len;
126     void *config;
127     uint16_t config_vector;
128     uint32_t generation;
129     int nvectors;
130     VirtQueue *vq;
131     MemoryListener listener;
132     uint16_t device_id;
133     /* @vm_running: current VM running state via virtio_vmstate_change() */
134     bool vm_running;
135     bool broken; /* device in invalid state, needs reset */
136     bool use_disabled_flag; /* allow use of 'disable' flag when needed */
137     bool disabled; /* device in temporarily disabled state */
138     /**
139      * @use_started: true if the @started flag should be used to check the
140      * current state of the VirtIO device. Otherwise status bits
141      * should be checked for a current status of the device.
142      * @use_started is only set via QMP and defaults to true for all
143      * modern machines (since 4.1).
144      */
145     bool use_started;
146     bool started;
147     bool start_on_kick; /* when virtio 1.0 feature has not been negotiated */
148     bool disable_legacy_check;
149     bool vhost_started;
150     VMChangeStateEntry *vmstate;
151     char *bus_name;
152     uint8_t device_endian;
153     bool use_guest_notifier_mask;
154     AddressSpace *dma_as;
155     QLIST_HEAD(, VirtQueue) *vector_queues;
156     QTAILQ_ENTRY(VirtIODevice) next;
157     EventNotifier config_notifier;
158     bool device_iotlb_enabled;
159 };
160 
161 struct VirtioDeviceClass {
162     /*< private >*/
163     DeviceClass parent;
164     /*< public >*/
165 
166     /* This is what a VirtioDevice must implement */
167     DeviceRealize realize;
168     DeviceUnrealize unrealize;
169     uint64_t (*get_features)(VirtIODevice *vdev,
170                              uint64_t requested_features,
171                              Error **errp);
172     uint64_t (*bad_features)(VirtIODevice *vdev);
173     void (*set_features)(VirtIODevice *vdev, uint64_t val);
174     int (*validate_features)(VirtIODevice *vdev);
175     void (*get_config)(VirtIODevice *vdev, uint8_t *config);
176     void (*set_config)(VirtIODevice *vdev, const uint8_t *config);
177     void (*reset)(VirtIODevice *vdev);
178     void (*set_status)(VirtIODevice *vdev, uint8_t val);
179     /* Device must validate queue_index.  */
180     void (*queue_reset)(VirtIODevice *vdev, uint32_t queue_index);
181     /* Device must validate queue_index.  */
182     void (*queue_enable)(VirtIODevice *vdev, uint32_t queue_index);
183     /* For transitional devices, this is a bitmap of features
184      * that are only exposed on the legacy interface but not
185      * the modern one.
186      */
187     uint64_t legacy_features;
188     /* Test and clear event pending status.
189      * Should be called after unmask to avoid losing events.
190      * If backend does not support masking,
191      * must check in frontend instead.
192      */
193     bool (*guest_notifier_pending)(VirtIODevice *vdev, int n);
194     /* Mask/unmask events from this vq. Any events reported
195      * while masked will become pending.
196      * If backend does not support masking,
197      * must mask in frontend instead.
198      */
199     void (*guest_notifier_mask)(VirtIODevice *vdev, int n, bool mask);
200     int (*start_ioeventfd)(VirtIODevice *vdev);
201     void (*stop_ioeventfd)(VirtIODevice *vdev);
202     /* Saving and loading of a device; trying to deprecate save/load
203      * use vmsd for new devices.
204      */
205     void (*save)(VirtIODevice *vdev, QEMUFile *f);
206     int (*load)(VirtIODevice *vdev, QEMUFile *f, int version_id);
207     /* Post load hook in vmsd is called early while device is processed, and
208      * when VirtIODevice isn't fully initialized.  Devices should use this instead,
209      * unless they specifically want to verify the migration stream as it's
210      * processed, e.g. for bounds checking.
