xref: /openbmc/qemu/hw/virtio/virtio.c (revision 954a6c4f)
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 #include "qemu/osdep.h"
15 #include "qapi/error.h"
16 #include "qapi/qmp/qdict.h"
17 #include "qapi/qapi-commands-virtio.h"
18 #include "qapi/qapi-commands-qom.h"
19 #include "qapi/qapi-visit-virtio.h"
20 #include "qapi/qmp/qjson.h"
21 #include "cpu.h"
22 #include "trace.h"
23 #include "qemu/error-report.h"
24 #include "qemu/log.h"
25 #include "qemu/main-loop.h"
26 #include "qemu/module.h"
27 #include "qom/object_interfaces.h"
28 #include "hw/virtio/virtio.h"
29 #include "migration/qemu-file-types.h"
30 #include "qemu/atomic.h"
31 #include "hw/virtio/virtio-bus.h"
32 #include "hw/qdev-properties.h"
33 #include "hw/virtio/virtio-access.h"
34 #include "sysemu/dma.h"
35 #include "sysemu/runstate.h"
36 #include "standard-headers/linux/virtio_ids.h"
37 #include "standard-headers/linux/vhost_types.h"
38 #include "standard-headers/linux/virtio_blk.h"
39 #include "standard-headers/linux/virtio_console.h"
40 #include "standard-headers/linux/virtio_gpu.h"
41 #include "standard-headers/linux/virtio_net.h"
42 #include "standard-headers/linux/virtio_scsi.h"
43 #include "standard-headers/linux/virtio_i2c.h"
44 #include "standard-headers/linux/virtio_balloon.h"
45 #include "standard-headers/linux/virtio_iommu.h"
46 #include "standard-headers/linux/virtio_mem.h"
47 #include "standard-headers/linux/virtio_vsock.h"
48 #include CONFIG_DEVICES
49 
50 /* QAPI list of realized VirtIODevices */
51 static QTAILQ_HEAD(, VirtIODevice) virtio_list;
52 
53 /*
54  * Maximum size of virtio device config space
55  */
56 #define VHOST_USER_MAX_CONFIG_SIZE 256
57 
58 #define FEATURE_ENTRY(name, desc) (qmp_virtio_feature_map_t) \
59     { .virtio_bit = name, .feature_desc = desc }
60 
61 enum VhostUserProtocolFeature {
62     VHOST_USER_PROTOCOL_F_MQ = 0,
63     VHOST_USER_PROTOCOL_F_LOG_SHMFD = 1,
64     VHOST_USER_PROTOCOL_F_RARP = 2,
65     VHOST_USER_PROTOCOL_F_REPLY_ACK = 3,
66     VHOST_USER_PROTOCOL_F_NET_MTU = 4,
67     VHOST_USER_PROTOCOL_F_SLAVE_REQ = 5,
68     VHOST_USER_PROTOCOL_F_CROSS_ENDIAN = 6,
69     VHOST_USER_PROTOCOL_F_CRYPTO_SESSION = 7,
70     VHOST_USER_PROTOCOL_F_PAGEFAULT = 8,
71     VHOST_USER_PROTOCOL_F_CONFIG = 9,
72     VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD = 10,
73     VHOST_USER_PROTOCOL_F_HOST_NOTIFIER = 11,
74     VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD = 12,
75     VHOST_USER_PROTOCOL_F_RESET_DEVICE = 13,
76     VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS = 14,
77     VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS = 15,
78     VHOST_USER_PROTOCOL_F_MAX
79 };
80 
81 /* Virtio transport features mapping */
82 static qmp_virtio_feature_map_t virtio_transport_map[] = {
83     /* Virtio device transport features */
84 #ifndef VIRTIO_CONFIG_NO_LEGACY
85     FEATURE_ENTRY(VIRTIO_F_NOTIFY_ON_EMPTY, \
86             "VIRTIO_F_NOTIFY_ON_EMPTY: Notify when device runs out of avail. "
87             "descs. on VQ"),
88     FEATURE_ENTRY(VIRTIO_F_ANY_LAYOUT, \
89             "VIRTIO_F_ANY_LAYOUT: Device accepts arbitrary desc. layouts"),
90 #endif /* !VIRTIO_CONFIG_NO_LEGACY */
91     FEATURE_ENTRY(VIRTIO_F_VERSION_1, \
92             "VIRTIO_F_VERSION_1: Device compliant for v1 spec (legacy)"),
93     FEATURE_ENTRY(VIRTIO_F_IOMMU_PLATFORM, \
94             "VIRTIO_F_IOMMU_PLATFORM: Device can be used on IOMMU platform"),
95     FEATURE_ENTRY(VIRTIO_F_RING_PACKED, \
96             "VIRTIO_F_RING_PACKED: Device supports packed VQ layout"),
97     FEATURE_ENTRY(VIRTIO_F_IN_ORDER, \
98             "VIRTIO_F_IN_ORDER: Device uses buffers in same order as made "
99             "available by driver"),
100     FEATURE_ENTRY(VIRTIO_F_ORDER_PLATFORM, \
101             "VIRTIO_F_ORDER_PLATFORM: Memory accesses ordered by platform"),
102     FEATURE_ENTRY(VIRTIO_F_SR_IOV, \
103             "VIRTIO_F_SR_IOV: Device supports single root I/O virtualization"),
104     /* Virtio ring transport features */
105     FEATURE_ENTRY(VIRTIO_RING_F_INDIRECT_DESC, \
106             "VIRTIO_RING_F_INDIRECT_DESC: Indirect descriptors supported"),
107     FEATURE_ENTRY(VIRTIO_RING_F_EVENT_IDX, \
108             "VIRTIO_RING_F_EVENT_IDX: Used & avail. event fields enabled"),
109     { -1, "" }
110 };
111 
112 /* Vhost-user protocol features mapping */
113 static qmp_virtio_feature_map_t vhost_user_protocol_map[] = {
114     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_MQ, \
115             "VHOST_USER_PROTOCOL_F_MQ: Multiqueue protocol supported"),
116     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_LOG_SHMFD, \
117             "VHOST_USER_PROTOCOL_F_LOG_SHMFD: Shared log memory fd supported"),
118     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_RARP, \
119             "VHOST_USER_PROTOCOL_F_RARP: Vhost-user back-end RARP broadcasting "
120             "supported"),
121     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_REPLY_ACK, \
122             "VHOST_USER_PROTOCOL_F_REPLY_ACK: Requested operation status ack. "
123             "supported"),
124     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_NET_MTU, \
125             "VHOST_USER_PROTOCOL_F_NET_MTU: Expose host MTU to guest supported"),
126     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_SLAVE_REQ, \
127             "VHOST_USER_PROTOCOL_F_SLAVE_REQ: Socket fd for back-end initiated "
128             "requests supported"),
129     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_CROSS_ENDIAN, \
130             "VHOST_USER_PROTOCOL_F_CROSS_ENDIAN: Endianness of VQs for legacy "
131             "devices supported"),
132     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_CRYPTO_SESSION, \
133             "VHOST_USER_PROTOCOL_F_CRYPTO_SESSION: Session creation for crypto "
134             "operations supported"),
135     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_PAGEFAULT, \
136             "VHOST_USER_PROTOCOL_F_PAGEFAULT: Request servicing on userfaultfd "
137             "for accessed pages supported"),
138     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_CONFIG, \
139             "VHOST_USER_PROTOCOL_F_CONFIG: Vhost-user messaging for virtio "
140             "device configuration space supported"),
141     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD, \
142             "VHOST_USER_PROTOCOL_F_SLAVE_SEND_FD: Slave fd communication "
143             "channel supported"),
144     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_HOST_NOTIFIER, \
145             "VHOST_USER_PROTOCOL_F_HOST_NOTIFIER: Host notifiers for specified "
146             "VQs supported"),
147     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD, \
148             "VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD: Shared inflight I/O buffers "
149             "supported"),
150     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_RESET_DEVICE, \
151             "VHOST_USER_PROTOCOL_F_RESET_DEVICE: Disabling all rings and "
152             "resetting internal device state supported"),
153     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS, \
154             "VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS: In-band messaging "
155             "supported"),
156     FEATURE_ENTRY(VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS, \
157             "VHOST_USER_PROTOCOL_F_CONFIGURE_MEM_SLOTS: Configuration for "
158             "memory slots supported"),
159     { -1, "" }
160 };
161 
162 /* virtio device configuration statuses */
163 static qmp_virtio_feature_map_t virtio_config_status_map[] = {
164     FEATURE_ENTRY(VIRTIO_CONFIG_S_DRIVER_OK, \
165             "VIRTIO_CONFIG_S_DRIVER_OK: Driver setup and ready"),
166     FEATURE_ENTRY(VIRTIO_CONFIG_S_FEATURES_OK, \
167             "VIRTIO_CONFIG_S_FEATURES_OK: Feature negotiation complete"),
168     FEATURE_ENTRY(VIRTIO_CONFIG_S_DRIVER, \
169             "VIRTIO_CONFIG_S_DRIVER: Guest OS compatible with device"),
170     FEATURE_ENTRY(VIRTIO_CONFIG_S_NEEDS_RESET, \
171             "VIRTIO_CONFIG_S_NEEDS_RESET: Irrecoverable error, device needs "
172             "reset"),
173     FEATURE_ENTRY(VIRTIO_CONFIG_S_FAILED, \
174             "VIRTIO_CONFIG_S_FAILED: Error in guest, device failed"),
175     FEATURE_ENTRY(VIRTIO_CONFIG_S_ACKNOWLEDGE, \
176             "VIRTIO_CONFIG_S_ACKNOWLEDGE: Valid virtio device found"),
177     { -1, "" }
178 };
179 
180 /* virtio-blk features mapping */
181 qmp_virtio_feature_map_t virtio_blk_feature_map[] = {
182     FEATURE_ENTRY(VIRTIO_BLK_F_SIZE_MAX, \
183             "VIRTIO_BLK_F_SIZE_MAX: Max segment size is size_max"),
184     FEATURE_ENTRY(VIRTIO_BLK_F_SEG_MAX, \
185             "VIRTIO_BLK_F_SEG_MAX: Max segments in a request is seg_max"),
186     FEATURE_ENTRY(VIRTIO_BLK_F_GEOMETRY, \
187             "VIRTIO_BLK_F_GEOMETRY: Legacy geometry available"),
188     FEATURE_ENTRY(VIRTIO_BLK_F_RO, \
189             "VIRTIO_BLK_F_RO: Device is read-only"),
190     FEATURE_ENTRY(VIRTIO_BLK_F_BLK_SIZE, \
191             "VIRTIO_BLK_F_BLK_SIZE: Block size of disk available"),
192     FEATURE_ENTRY(VIRTIO_BLK_F_TOPOLOGY, \
193             "VIRTIO_BLK_F_TOPOLOGY: Topology information available"),
194     FEATURE_ENTRY(VIRTIO_BLK_F_MQ, \
195             "VIRTIO_BLK_F_MQ: Multiqueue supported"),
196     FEATURE_ENTRY(VIRTIO_BLK_F_DISCARD, \
197             "VIRTIO_BLK_F_DISCARD: Discard command supported"),
198     FEATURE_ENTRY(VIRTIO_BLK_F_WRITE_ZEROES, \
199             "VIRTIO_BLK_F_WRITE_ZEROES: Write zeroes command supported"),
200 #ifndef VIRTIO_BLK_NO_LEGACY
201     FEATURE_ENTRY(VIRTIO_BLK_F_BARRIER, \
202             "VIRTIO_BLK_F_BARRIER: Request barriers supported"),
203     FEATURE_ENTRY(VIRTIO_BLK_F_SCSI, \
204             "VIRTIO_BLK_F_SCSI: SCSI packet commands supported"),
205     FEATURE_ENTRY(VIRTIO_BLK_F_FLUSH, \
206             "VIRTIO_BLK_F_FLUSH: Flush command supported"),
207     FEATURE_ENTRY(VIRTIO_BLK_F_CONFIG_WCE, \
208             "VIRTIO_BLK_F_CONFIG_WCE: Cache writeback and writethrough modes "
209             "supported"),
210 #endif /* !VIRTIO_BLK_NO_LEGACY */
211     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
212             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
213     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
214             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
215             "negotiation supported"),
216     { -1, "" }
217 };
218 
219 /* virtio-serial features mapping */
220 qmp_virtio_feature_map_t virtio_serial_feature_map[] = {
221     FEATURE_ENTRY(VIRTIO_CONSOLE_F_SIZE, \
222             "VIRTIO_CONSOLE_F_SIZE: Host providing console size"),
223     FEATURE_ENTRY(VIRTIO_CONSOLE_F_MULTIPORT, \
224             "VIRTIO_CONSOLE_F_MULTIPORT: Multiple ports for device supported"),
225     FEATURE_ENTRY(VIRTIO_CONSOLE_F_EMERG_WRITE, \
226             "VIRTIO_CONSOLE_F_EMERG_WRITE: Emergency write supported"),
227     { -1, "" }
228 };
229 
230 /* virtio-gpu features mapping */
231 qmp_virtio_feature_map_t virtio_gpu_feature_map[] = {
232     FEATURE_ENTRY(VIRTIO_GPU_F_VIRGL, \
233             "VIRTIO_GPU_F_VIRGL: Virgl 3D mode supported"),
234     FEATURE_ENTRY(VIRTIO_GPU_F_EDID, \
235             "VIRTIO_GPU_F_EDID: EDID metadata supported"),
236     FEATURE_ENTRY(VIRTIO_GPU_F_RESOURCE_UUID, \
237             "VIRTIO_GPU_F_RESOURCE_UUID: Resource UUID assigning supported"),
238     FEATURE_ENTRY(VIRTIO_GPU_F_RESOURCE_BLOB, \
239             "VIRTIO_GPU_F_RESOURCE_BLOB: Size-based blob resources supported"),
240     FEATURE_ENTRY(VIRTIO_GPU_F_CONTEXT_INIT, \
241             "VIRTIO_GPU_F_CONTEXT_INIT: Context types and synchronization "
242             "timelines supported"),
243     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
244             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
245     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
246             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
247             "negotiation supported"),
248     { -1, "" }
249 };
250 
251 /* virtio-input features mapping */
252 qmp_virtio_feature_map_t virtio_input_feature_map[] = {
253     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
254             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
255     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
256             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
257             "negotiation supported"),
258     { -1, "" }
259 };
260 
261 /* virtio-net features mapping */
262 qmp_virtio_feature_map_t virtio_net_feature_map[] = {
263     FEATURE_ENTRY(VIRTIO_NET_F_CSUM, \
264             "VIRTIO_NET_F_CSUM: Device handling packets with partial checksum "
265             "supported"),
266     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_CSUM, \
267             "VIRTIO_NET_F_GUEST_CSUM: Driver handling packets with partial "
268             "checksum supported"),
269     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
270             "VIRTIO_NET_F_CTRL_GUEST_OFFLOADS: Control channel offloading "
271             "reconfig. supported"),
272     FEATURE_ENTRY(VIRTIO_NET_F_MTU, \
273             "VIRTIO_NET_F_MTU: Device max MTU reporting supported"),
274     FEATURE_ENTRY(VIRTIO_NET_F_MAC, \
275             "VIRTIO_NET_F_MAC: Device has given MAC address"),
276     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_TSO4, \
277             "VIRTIO_NET_F_GUEST_TSO4: Driver can receive TSOv4"),
278     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_TSO6, \
279             "VIRTIO_NET_F_GUEST_TSO6: Driver can receive TSOv6"),
280     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_ECN, \
281             "VIRTIO_NET_F_GUEST_ECN: Driver can receive TSO with ECN"),
282     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_UFO, \
283             "VIRTIO_NET_F_GUEST_UFO: Driver can receive UFO"),
284     FEATURE_ENTRY(VIRTIO_NET_F_HOST_TSO4, \
285             "VIRTIO_NET_F_HOST_TSO4: Device can receive TSOv4"),
286     FEATURE_ENTRY(VIRTIO_NET_F_HOST_TSO6, \
287             "VIRTIO_NET_F_HOST_TSO6: Device can receive TSOv6"),
288     FEATURE_ENTRY(VIRTIO_NET_F_HOST_ECN, \
289             "VIRTIO_NET_F_HOST_ECN: Device can receive TSO with ECN"),
290     FEATURE_ENTRY(VIRTIO_NET_F_HOST_UFO, \
291             "VIRTIO_NET_F_HOST_UFO: Device can receive UFO"),
292     FEATURE_ENTRY(VIRTIO_NET_F_MRG_RXBUF, \
293             "VIRTIO_NET_F_MRG_RXBUF: Driver can merge receive buffers"),
294     FEATURE_ENTRY(VIRTIO_NET_F_STATUS, \
295             "VIRTIO_NET_F_STATUS: Configuration status field available"),
296     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_VQ, \
297             "VIRTIO_NET_F_CTRL_VQ: Control channel available"),
298     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_RX, \
299             "VIRTIO_NET_F_CTRL_RX: Control channel RX mode supported"),
300     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_VLAN, \
301             "VIRTIO_NET_F_CTRL_VLAN: Control channel VLAN filtering supported"),
302     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_RX_EXTRA, \
303             "VIRTIO_NET_F_CTRL_RX_EXTRA: Extra RX mode control supported"),
304     FEATURE_ENTRY(VIRTIO_NET_F_GUEST_ANNOUNCE, \
305             "VIRTIO_NET_F_GUEST_ANNOUNCE: Driver sending gratuitous packets "
306             "supported"),
307     FEATURE_ENTRY(VIRTIO_NET_F_MQ, \
308             "VIRTIO_NET_F_MQ: Multiqueue with automatic receive steering "
309             "supported"),
310     FEATURE_ENTRY(VIRTIO_NET_F_CTRL_MAC_ADDR, \
311             "VIRTIO_NET_F_CTRL_MAC_ADDR: MAC address set through control "
312             "channel"),
313     FEATURE_ENTRY(VIRTIO_NET_F_HASH_REPORT, \
314             "VIRTIO_NET_F_HASH_REPORT: Hash reporting supported"),
315     FEATURE_ENTRY(VIRTIO_NET_F_RSS, \
316             "VIRTIO_NET_F_RSS: RSS RX steering supported"),
317     FEATURE_ENTRY(VIRTIO_NET_F_RSC_EXT, \
318             "VIRTIO_NET_F_RSC_EXT: Extended coalescing info supported"),
319     FEATURE_ENTRY(VIRTIO_NET_F_STANDBY, \
320             "VIRTIO_NET_F_STANDBY: Device acting as standby for primary "
321             "device with same MAC addr. supported"),
322     FEATURE_ENTRY(VIRTIO_NET_F_SPEED_DUPLEX, \
323             "VIRTIO_NET_F_SPEED_DUPLEX: Device set linkspeed and duplex"),
324 #ifndef VIRTIO_NET_NO_LEGACY
325     FEATURE_ENTRY(VIRTIO_NET_F_GSO, \
326             "VIRTIO_NET_F_GSO: Handling GSO-type packets supported"),
327 #endif /* !VIRTIO_NET_NO_LEGACY */
328     FEATURE_ENTRY(VHOST_NET_F_VIRTIO_NET_HDR, \
329             "VHOST_NET_F_VIRTIO_NET_HDR: Virtio-net headers for RX and TX "
330             "packets supported"),
331     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
332             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
333     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
334             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
335             "negotiation supported"),
336     { -1, "" }
337 };
338 
339 /* virtio-scsi features mapping */
340 qmp_virtio_feature_map_t virtio_scsi_feature_map[] = {
341     FEATURE_ENTRY(VIRTIO_SCSI_F_INOUT, \
342             "VIRTIO_SCSI_F_INOUT: Requests including read and writable data "
343             "buffers suppoted"),
344     FEATURE_ENTRY(VIRTIO_SCSI_F_HOTPLUG, \
345             "VIRTIO_SCSI_F_HOTPLUG: Reporting and handling hot-plug events "
346             "supported"),
347     FEATURE_ENTRY(VIRTIO_SCSI_F_CHANGE, \
348             "VIRTIO_SCSI_F_CHANGE: Reporting and handling LUN changes "
349             "supported"),
350     FEATURE_ENTRY(VIRTIO_SCSI_F_T10_PI, \
351             "VIRTIO_SCSI_F_T10_PI: T10 info included in request header"),
352     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
353             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
354     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
355             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
356             "negotiation supported"),
357     { -1, "" }
358 };
359 
360 /* virtio/vhost-user-fs features mapping */
361 qmp_virtio_feature_map_t virtio_fs_feature_map[] = {
362     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
363             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
364     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
365             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
366             "negotiation supported"),
367     { -1, "" }
368 };
369 
370 /* virtio/vhost-user-i2c features mapping */
371 qmp_virtio_feature_map_t virtio_i2c_feature_map[] = {
372     FEATURE_ENTRY(VIRTIO_I2C_F_ZERO_LENGTH_REQUEST, \
373             "VIRTIO_I2C_F_ZERO_LEGNTH_REQUEST: Zero length requests supported"),
374     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
375             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
376     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
377             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
378             "negotiation supported"),
379     { -1, "" }
380 };
381 
382 /* virtio/vhost-vsock features mapping */
383 qmp_virtio_feature_map_t virtio_vsock_feature_map[] = {
384     FEATURE_ENTRY(VIRTIO_VSOCK_F_SEQPACKET, \
385             "VIRTIO_VSOCK_F_SEQPACKET: SOCK_SEQPACKET supported"),
386     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
387             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
388     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
389             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
390             "negotiation supported"),
391     { -1, "" }
392 };
393 
394 /* virtio-balloon features mapping */
395 qmp_virtio_feature_map_t virtio_balloon_feature_map[] = {
396     FEATURE_ENTRY(VIRTIO_BALLOON_F_MUST_TELL_HOST, \
397             "VIRTIO_BALLOON_F_MUST_TELL_HOST: Tell host before reclaiming "
398             "pages"),
399     FEATURE_ENTRY(VIRTIO_BALLOON_F_STATS_VQ, \
400             "VIRTIO_BALLOON_F_STATS_VQ: Guest memory stats VQ available"),
401     FEATURE_ENTRY(VIRTIO_BALLOON_F_DEFLATE_ON_OOM, \
402             "VIRTIO_BALLOON_F_DEFLATE_ON_OOM: Deflate balloon when guest OOM"),
403     FEATURE_ENTRY(VIRTIO_BALLOON_F_FREE_PAGE_HINT, \
404             "VIRTIO_BALLOON_F_FREE_PAGE_HINT: VQ reporting free pages enabled"),
405     FEATURE_ENTRY(VIRTIO_BALLOON_F_PAGE_POISON, \
406             "VIRTIO_BALLOON_F_PAGE_POISON: Guest page poisoning enabled"),
407     FEATURE_ENTRY(VIRTIO_BALLOON_F_REPORTING, \
408             "VIRTIO_BALLOON_F_REPORTING: Page reporting VQ enabled"),
409     { -1, "" }
410 };
411 
412 /* virtio-crypto features mapping */
413 qmp_virtio_feature_map_t virtio_crypto_feature_map[] = {
414     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
415             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
416     { -1, "" }
417 };
418 
419 /* virtio-iommu features mapping */
420 qmp_virtio_feature_map_t virtio_iommu_feature_map[] = {
421     FEATURE_ENTRY(VIRTIO_IOMMU_F_INPUT_RANGE, \
422             "VIRTIO_IOMMU_F_INPUT_RANGE: Range of available virtual addrs. "
423             "available"),
424     FEATURE_ENTRY(VIRTIO_IOMMU_F_DOMAIN_RANGE, \
425             "VIRTIO_IOMMU_F_DOMAIN_RANGE: Number of supported domains "
426             "available"),
427     FEATURE_ENTRY(VIRTIO_IOMMU_F_MAP_UNMAP, \
428             "VIRTIO_IOMMU_F_MAP_UNMAP: Map and unmap requests available"),
429     FEATURE_ENTRY(VIRTIO_IOMMU_F_BYPASS, \
430             "VIRTIO_IOMMU_F_BYPASS: Endpoints not attached to domains are in "
431             "bypass mode"),
432     FEATURE_ENTRY(VIRTIO_IOMMU_F_PROBE, \
433             "VIRTIO_IOMMU_F_PROBE: Probe requests available"),
434     FEATURE_ENTRY(VIRTIO_IOMMU_F_MMIO, \
435             "VIRTIO_IOMMU_F_MMIO: VIRTIO_IOMMU_MAP_F_MMIO flag available"),
436     FEATURE_ENTRY(VIRTIO_IOMMU_F_BYPASS_CONFIG, \
437             "VIRTIO_IOMMU_F_BYPASS_CONFIG: Bypass field of IOMMU config "
438             "available"),
439     { -1, "" }
440 };
441 
442 /* virtio-mem features mapping */
443 qmp_virtio_feature_map_t virtio_mem_feature_map[] = {
444 #ifndef CONFIG_ACPI
445     FEATURE_ENTRY(VIRTIO_MEM_F_ACPI_PXM, \
446             "VIRTIO_MEM_F_ACPI_PXM: node_id is an ACPI PXM and is valid"),
447 #endif /* !CONFIG_ACPI */
448     FEATURE_ENTRY(VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE, \
449             "VIRTIO_MEM_F_UNPLUGGED_INACCESSIBLE: Unplugged memory cannot be "
450             "accessed"),
451     { -1, "" }
452 };
453 
454 /* virtio-rng features mapping */
455 qmp_virtio_feature_map_t virtio_rng_feature_map[] = {
456     FEATURE_ENTRY(VHOST_F_LOG_ALL, \
457             "VHOST_F_LOG_ALL: Logging write descriptors supported"),
458     FEATURE_ENTRY(VHOST_USER_F_PROTOCOL_FEATURES, \
459             "VHOST_USER_F_PROTOCOL_FEATURES: Vhost-user protocol features "
460             "negotiation supported"),
461     { -1, "" }
462 };
463 
464 /*
465  * The alignment to use between consumer and producer parts of vring.
466  * x86 pagesize again. This is the default, used by transports like PCI
467  * which don't provide a means for the guest to tell the host the alignment.
468  */
469 #define VIRTIO_PCI_VRING_ALIGN         4096
470 
471 typedef struct VRingDesc
472 {
473     uint64_t addr;
474     uint32_t len;
475     uint16_t flags;
476     uint16_t next;
477 } VRingDesc;
478 
479 typedef struct VRingPackedDesc {
480     uint64_t addr;
481     uint32_t len;
482     uint16_t id;
483     uint16_t flags;
484 } VRingPackedDesc;
485 
486 typedef struct VRingAvail
487 {
488     uint16_t flags;
489     uint16_t idx;
490     uint16_t ring[];
491 } VRingAvail;
492 
493 typedef struct VRingUsedElem
494 {
495     uint32_t id;
496     uint32_t len;
497 } VRingUsedElem;
498 
499 typedef struct VRingUsed
500 {
501     uint16_t flags;
502     uint16_t idx;
503     VRingUsedElem ring[];
504 } VRingUsed;
505 
506 typedef struct VRingMemoryRegionCaches {
507     struct rcu_head rcu;
508     MemoryRegionCache desc;
509     MemoryRegionCache avail;
510     MemoryRegionCache used;
511 } VRingMemoryRegionCaches;
512 
513 typedef struct VRing
514 {
515     unsigned int num;
516     unsigned int num_default;
517     unsigned int align;
518     hwaddr desc;
519     hwaddr avail;
520     hwaddr used;
521     VRingMemoryRegionCaches *caches;
522 } VRing;
523 
524 typedef struct VRingPackedDescEvent {
525     uint16_t off_wrap;
526     uint16_t flags;
527 } VRingPackedDescEvent ;
528 
529 struct VirtQueue
530 {
531     VRing vring;
532     VirtQueueElement *used_elems;
533 
534     /* Next head to pop */
535     uint16_t last_avail_idx;
536     bool last_avail_wrap_counter;
537 
538     /* Last avail_idx read from VQ. */
539     uint16_t shadow_avail_idx;
540     bool shadow_avail_wrap_counter;
541 
542     uint16_t used_idx;
543     bool used_wrap_counter;
544 
545     /* Last used index value we have signalled on */
546     uint16_t signalled_used;
547 
548     /* Last used index value we have signalled on */
549     bool signalled_used_valid;
550 
551     /* Notification enabled? */
552     bool notification;
553 
554     uint16_t queue_index;
555 
556     unsigned int inuse;
557 
558     uint16_t vector;
559     VirtIOHandleOutput handle_output;
560     VirtIODevice *vdev;
561     EventNotifier guest_notifier;
562     EventNotifier host_notifier;
563     bool host_notifier_enabled;
564     QLIST_ENTRY(VirtQueue) node;
565 };
566 
567 const char *virtio_device_names[] = {
568     [VIRTIO_ID_NET] = "virtio-net",
569     [VIRTIO_ID_BLOCK] = "virtio-blk",
570     [VIRTIO_ID_CONSOLE] = "virtio-serial",
571     [VIRTIO_ID_RNG] = "virtio-rng",
572     [VIRTIO_ID_BALLOON] = "virtio-balloon",
573     [VIRTIO_ID_IOMEM] = "virtio-iomem",
574     [VIRTIO_ID_RPMSG] = "virtio-rpmsg",
575     [VIRTIO_ID_SCSI] = "virtio-scsi",
576     [VIRTIO_ID_9P] = "virtio-9p",
577     [VIRTIO_ID_MAC80211_WLAN] = "virtio-mac-wlan",
578     [VIRTIO_ID_RPROC_SERIAL] = "virtio-rproc-serial",
579     [VIRTIO_ID_CAIF] = "virtio-caif",
580     [VIRTIO_ID_MEMORY_BALLOON] = "virtio-mem-balloon",
581     [VIRTIO_ID_GPU] = "virtio-gpu",
582     [VIRTIO_ID_CLOCK] = "virtio-clk",
583     [VIRTIO_ID_INPUT] = "virtio-input",
584     [VIRTIO_ID_VSOCK] = "vhost-vsock",
585     [VIRTIO_ID_CRYPTO] = "virtio-crypto",
586     [VIRTIO_ID_SIGNAL_DIST] = "virtio-signal",
587     [VIRTIO_ID_PSTORE] = "virtio-pstore",
588     [VIRTIO_ID_IOMMU] = "virtio-iommu",
589     [VIRTIO_ID_MEM] = "virtio-mem",
590     [VIRTIO_ID_SOUND] = "virtio-sound",
591     [VIRTIO_ID_FS] = "virtio-user-fs",
592     [VIRTIO_ID_PMEM] = "virtio-pmem",
593     [VIRTIO_ID_RPMB] = "virtio-rpmb",
594     [VIRTIO_ID_MAC80211_HWSIM] = "virtio-mac-hwsim",
595     [VIRTIO_ID_VIDEO_ENCODER] = "virtio-vid-encoder",
596     [VIRTIO_ID_VIDEO_DECODER] = "virtio-vid-decoder",
597     [VIRTIO_ID_SCMI] = "virtio-scmi",
598     [VIRTIO_ID_NITRO_SEC_MOD] = "virtio-nitro-sec-mod",
599     [VIRTIO_ID_I2C_ADAPTER] = "vhost-user-i2c",
600     [VIRTIO_ID_WATCHDOG] = "virtio-watchdog",
601     [VIRTIO_ID_CAN] = "virtio-can",
602     [VIRTIO_ID_DMABUF] = "virtio-dmabuf",
603     [VIRTIO_ID_PARAM_SERV] = "virtio-param-serv",
604     [VIRTIO_ID_AUDIO_POLICY] = "virtio-audio-pol",
605     [VIRTIO_ID_BT] = "virtio-bluetooth",
606     [VIRTIO_ID_GPIO] = "virtio-gpio"
607 };
608 
609 static const char *virtio_id_to_name(uint16_t device_id)
610 {
611     assert(device_id < G_N_ELEMENTS(virtio_device_names));
612     const char *name = virtio_device_names[device_id];
613     assert(name != NULL);
614     return name;
615 }
616 
617 /* Called within call_rcu().  */
618 static void virtio_free_region_cache(VRingMemoryRegionCaches *caches)
619 {
620     assert(caches != NULL);
621     address_space_cache_destroy(&caches->desc);
622     address_space_cache_destroy(&caches->avail);
623     address_space_cache_destroy(&caches->used);
624     g_free(caches);
625 }
626 
627 static void virtio_virtqueue_reset_region_cache(struct VirtQueue *vq)
628 {
629     VRingMemoryRegionCaches *caches;
630 
631     caches = qatomic_read(&vq->vring.caches);
632     qatomic_rcu_set(&vq->vring.caches, NULL);
633     if (caches) {
634         call_rcu(caches, virtio_free_region_cache, rcu);
635     }
636 }
637 
638 static void virtio_init_region_cache(VirtIODevice *vdev, int n)
639 {
640     VirtQueue *vq = &vdev->vq[n];
641     VRingMemoryRegionCaches *old = vq->vring.caches;
642     VRingMemoryRegionCaches *new = NULL;
643     hwaddr addr, size;
644     int64_t len;
645     bool packed;
646 
647 
648     addr = vq->vring.desc;
649     if (!addr) {
650         goto out_no_cache;
651     }
652     new = g_new0(VRingMemoryRegionCaches, 1);
653     size = virtio_queue_get_desc_size(vdev, n);
654     packed = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
655                                    true : false;
656     len = address_space_cache_init(&new->desc, vdev->dma_as,
657                                    addr, size, packed);
658     if (len < size) {
659         virtio_error(vdev, "Cannot map desc");
660         goto err_desc;
661     }
662 
663     size = virtio_queue_get_used_size(vdev, n);
664     len = address_space_cache_init(&new->used, vdev->dma_as,
665                                    vq->vring.used, size, true);
666     if (len < size) {
667         virtio_error(vdev, "Cannot map used");
668         goto err_used;
669     }
670 
671     size = virtio_queue_get_avail_size(vdev, n);
672     len = address_space_cache_init(&new->avail, vdev->dma_as,
673                                    vq->vring.avail, size, false);
674     if (len < size) {
675         virtio_error(vdev, "Cannot map avail");
676         goto err_avail;
677     }
678 
679     qatomic_rcu_set(&vq->vring.caches, new);
680     if (old) {
681         call_rcu(old, virtio_free_region_cache, rcu);
682     }
683     return;
684 
685 err_avail:
686     address_space_cache_destroy(&new->avail);
687 err_used:
688     address_space_cache_destroy(&new->used);
689 err_desc:
690     address_space_cache_destroy(&new->desc);
691 out_no_cache:
692     g_free(new);
693     virtio_virtqueue_reset_region_cache(vq);
694 }
695 
696 /* virt queue functions */
697 void virtio_queue_update_rings(VirtIODevice *vdev, int n)
698 {
699     VRing *vring = &vdev->vq[n].vring;
700 
701     if (!vring->num || !vring->desc || !vring->align) {
702         /* not yet setup -> nothing to do */
703         return;
704     }
705     vring->avail = vring->desc + vring->num * sizeof(VRingDesc);
706     vring->used = vring_align(vring->avail +
707                               offsetof(VRingAvail, ring[vring->num]),
708                               vring->align);
709     virtio_init_region_cache(vdev, n);
710 }
711 
712 /* Called within rcu_read_lock().  */
713 static void vring_split_desc_read(VirtIODevice *vdev, VRingDesc *desc,
714                                   MemoryRegionCache *cache, int i)
715 {
716     address_space_read_cached(cache, i * sizeof(VRingDesc),
717                               desc, sizeof(VRingDesc));
718     virtio_tswap64s(vdev, &desc->addr);
719     virtio_tswap32s(vdev, &desc->len);
720     virtio_tswap16s(vdev, &desc->flags);
721     virtio_tswap16s(vdev, &desc->next);
722 }
723 
724 static void vring_packed_event_read(VirtIODevice *vdev,
725                                     MemoryRegionCache *cache,
726                                     VRingPackedDescEvent *e)
727 {
728     hwaddr off_off = offsetof(VRingPackedDescEvent, off_wrap);
729     hwaddr off_flags = offsetof(VRingPackedDescEvent, flags);
730 
731     e->flags = virtio_lduw_phys_cached(vdev, cache, off_flags);
732     /* Make sure flags is seen before off_wrap */
733     smp_rmb();
734     e->off_wrap = virtio_lduw_phys_cached(vdev, cache, off_off);
735     virtio_tswap16s(vdev, &e->flags);
736 }
737 
738 static void vring_packed_off_wrap_write(VirtIODevice *vdev,
739                                         MemoryRegionCache *cache,
740                                         uint16_t off_wrap)
741 {
742     hwaddr off = offsetof(VRingPackedDescEvent, off_wrap);
743 
744     virtio_stw_phys_cached(vdev, cache, off, off_wrap);
745     address_space_cache_invalidate(cache, off, sizeof(off_wrap));
746 }
747 
748 static void vring_packed_flags_write(VirtIODevice *vdev,
749                                      MemoryRegionCache *cache, uint16_t flags)
750 {
751     hwaddr off = offsetof(VRingPackedDescEvent, flags);
752 
753     virtio_stw_phys_cached(vdev, cache, off, flags);
754     address_space_cache_invalidate(cache, off, sizeof(flags));
755 }
756 
757 /* Called within rcu_read_lock().  */
758 static VRingMemoryRegionCaches *vring_get_region_caches(struct VirtQueue *vq)
759 {
760     return qatomic_rcu_read(&vq->vring.caches);
761 }
762 
763 /* Called within rcu_read_lock().  */
764 static inline uint16_t vring_avail_flags(VirtQueue *vq)
765 {
766     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
767     hwaddr pa = offsetof(VRingAvail, flags);
768 
769     if (!caches) {
770         return 0;
771     }
772 
773     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
774 }
775 
776 /* Called within rcu_read_lock().  */
777 static inline uint16_t vring_avail_idx(VirtQueue *vq)
778 {
779     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
780     hwaddr pa = offsetof(VRingAvail, idx);
781 
782     if (!caches) {
783         return 0;
784     }
785 
786     vq->shadow_avail_idx = virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
787     return vq->shadow_avail_idx;
788 }
789 
790 /* Called within rcu_read_lock().  */
791 static inline uint16_t vring_avail_ring(VirtQueue *vq, int i)
792 {
793     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
794     hwaddr pa = offsetof(VRingAvail, ring[i]);
795 
796     if (!caches) {
797         return 0;
798     }
799 
800     return virtio_lduw_phys_cached(vq->vdev, &caches->avail, pa);
801 }
802 
803 /* Called within rcu_read_lock().  */
804 static inline uint16_t vring_get_used_event(VirtQueue *vq)
805 {
806     return vring_avail_ring(vq, vq->vring.num);
807 }
808 
809 /* Called within rcu_read_lock().  */
810 static inline void vring_used_write(VirtQueue *vq, VRingUsedElem *uelem,
811                                     int i)
812 {
813     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
814     hwaddr pa = offsetof(VRingUsed, ring[i]);
815 
816     if (!caches) {
817         return;
818     }
819 
820     virtio_tswap32s(vq->vdev, &uelem->id);
821     virtio_tswap32s(vq->vdev, &uelem->len);
822     address_space_write_cached(&caches->used, pa, uelem, sizeof(VRingUsedElem));
823     address_space_cache_invalidate(&caches->used, pa, sizeof(VRingUsedElem));
824 }
825 
826 /* Called within rcu_read_lock(). */
827 static inline uint16_t vring_used_flags(VirtQueue *vq)
828 {
829     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
830     hwaddr pa = offsetof(VRingUsed, flags);
831 
832     if (!caches) {
833         return 0;
834     }
835 
836     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
837 }
838 
839 /* Called within rcu_read_lock().  */
840 static uint16_t vring_used_idx(VirtQueue *vq)
841 {
842     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
843     hwaddr pa = offsetof(VRingUsed, idx);
844 
845     if (!caches) {
846         return 0;
847     }
848 
849     return virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
850 }
851 
852 /* Called within rcu_read_lock().  */
853 static inline void vring_used_idx_set(VirtQueue *vq, uint16_t val)
854 {
855     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
856     hwaddr pa = offsetof(VRingUsed, idx);
857 
858     if (caches) {
859         virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
860         address_space_cache_invalidate(&caches->used, pa, sizeof(val));
861     }
862 
863     vq->used_idx = val;
864 }
865 
866 /* Called within rcu_read_lock().  */
867 static inline void vring_used_flags_set_bit(VirtQueue *vq, int mask)
868 {
869     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
870     VirtIODevice *vdev = vq->vdev;
871     hwaddr pa = offsetof(VRingUsed, flags);
872     uint16_t flags;
873 
874     if (!caches) {
875         return;