211      */
212     int (*post_load)(VirtIODevice *vdev);
213     const VMStateDescription *vmsd;
214     bool (*primary_unplug_pending)(void *opaque);
215     struct vhost_dev *(*get_vhost)(VirtIODevice *vdev);
216     void (*toggle_device_iotlb)(VirtIODevice *vdev);
217 };
218 
219 void virtio_instance_init_common(Object *proxy_obj, void *data,
220                                  size_t vdev_size, const char *vdev_name);
221 
222 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size);
223 
224 void virtio_cleanup(VirtIODevice *vdev);
225 
226 void virtio_error(VirtIODevice *vdev, const char *fmt, ...) G_GNUC_PRINTF(2, 3);
227 
228 /* Set the child bus name. */
229 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name);
230 
231 typedef void (*VirtIOHandleOutput)(VirtIODevice *, VirtQueue *);
232 
233 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
234                             VirtIOHandleOutput handle_output);
235 
236 void virtio_del_queue(VirtIODevice *vdev, int n);
237 
238 void virtio_delete_queue(VirtQueue *vq);
239 
240 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
241                     unsigned int len);
242 void virtqueue_flush(VirtQueue *vq, unsigned int count);
243 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
244                               unsigned int len);
245 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
246                      unsigned int len);
247 bool virtqueue_rewind(VirtQueue *vq, unsigned int num);
248 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
249                     unsigned int len, unsigned int idx);
250 
251 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem);
252 void *virtqueue_pop(VirtQueue *vq, size_t sz);
253 unsigned int virtqueue_drop_all(VirtQueue *vq);
254 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz);
255 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
256                                 VirtQueueElement *elem);
257 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
258                           unsigned int out_bytes);
259 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
260                                unsigned int *out_bytes,
261                                unsigned max_in_bytes, unsigned max_out_bytes);
262 
263 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq);
264 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq);
265 
266 int virtio_save(VirtIODevice *vdev, QEMUFile *f);
267 
268 extern const VMStateInfo virtio_vmstate_info;
269 
270 #define VMSTATE_VIRTIO_DEVICE \
271     {                                         \
272         .name = "virtio",                     \
273         .info = &virtio_vmstate_info,         \
274         .flags = VMS_SINGLE,                  \
275     }
276 
277 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id);
278 
279 void virtio_notify_config(VirtIODevice *vdev);
280 
281 bool virtio_queue_get_notification(VirtQueue *vq);
282 void virtio_queue_set_notification(VirtQueue *vq, int enable);
283 
284 int virtio_queue_ready(VirtQueue *vq);
285 
286 int virtio_queue_empty(VirtQueue *vq);
287 
288 /* Host binding interface.  */
289 
290 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr);
291 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr);
292 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr);
293 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data);
294 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data);
295 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data);
296 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr);
297 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr);
298 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr);
299 void virtio_config_modern_writeb(VirtIODevice *vdev,
300                                  uint32_t addr, uint32_t data);
301 void virtio_config_modern_writew(VirtIODevice *vdev,
302                                  uint32_t addr, uint32_t data);
303 void virtio_config_modern_writel(VirtIODevice *vdev,
304                                  uint32_t addr, uint32_t data);
305 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr);
306 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n);
307 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num);
308 int virtio_queue_get_num(VirtIODevice *vdev, int n);
309 int virtio_queue_get_max_num(VirtIODevice *vdev, int n);
310 int virtio_get_num_queues(VirtIODevice *vdev);
311 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
312                             hwaddr avail, hwaddr used);
313 void virtio_queue_update_rings(VirtIODevice *vdev, int n);
314 void virtio_init_region_cache(VirtIODevice *vdev, int n);
315 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align);
316 void virtio_queue_notify(VirtIODevice *vdev, int n);
317 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n);
318 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector);
319 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
320                                       MemoryRegion *mr, bool assign);
321 int virtio_set_status(VirtIODevice *vdev, uint8_t val);
322 void virtio_reset(void *opaque);
323 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index);
324 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index);
325 void virtio_update_irq(VirtIODevice *vdev);
326 int virtio_set_features(VirtIODevice *vdev, uint64_t val);
327 
328 /* Base devices.  */
329 typedef struct VirtIOBlkConf VirtIOBlkConf;
330 struct virtio_net_conf;
331 typedef struct virtio_serial_conf virtio_serial_conf;
332 typedef struct virtio_input_conf virtio_input_conf;
333 typedef struct VirtIOSCSIConf VirtIOSCSIConf;
334 typedef struct VirtIORNGConf VirtIORNGConf;
335 
336 #define DEFINE_VIRTIO_COMMON_FEATURES(_state, _field) \
337     DEFINE_PROP_BIT64("indirect_desc", _state, _field,    \
338                       VIRTIO_RING_F_INDIRECT_DESC, true), \
339     DEFINE_PROP_BIT64("event_idx", _state, _field,        \