876     }
877 
878     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
879     virtio_stw_phys_cached(vdev, &caches->used, pa, flags | mask);
880     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
881 }
882 
883 /* Called within rcu_read_lock().  */
884 static inline void vring_used_flags_unset_bit(VirtQueue *vq, int mask)
885 {
886     VRingMemoryRegionCaches *caches = vring_get_region_caches(vq);
887     VirtIODevice *vdev = vq->vdev;
888     hwaddr pa = offsetof(VRingUsed, flags);
889     uint16_t flags;
890 
891     if (!caches) {
892         return;
893     }
894 
895     flags = virtio_lduw_phys_cached(vq->vdev, &caches->used, pa);
896     virtio_stw_phys_cached(vdev, &caches->used, pa, flags & ~mask);
897     address_space_cache_invalidate(&caches->used, pa, sizeof(flags));
898 }
899 
900 /* Called within rcu_read_lock().  */
901 static inline void vring_set_avail_event(VirtQueue *vq, uint16_t val)
902 {
903     VRingMemoryRegionCaches *caches;
904     hwaddr pa;
905     if (!vq->notification) {
906         return;
907     }
908 
909     caches = vring_get_region_caches(vq);
910     if (!caches) {
911         return;
912     }
913 
914     pa = offsetof(VRingUsed, ring[vq->vring.num]);
915     virtio_stw_phys_cached(vq->vdev, &caches->used, pa, val);
916     address_space_cache_invalidate(&caches->used, pa, sizeof(val));
917 }
918 
919 static void virtio_queue_split_set_notification(VirtQueue *vq, int enable)
920 {
921     RCU_READ_LOCK_GUARD();
922 
923     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
924         vring_set_avail_event(vq, vring_avail_idx(vq));
925     } else if (enable) {
926         vring_used_flags_unset_bit(vq, VRING_USED_F_NO_NOTIFY);
927     } else {
928         vring_used_flags_set_bit(vq, VRING_USED_F_NO_NOTIFY);
929     }
930     if (enable) {
931         /* Expose avail event/used flags before caller checks the avail idx. */
932         smp_mb();
933     }
934 }
935 
936 static void virtio_queue_packed_set_notification(VirtQueue *vq, int enable)
937 {
938     uint16_t off_wrap;
939     VRingPackedDescEvent e;
940     VRingMemoryRegionCaches *caches;
941 
942     RCU_READ_LOCK_GUARD();
943     caches = vring_get_region_caches(vq);
944     if (!caches) {
945         return;
946     }
947 
948     vring_packed_event_read(vq->vdev, &caches->used, &e);
949 
950     if (!enable) {
951         e.flags = VRING_PACKED_EVENT_FLAG_DISABLE;
952     } else if (virtio_vdev_has_feature(vq->vdev, VIRTIO_RING_F_EVENT_IDX)) {
953         off_wrap = vq->shadow_avail_idx | vq->shadow_avail_wrap_counter << 15;
954         vring_packed_off_wrap_write(vq->vdev, &caches->used, off_wrap);
955         /* Make sure off_wrap is wrote before flags */
956         smp_wmb();
957         e.flags = VRING_PACKED_EVENT_FLAG_DESC;
958     } else {
959         e.flags = VRING_PACKED_EVENT_FLAG_ENABLE;
960     }
961 
962     vring_packed_flags_write(vq->vdev, &caches->used, e.flags);
963     if (enable) {
964         /* Expose avail event/used flags before caller checks the avail idx. */
965         smp_mb();
966     }
967 }
968 
969 bool virtio_queue_get_notification(VirtQueue *vq)
970 {
971     return vq->notification;
972 }
973 
974 void virtio_queue_set_notification(VirtQueue *vq, int enable)
975 {
976     vq->notification = enable;
977 
978     if (!vq->vring.desc) {
979         return;
980     }
981 
982     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
983         virtio_queue_packed_set_notification(vq, enable);
984     } else {
985         virtio_queue_split_set_notification(vq, enable);
986     }
987 }
988 
989 int virtio_queue_ready(VirtQueue *vq)
990 {
991     return vq->vring.avail != 0;
992 }
993 
994 static void vring_packed_desc_read_flags(VirtIODevice *vdev,
995                                          uint16_t *flags,
996                                          MemoryRegionCache *cache,
997                                          int i)
998 {
999     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
1000 
1001     *flags = virtio_lduw_phys_cached(vdev, cache, off);
1002 }
1003 
1004 static void vring_packed_desc_read(VirtIODevice *vdev,
1005                                    VRingPackedDesc *desc,
1006                                    MemoryRegionCache *cache,
1007                                    int i, bool strict_order)
1008 {
1009     hwaddr off = i * sizeof(VRingPackedDesc);
1010 
1011     vring_packed_desc_read_flags(vdev, &desc->flags, cache, i);
1012 
1013     if (strict_order) {
1014         /* Make sure flags is read before the rest fields. */
1015         smp_rmb();
1016     }
1017 
1018     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, addr),
1019                               &desc->addr, sizeof(desc->addr));
1020     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, id),
1021                               &desc->id, sizeof(desc->id));
1022     address_space_read_cached(cache, off + offsetof(VRingPackedDesc, len),
1023                               &desc->len, sizeof(desc->len));
1024     virtio_tswap64s(vdev, &desc->addr);
1025     virtio_tswap16s(vdev, &desc->id);
1026     virtio_tswap32s(vdev, &desc->len);
1027 }
1028 
1029 static void vring_packed_desc_write_data(VirtIODevice *vdev,
1030                                          VRingPackedDesc *desc,
1031                                          MemoryRegionCache *cache,
1032                                          int i)
1033 {
1034     hwaddr off_id = i * sizeof(VRingPackedDesc) +
1035                     offsetof(VRingPackedDesc, id);
1036     hwaddr off_len = i * sizeof(VRingPackedDesc) +
1037                     offsetof(VRingPackedDesc, len);
1038 
1039     virtio_tswap32s(vdev, &desc->len);
1040     virtio_tswap16s(vdev, &desc->id);
1041     address_space_write_cached(cache, off_id, &desc->id, sizeof(desc->id));
1042     address_space_cache_invalidate(cache, off_id, sizeof(desc->id));
1043     address_space_write_cached(cache, off_len, &desc->len, sizeof(desc->len));
1044     address_space_cache_invalidate(cache, off_len, sizeof(desc->len));
1045 }
1046 
1047 static void vring_packed_desc_write_flags(VirtIODevice *vdev,
1048                                           VRingPackedDesc *desc,
1049                                           MemoryRegionCache *cache,
1050                                           int i)
1051 {
1052     hwaddr off = i * sizeof(VRingPackedDesc) + offsetof(VRingPackedDesc, flags);
1053 
1054     virtio_stw_phys_cached(vdev, cache, off, desc->flags);
1055     address_space_cache_invalidate(cache, off, sizeof(desc->flags));
1056 }
1057 
1058 static void vring_packed_desc_write(VirtIODevice *vdev,
1059                                     VRingPackedDesc *desc,
1060                                     MemoryRegionCache *cache,
1061                                     int i, bool strict_order)
1062 {
1063     vring_packed_desc_write_data(vdev, desc, cache, i);
1064     if (strict_order) {
1065         /* Make sure data is wrote before flags. */
1066         smp_wmb();
1067     }
1068     vring_packed_desc_write_flags(vdev, desc, cache, i);
1069 }
1070 
1071 static inline bool is_desc_avail(uint16_t flags, bool wrap_counter)
1072 {
1073     bool avail, used;
1074 
1075     avail = !!(flags & (1 << VRING_PACKED_DESC_F_AVAIL));
1076     used = !!(flags & (1 << VRING_PACKED_DESC_F_USED));
1077     return (avail != used) && (avail == wrap_counter);
1078 }
1079 
1080 /* Fetch avail_idx from VQ memory only when we really need to know if
1081  * guest has added some buffers.
1082  * Called within rcu_read_lock().  */
1083 static int virtio_queue_empty_rcu(VirtQueue *vq)
1084 {
1085     if (virtio_device_disabled(vq->vdev)) {
1086         return 1;
1087     }
1088 
1089     if (unlikely(!vq->vring.avail)) {
1090         return 1;
1091     }
1092 
1093     if (vq->shadow_avail_idx != vq->last_avail_idx) {
1094         return 0;
1095     }
1096 
1097     return vring_avail_idx(vq) == vq->last_avail_idx;
1098 }
1099 
1100 static int virtio_queue_split_empty(VirtQueue *vq)
1101 {
1102     bool empty;
1103 
1104     if (virtio_device_disabled(vq->vdev)) {
1105         return 1;
1106     }
1107 
1108     if (unlikely(!vq->vring.avail)) {
1109         return 1;
1110     }
1111 
1112     if (vq->shadow_avail_idx != vq->last_avail_idx) {
1113         return 0;
1114     }
1115 
1116     RCU_READ_LOCK_GUARD();
1117     empty = vring_avail_idx(vq) == vq->last_avail_idx;
1118     return empty;
1119 }
1120 
1121 /* Called within rcu_read_lock().  */
1122 static int virtio_queue_packed_empty_rcu(VirtQueue *vq)
1123 {
1124     struct VRingPackedDesc desc;
1125     VRingMemoryRegionCaches *cache;
1126 
1127     if (unlikely(!vq->vring.desc)) {
1128         return 1;
1129     }
1130 
1131     cache = vring_get_region_caches(vq);
1132     if (!cache) {
1133         return 1;
1134     }
1135 
1136     vring_packed_desc_read_flags(vq->vdev, &desc.flags, &cache->desc,
1137                                  vq->last_avail_idx);
1138 
1139     return !is_desc_avail(desc.flags, vq->last_avail_wrap_counter);
1140 }
1141 
1142 static int virtio_queue_packed_empty(VirtQueue *vq)
1143 {
1144     RCU_READ_LOCK_GUARD();
1145     return virtio_queue_packed_empty_rcu(vq);
1146 }
1147 
1148 int virtio_queue_empty(VirtQueue *vq)
1149 {
1150     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1151         return virtio_queue_packed_empty(vq);
1152     } else {
1153         return virtio_queue_split_empty(vq);
1154     }
1155 }
1156 
1157 static void virtqueue_unmap_sg(VirtQueue *vq, const VirtQueueElement *elem,
1158                                unsigned int len)
1159 {
1160     AddressSpace *dma_as = vq->vdev->dma_as;
1161     unsigned int offset;
1162     int i;
1163 
1164     offset = 0;
1165     for (i = 0; i < elem->in_num; i++) {
1166         size_t size = MIN(len - offset, elem->in_sg[i].iov_len);
1167 
1168         dma_memory_unmap(dma_as, elem->in_sg[i].iov_base,
1169                          elem->in_sg[i].iov_len,
1170                          DMA_DIRECTION_FROM_DEVICE, size);
1171 
1172         offset += size;
1173     }
1174 
1175     for (i = 0; i < elem->out_num; i++)
1176         dma_memory_unmap(dma_as, elem->out_sg[i].iov_base,
1177                          elem->out_sg[i].iov_len,
1178                          DMA_DIRECTION_TO_DEVICE,
1179                          elem->out_sg[i].iov_len);
1180 }
1181 
1182 /* virtqueue_detach_element:
1183  * @vq: The #VirtQueue
1184  * @elem: The #VirtQueueElement
1185  * @len: number of bytes written
1186  *
1187  * Detach the element from the virtqueue.  This function is suitable for device
1188  * reset or other situations where a #VirtQueueElement is simply freed and will
1189  * not be pushed or discarded.
1190  */
1191 void virtqueue_detach_element(VirtQueue *vq, const VirtQueueElement *elem,
1192                               unsigned int len)
1193 {
1194     vq->inuse -= elem->ndescs;
1195     virtqueue_unmap_sg(vq, elem, len);
1196 }
1197 
1198 static void virtqueue_split_rewind(VirtQueue *vq, unsigned int num)
1199 {
1200     vq->last_avail_idx -= num;
1201 }
1202 
1203 static void virtqueue_packed_rewind(VirtQueue *vq, unsigned int num)
1204 {
1205     if (vq->last_avail_idx < num) {
1206         vq->last_avail_idx = vq->vring.num + vq->last_avail_idx - num;
1207         vq->last_avail_wrap_counter ^= 1;
1208     } else {
1209         vq->last_avail_idx -= num;
1210     }
1211 }
1212 
1213 /* virtqueue_unpop:
1214  * @vq: The #VirtQueue
1215  * @elem: The #VirtQueueElement
1216  * @len: number of bytes written
1217  *
1218  * Pretend the most recent element wasn't popped from the virtqueue.  The next
1219  * call to virtqueue_pop() will refetch the element.
1220  */
1221 void virtqueue_unpop(VirtQueue *vq, const VirtQueueElement *elem,
1222                      unsigned int len)
1223 {
1224 
1225     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1226         virtqueue_packed_rewind(vq, 1);
1227     } else {
1228         virtqueue_split_rewind(vq, 1);
1229     }
1230 
1231     virtqueue_detach_element(vq, elem, len);
1232 }
1233 
1234 /* virtqueue_rewind:
1235  * @vq: The #VirtQueue
1236  * @num: Number of elements to push back
1237  *
1238  * Pretend that elements weren't popped from the virtqueue.  The next
1239  * virtqueue_pop() will refetch the oldest element.
1240  *
1241  * Use virtqueue_unpop() instead if you have a VirtQueueElement.
1242  *
1243  * Returns: true on success, false if @num is greater than the number of in use
1244  * elements.
1245  */
1246 bool virtqueue_rewind(VirtQueue *vq, unsigned int num)
1247 {
1248     if (num > vq->inuse) {
1249         return false;
1250     }
1251 
1252     vq->inuse -= num;
1253     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1254         virtqueue_packed_rewind(vq, num);
1255     } else {
1256         virtqueue_split_rewind(vq, num);
1257     }
1258     return true;
1259 }
1260 
1261 static void virtqueue_split_fill(VirtQueue *vq, const VirtQueueElement *elem,
1262                     unsigned int len, unsigned int idx)
1263 {
1264     VRingUsedElem uelem;
1265 
1266     if (unlikely(!vq->vring.used)) {
1267         return;
1268     }
1269 
1270     idx = (idx + vq->used_idx) % vq->vring.num;
1271 
1272     uelem.id = elem->index;
1273     uelem.len = len;
1274     vring_used_write(vq, &uelem, idx);
1275 }
1276 
1277 static void virtqueue_packed_fill(VirtQueue *vq, const VirtQueueElement *elem,
1278                                   unsigned int len, unsigned int idx)
1279 {
1280     vq->used_elems[idx].index = elem->index;
1281     vq->used_elems[idx].len = len;
1282     vq->used_elems[idx].ndescs = elem->ndescs;
1283 }
1284 
1285 static void virtqueue_packed_fill_desc(VirtQueue *vq,
1286                                        const VirtQueueElement *elem,
1287                                        unsigned int idx,
1288                                        bool strict_order)
1289 {
1290     uint16_t head;
1291     VRingMemoryRegionCaches *caches;
1292     VRingPackedDesc desc = {
1293         .id = elem->index,
1294         .len = elem->len,
1295     };
1296     bool wrap_counter = vq->used_wrap_counter;
1297 
1298     if (unlikely(!vq->vring.desc)) {
1299         return;
1300     }
1301 
1302     head = vq->used_idx + idx;
1303     if (head >= vq->vring.num) {
1304         head -= vq->vring.num;
1305         wrap_counter ^= 1;
1306     }
1307     if (wrap_counter) {
1308         desc.flags |= (1 << VRING_PACKED_DESC_F_AVAIL);
1309         desc.flags |= (1 << VRING_PACKED_DESC_F_USED);
1310     } else {
1311         desc.flags &= ~(1 << VRING_PACKED_DESC_F_AVAIL);
1312         desc.flags &= ~(1 << VRING_PACKED_DESC_F_USED);
1313     }
1314 
1315     caches = vring_get_region_caches(vq);
1316     if (!caches) {
1317         return;
1318     }
1319 
1320     vring_packed_desc_write(vq->vdev, &desc, &caches->desc, head, strict_order);
1321 }
1322 
1323 /* Called within rcu_read_lock().  */
1324 void virtqueue_fill(VirtQueue *vq, const VirtQueueElement *elem,
1325                     unsigned int len, unsigned int idx)
1326 {
1327     trace_virtqueue_fill(vq, elem, len, idx);
1328 
1329     virtqueue_unmap_sg(vq, elem, len);
1330 
1331     if (virtio_device_disabled(vq->vdev)) {
1332         return;
1333     }
1334 
1335     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1336         virtqueue_packed_fill(vq, elem, len, idx);
1337     } else {
1338         virtqueue_split_fill(vq, elem, len, idx);
1339     }
1340 }
1341 
1342 /* Called within rcu_read_lock().  */
1343 static void virtqueue_split_flush(VirtQueue *vq, unsigned int count)
1344 {
1345     uint16_t old, new;
1346 
1347     if (unlikely(!vq->vring.used)) {
1348         return;
1349     }
1350 
1351     /* Make sure buffer is written before we update index. */
1352     smp_wmb();
1353     trace_virtqueue_flush(vq, count);
1354     old = vq->used_idx;
1355     new = old + count;
1356     vring_used_idx_set(vq, new);
1357     vq->inuse -= count;
1358     if (unlikely((int16_t)(new - vq->signalled_used) < (uint16_t)(new - old)))
1359         vq->signalled_used_valid = false;
1360 }
1361 
1362 static void virtqueue_packed_flush(VirtQueue *vq, unsigned int count)
1363 {
1364     unsigned int i, ndescs = 0;
1365 
1366     if (unlikely(!vq->vring.desc)) {
1367         return;
1368     }
1369 
1370     for (i = 1; i < count; i++) {
1371         virtqueue_packed_fill_desc(vq, &vq->used_elems[i], i, false);
1372         ndescs += vq->used_elems[i].ndescs;
1373     }
1374     virtqueue_packed_fill_desc(vq, &vq->used_elems[0], 0, true);
1375     ndescs += vq->used_elems[0].ndescs;
1376 
1377     vq->inuse -= ndescs;
1378     vq->used_idx += ndescs;
1379     if (vq->used_idx >= vq->vring.num) {
1380         vq->used_idx -= vq->vring.num;
1381         vq->used_wrap_counter ^= 1;
1382         vq->signalled_used_valid = false;
1383     }
1384 }
1385 
1386 void virtqueue_flush(VirtQueue *vq, unsigned int count)
1387 {
1388     if (virtio_device_disabled(vq->vdev)) {
1389         vq->inuse -= count;
1390         return;
1391     }
1392 
1393     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1394         virtqueue_packed_flush(vq, count);
1395     } else {
1396         virtqueue_split_flush(vq, count);
1397     }
1398 }
1399 
1400 void virtqueue_push(VirtQueue *vq, const VirtQueueElement *elem,
1401                     unsigned int len)
1402 {
1403     RCU_READ_LOCK_GUARD();
1404     virtqueue_fill(vq, elem, len, 0);
1405     virtqueue_flush(vq, 1);
1406 }
1407 
1408 /* Called within rcu_read_lock().  */
1409 static int virtqueue_num_heads(VirtQueue *vq, unsigned int idx)
1410 {
1411     uint16_t num_heads = vring_avail_idx(vq) - idx;
1412 
1413     /* Check it isn't doing very strange things with descriptor numbers. */
1414     if (num_heads > vq->vring.num) {
1415         virtio_error(vq->vdev, "Guest moved used index from %u to %u",
1416                      idx, vq->shadow_avail_idx);
1417         return -EINVAL;
1418     }
1419     /* On success, callers read a descriptor at vq->last_avail_idx.
1420      * Make sure descriptor read does not bypass avail index read. */
1421     if (num_heads) {
1422         smp_rmb();
1423     }
1424 
1425     return num_heads;
1426 }
1427 
1428 /* Called within rcu_read_lock().  */
1429 static bool virtqueue_get_head(VirtQueue *vq, unsigned int idx,
1430                                unsigned int *head)
1431 {
1432     /* Grab the next descriptor number they're advertising, and increment
1433      * the index we've seen. */
1434     *head = vring_avail_ring(vq, idx % vq->vring.num);
1435 
1436     /* If their number is silly, that's a fatal mistake. */
1437     if (*head >= vq->vring.num) {
1438         virtio_error(vq->vdev, "Guest says index %u is available", *head);
1439         return false;
1440     }
1441 
1442     return true;
1443 }
1444 
1445 enum {
1446     VIRTQUEUE_READ_DESC_ERROR = -1,
1447     VIRTQUEUE_READ_DESC_DONE = 0,   /* end of chain */
1448     VIRTQUEUE_READ_DESC_MORE = 1,   /* more buffers in chain */
1449 };
1450 
1451 static int virtqueue_split_read_next_desc(VirtIODevice *vdev, VRingDesc *desc,
1452                                           MemoryRegionCache *desc_cache,
1453                                           unsigned int max, unsigned int *next)
1454 {
1455     /* If this descriptor says it doesn't chain, we're done. */
1456     if (!(desc->flags & VRING_DESC_F_NEXT)) {
1457         return VIRTQUEUE_READ_DESC_DONE;
1458     }
1459 
1460     /* Check they're not leading us off end of descriptors. */
1461     *next = desc->next;
1462     /* Make sure compiler knows to grab that: we don't want it changing! */
1463     smp_wmb();
1464 
1465     if (*next >= max) {
1466         virtio_error(vdev, "Desc next is %u", *next);
1467         return VIRTQUEUE_READ_DESC_ERROR;
1468     }
1469 
1470     vring_split_desc_read(vdev, desc, desc_cache, *next);
1471     return VIRTQUEUE_READ_DESC_MORE;
1472 }
1473 
1474 /* Called within rcu_read_lock().  */
1475 static void virtqueue_split_get_avail_bytes(VirtQueue *vq,
1476                             unsigned int *in_bytes, unsigned int *out_bytes,
1477                             unsigned max_in_bytes, unsigned max_out_bytes,
1478                             VRingMemoryRegionCaches *caches)
1479 {
1480     VirtIODevice *vdev = vq->vdev;
1481     unsigned int max, idx;
1482     unsigned int total_bufs, in_total, out_total;
1483     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1484     int64_t len = 0;
1485     int rc;
1486 
1487     idx = vq->last_avail_idx;
1488     total_bufs = in_total = out_total = 0;
1489 
1490     max = vq->vring.num;
1491 
1492     while ((rc = virtqueue_num_heads(vq, idx)) > 0) {
1493         MemoryRegionCache *desc_cache = &caches->desc;
1494         unsigned int num_bufs;
1495         VRingDesc desc;
1496         unsigned int i;
1497 
1498         num_bufs = total_bufs;
1499 
1500         if (!virtqueue_get_head(vq, idx++, &i)) {
1501             goto err;
1502         }
1503 
1504         vring_split_desc_read(vdev, &desc, desc_cache, i);
1505 
1506         if (desc.flags & VRING_DESC_F_INDIRECT) {
1507             if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1508                 virtio_error(vdev, "Invalid size for indirect buffer table");
1509                 goto err;
1510             }
1511 
1512             /* If we've got too many, that implies a descriptor loop. */
1513             if (num_bufs >= max) {
1514                 virtio_error(vdev, "Looped descriptor");
1515                 goto err;
1516             }
1517 
1518             /* loop over the indirect descriptor table */
1519             len = address_space_cache_init(&indirect_desc_cache,
1520                                            vdev->dma_as,
1521                                            desc.addr, desc.len, false);
1522             desc_cache = &indirect_desc_cache;
1523             if (len < desc.len) {
1524                 virtio_error(vdev, "Cannot map indirect buffer");
1525                 goto err;
1526             }
1527 
1528             max = desc.len / sizeof(VRingDesc);
1529             num_bufs = i = 0;
1530             vring_split_desc_read(vdev, &desc, desc_cache, i);
1531         }
1532 
1533         do {
1534             /* If we've got too many, that implies a descriptor loop. */
1535             if (++num_bufs > max) {
1536                 virtio_error(vdev, "Looped descriptor");
1537                 goto err;
1538             }
1539 
1540             if (desc.flags & VRING_DESC_F_WRITE) {
1541                 in_total += desc.len;
1542             } else {
1543                 out_total += desc.len;
1544             }
1545             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1546                 goto done;
1547             }
1548 
1549             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
1550         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1551 
1552         if (rc == VIRTQUEUE_READ_DESC_ERROR) {
1553             goto err;
1554         }
1555 
1556         if (desc_cache == &indirect_desc_cache) {
1557             address_space_cache_destroy(&indirect_desc_cache);
1558             total_bufs++;
1559         } else {
1560             total_bufs = num_bufs;
1561         }
1562     }
1563 
1564     if (rc < 0) {
1565         goto err;
1566     }
1567 
1568 done:
1569     address_space_cache_destroy(&indirect_desc_cache);
1570     if (in_bytes) {
1571         *in_bytes = in_total;
1572     }
1573     if (out_bytes) {
1574         *out_bytes = out_total;
1575     }
1576     return;
1577 
1578 err:
1579     in_total = out_total = 0;
1580     goto done;
1581 }
1582 
1583 static int virtqueue_packed_read_next_desc(VirtQueue *vq,
1584                                            VRingPackedDesc *desc,
1585                                            MemoryRegionCache
1586                                            *desc_cache,
1587                                            unsigned int max,
1588                                            unsigned int *next,
1589                                            bool indirect)
1590 {
1591     /* If this descriptor says it doesn't chain, we're done. */
1592     if (!indirect && !(desc->flags & VRING_DESC_F_NEXT)) {
1593         return VIRTQUEUE_READ_DESC_DONE;
1594     }
1595 
1596     ++*next;
1597     if (*next == max) {
1598         if (indirect) {
1599             return VIRTQUEUE_READ_DESC_DONE;
1600         } else {
1601             (*next) -= vq->vring.num;
1602         }
1603     }
1604 
1605     vring_packed_desc_read(vq->vdev, desc, desc_cache, *next, false);
1606     return VIRTQUEUE_READ_DESC_MORE;
1607 }
1608 
1609 /* Called within rcu_read_lock().  */
1610 static void virtqueue_packed_get_avail_bytes(VirtQueue *vq,
1611                                              unsigned int *in_bytes,
1612                                              unsigned int *out_bytes,
1613                                              unsigned max_in_bytes,
1614                                              unsigned max_out_bytes,
1615                                              VRingMemoryRegionCaches *caches)
1616 {
1617     VirtIODevice *vdev = vq->vdev;
1618     unsigned int max, idx;
1619     unsigned int total_bufs, in_total, out_total;
1620     MemoryRegionCache *desc_cache;
1621     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1622     int64_t len = 0;
1623     VRingPackedDesc desc;
1624     bool wrap_counter;
1625 
1626     idx = vq->last_avail_idx;
1627     wrap_counter = vq->last_avail_wrap_counter;
1628     total_bufs = in_total = out_total = 0;
1629 
1630     max = vq->vring.num;
1631 
1632     for (;;) {
1633         unsigned int num_bufs = total_bufs;
1634         unsigned int i = idx;
1635         int rc;
1636 
1637         desc_cache = &caches->desc;
1638         vring_packed_desc_read(vdev, &desc, desc_cache, idx, true);
1639         if (!is_desc_avail(desc.flags, wrap_counter)) {
1640             break;
1641         }
1642 
1643         if (desc.flags & VRING_DESC_F_INDIRECT) {
1644             if (desc.len % sizeof(VRingPackedDesc)) {
1645                 virtio_error(vdev, "Invalid size for indirect buffer table");
1646                 goto err;
1647             }
1648 
1649             /* If we've got too many, that implies a descriptor loop. */
1650             if (num_bufs >= max) {
1651                 virtio_error(vdev, "Looped descriptor");
1652                 goto err;
1653             }
1654 
1655             /* loop over the indirect descriptor table */
1656             len = address_space_cache_init(&indirect_desc_cache,
1657                                            vdev->dma_as,
1658                                            desc.addr, desc.len, false);
1659             desc_cache = &indirect_desc_cache;
1660             if (len < desc.len) {
1661                 virtio_error(vdev, "Cannot map indirect buffer");
1662                 goto err;
1663             }
1664 
1665             max = desc.len / sizeof(VRingPackedDesc);
1666             num_bufs = i = 0;
1667             vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
1668         }
1669 
1670         do {
1671             /* If we've got too many, that implies a descriptor loop. */
1672             if (++num_bufs > max) {
1673                 virtio_error(vdev, "Looped descriptor");
1674                 goto err;
1675             }
1676 
1677             if (desc.flags & VRING_DESC_F_WRITE) {
1678                 in_total += desc.len;
1679             } else {
1680                 out_total += desc.len;
1681             }
1682             if (in_total >= max_in_bytes && out_total >= max_out_bytes) {
1683                 goto done;
1684             }
1685 
1686             rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max,
1687                                                  &i, desc_cache ==
1688                                                  &indirect_desc_cache);
1689         } while (rc == VIRTQUEUE_READ_DESC_MORE);
1690 
1691         if (desc_cache == &indirect_desc_cache) {
1692             address_space_cache_destroy(&indirect_desc_cache);
1693             total_bufs++;
1694             idx++;
1695         } else {
1696             idx += num_bufs - total_bufs;
1697             total_bufs = num_bufs;
1698         }
1699 
1700         if (idx >= vq->vring.num) {
1701             idx -= vq->vring.num;
1702             wrap_counter ^= 1;
1703         }
1704     }
1705 
1706     /* Record the index and wrap counter for a kick we want */
1707     vq->shadow_avail_idx = idx;
1708     vq->shadow_avail_wrap_counter = wrap_counter;
1709 done:
1710     address_space_cache_destroy(&indirect_desc_cache);
1711     if (in_bytes) {
1712         *in_bytes = in_total;
1713     }
1714     if (out_bytes) {
1715         *out_bytes = out_total;
1716     }
1717     return;
1718 
1719 err:
1720     in_total = out_total = 0;
1721     goto done;
1722 }
1723 
1724 void virtqueue_get_avail_bytes(VirtQueue *vq, unsigned int *in_bytes,
1725                                unsigned int *out_bytes,
1726                                unsigned max_in_bytes, unsigned max_out_bytes)
1727 {
1728     uint16_t desc_size;
1729     VRingMemoryRegionCaches *caches;
1730 
1731     RCU_READ_LOCK_GUARD();
1732 
1733     if (unlikely(!vq->vring.desc)) {
1734         goto err;
1735     }
1736 
1737     caches = vring_get_region_caches(vq);
1738     if (!caches) {
1739         goto err;
1740     }
1741 
1742     desc_size = virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED) ?
1743                                 sizeof(VRingPackedDesc) : sizeof(VRingDesc);
1744     if (caches->desc.len < vq->vring.num * desc_size) {
1745         virtio_error(vq->vdev, "Cannot map descriptor ring");
1746         goto err;
1747     }
1748 
1749     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
1750         virtqueue_packed_get_avail_bytes(vq, in_bytes, out_bytes,
1751                                          max_in_bytes, max_out_bytes,
1752                                          caches);
1753     } else {
1754         virtqueue_split_get_avail_bytes(vq, in_bytes, out_bytes,
1755                                         max_in_bytes, max_out_bytes,
1756                                         caches);
1757     }
1758 
1759     return;
1760 err:
1761     if (in_bytes) {
1762         *in_bytes = 0;
1763     }
1764     if (out_bytes) {
1765         *out_bytes = 0;
1766     }
1767 }
1768 
1769 int virtqueue_avail_bytes(VirtQueue *vq, unsigned int in_bytes,
1770                           unsigned int out_bytes)
1771 {
1772     unsigned int in_total, out_total;
1773 
1774     virtqueue_get_avail_bytes(vq, &in_total, &out_total, in_bytes, out_bytes);
1775     return in_bytes <= in_total && out_bytes <= out_total;
1776 }
1777 
1778 static bool virtqueue_map_desc(VirtIODevice *vdev, unsigned int *p_num_sg,
1779                                hwaddr *addr, struct iovec *iov,
1780                                unsigned int max_num_sg, bool is_write,
1781                                hwaddr pa, size_t sz)
1782 {
1783     bool ok = false;
1784     unsigned num_sg = *p_num_sg;
1785     assert(num_sg <= max_num_sg);
1786 
1787     if (!sz) {
1788         virtio_error(vdev, "virtio: zero sized buffers are not allowed");
1789         goto out;
1790     }
1791 
1792     while (sz) {
1793         hwaddr len = sz;
1794 
1795         if (num_sg == max_num_sg) {
1796             virtio_error(vdev, "virtio: too many write descriptors in "
1797                                "indirect table");
1798             goto out;
1799         }
1800 
1801         iov[num_sg].iov_base = dma_memory_map(vdev->dma_as, pa, &len,
1802                                               is_write ?
1803                                               DMA_DIRECTION_FROM_DEVICE :
1804                                               DMA_DIRECTION_TO_DEVICE,
1805                                               MEMTXATTRS_UNSPECIFIED);
1806         if (!iov[num_sg].iov_base) {
1807             virtio_error(vdev, "virtio: bogus descriptor or out of resources");
1808             goto out;
1809         }
1810 
1811         iov[num_sg].iov_len = len;
1812         addr[num_sg] = pa;
1813 
1814         sz -= len;
1815         pa += len;
1816         num_sg++;
1817     }
1818     ok = true;
1819 
1820 out:
1821     *p_num_sg = num_sg;
1822     return ok;
1823 }
1824 
1825 /* Only used by error code paths before we have a VirtQueueElement (therefore
1826  * virtqueue_unmap_sg() can't be used).  Assumes buffers weren't written to
1827  * yet.
1828  */
1829 static void virtqueue_undo_map_desc(unsigned int out_num, unsigned int in_num,
1830                                     struct iovec *iov)
1831 {
1832     unsigned int i;
1833 
1834     for (i = 0; i < out_num + in_num; i++) {
1835         int is_write = i >= out_num;
1836 
1837         cpu_physical_memory_unmap(iov->iov_base, iov->iov_len, is_write, 0);
1838         iov++;
1839     }
1840 }
1841 
1842 static void virtqueue_map_iovec(VirtIODevice *vdev, struct iovec *sg,
1843                                 hwaddr *addr, unsigned int num_sg,
1844                                 bool is_write)
1845 {
1846     unsigned int i;
1847     hwaddr len;
1848 
1849     for (i = 0; i < num_sg; i++) {
1850         len = sg[i].iov_len;
1851         sg[i].iov_base = dma_memory_map(vdev->dma_as,
1852                                         addr[i], &len, is_write ?