340                       VIRTIO_RING_F_EVENT_IDX, true),     \
341     DEFINE_PROP_BIT64("notify_on_empty", _state, _field,  \
342                       VIRTIO_F_NOTIFY_ON_EMPTY, true), \
343     DEFINE_PROP_BIT64("any_layout", _state, _field, \
344                       VIRTIO_F_ANY_LAYOUT, true), \
345     DEFINE_PROP_BIT64("iommu_platform", _state, _field, \
346                       VIRTIO_F_IOMMU_PLATFORM, false), \
347     DEFINE_PROP_BIT64("packed", _state, _field, \
348                       VIRTIO_F_RING_PACKED, false), \
349     DEFINE_PROP_BIT64("queue_reset", _state, _field, \
350                       VIRTIO_F_RING_RESET, true)
351 
352 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n);
353 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n);
354 bool virtio_queue_enabled(VirtIODevice *vdev, int n);
355 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n);
356 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n);
357 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n);
358 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n);
359 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n);
360 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n);
361 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
362                                      unsigned int idx);
363 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n);
364 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n);
365 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n);
366 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n);
367 uint16_t virtio_get_queue_index(VirtQueue *vq);
368 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq);
369 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
370                                                 bool with_irqfd);
371 int virtio_device_start_ioeventfd(VirtIODevice *vdev);
372 int virtio_device_grab_ioeventfd(VirtIODevice *vdev);
373 void virtio_device_release_ioeventfd(VirtIODevice *vdev);
374 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev);
375 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq);
376 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled);
377 void virtio_queue_host_notifier_read(EventNotifier *n);
378 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx);
379 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx);
380 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx);
381 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector);
382 VirtQueue *virtio_vector_next_queue(VirtQueue *vq);
383 EventNotifier *virtio_config_get_guest_notifier(VirtIODevice *vdev);
384 void virtio_config_set_guest_notifier_fd_handler(VirtIODevice *vdev,
385                                                  bool assign, bool with_irqfd);
386 
387 static inline void virtio_add_feature(uint64_t *features, unsigned int fbit)
388 {
389     assert(fbit < 64);
390     *features |= (1ULL << fbit);
391 }
392 
393 static inline void virtio_clear_feature(uint64_t *features, unsigned int fbit)
394 {
395     assert(fbit < 64);
396     *features &= ~(1ULL << fbit);
397 }
398 
399 static inline bool virtio_has_feature(uint64_t features, unsigned int fbit)
400 {
401     assert(fbit < 64);
402     return !!(features & (1ULL << fbit));
403 }
404 
405 static inline bool virtio_vdev_has_feature(const VirtIODevice *vdev,
406                                            unsigned int fbit)
407 {
408     return virtio_has_feature(vdev->guest_features, fbit);
409 }
410 
411 static inline bool virtio_host_has_feature(VirtIODevice *vdev,
412                                            unsigned int fbit)
413 {
414     return virtio_has_feature(vdev->host_features, fbit);
415 }
416 
417 static inline bool virtio_is_big_endian(VirtIODevice *vdev)
418 {
419     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
420         assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
421         return vdev->device_endian == VIRTIO_DEVICE_ENDIAN_BIG;
422     }
423     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
424     return false;
425 }
426 
427 /**
428  * virtio_device_started() - check if device started
429  * @vdev - the VirtIO device
430  * @status - the devices status bits
431  *
432  * Check if the device is started. For most modern machines this is
433  * tracked via the @vdev->started field (to support migration),
434  * otherwise we check for the final negotiated status bit that
435  * indicates everything is ready.
436  */
437 static inline bool virtio_device_started(VirtIODevice *vdev, uint8_t status)
438 {
439     if (vdev->use_started) {
440         return vdev->started;
441     }
442 
443     return status & VIRTIO_CONFIG_S_DRIVER_OK;
444 }
445 
446 /**
447  * virtio_device_should_start() - check if device startable
448  * @vdev - the VirtIO device
449  * @status - the devices status bits
450  *
451  * This is similar to virtio_device_started() but also encapsulates a
452  * check on the VM status which would prevent a device starting
453  * anyway.
454  */
455 static inline bool virtio_device_should_start(VirtIODevice *vdev, uint8_t status)
456 {
457     if (!vdev->vm_running) {
458         return false;
459     }
460 
461     return virtio_device_started(vdev, status);
462 }
463 
464 static inline void virtio_set_started(VirtIODevice *vdev, bool started)
465 {
466     if (started) {
467         vdev->start_on_kick = false;
468     }
469 
470     if (vdev->use_started) {
471         vdev->started = started;
472     }
473 }
474 
475 static inline void virtio_set_disabled(VirtIODevice *vdev, bool disable)
476 {
477     if (vdev->use_disabled_flag) {
478         vdev->disabled = disable;
479     }
480 }
481 
482 static inline bool virtio_device_disabled(VirtIODevice *vdev)
483 {
484     return unlikely(vdev->disabled || vdev->broken);
485 }
486 
487 bool virtio_legacy_allowed(VirtIODevice *vdev);
488 bool virtio_legacy_check_disabled(VirtIODevice *vdev);
489 
490 #endif
491