1853                                         DMA_DIRECTION_FROM_DEVICE :
1854                                         DMA_DIRECTION_TO_DEVICE,
1855                                         MEMTXATTRS_UNSPECIFIED);
1856         if (!sg[i].iov_base) {
1857             error_report("virtio: error trying to map MMIO memory");
1858             exit(1);
1859         }
1860         if (len != sg[i].iov_len) {
1861             error_report("virtio: unexpected memory split");
1862             exit(1);
1863         }
1864     }
1865 }
1866 
1867 void virtqueue_map(VirtIODevice *vdev, VirtQueueElement *elem)
1868 {
1869     virtqueue_map_iovec(vdev, elem->in_sg, elem->in_addr, elem->in_num, true);
1870     virtqueue_map_iovec(vdev, elem->out_sg, elem->out_addr, elem->out_num,
1871                                                                         false);
1872 }
1873 
1874 static void *virtqueue_alloc_element(size_t sz, unsigned out_num, unsigned in_num)
1875 {
1876     VirtQueueElement *elem;
1877     size_t in_addr_ofs = QEMU_ALIGN_UP(sz, __alignof__(elem->in_addr[0]));
1878     size_t out_addr_ofs = in_addr_ofs + in_num * sizeof(elem->in_addr[0]);
1879     size_t out_addr_end = out_addr_ofs + out_num * sizeof(elem->out_addr[0]);
1880     size_t in_sg_ofs = QEMU_ALIGN_UP(out_addr_end, __alignof__(elem->in_sg[0]));
1881     size_t out_sg_ofs = in_sg_ofs + in_num * sizeof(elem->in_sg[0]);
1882     size_t out_sg_end = out_sg_ofs + out_num * sizeof(elem->out_sg[0]);
1883 
1884     assert(sz >= sizeof(VirtQueueElement));
1885     elem = g_malloc(out_sg_end);
1886     trace_virtqueue_alloc_element(elem, sz, in_num, out_num);
1887     elem->out_num = out_num;
1888     elem->in_num = in_num;
1889     elem->in_addr = (void *)elem + in_addr_ofs;
1890     elem->out_addr = (void *)elem + out_addr_ofs;
1891     elem->in_sg = (void *)elem + in_sg_ofs;
1892     elem->out_sg = (void *)elem + out_sg_ofs;
1893     return elem;
1894 }
1895 
1896 static void *virtqueue_split_pop(VirtQueue *vq, size_t sz)
1897 {
1898     unsigned int i, head, max;
1899     VRingMemoryRegionCaches *caches;
1900     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
1901     MemoryRegionCache *desc_cache;
1902     int64_t len;
1903     VirtIODevice *vdev = vq->vdev;
1904     VirtQueueElement *elem = NULL;
1905     unsigned out_num, in_num, elem_entries;
1906     hwaddr addr[VIRTQUEUE_MAX_SIZE];
1907     struct iovec iov[VIRTQUEUE_MAX_SIZE];
1908     VRingDesc desc;
1909     int rc;
1910 
1911     RCU_READ_LOCK_GUARD();
1912     if (virtio_queue_empty_rcu(vq)) {
1913         goto done;
1914     }
1915     /* Needed after virtio_queue_empty(), see comment in
1916      * virtqueue_num_heads(). */
1917     smp_rmb();
1918 
1919     /* When we start there are none of either input nor output. */
1920     out_num = in_num = elem_entries = 0;
1921 
1922     max = vq->vring.num;
1923 
1924     if (vq->inuse >= vq->vring.num) {
1925         virtio_error(vdev, "Virtqueue size exceeded");
1926         goto done;
1927     }
1928 
1929     if (!virtqueue_get_head(vq, vq->last_avail_idx++, &head)) {
1930         goto done;
1931     }
1932 
1933     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
1934         vring_set_avail_event(vq, vq->last_avail_idx);
1935     }
1936 
1937     i = head;
1938 
1939     caches = vring_get_region_caches(vq);
1940     if (!caches) {
1941         virtio_error(vdev, "Region caches not initialized");
1942         goto done;
1943     }
1944 
1945     if (caches->desc.len < max * sizeof(VRingDesc)) {
1946         virtio_error(vdev, "Cannot map descriptor ring");
1947         goto done;
1948     }
1949 
1950     desc_cache = &caches->desc;
1951     vring_split_desc_read(vdev, &desc, desc_cache, i);
1952     if (desc.flags & VRING_DESC_F_INDIRECT) {
1953         if (!desc.len || (desc.len % sizeof(VRingDesc))) {
1954             virtio_error(vdev, "Invalid size for indirect buffer table");
1955             goto done;
1956         }
1957 
1958         /* loop over the indirect descriptor table */
1959         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
1960                                        desc.addr, desc.len, false);
1961         desc_cache = &indirect_desc_cache;
1962         if (len < desc.len) {
1963             virtio_error(vdev, "Cannot map indirect buffer");
1964             goto done;
1965         }
1966 
1967         max = desc.len / sizeof(VRingDesc);
1968         i = 0;
1969         vring_split_desc_read(vdev, &desc, desc_cache, i);
1970     }
1971 
1972     /* Collect all the descriptors */
1973     do {
1974         bool map_ok;
1975 
1976         if (desc.flags & VRING_DESC_F_WRITE) {
1977             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
1978                                         iov + out_num,
1979                                         VIRTQUEUE_MAX_SIZE - out_num, true,
1980                                         desc.addr, desc.len);
1981         } else {
1982             if (in_num) {
1983                 virtio_error(vdev, "Incorrect order for descriptors");
1984                 goto err_undo_map;
1985             }
1986             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
1987                                         VIRTQUEUE_MAX_SIZE, false,
1988                                         desc.addr, desc.len);
1989         }
1990         if (!map_ok) {
1991             goto err_undo_map;
1992         }
1993 
1994         /* If we've got too many, that implies a descriptor loop. */
1995         if (++elem_entries > max) {
1996             virtio_error(vdev, "Looped descriptor");
1997             goto err_undo_map;
1998         }
1999 
2000         rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache, max, &i);
2001     } while (rc == VIRTQUEUE_READ_DESC_MORE);
2002 
2003     if (rc == VIRTQUEUE_READ_DESC_ERROR) {
2004         goto err_undo_map;
2005     }
2006 
2007     /* Now copy what we have collected and mapped */
2008     elem = virtqueue_alloc_element(sz, out_num, in_num);
2009     elem->index = head;
2010     elem->ndescs = 1;
2011     for (i = 0; i < out_num; i++) {
2012         elem->out_addr[i] = addr[i];
2013         elem->out_sg[i] = iov[i];
2014     }
2015     for (i = 0; i < in_num; i++) {
2016         elem->in_addr[i] = addr[out_num + i];
2017         elem->in_sg[i] = iov[out_num + i];
2018     }
2019 
2020     vq->inuse++;
2021 
2022     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
2023 done:
2024     address_space_cache_destroy(&indirect_desc_cache);
2025 
2026     return elem;
2027 
2028 err_undo_map:
2029     virtqueue_undo_map_desc(out_num, in_num, iov);
2030     goto done;
2031 }
2032 
2033 static void *virtqueue_packed_pop(VirtQueue *vq, size_t sz)
2034 {
2035     unsigned int i, max;
2036     VRingMemoryRegionCaches *caches;
2037     MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
2038     MemoryRegionCache *desc_cache;
2039     int64_t len;
2040     VirtIODevice *vdev = vq->vdev;
2041     VirtQueueElement *elem = NULL;
2042     unsigned out_num, in_num, elem_entries;
2043     hwaddr addr[VIRTQUEUE_MAX_SIZE];
2044     struct iovec iov[VIRTQUEUE_MAX_SIZE];
2045     VRingPackedDesc desc;
2046     uint16_t id;
2047     int rc;
2048 
2049     RCU_READ_LOCK_GUARD();
2050     if (virtio_queue_packed_empty_rcu(vq)) {
2051         goto done;
2052     }
2053 
2054     /* When we start there are none of either input nor output. */
2055     out_num = in_num = elem_entries = 0;
2056 
2057     max = vq->vring.num;
2058 
2059     if (vq->inuse >= vq->vring.num) {
2060         virtio_error(vdev, "Virtqueue size exceeded");
2061         goto done;
2062     }
2063 
2064     i = vq->last_avail_idx;
2065 
2066     caches = vring_get_region_caches(vq);
2067     if (!caches) {
2068         virtio_error(vdev, "Region caches not initialized");
2069         goto done;
2070     }
2071 
2072     if (caches->desc.len < max * sizeof(VRingDesc)) {
2073         virtio_error(vdev, "Cannot map descriptor ring");
2074         goto done;
2075     }
2076 
2077     desc_cache = &caches->desc;
2078     vring_packed_desc_read(vdev, &desc, desc_cache, i, true);
2079     id = desc.id;
2080     if (desc.flags & VRING_DESC_F_INDIRECT) {
2081         if (desc.len % sizeof(VRingPackedDesc)) {
2082             virtio_error(vdev, "Invalid size for indirect buffer table");
2083             goto done;
2084         }
2085 
2086         /* loop over the indirect descriptor table */
2087         len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
2088                                        desc.addr, desc.len, false);
2089         desc_cache = &indirect_desc_cache;
2090         if (len < desc.len) {
2091             virtio_error(vdev, "Cannot map indirect buffer");
2092             goto done;
2093         }
2094 
2095         max = desc.len / sizeof(VRingPackedDesc);
2096         i = 0;
2097         vring_packed_desc_read(vdev, &desc, desc_cache, i, false);
2098     }
2099 
2100     /* Collect all the descriptors */
2101     do {
2102         bool map_ok;
2103 
2104         if (desc.flags & VRING_DESC_F_WRITE) {
2105             map_ok = virtqueue_map_desc(vdev, &in_num, addr + out_num,
2106                                         iov + out_num,
2107                                         VIRTQUEUE_MAX_SIZE - out_num, true,
2108                                         desc.addr, desc.len);
2109         } else {
2110             if (in_num) {
2111                 virtio_error(vdev, "Incorrect order for descriptors");
2112                 goto err_undo_map;
2113             }
2114             map_ok = virtqueue_map_desc(vdev, &out_num, addr, iov,
2115                                         VIRTQUEUE_MAX_SIZE, false,
2116                                         desc.addr, desc.len);
2117         }
2118         if (!map_ok) {
2119             goto err_undo_map;
2120         }
2121 
2122         /* If we've got too many, that implies a descriptor loop. */
2123         if (++elem_entries > max) {
2124             virtio_error(vdev, "Looped descriptor");
2125             goto err_undo_map;
2126         }
2127 
2128         rc = virtqueue_packed_read_next_desc(vq, &desc, desc_cache, max, &i,
2129                                              desc_cache ==
2130                                              &indirect_desc_cache);
2131     } while (rc == VIRTQUEUE_READ_DESC_MORE);
2132 
2133     /* Now copy what we have collected and mapped */
2134     elem = virtqueue_alloc_element(sz, out_num, in_num);
2135     for (i = 0; i < out_num; i++) {
2136         elem->out_addr[i] = addr[i];
2137         elem->out_sg[i] = iov[i];
2138     }
2139     for (i = 0; i < in_num; i++) {
2140         elem->in_addr[i] = addr[out_num + i];
2141         elem->in_sg[i] = iov[out_num + i];
2142     }
2143 
2144     elem->index = id;
2145     elem->ndescs = (desc_cache == &indirect_desc_cache) ? 1 : elem_entries;
2146     vq->last_avail_idx += elem->ndescs;
2147     vq->inuse += elem->ndescs;
2148 
2149     if (vq->last_avail_idx >= vq->vring.num) {
2150         vq->last_avail_idx -= vq->vring.num;
2151         vq->last_avail_wrap_counter ^= 1;
2152     }
2153 
2154     vq->shadow_avail_idx = vq->last_avail_idx;
2155     vq->shadow_avail_wrap_counter = vq->last_avail_wrap_counter;
2156 
2157     trace_virtqueue_pop(vq, elem, elem->in_num, elem->out_num);
2158 done:
2159     address_space_cache_destroy(&indirect_desc_cache);
2160 
2161     return elem;
2162 
2163 err_undo_map:
2164     virtqueue_undo_map_desc(out_num, in_num, iov);
2165     goto done;
2166 }
2167 
2168 void *virtqueue_pop(VirtQueue *vq, size_t sz)
2169 {
2170     if (virtio_device_disabled(vq->vdev)) {
2171         return NULL;
2172     }
2173 
2174     if (virtio_vdev_has_feature(vq->vdev, VIRTIO_F_RING_PACKED)) {
2175         return virtqueue_packed_pop(vq, sz);
2176     } else {
2177         return virtqueue_split_pop(vq, sz);
2178     }
2179 }
2180 
2181 static unsigned int virtqueue_packed_drop_all(VirtQueue *vq)
2182 {
2183     VRingMemoryRegionCaches *caches;
2184     MemoryRegionCache *desc_cache;
2185     unsigned int dropped = 0;
2186     VirtQueueElement elem = {};
2187     VirtIODevice *vdev = vq->vdev;
2188     VRingPackedDesc desc;
2189 
2190     RCU_READ_LOCK_GUARD();
2191 
2192     caches = vring_get_region_caches(vq);
2193     if (!caches) {
2194         return 0;
2195     }
2196 
2197     desc_cache = &caches->desc;
2198 
2199     virtio_queue_set_notification(vq, 0);
2200 
2201     while (vq->inuse < vq->vring.num) {
2202         unsigned int idx = vq->last_avail_idx;
2203         /*
2204          * works similar to virtqueue_pop but does not map buffers
2205          * and does not allocate any memory.
2206          */
2207         vring_packed_desc_read(vdev, &desc, desc_cache,
2208                                vq->last_avail_idx , true);
2209         if (!is_desc_avail(desc.flags, vq->last_avail_wrap_counter)) {
2210             break;
2211         }
2212         elem.index = desc.id;
2213         elem.ndescs = 1;
2214         while (virtqueue_packed_read_next_desc(vq, &desc, desc_cache,
2215                                                vq->vring.num, &idx, false)) {
2216             ++elem.ndescs;
2217         }
2218         /*
2219          * immediately push the element, nothing to unmap
2220          * as both in_num and out_num are set to 0.
2221          */
2222         virtqueue_push(vq, &elem, 0);
2223         dropped++;
2224         vq->last_avail_idx += elem.ndescs;
2225         if (vq->last_avail_idx >= vq->vring.num) {
2226             vq->last_avail_idx -= vq->vring.num;
2227             vq->last_avail_wrap_counter ^= 1;
2228         }
2229     }
2230 
2231     return dropped;
2232 }
2233 
2234 static unsigned int virtqueue_split_drop_all(VirtQueue *vq)
2235 {
2236     unsigned int dropped = 0;
2237     VirtQueueElement elem = {};
2238     VirtIODevice *vdev = vq->vdev;
2239     bool fEventIdx = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX);
2240 
2241     while (!virtio_queue_empty(vq) && vq->inuse < vq->vring.num) {
2242         /* works similar to virtqueue_pop but does not map buffers
2243         * and does not allocate any memory */
2244         smp_rmb();
2245         if (!virtqueue_get_head(vq, vq->last_avail_idx, &elem.index)) {
2246             break;
2247         }
2248         vq->inuse++;
2249         vq->last_avail_idx++;
2250         if (fEventIdx) {
2251             vring_set_avail_event(vq, vq->last_avail_idx);
2252         }
2253         /* immediately push the element, nothing to unmap
2254          * as both in_num and out_num are set to 0 */
2255         virtqueue_push(vq, &elem, 0);
2256         dropped++;
2257     }
2258 
2259     return dropped;
2260 }
2261 
2262 /* virtqueue_drop_all:
2263  * @vq: The #VirtQueue
2264  * Drops all queued buffers and indicates them to the guest
2265  * as if they are done. Useful when buffers can not be
2266  * processed but must be returned to the guest.
2267  */
2268 unsigned int virtqueue_drop_all(VirtQueue *vq)
2269 {
2270     struct VirtIODevice *vdev = vq->vdev;
2271 
2272     if (virtio_device_disabled(vq->vdev)) {
2273         return 0;
2274     }
2275 
2276     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2277         return virtqueue_packed_drop_all(vq);
2278     } else {
2279         return virtqueue_split_drop_all(vq);
2280     }
2281 }
2282 
2283 /* Reading and writing a structure directly to QEMUFile is *awful*, but
2284  * it is what QEMU has always done by mistake.  We can change it sooner
2285  * or later by bumping the version number of the affected vm states.
2286  * In the meanwhile, since the in-memory layout of VirtQueueElement
2287  * has changed, we need to marshal to and from the layout that was
2288  * used before the change.
2289  */
2290 typedef struct VirtQueueElementOld {
2291     unsigned int index;
2292     unsigned int out_num;
2293     unsigned int in_num;
2294     hwaddr in_addr[VIRTQUEUE_MAX_SIZE];
2295     hwaddr out_addr[VIRTQUEUE_MAX_SIZE];
2296     struct iovec in_sg[VIRTQUEUE_MAX_SIZE];
2297     struct iovec out_sg[VIRTQUEUE_MAX_SIZE];
2298 } VirtQueueElementOld;
2299 
2300 void *qemu_get_virtqueue_element(VirtIODevice *vdev, QEMUFile *f, size_t sz)
2301 {
2302     VirtQueueElement *elem;
2303     VirtQueueElementOld data;
2304     int i;
2305 
2306     qemu_get_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
2307 
2308     /* TODO: teach all callers that this can fail, and return failure instead
2309      * of asserting here.
2310      * This is just one thing (there are probably more) that must be
2311      * fixed before we can allow NDEBUG compilation.
2312      */
2313     assert(ARRAY_SIZE(data.in_addr) >= data.in_num);
2314     assert(ARRAY_SIZE(data.out_addr) >= data.out_num);
2315 
2316     elem = virtqueue_alloc_element(sz, data.out_num, data.in_num);
2317     elem->index = data.index;
2318 
2319     for (i = 0; i < elem->in_num; i++) {
2320         elem->in_addr[i] = data.in_addr[i];
2321     }
2322 
2323     for (i = 0; i < elem->out_num; i++) {
2324         elem->out_addr[i] = data.out_addr[i];
2325     }
2326 
2327     for (i = 0; i < elem->in_num; i++) {
2328         /* Base is overwritten by virtqueue_map.  */
2329         elem->in_sg[i].iov_base = 0;
2330         elem->in_sg[i].iov_len = data.in_sg[i].iov_len;
2331     }
2332 
2333     for (i = 0; i < elem->out_num; i++) {
2334         /* Base is overwritten by virtqueue_map.  */
2335         elem->out_sg[i].iov_base = 0;
2336         elem->out_sg[i].iov_len = data.out_sg[i].iov_len;
2337     }
2338 
2339     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2340         qemu_get_be32s(f, &elem->ndescs);
2341     }
2342 
2343     virtqueue_map(vdev, elem);
2344     return elem;
2345 }
2346 
2347 void qemu_put_virtqueue_element(VirtIODevice *vdev, QEMUFile *f,
2348                                 VirtQueueElement *elem)
2349 {
2350     VirtQueueElementOld data;
2351     int i;
2352 
2353     memset(&data, 0, sizeof(data));
2354     data.index = elem->index;
2355     data.in_num = elem->in_num;
2356     data.out_num = elem->out_num;
2357 
2358     for (i = 0; i < elem->in_num; i++) {
2359         data.in_addr[i] = elem->in_addr[i];
2360     }
2361 
2362     for (i = 0; i < elem->out_num; i++) {
2363         data.out_addr[i] = elem->out_addr[i];
2364     }
2365 
2366     for (i = 0; i < elem->in_num; i++) {
2367         /* Base is overwritten by virtqueue_map when loading.  Do not
2368          * save it, as it would leak the QEMU address space layout.  */
2369         data.in_sg[i].iov_len = elem->in_sg[i].iov_len;
2370     }
2371 
2372     for (i = 0; i < elem->out_num; i++) {
2373         /* Do not save iov_base as above.  */
2374         data.out_sg[i].iov_len = elem->out_sg[i].iov_len;
2375     }
2376 
2377     if (virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
2378         qemu_put_be32s(f, &elem->ndescs);
2379     }
2380 
2381     qemu_put_buffer(f, (uint8_t *)&data, sizeof(VirtQueueElementOld));
2382 }
2383 
2384 /* virtio device */
2385 static void virtio_notify_vector(VirtIODevice *vdev, uint16_t vector)
2386 {
2387     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2388     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2389 
2390     if (virtio_device_disabled(vdev)) {
2391         return;
2392     }
2393 
2394     if (k->notify) {
2395         k->notify(qbus->parent, vector);
2396     }
2397 }
2398 
2399 void virtio_update_irq(VirtIODevice *vdev)
2400 {
2401     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
2402 }
2403 
2404 static int virtio_validate_features(VirtIODevice *vdev)
2405 {
2406     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2407 
2408     if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM) &&
2409         !virtio_vdev_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
2410         return -EFAULT;
2411     }
2412 
2413     if (k->validate_features) {
2414         return k->validate_features(vdev);
2415     } else {
2416         return 0;
2417     }
2418 }
2419 
2420 int virtio_set_status(VirtIODevice *vdev, uint8_t val)
2421 {
2422     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2423     trace_virtio_set_status(vdev, val);
2424 
2425     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2426         if (!(vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) &&
2427             val & VIRTIO_CONFIG_S_FEATURES_OK) {
2428             int ret = virtio_validate_features(vdev);
2429 
2430             if (ret) {
2431                 return ret;
2432             }
2433         }
2434     }
2435 
2436     if ((vdev->status & VIRTIO_CONFIG_S_DRIVER_OK) !=
2437         (val & VIRTIO_CONFIG_S_DRIVER_OK)) {
2438         virtio_set_started(vdev, val & VIRTIO_CONFIG_S_DRIVER_OK);
2439     }
2440 
2441     if (k->set_status) {
2442         k->set_status(vdev, val);
2443     }
2444     vdev->status = val;
2445 
2446     return 0;
2447 }
2448 
2449 static enum virtio_device_endian virtio_default_endian(void)
2450 {
2451     if (target_words_bigendian()) {
2452         return VIRTIO_DEVICE_ENDIAN_BIG;
2453     } else {
2454         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2455     }
2456 }
2457 
2458 static enum virtio_device_endian virtio_current_cpu_endian(void)
2459 {
2460     if (cpu_virtio_is_big_endian(current_cpu)) {
2461         return VIRTIO_DEVICE_ENDIAN_BIG;
2462     } else {
2463         return VIRTIO_DEVICE_ENDIAN_LITTLE;
2464     }
2465 }
2466 
2467 static void __virtio_queue_reset(VirtIODevice *vdev, uint32_t i)
2468 {
2469     vdev->vq[i].vring.desc = 0;
2470     vdev->vq[i].vring.avail = 0;
2471     vdev->vq[i].vring.used = 0;
2472     vdev->vq[i].last_avail_idx = 0;
2473     vdev->vq[i].shadow_avail_idx = 0;
2474     vdev->vq[i].used_idx = 0;
2475     vdev->vq[i].last_avail_wrap_counter = true;
2476     vdev->vq[i].shadow_avail_wrap_counter = true;
2477     vdev->vq[i].used_wrap_counter = true;
2478     virtio_queue_set_vector(vdev, i, VIRTIO_NO_VECTOR);
2479     vdev->vq[i].signalled_used = 0;
2480     vdev->vq[i].signalled_used_valid = false;
2481     vdev->vq[i].notification = true;
2482     vdev->vq[i].vring.num = vdev->vq[i].vring.num_default;
2483     vdev->vq[i].inuse = 0;
2484     virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
2485 }
2486 
2487 void virtio_queue_reset(VirtIODevice *vdev, uint32_t queue_index)
2488 {
2489     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2490 
2491     if (k->queue_reset) {
2492         k->queue_reset(vdev, queue_index);
2493     }
2494 
2495     __virtio_queue_reset(vdev, queue_index);
2496 }
2497 
2498 void virtio_queue_enable(VirtIODevice *vdev, uint32_t queue_index)
2499 {
2500     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2501 
2502     /*
2503      * TODO: Seabios is currently out of spec and triggering this error.
2504      * So this needs to be fixed in Seabios, then this can
2505      * be re-enabled for new machine types only, and also after
2506      * being converted to LOG_GUEST_ERROR.
2507      *
2508     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2509         error_report("queue_enable is only suppported in devices of virtio "
2510                      "1.0 or later.");
2511     }
2512     */
2513 
2514     if (k->queue_enable) {
2515         k->queue_enable(vdev, queue_index);
2516     }
2517 }
2518 
2519 void virtio_reset(void *opaque)
2520 {
2521     VirtIODevice *vdev = opaque;
2522     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2523     int i;
2524 
2525     virtio_set_status(vdev, 0);
2526     if (current_cpu) {
2527         /* Guest initiated reset */
2528         vdev->device_endian = virtio_current_cpu_endian();
2529     } else {
2530         /* System reset */
2531         vdev->device_endian = virtio_default_endian();
2532     }
2533 
2534     if (k->reset) {
2535         k->reset(vdev);
2536     }
2537 
2538     vdev->start_on_kick = false;
2539     vdev->started = false;
2540     vdev->broken = false;
2541     vdev->guest_features = 0;
2542     vdev->queue_sel = 0;
2543     vdev->status = 0;
2544     vdev->disabled = false;
2545     qatomic_set(&vdev->isr, 0);
2546     vdev->config_vector = VIRTIO_NO_VECTOR;
2547     virtio_notify_vector(vdev, vdev->config_vector);
2548 
2549     for(i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2550         __virtio_queue_reset(vdev, i);
2551     }
2552 }
2553 
2554 uint32_t virtio_config_readb(VirtIODevice *vdev, uint32_t addr)
2555 {
2556     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2557     uint8_t val;
2558 
2559     if (addr + sizeof(val) > vdev->config_len) {
2560         return (uint32_t)-1;
2561     }
2562 
2563     k->get_config(vdev, vdev->config);
2564 
2565     val = ldub_p(vdev->config + addr);
2566     return val;
2567 }
2568 
2569 uint32_t virtio_config_readw(VirtIODevice *vdev, uint32_t addr)
2570 {
2571     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2572     uint16_t val;
2573 
2574     if (addr + sizeof(val) > vdev->config_len) {
2575         return (uint32_t)-1;
2576     }
2577 
2578     k->get_config(vdev, vdev->config);
2579 
2580     val = lduw_p(vdev->config + addr);
2581     return val;
2582 }
2583 
2584 uint32_t virtio_config_readl(VirtIODevice *vdev, uint32_t addr)
2585 {
2586     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2587     uint32_t val;
2588 
2589     if (addr + sizeof(val) > vdev->config_len) {
2590         return (uint32_t)-1;
2591     }
2592 
2593     k->get_config(vdev, vdev->config);
2594 
2595     val = ldl_p(vdev->config + addr);
2596     return val;
2597 }
2598 
2599 void virtio_config_writeb(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2600 {
2601     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2602     uint8_t val = data;
2603 
2604     if (addr + sizeof(val) > vdev->config_len) {
2605         return;
2606     }
2607 
2608     stb_p(vdev->config + addr, val);
2609 
2610     if (k->set_config) {
2611         k->set_config(vdev, vdev->config);
2612     }
2613 }
2614 
2615 void virtio_config_writew(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2616 {
2617     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2618     uint16_t val = data;
2619 
2620     if (addr + sizeof(val) > vdev->config_len) {
2621         return;
2622     }
2623 
2624     stw_p(vdev->config + addr, val);
2625 
2626     if (k->set_config) {
2627         k->set_config(vdev, vdev->config);
2628     }
2629 }
2630 
2631 void virtio_config_writel(VirtIODevice *vdev, uint32_t addr, uint32_t data)
2632 {
2633     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2634     uint32_t val = data;
2635 
2636     if (addr + sizeof(val) > vdev->config_len) {
2637         return;
2638     }
2639 
2640     stl_p(vdev->config + addr, val);
2641 
2642     if (k->set_config) {
2643         k->set_config(vdev, vdev->config);
2644     }
2645 }
2646 
2647 uint32_t virtio_config_modern_readb(VirtIODevice *vdev, uint32_t addr)
2648 {
2649     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2650     uint8_t val;
2651 
2652     if (addr + sizeof(val) > vdev->config_len) {
2653         return (uint32_t)-1;
2654     }
2655 
2656     k->get_config(vdev, vdev->config);
2657 
2658     val = ldub_p(vdev->config + addr);
2659     return val;
2660 }
2661 
2662 uint32_t virtio_config_modern_readw(VirtIODevice *vdev, uint32_t addr)
2663 {
2664     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2665     uint16_t val;
2666 
2667     if (addr + sizeof(val) > vdev->config_len) {
2668         return (uint32_t)-1;
2669     }
2670 
2671     k->get_config(vdev, vdev->config);
2672 
2673     val = lduw_le_p(vdev->config + addr);
2674     return val;
2675 }
2676 
2677 uint32_t virtio_config_modern_readl(VirtIODevice *vdev, uint32_t addr)
2678 {
2679     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2680     uint32_t val;
2681 
2682     if (addr + sizeof(val) > vdev->config_len) {
2683         return (uint32_t)-1;
2684     }
2685 
2686     k->get_config(vdev, vdev->config);
2687 
2688     val = ldl_le_p(vdev->config + addr);
2689     return val;
2690 }
2691 
2692 void virtio_config_modern_writeb(VirtIODevice *vdev,
2693                                  uint32_t addr, uint32_t data)
2694 {
2695     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2696     uint8_t val = data;
2697 
2698     if (addr + sizeof(val) > vdev->config_len) {
2699         return;
2700     }
2701 
2702     stb_p(vdev->config + addr, val);
2703 
2704     if (k->set_config) {
2705         k->set_config(vdev, vdev->config);
2706     }
2707 }
2708 
2709 void virtio_config_modern_writew(VirtIODevice *vdev,
2710                                  uint32_t addr, uint32_t data)
2711 {
2712     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2713     uint16_t val = data;
2714 
2715     if (addr + sizeof(val) > vdev->config_len) {
2716         return;
2717     }
2718 
2719     stw_le_p(vdev->config + addr, val);
2720 
2721     if (k->set_config) {
2722         k->set_config(vdev, vdev->config);
2723     }
2724 }
2725 
2726 void virtio_config_modern_writel(VirtIODevice *vdev,
2727                                  uint32_t addr, uint32_t data)
2728 {
2729     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
2730     uint32_t val = data;
2731 
2732     if (addr + sizeof(val) > vdev->config_len) {
2733         return;
2734     }
2735 
2736     stl_le_p(vdev->config + addr, val);
2737 
2738     if (k->set_config) {
2739         k->set_config(vdev, vdev->config);
2740     }
2741 }
2742 
2743 void virtio_queue_set_addr(VirtIODevice *vdev, int n, hwaddr addr)
2744 {
2745     if (!vdev->vq[n].vring.num) {
2746         return;
2747     }
2748     vdev->vq[n].vring.desc = addr;
2749     virtio_queue_update_rings(vdev, n);
2750 }
2751 
2752 hwaddr virtio_queue_get_addr(VirtIODevice *vdev, int n)
2753 {
2754     return vdev->vq[n].vring.desc;
2755 }
2756 
2757 void virtio_queue_set_rings(VirtIODevice *vdev, int n, hwaddr desc,
2758                             hwaddr avail, hwaddr used)
2759 {
2760     if (!vdev->vq[n].vring.num) {
2761         return;
2762     }
2763     vdev->vq[n].vring.desc = desc;
2764     vdev->vq[n].vring.avail = avail;
2765     vdev->vq[n].vring.used = used;
2766     virtio_init_region_cache(vdev, n);
2767 }
2768 
2769 void virtio_queue_set_num(VirtIODevice *vdev, int n, int num)
2770 {
2771     /* Don't allow guest to flip queue between existent and
2772      * nonexistent states, or to set it to an invalid size.
2773      */
2774     if (!!num != !!vdev->vq[n].vring.num ||
2775         num > VIRTQUEUE_MAX_SIZE ||
2776         num < 0) {
2777         return;
2778     }
2779     vdev->vq[n].vring.num = num;
2780 }
2781 
2782 VirtQueue *virtio_vector_first_queue(VirtIODevice *vdev, uint16_t vector)
2783 {
2784     return QLIST_FIRST(&vdev->vector_queues[vector]);
2785 }
2786 
2787 VirtQueue *virtio_vector_next_queue(VirtQueue *vq)
2788 {
2789     return QLIST_NEXT(vq, node);
2790 }
2791 
2792 int virtio_queue_get_num(VirtIODevice *vdev, int n)
2793 {
2794     return vdev->vq[n].vring.num;
2795 }
2796 
2797 int virtio_queue_get_max_num(VirtIODevice *vdev, int n)
2798 {
2799     return vdev->vq[n].vring.num_default;
2800 }
2801 
2802 int virtio_get_num_queues(VirtIODevice *vdev)
2803 {
2804     int i;
2805 
2806     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2807         if (!virtio_queue_get_num(vdev, i)) {
2808             break;
2809         }
2810     }
2811 
2812     return i;
2813 }
2814 
2815 void virtio_queue_set_align(VirtIODevice *vdev, int n, int align)
2816 {
2817     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
2818     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
2819 
2820     /* virtio-1 compliant devices cannot change the alignment */
2821     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
2822         error_report("tried to modify queue alignment for virtio-1 device");
2823         return;
2824     }
2825     /* Check that the transport told us it was going to do this
2826      * (so a buggy transport will immediately assert rather than
2827      * silently failing to migrate this state)
2828      */
2829     assert(k->has_variable_vring_alignment);
2830 
2831     if (align) {
2832         vdev->vq[n].vring.align = align;
2833         virtio_queue_update_rings(vdev, n);
2834     }
2835 }
2836 
2837 static void virtio_queue_notify_vq(VirtQueue *vq)
2838 {
2839     if (vq->vring.desc && vq->handle_output) {
2840         VirtIODevice *vdev = vq->vdev;
2841 
2842         if (unlikely(vdev->broken)) {
2843             return;
2844         }
2845 
2846         trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2847         vq->handle_output(vdev, vq);
2848 
2849         if (unlikely(vdev->start_on_kick)) {
2850             virtio_set_started(vdev, true);
2851         }
2852     }
2853 }
2854 
2855 void virtio_queue_notify(VirtIODevice *vdev, int n)
2856 {
2857     VirtQueue *vq = &vdev->vq[n];
2858 
2859     if (unlikely(!vq->vring.desc || vdev->broken)) {
2860         return;
2861     }
2862 
2863     trace_virtio_queue_notify(vdev, vq - vdev->vq, vq);
2864     if (vq->host_notifier_enabled) {
2865         event_notifier_set(&vq->host_notifier);
2866     } else if (vq->handle_output) {
2867         vq->handle_output(vdev, vq);
2868 
2869         if (unlikely(vdev->start_on_kick)) {
2870             virtio_set_started(vdev, true);
2871         }
2872     }
2873 }
2874 
2875 uint16_t virtio_queue_vector(VirtIODevice *vdev, int n)
2876 {
2877     return n < VIRTIO_QUEUE_MAX ? vdev->vq[n].vector :
2878         VIRTIO_NO_VECTOR;
2879 }
2880 
2881 void virtio_queue_set_vector(VirtIODevice *vdev, int n, uint16_t vector)
2882 {
2883     VirtQueue *vq = &vdev->vq[n];
2884 
2885     if (n < VIRTIO_QUEUE_MAX) {
2886         if (vdev->vector_queues &&
2887             vdev->vq[n].vector != VIRTIO_NO_VECTOR) {
2888             QLIST_REMOVE(vq, node);
2889         }
2890         vdev->vq[n].vector = vector;
2891         if (vdev->vector_queues &&
2892             vector != VIRTIO_NO_VECTOR) {
2893             QLIST_INSERT_HEAD(&vdev->vector_queues[vector], vq, node);
2894         }
2895     }
2896 }
2897 
2898 VirtQueue *virtio_add_queue(VirtIODevice *vdev, int queue_size,
2899                             VirtIOHandleOutput handle_output)
2900 {
2901     int i;
2902 
2903     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2904         if (vdev->vq[i].vring.num == 0)
2905             break;
2906     }
2907 
2908     if (i == VIRTIO_QUEUE_MAX || queue_size > VIRTQUEUE_MAX_SIZE)
2909         abort();
2910 
2911     vdev->vq[i].vring.num = queue_size;
2912     vdev->vq[i].vring.num_default = queue_size;
2913     vdev->vq[i].vring.align = VIRTIO_PCI_VRING_ALIGN;
2914     vdev->vq[i].handle_output = handle_output;
2915     vdev->vq[i].used_elems = g_new0(VirtQueueElement, queue_size);
2916 
2917     return &vdev->vq[i];
2918 }
2919 
2920 void virtio_delete_queue(VirtQueue *vq)
2921 {
2922     vq->vring.num = 0;
2923     vq->vring.num_default = 0;
2924     vq->handle_output = NULL;
2925     g_free(vq->used_elems);
2926     vq->used_elems = NULL;
2927     virtio_virtqueue_reset_region_cache(vq);
2928 }
2929 
2930 void virtio_del_queue(VirtIODevice *vdev, int n)
2931 {
2932     if (n < 0 || n >= VIRTIO_QUEUE_MAX) {
2933         abort();
2934     }
2935 
2936     virtio_delete_queue(&vdev->vq[n]);
2937 }
2938 
2939 static void virtio_set_isr(VirtIODevice *vdev, int value)
2940 {
2941     uint8_t old = qatomic_read(&vdev->isr);
2942 
2943     /* Do not write ISR if it does not change, so that its cacheline remains
2944      * shared in the common case where the guest does not read it.
2945      */
2946     if ((old & value) != value) {
2947         qatomic_or(&vdev->isr, value);
2948     }
2949 }
2950 
2951 /* Called within rcu_read_lock(). */
2952 static bool virtio_split_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2953 {
2954     uint16_t old, new;
2955     bool v;
2956     /* We need to expose used array entries before checking used event. */
2957     smp_mb();
2958     /* Always notify when queue is empty (when feature acknowledge) */
2959     if (virtio_vdev_has_feature(vdev, VIRTIO_F_NOTIFY_ON_EMPTY) &&
2960         !vq->inuse && virtio_queue_empty(vq)) {
2961         return true;
2962     }
2963 
2964     if (!virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
2965         return !(vring_avail_flags(vq) & VRING_AVAIL_F_NO_INTERRUPT);
2966     }
2967 
2968     v = vq->signalled_used_valid;
2969     vq->signalled_used_valid = true;
2970     old = vq->signalled_used;
2971     new = vq->signalled_used = vq->used_idx;
2972     return !v || vring_need_event(vring_get_used_event(vq), new, old);
2973 }
2974 
2975 static bool vring_packed_need_event(VirtQueue *vq, bool wrap,
2976                                     uint16_t off_wrap, uint16_t new,
2977                                     uint16_t old)
2978 {
2979     int off = off_wrap & ~(1 << 15);
2980 
2981     if (wrap != off_wrap >> 15) {
2982         off -= vq->vring.num;
2983     }
2984 
2985     return vring_need_event(off, new, old);
2986 }
2987 
2988 /* Called within rcu_read_lock(). */
2989 static bool virtio_packed_should_notify(VirtIODevice *vdev, VirtQueue *vq)
2990 {
2991     VRingPackedDescEvent e;
2992     uint16_t old, new;
2993     bool v;
2994     VRingMemoryRegionCaches *caches;
2995 
2996     caches = vring_get_region_caches(vq);
2997     if (!caches) {
2998         return false;
2999     }
3000 
3001     vring_packed_event_read(vdev, &caches->avail, &e);
3002 
3003     old = vq->signalled_used;
3004     new = vq->signalled_used = vq->used_idx;
3005     v = vq->signalled_used_valid;
3006     vq->signalled_used_valid = true;
3007 
3008     if (e.flags == VRING_PACKED_EVENT_FLAG_DISABLE) {
3009         return false;
3010     } else if (e.flags == VRING_PACKED_EVENT_FLAG_ENABLE) {
3011         return true;
3012     }
3013 
3014     return !v || vring_packed_need_event(vq, vq->used_wrap_counter,
3015                                          e.off_wrap, new, old);
3016 }
3017 
3018 /* Called within rcu_read_lock().  */
3019 static bool virtio_should_notify(VirtIODevice *vdev, VirtQueue *vq)
3020 {
3021     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3022         return virtio_packed_should_notify(vdev, vq);
3023     } else {
3024         return virtio_split_should_notify(vdev, vq);
3025     }
3026 }
3027 
3028 void virtio_notify_irqfd(VirtIODevice *vdev, VirtQueue *vq)
3029 {
3030     WITH_RCU_READ_LOCK_GUARD() {
3031         if (!virtio_should_notify(vdev, vq)) {
3032             return;
3033         }
3034     }
3035 
3036     trace_virtio_notify_irqfd(vdev, vq);
3037 
3038     /*
3039      * virtio spec 1.0 says ISR bit 0 should be ignored with MSI, but
3040      * windows drivers included in virtio-win 1.8.0 (circa 2015) are
3041      * incorrectly polling this bit during crashdump and hibernation
3042      * in MSI mode, causing a hang if this bit is never updated.
3043      * Recent releases of Windows do not really shut down, but rather
3044      * log out and hibernate to make the next startup faster.  Hence,
3045      * this manifested as a more serious hang during shutdown with
3046      *
3047      * Next driver release from 2016 fixed this problem, so working around it
3048      * is not a must, but it's easy to do so let's do it here.
3049      *
3050      * Note: it's safe to update ISR from any thread as it was switched
3051      * to an atomic operation.
3052      */
3053     virtio_set_isr(vq->vdev, 0x1);
3054     event_notifier_set(&vq->guest_notifier);
3055 }
3056 
3057 static void virtio_irq(VirtQueue *vq)
3058 {
3059     virtio_set_isr(vq->vdev, 0x1);
3060     virtio_notify_vector(vq->vdev, vq->vector);
3061 }
3062 
3063 void virtio_notify(VirtIODevice *vdev, VirtQueue *vq)
3064 {
3065     WITH_RCU_READ_LOCK_GUARD() {
3066         if (!virtio_should_notify(vdev, vq)) {
3067             return;
3068         }
3069     }
3070 
3071     trace_virtio_notify(vdev, vq);
3072     virtio_irq(vq);
3073 }
3074 
3075 void virtio_notify_config(VirtIODevice *vdev)
3076 {
3077     if (!(vdev->status & VIRTIO_CONFIG_S_DRIVER_OK))
3078         return;
3079 
3080     virtio_set_isr(vdev, 0x3);
3081     vdev->generation++;
3082     virtio_notify_vector(vdev, vdev->config_vector);
3083 }
3084 
3085 static bool virtio_device_endian_needed(void *opaque)
3086 {
3087     VirtIODevice *vdev = opaque;
3088 
3089     assert(vdev->device_endian != VIRTIO_DEVICE_ENDIAN_UNKNOWN);
3090     if (!virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3091         return vdev->device_endian != virtio_default_endian();
3092     }
3093     /* Devices conforming to VIRTIO 1.0 or later are always LE. */
3094     return vdev->device_endian != VIRTIO_DEVICE_ENDIAN_LITTLE;
3095 }
3096 
3097 static bool virtio_64bit_features_needed(void *opaque)
3098 {
3099     VirtIODevice *vdev = opaque;
3100 
3101     return (vdev->host_features >> 32) != 0;
3102 }
3103 
3104 static bool virtio_virtqueue_needed(void *opaque)
3105 {
3106     VirtIODevice *vdev = opaque;
3107 
3108     return virtio_host_has_feature(vdev, VIRTIO_F_VERSION_1);
3109 }
3110 
3111 static bool virtio_packed_virtqueue_needed(void *opaque)
3112 {
3113     VirtIODevice *vdev = opaque;
3114 
3115     return virtio_host_has_feature(vdev, VIRTIO_F_RING_PACKED);
3116 }
3117 
3118 static bool virtio_ringsize_needed(void *opaque)
3119 {
3120     VirtIODevice *vdev = opaque;
3121     int i;
3122 
3123     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3124         if (vdev->vq[i].vring.num != vdev->vq[i].vring.num_default) {
3125             return true;
3126         }
3127     }
3128     return false;
3129 }
3130 
3131 static bool virtio_extra_state_needed(void *opaque)
3132 {
3133     VirtIODevice *vdev = opaque;
3134     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3135     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3136 
3137     return k->has_extra_state &&
3138         k->has_extra_state(qbus->parent);
3139 }
3140 
3141 static bool virtio_broken_needed(void *opaque)
3142 {
3143     VirtIODevice *vdev = opaque;
3144 
3145     return vdev->broken;
3146 }
3147 
3148 static bool virtio_started_needed(void *opaque)
3149 {
3150     VirtIODevice *vdev = opaque;
3151 
3152     return vdev->started;
3153 }
3154 
3155 static bool virtio_disabled_needed(void *opaque)
3156 {
3157     VirtIODevice *vdev = opaque;
3158 
3159     return vdev->disabled;
3160 }
3161 
3162 static const VMStateDescription vmstate_virtqueue = {
3163     .name = "virtqueue_state",
3164     .version_id = 1,
3165     .minimum_version_id = 1,
3166     .fields = (VMStateField[]) {
3167         VMSTATE_UINT64(vring.avail, struct VirtQueue),
3168         VMSTATE_UINT64(vring.used, struct VirtQueue),
3169         VMSTATE_END_OF_LIST()
3170     }
3171 };
3172 
3173 static const VMStateDescription vmstate_packed_virtqueue = {
3174     .name = "packed_virtqueue_state",
3175     .version_id = 1,
3176     .minimum_version_id = 1,
3177     .fields = (VMStateField[]) {
3178         VMSTATE_UINT16(last_avail_idx, struct VirtQueue),
3179         VMSTATE_BOOL(last_avail_wrap_counter, struct VirtQueue),
3180         VMSTATE_UINT16(used_idx, struct VirtQueue),
3181         VMSTATE_BOOL(used_wrap_counter, struct VirtQueue),
3182         VMSTATE_UINT32(inuse, struct VirtQueue),
3183         VMSTATE_END_OF_LIST()
3184     }
3185 };
3186 
3187 static const VMStateDescription vmstate_virtio_virtqueues = {
3188     .name = "virtio/virtqueues",
3189     .version_id = 1,
3190     .minimum_version_id = 1,
3191     .needed = &virtio_virtqueue_needed,
3192     .fields = (VMStateField[]) {
3193         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
3194                       VIRTIO_QUEUE_MAX, 0, vmstate_virtqueue, VirtQueue),
3195         VMSTATE_END_OF_LIST()
3196     }
3197 };
3198 
3199 static const VMStateDescription vmstate_virtio_packed_virtqueues = {
3200     .name = "virtio/packed_virtqueues",
3201     .version_id = 1,
3202     .minimum_version_id = 1,
3203     .needed = &virtio_packed_virtqueue_needed,
3204     .fields = (VMStateField[]) {
3205         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
3206                       VIRTIO_QUEUE_MAX, 0, vmstate_packed_virtqueue, VirtQueue),
3207         VMSTATE_END_OF_LIST()
3208     }
3209 };
3210 
3211 static const VMStateDescription vmstate_ringsize = {
3212     .name = "ringsize_state",
3213     .version_id = 1,
3214     .minimum_version_id = 1,
3215     .fields = (VMStateField[]) {
3216         VMSTATE_UINT32(vring.num_default, struct VirtQueue),
3217         VMSTATE_END_OF_LIST()
3218     }
3219 };
3220 
3221 static const VMStateDescription vmstate_virtio_ringsize = {
3222     .name = "virtio/ringsize",
3223     .version_id = 1,
3224     .minimum_version_id = 1,
3225     .needed = &virtio_ringsize_needed,
3226     .fields = (VMStateField[]) {
3227         VMSTATE_STRUCT_VARRAY_POINTER_KNOWN(vq, struct VirtIODevice,
3228                       VIRTIO_QUEUE_MAX, 0, vmstate_ringsize, VirtQueue),
3229         VMSTATE_END_OF_LIST()
3230     }
3231 };
3232 
3233 static int get_extra_state(QEMUFile *f, void *pv, size_t size,
3234                            const VMStateField *field)
3235 {
3236     VirtIODevice *vdev = pv;
3237     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3238     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3239 
3240     if (!k->load_extra_state) {
3241         return -1;
3242     } else {
3243         return k->load_extra_state(qbus->parent, f);
3244     }
3245 }
3246 
3247 static int put_extra_state(QEMUFile *f, void *pv, size_t size,
3248                            const VMStateField *field, JSONWriter *vmdesc)
3249 {
3250     VirtIODevice *vdev = pv;
3251     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3252     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3253 
3254     k->save_extra_state(qbus->parent, f);
3255     return 0;
3256 }
3257 
3258 static const VMStateInfo vmstate_info_extra_state = {
3259     .name = "virtqueue_extra_state",
3260     .get = get_extra_state,
3261     .put = put_extra_state,
3262 };
3263 
3264 static const VMStateDescription vmstate_virtio_extra_state = {
3265     .name = "virtio/extra_state",
3266     .version_id = 1,
3267     .minimum_version_id = 1,
3268     .needed = &virtio_extra_state_needed,
3269     .fields = (VMStateField[]) {
3270         {
3271             .name         = "extra_state",
3272             .version_id   = 0,
3273             .field_exists = NULL,
3274             .size         = 0,
3275             .info         = &vmstate_info_extra_state,
3276             .flags        = VMS_SINGLE,
3277             .offset       = 0,
3278         },
3279         VMSTATE_END_OF_LIST()
3280     }
3281 };
3282 
3283 static const VMStateDescription vmstate_virtio_device_endian = {
3284     .name = "virtio/device_endian",
3285     .version_id = 1,
3286     .minimum_version_id = 1,
3287     .needed = &virtio_device_endian_needed,
3288     .fields = (VMStateField[]) {
3289         VMSTATE_UINT8(device_endian, VirtIODevice),
3290         VMSTATE_END_OF_LIST()
3291     }
3292 };
3293 
3294 static const VMStateDescription vmstate_virtio_64bit_features = {
3295     .name = "virtio/64bit_features",
3296     .version_id = 1,
3297     .minimum_version_id = 1,
3298     .needed = &virtio_64bit_features_needed,
3299     .fields = (VMStateField[]) {
3300         VMSTATE_UINT64(guest_features, VirtIODevice),
3301         VMSTATE_END_OF_LIST()
3302     }
3303 };
3304 
3305 static const VMStateDescription vmstate_virtio_broken = {
3306     .name = "virtio/broken",
3307     .version_id = 1,
3308     .minimum_version_id = 1,
3309     .needed = &virtio_broken_needed,
3310     .fields = (VMStateField[]) {
3311         VMSTATE_BOOL(broken, VirtIODevice),
3312         VMSTATE_END_OF_LIST()
3313     }
3314 };
3315 
3316 static const VMStateDescription vmstate_virtio_started = {
3317     .name = "virtio/started",
3318     .version_id = 1,
3319     .minimum_version_id = 1,
3320     .needed = &virtio_started_needed,
3321     .fields = (VMStateField[]) {
3322         VMSTATE_BOOL(started, VirtIODevice),
3323         VMSTATE_END_OF_LIST()
3324     }
3325 };
3326 
3327 static const VMStateDescription vmstate_virtio_disabled = {
3328     .name = "virtio/disabled",
3329     .version_id = 1,
3330     .minimum_version_id = 1,
3331     .needed = &virtio_disabled_needed,
3332     .fields = (VMStateField[]) {
3333         VMSTATE_BOOL(disabled, VirtIODevice),
3334         VMSTATE_END_OF_LIST()
3335     }
3336 };
3337 
3338 static const VMStateDescription vmstate_virtio = {
3339     .name = "virtio",
3340     .version_id = 1,
3341     .minimum_version_id = 1,
3342     .fields = (VMStateField[]) {
3343         VMSTATE_END_OF_LIST()
3344     },
3345     .subsections = (const VMStateDescription*[]) {
3346         &vmstate_virtio_device_endian,
3347         &vmstate_virtio_64bit_features,
3348         &vmstate_virtio_virtqueues,
3349         &vmstate_virtio_ringsize,
3350         &vmstate_virtio_broken,
3351         &vmstate_virtio_extra_state,
3352         &vmstate_virtio_started,
3353         &vmstate_virtio_packed_virtqueues,
3354         &vmstate_virtio_disabled,
3355         NULL
3356     }
3357 };
3358 
3359 int virtio_save(VirtIODevice *vdev, QEMUFile *f)
3360 {
3361     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3362     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3363     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
3364     uint32_t guest_features_lo = (vdev->guest_features & 0xffffffff);
3365     int i;
3366 
3367     if (k->save_config) {
3368         k->save_config(qbus->parent, f);
3369     }
3370 
3371     qemu_put_8s(f, &vdev->status);
3372     qemu_put_8s(f, &vdev->isr);
3373     qemu_put_be16s(f, &vdev->queue_sel);
3374     qemu_put_be32s(f, &guest_features_lo);
3375     qemu_put_be32(f, vdev->config_len);
3376     qemu_put_buffer(f, vdev->config, vdev->config_len);
3377 
3378     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3379         if (vdev->vq[i].vring.num == 0)
3380             break;
3381     }
3382 
3383     qemu_put_be32(f, i);
3384 
3385     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3386         if (vdev->vq[i].vring.num == 0)
3387             break;
3388 
3389         qemu_put_be32(f, vdev->vq[i].vring.num);
3390         if (k->has_variable_vring_alignment) {
3391             qemu_put_be32(f, vdev->vq[i].vring.align);
3392         }
3393         /*
3394          * Save desc now, the rest of the ring addresses are saved in
3395          * subsections for VIRTIO-1 devices.
3396          */
3397         qemu_put_be64(f, vdev->vq[i].vring.desc);
3398         qemu_put_be16s(f, &vdev->vq[i].last_avail_idx);
3399         if (k->save_queue) {
3400             k->save_queue(qbus->parent, i, f);
3401         }
3402     }
3403 
3404     if (vdc->save != NULL) {
3405         vdc->save(vdev, f);
3406     }
3407 
3408     if (vdc->vmsd) {
3409         int ret = vmstate_save_state(f, vdc->vmsd, vdev, NULL);
3410         if (ret) {
3411             return ret;
3412         }
3413     }
3414 
3415     /* Subsections */
3416     return vmstate_save_state(f, &vmstate_virtio, vdev, NULL);
3417 }
3418 
3419 /* A wrapper for use as a VMState .put function */
3420 static int virtio_device_put(QEMUFile *f, void *opaque, size_t size,
3421                               const VMStateField *field, JSONWriter *vmdesc)
3422 {
3423     return virtio_save(VIRTIO_DEVICE(opaque), f);
3424 }
3425 
3426 /* A wrapper for use as a VMState .get function */
3427 static int virtio_device_get(QEMUFile *f, void *opaque, size_t size,
3428                              const VMStateField *field)
3429 {
3430     VirtIODevice *vdev = VIRTIO_DEVICE(opaque);
3431     DeviceClass *dc = DEVICE_CLASS(VIRTIO_DEVICE_GET_CLASS(vdev));
3432 
3433     return virtio_load(vdev, f, dc->vmsd->version_id);
3434 }
3435 
3436 const VMStateInfo  virtio_vmstate_info = {
3437     .name = "virtio",
3438     .get = virtio_device_get,
3439     .put = virtio_device_put,
3440 };
3441 
3442 static int virtio_set_features_nocheck(VirtIODevice *vdev, uint64_t val)
3443 {
3444     VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
3445     bool bad = (val & ~(vdev->host_features)) != 0;
3446 
3447     val &= vdev->host_features;
3448     if (k->set_features) {
3449         k->set_features(vdev, val);
3450     }
3451     vdev->guest_features = val;
3452     return bad ? -1 : 0;
3453 }
3454 
3455 int virtio_set_features(VirtIODevice *vdev, uint64_t val)
3456 {
3457     int ret;
3458     /*
3459      * The driver must not attempt to set features after feature negotiation
3460      * has finished.
3461      */
3462     if (vdev->status & VIRTIO_CONFIG_S_FEATURES_OK) {
3463         return -EINVAL;
3464     }
3465 
3466     if (val & (1ull << VIRTIO_F_BAD_FEATURE)) {
3467         qemu_log_mask(LOG_GUEST_ERROR,
3468                       "%s: guest driver for %s has enabled UNUSED(30) feature bit!\n",
3469                       __func__, vdev->name);
3470     }
3471 
3472     ret = virtio_set_features_nocheck(vdev, val);
3473     if (virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX)) {
3474         /* VIRTIO_RING_F_EVENT_IDX changes the size of the caches.  */
3475         int i;
3476         for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3477             if (vdev->vq[i].vring.num != 0) {
3478                 virtio_init_region_cache(vdev, i);
3479             }
3480         }
3481     }
3482     if (!ret) {
3483         if (!virtio_device_started(vdev, vdev->status) &&
3484             !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3485             vdev->start_on_kick = true;
3486         }
3487     }
3488     return ret;
3489 }
3490 
3491 size_t virtio_get_config_size(const VirtIOConfigSizeParams *params,
3492                               uint64_t host_features)
3493 {
3494     size_t config_size = params->min_size;
3495     const VirtIOFeature *feature_sizes = params->feature_sizes;
3496     size_t i;
3497 
3498     for (i = 0; feature_sizes[i].flags != 0; i++) {
3499         if (host_features & feature_sizes[i].flags) {
3500             config_size = MAX(feature_sizes[i].end, config_size);
3501         }
3502     }
3503 
3504     assert(config_size <= params->max_size);
3505     return config_size;
3506 }
3507 
3508 int virtio_load(VirtIODevice *vdev, QEMUFile *f, int version_id)
3509 {
3510     int i, ret;
3511     int32_t config_len;
3512     uint32_t num;
3513     uint32_t features;
3514     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3515     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3516     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
3517 
3518     /*
3519      * We poison the endianness to ensure it does not get used before
3520      * subsections have been loaded.
3521      */
3522     vdev->device_endian = VIRTIO_DEVICE_ENDIAN_UNKNOWN;
3523 
3524     if (k->load_config) {
3525         ret = k->load_config(qbus->parent, f);
3526         if (ret)
3527             return ret;
3528     }
3529 
3530     qemu_get_8s(f, &vdev->status);
3531     qemu_get_8s(f, &vdev->isr);
3532     qemu_get_be16s(f, &vdev->queue_sel);
3533     if (vdev->queue_sel >= VIRTIO_QUEUE_MAX) {
3534         return -1;
3535     }
3536     qemu_get_be32s(f, &features);
3537 
3538     /*
3539      * Temporarily set guest_features low bits - needed by
3540      * virtio net load code testing for VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
3541      * VIRTIO_NET_F_GUEST_ANNOUNCE and VIRTIO_NET_F_CTRL_VQ.
3542      *
3543      * Note: devices should always test host features in future - don't create
3544      * new dependencies like this.
3545      */
3546     vdev->guest_features = features;
3547 
3548     config_len = qemu_get_be32(f);
3549 
3550     /*
3551      * There are cases where the incoming config can be bigger or smaller
3552      * than what we have; so load what we have space for, and skip
3553      * any excess that's in the stream.
3554      */
3555     qemu_get_buffer(f, vdev->config, MIN(config_len, vdev->config_len));
3556 
3557     while (config_len > vdev->config_len) {
3558         qemu_get_byte(f);
3559         config_len--;
3560     }
3561 
3562     num = qemu_get_be32(f);
3563 
3564     if (num > VIRTIO_QUEUE_MAX) {
3565         error_report("Invalid number of virtqueues: 0x%x", num);
3566         return -1;
3567     }
3568 
3569     for (i = 0; i < num; i++) {
3570         vdev->vq[i].vring.num = qemu_get_be32(f);
3571         if (k->has_variable_vring_alignment) {
3572             vdev->vq[i].vring.align = qemu_get_be32(f);
3573         }
3574         vdev->vq[i].vring.desc = qemu_get_be64(f);
3575         qemu_get_be16s(f, &vdev->vq[i].last_avail_idx);
3576         vdev->vq[i].signalled_used_valid = false;
3577         vdev->vq[i].notification = true;
3578 
3579         if (!vdev->vq[i].vring.desc && vdev->vq[i].last_avail_idx) {
3580             error_report("VQ %d address 0x0 "
3581                          "inconsistent with Host index 0x%x",
3582                          i, vdev->vq[i].last_avail_idx);
3583             return -1;
3584         }
3585         if (k->load_queue) {
3586             ret = k->load_queue(qbus->parent, i, f);
3587             if (ret)
3588                 return ret;
3589         }
3590     }
3591 
3592     virtio_notify_vector(vdev, VIRTIO_NO_VECTOR);
3593 
3594     if (vdc->load != NULL) {
3595         ret = vdc->load(vdev, f, version_id);
3596         if (ret) {
3597             return ret;
3598         }
3599     }
3600 
3601     if (vdc->vmsd) {
3602         ret = vmstate_load_state(f, vdc->vmsd, vdev, version_id);
3603         if (ret) {
3604             return ret;
3605         }
3606     }
3607 
3608     /* Subsections */
3609     ret = vmstate_load_state(f, &vmstate_virtio, vdev, 1);
3610     if (ret) {
3611         return ret;
3612     }
3613 
3614     if (vdev->device_endian == VIRTIO_DEVICE_ENDIAN_UNKNOWN) {
3615         vdev->device_endian = virtio_default_endian();
3616     }
3617 
3618     if (virtio_64bit_features_needed(vdev)) {
3619         /*
3620          * Subsection load filled vdev->guest_features.  Run them
3621          * through virtio_set_features to sanity-check them against
3622          * host_features.
3623          */
3624         uint64_t features64 = vdev->guest_features;
3625         if (virtio_set_features_nocheck(vdev, features64) < 0) {
3626             error_report("Features 0x%" PRIx64 " unsupported. "
3627                          "Allowed features: 0x%" PRIx64,
3628                          features64, vdev->host_features);
3629             return -1;
3630         }
3631     } else {
3632         if (virtio_set_features_nocheck(vdev, features) < 0) {
3633             error_report("Features 0x%x unsupported. "
3634                          "Allowed features: 0x%" PRIx64,
3635                          features, vdev->host_features);
3636             return -1;
3637         }
3638     }
3639 
3640     if (!virtio_device_started(vdev, vdev->status) &&
3641         !virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3642         vdev->start_on_kick = true;
3643     }
3644 
3645     RCU_READ_LOCK_GUARD();
3646     for (i = 0; i < num; i++) {
3647         if (vdev->vq[i].vring.desc) {
3648             uint16_t nheads;
3649 
3650             /*
3651              * VIRTIO-1 devices migrate desc, used, and avail ring addresses so
3652              * only the region cache needs to be set up.  Legacy devices need
3653              * to calculate used and avail ring addresses based on the desc
3654              * address.
3655              */
3656             if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
3657                 virtio_init_region_cache(vdev, i);
3658             } else {
3659                 virtio_queue_update_rings(vdev, i);
3660             }
3661 
3662             if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3663                 vdev->vq[i].shadow_avail_idx = vdev->vq[i].last_avail_idx;
3664                 vdev->vq[i].shadow_avail_wrap_counter =
3665                                         vdev->vq[i].last_avail_wrap_counter;
3666                 continue;
3667             }
3668 
3669             nheads = vring_avail_idx(&vdev->vq[i]) - vdev->vq[i].last_avail_idx;
3670             /* Check it isn't doing strange things with descriptor numbers. */
3671             if (nheads > vdev->vq[i].vring.num) {
3672                 virtio_error(vdev, "VQ %d size 0x%x Guest index 0x%x "
3673                              "inconsistent with Host index 0x%x: delta 0x%x",
3674                              i, vdev->vq[i].vring.num,
3675                              vring_avail_idx(&vdev->vq[i]),
3676                              vdev->vq[i].last_avail_idx, nheads);
3677                 vdev->vq[i].used_idx = 0;
3678                 vdev->vq[i].shadow_avail_idx = 0;
3679                 vdev->vq[i].inuse = 0;
3680                 continue;
3681             }
3682             vdev->vq[i].used_idx = vring_used_idx(&vdev->vq[i]);
3683             vdev->vq[i].shadow_avail_idx = vring_avail_idx(&vdev->vq[i]);
3684 
3685             /*
3686              * Some devices migrate VirtQueueElements that have been popped
3687              * from the avail ring but not yet returned to the used ring.
3688              * Since max ring size < UINT16_MAX it's safe to use modulo
3689              * UINT16_MAX + 1 subtraction.
3690              */
3691             vdev->vq[i].inuse = (uint16_t)(vdev->vq[i].last_avail_idx -
3692                                 vdev->vq[i].used_idx);
3693             if (vdev->vq[i].inuse > vdev->vq[i].vring.num) {
3694                 error_report("VQ %d size 0x%x < last_avail_idx 0x%x - "
3695                              "used_idx 0x%x",
3696                              i, vdev->vq[i].vring.num,
3697                              vdev->vq[i].last_avail_idx,
3698                              vdev->vq[i].used_idx);
3699                 return -1;
3700             }
3701         }
3702     }
3703 
3704     if (vdc->post_load) {
3705         ret = vdc->post_load(vdev);
3706         if (ret) {
3707             return ret;
3708         }
3709     }
3710 
3711     return 0;
3712 }
3713 
3714 void virtio_cleanup(VirtIODevice *vdev)
3715 {
3716     qemu_del_vm_change_state_handler(vdev->vmstate);
3717 }
3718 
3719 static void virtio_vmstate_change(void *opaque, bool running, RunState state)
3720 {
3721     VirtIODevice *vdev = opaque;
3722     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3723     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3724     bool backend_run = running && virtio_device_started(vdev, vdev->status);
3725     vdev->vm_running = running;
3726 
3727     if (backend_run) {
3728         virtio_set_status(vdev, vdev->status);
3729     }
3730 
3731     if (k->vmstate_change) {
3732         k->vmstate_change(qbus->parent, backend_run);
3733     }
3734 
3735     if (!backend_run) {
3736         virtio_set_status(vdev, vdev->status);
3737     }
3738 }
3739 
3740 void virtio_instance_init_common(Object *proxy_obj, void *data,
3741                                  size_t vdev_size, const char *vdev_name)
3742 {
3743     DeviceState *vdev = data;
3744 
3745     object_initialize_child_with_props(proxy_obj, "virtio-backend", vdev,
3746                                        vdev_size, vdev_name, &error_abort,
3747                                        NULL);
3748     qdev_alias_all_properties(vdev, proxy_obj);
3749 }
3750 
3751 void virtio_init(VirtIODevice *vdev, uint16_t device_id, size_t config_size)
3752 {
3753     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3754     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3755     int i;
3756     int nvectors = k->query_nvectors ? k->query_nvectors(qbus->parent) : 0;
3757 
3758     if (nvectors) {
3759         vdev->vector_queues =
3760             g_malloc0(sizeof(*vdev->vector_queues) * nvectors);
3761     }
3762 
3763     vdev->start_on_kick = false;
3764     vdev->started = false;
3765     vdev->vhost_started = false;
3766     vdev->device_id = device_id;
3767     vdev->status = 0;
3768     qatomic_set(&vdev->isr, 0);
3769     vdev->queue_sel = 0;
3770     vdev->config_vector = VIRTIO_NO_VECTOR;
3771     vdev->vq = g_new0(VirtQueue, VIRTIO_QUEUE_MAX);
3772     vdev->vm_running = runstate_is_running();
3773     vdev->broken = false;
3774     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
3775         vdev->vq[i].vector = VIRTIO_NO_VECTOR;
3776         vdev->vq[i].vdev = vdev;
3777         vdev->vq[i].queue_index = i;
3778         vdev->vq[i].host_notifier_enabled = false;
3779     }
3780 
3781     vdev->name = virtio_id_to_name(device_id);
3782     vdev->config_len = config_size;
3783     if (vdev->config_len) {
3784         vdev->config = g_malloc0(config_size);
3785     } else {
3786         vdev->config = NULL;
3787     }
3788     vdev->vmstate = qdev_add_vm_change_state_handler(DEVICE(vdev),
3789             virtio_vmstate_change, vdev);
3790     vdev->device_endian = virtio_default_endian();
3791     vdev->use_guest_notifier_mask = true;
3792 }
3793 
3794 /*
3795  * Only devices that have already been around prior to defining the virtio
3796  * standard support legacy mode; this includes devices not specified in the
3797  * standard. All newer devices conform to the virtio standard only.
3798  */
3799 bool virtio_legacy_allowed(VirtIODevice *vdev)
3800 {
3801     switch (vdev->device_id) {
3802     case VIRTIO_ID_NET:
3803     case VIRTIO_ID_BLOCK:
3804     case VIRTIO_ID_CONSOLE:
3805     case VIRTIO_ID_RNG:
3806     case VIRTIO_ID_BALLOON:
3807     case VIRTIO_ID_RPMSG:
3808     case VIRTIO_ID_SCSI:
3809     case VIRTIO_ID_9P:
3810     case VIRTIO_ID_RPROC_SERIAL:
3811     case VIRTIO_ID_CAIF:
3812         return true;
3813     default:
3814         return false;
3815     }
3816 }
3817 
3818 bool virtio_legacy_check_disabled(VirtIODevice *vdev)
3819 {
3820     return vdev->disable_legacy_check;
3821 }
3822 
3823 hwaddr virtio_queue_get_desc_addr(VirtIODevice *vdev, int n)
3824 {
3825     return vdev->vq[n].vring.desc;
3826 }
3827 
3828 bool virtio_queue_enabled_legacy(VirtIODevice *vdev, int n)
3829 {
3830     return virtio_queue_get_desc_addr(vdev, n) != 0;
3831 }
3832 
3833 bool virtio_queue_enabled(VirtIODevice *vdev, int n)
3834 {
3835     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
3836     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
3837 
3838     if (k->queue_enabled) {
3839         return k->queue_enabled(qbus->parent, n);
3840     }
3841     return virtio_queue_enabled_legacy(vdev, n);
3842 }
3843 
3844 hwaddr virtio_queue_get_avail_addr(VirtIODevice *vdev, int n)
3845 {
3846     return vdev->vq[n].vring.avail;
3847 }
3848 
3849 hwaddr virtio_queue_get_used_addr(VirtIODevice *vdev, int n)
3850 {
3851     return vdev->vq[n].vring.used;
3852 }
3853 
3854 hwaddr virtio_queue_get_desc_size(VirtIODevice *vdev, int n)
3855 {
3856     return sizeof(VRingDesc) * vdev->vq[n].vring.num;
3857 }
3858 
3859 hwaddr virtio_queue_get_avail_size(VirtIODevice *vdev, int n)
3860 {
3861     int s;
3862 
3863     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3864         return sizeof(struct VRingPackedDescEvent);
3865     }
3866 
3867     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3868     return offsetof(VRingAvail, ring) +
3869         sizeof(uint16_t) * vdev->vq[n].vring.num + s;
3870 }
3871 
3872 hwaddr virtio_queue_get_used_size(VirtIODevice *vdev, int n)
3873 {
3874     int s;
3875 
3876     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3877         return sizeof(struct VRingPackedDescEvent);
3878     }
3879 
3880     s = virtio_vdev_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX) ? 2 : 0;
3881     return offsetof(VRingUsed, ring) +
3882         sizeof(VRingUsedElem) * vdev->vq[n].vring.num + s;
3883 }
3884 
3885 static unsigned int virtio_queue_packed_get_last_avail_idx(VirtIODevice *vdev,
3886                                                            int n)
3887 {
3888     unsigned int avail, used;
3889 
3890     avail = vdev->vq[n].last_avail_idx;
3891     avail |= ((uint16_t)vdev->vq[n].last_avail_wrap_counter) << 15;
3892 
3893     used = vdev->vq[n].used_idx;
3894     used |= ((uint16_t)vdev->vq[n].used_wrap_counter) << 15;
3895 
3896     return avail | used << 16;
3897 }
3898 
3899 static uint16_t virtio_queue_split_get_last_avail_idx(VirtIODevice *vdev,
3900                                                       int n)
3901 {
3902     return vdev->vq[n].last_avail_idx;
3903 }
3904 
3905 unsigned int virtio_queue_get_last_avail_idx(VirtIODevice *vdev, int n)
3906 {
3907     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3908         return virtio_queue_packed_get_last_avail_idx(vdev, n);
3909     } else {
3910         return virtio_queue_split_get_last_avail_idx(vdev, n);
3911     }
3912 }
3913 
3914 static void virtio_queue_packed_set_last_avail_idx(VirtIODevice *vdev,
3915                                                    int n, unsigned int idx)
3916 {
3917     struct VirtQueue *vq = &vdev->vq[n];
3918 
3919     vq->last_avail_idx = vq->shadow_avail_idx = idx & 0x7fff;
3920     vq->last_avail_wrap_counter =
3921         vq->shadow_avail_wrap_counter = !!(idx & 0x8000);
3922     idx >>= 16;
3923     vq->used_idx = idx & 0x7ffff;
3924     vq->used_wrap_counter = !!(idx & 0x8000);
3925 }
3926 
3927 static void virtio_queue_split_set_last_avail_idx(VirtIODevice *vdev,
3928                                                   int n, unsigned int idx)
3929 {
3930         vdev->vq[n].last_avail_idx = idx;
3931         vdev->vq[n].shadow_avail_idx = idx;
3932 }
3933 
3934 void virtio_queue_set_last_avail_idx(VirtIODevice *vdev, int n,
3935                                      unsigned int idx)
3936 {
3937     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3938         virtio_queue_packed_set_last_avail_idx(vdev, n, idx);
3939     } else {
3940         virtio_queue_split_set_last_avail_idx(vdev, n, idx);
3941     }
3942 }
3943 
3944 static void virtio_queue_packed_restore_last_avail_idx(VirtIODevice *vdev,
3945                                                        int n)
3946 {
3947     /* We don't have a reference like avail idx in shared memory */
3948     return;
3949 }
3950 
3951 static void virtio_queue_split_restore_last_avail_idx(VirtIODevice *vdev,
3952                                                       int n)
3953 {
3954     RCU_READ_LOCK_GUARD();
3955     if (vdev->vq[n].vring.desc) {
3956         vdev->vq[n].last_avail_idx = vring_used_idx(&vdev->vq[n]);
3957         vdev->vq[n].shadow_avail_idx = vdev->vq[n].last_avail_idx;
3958     }
3959 }
3960 
3961 void virtio_queue_restore_last_avail_idx(VirtIODevice *vdev, int n)
3962 {
3963     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3964         virtio_queue_packed_restore_last_avail_idx(vdev, n);
3965     } else {
3966         virtio_queue_split_restore_last_avail_idx(vdev, n);
3967     }
3968 }
3969 
3970 static void virtio_queue_packed_update_used_idx(VirtIODevice *vdev, int n)
3971 {
3972     /* used idx was updated through set_last_avail_idx() */
3973     return;
3974 }
3975 
3976 static void virtio_split_packed_update_used_idx(VirtIODevice *vdev, int n)
3977 {
3978     RCU_READ_LOCK_GUARD();
3979     if (vdev->vq[n].vring.desc) {
3980         vdev->vq[n].used_idx = vring_used_idx(&vdev->vq[n]);
3981     }
3982 }
3983 
3984 void virtio_queue_update_used_idx(VirtIODevice *vdev, int n)
3985 {
3986     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
3987         return virtio_queue_packed_update_used_idx(vdev, n);
3988     } else {
3989         return virtio_split_packed_update_used_idx(vdev, n);
3990     }
3991 }
3992 
3993 void virtio_queue_invalidate_signalled_used(VirtIODevice *vdev, int n)
3994 {
3995     vdev->vq[n].signalled_used_valid = false;
3996 }
3997 
3998 VirtQueue *virtio_get_queue(VirtIODevice *vdev, int n)
3999 {
4000     return vdev->vq + n;
4001 }
4002 
4003 uint16_t virtio_get_queue_index(VirtQueue *vq)
4004 {
4005     return vq->queue_index;
4006 }
4007 
4008 static void virtio_queue_guest_notifier_read(EventNotifier *n)
4009 {
4010     VirtQueue *vq = container_of(n, VirtQueue, guest_notifier);
4011     if (event_notifier_test_and_clear(n)) {
4012         virtio_irq(vq);
4013     }
4014 }
4015 
4016 void virtio_queue_set_guest_notifier_fd_handler(VirtQueue *vq, bool assign,
4017                                                 bool with_irqfd)
4018 {
4019     if (assign && !with_irqfd) {
4020         event_notifier_set_handler(&vq->guest_notifier,
4021                                    virtio_queue_guest_notifier_read);
4022     } else {
4023         event_notifier_set_handler(&vq->guest_notifier, NULL);
4024     }
4025     if (!assign) {
4026         /* Test and clear notifier before closing it,
4027          * in case poll callback didn't have time to run. */
4028         virtio_queue_guest_notifier_read(&vq->guest_notifier);
4029     }
4030 }
4031 
4032 EventNotifier *virtio_queue_get_guest_notifier(VirtQueue *vq)
4033 {
4034     return &vq->guest_notifier;
4035 }
4036 
4037 static void virtio_queue_host_notifier_aio_poll_begin(EventNotifier *n)
4038 {
4039     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
4040 
4041     virtio_queue_set_notification(vq, 0);
4042 }
4043 
4044 static bool virtio_queue_host_notifier_aio_poll(void *opaque)
4045 {
4046     EventNotifier *n = opaque;
4047     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
4048 
4049     return vq->vring.desc && !virtio_queue_empty(vq);
4050 }
4051 
4052 static void virtio_queue_host_notifier_aio_poll_ready(EventNotifier *n)
4053 {
4054     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
4055 
4056     virtio_queue_notify_vq(vq);
4057 }
4058 
4059 static void virtio_queue_host_notifier_aio_poll_end(EventNotifier *n)
4060 {
4061     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
4062 
4063     /* Caller polls once more after this to catch requests that race with us */
4064     virtio_queue_set_notification(vq, 1);
4065 }
4066 
4067 void virtio_queue_aio_attach_host_notifier(VirtQueue *vq, AioContext *ctx)
4068 {
4069     aio_set_event_notifier(ctx, &vq->host_notifier, true,
4070                            virtio_queue_host_notifier_read,
4071                            virtio_queue_host_notifier_aio_poll,
4072                            virtio_queue_host_notifier_aio_poll_ready);
4073     aio_set_event_notifier_poll(ctx, &vq->host_notifier,
4074                                 virtio_queue_host_notifier_aio_poll_begin,
4075                                 virtio_queue_host_notifier_aio_poll_end);
4076 }
4077 
4078 /*
4079  * Same as virtio_queue_aio_attach_host_notifier() but without polling. Use
4080  * this for rx virtqueues and similar cases where the virtqueue handler
4081  * function does not pop all elements. When the virtqueue is left non-empty
4082  * polling consumes CPU cycles and should not be used.
4083  */
4084 void virtio_queue_aio_attach_host_notifier_no_poll(VirtQueue *vq, AioContext *ctx)
4085 {
4086     aio_set_event_notifier(ctx, &vq->host_notifier, true,
4087                            virtio_queue_host_notifier_read,
4088                            NULL, NULL);
4089 }
4090 
4091 void virtio_queue_aio_detach_host_notifier(VirtQueue *vq, AioContext *ctx)
4092 {
4093     aio_set_event_notifier(ctx, &vq->host_notifier, true, NULL, NULL, NULL);
4094     /* Test and clear notifier before after disabling event,
4095      * in case poll callback didn't have time to run. */
4096     virtio_queue_host_notifier_read(&vq->host_notifier);
4097 }
4098 
4099 void virtio_queue_host_notifier_read(EventNotifier *n)
4100 {
4101     VirtQueue *vq = container_of(n, VirtQueue, host_notifier);
4102     if (event_notifier_test_and_clear(n)) {
4103         virtio_queue_notify_vq(vq);
4104     }
4105 }
4106 
4107 EventNotifier *virtio_queue_get_host_notifier(VirtQueue *vq)
4108 {
4109     return &vq->host_notifier;
4110 }
4111 
4112 void virtio_queue_set_host_notifier_enabled(VirtQueue *vq, bool enabled)
4113 {
4114     vq->host_notifier_enabled = enabled;
4115 }
4116 
4117 int virtio_queue_set_host_notifier_mr(VirtIODevice *vdev, int n,
4118                                       MemoryRegion *mr, bool assign)
4119 {
4120     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
4121     VirtioBusClass *k = VIRTIO_BUS_GET_CLASS(qbus);
4122 
4123     if (k->set_host_notifier_mr) {
4124         return k->set_host_notifier_mr(qbus->parent, n, mr, assign);
4125     }
4126 
4127     return -1;
4128 }
4129 
4130 void virtio_device_set_child_bus_name(VirtIODevice *vdev, char *bus_name)
4131 {
4132     g_free(vdev->bus_name);
4133     vdev->bus_name = g_strdup(bus_name);
4134 }
4135 
4136 void G_GNUC_PRINTF(2, 3) virtio_error(VirtIODevice *vdev, const char *fmt, ...)
4137 {
4138     va_list ap;
4139 
4140     va_start(ap, fmt);
4141     error_vreport(fmt, ap);
4142     va_end(ap);
4143 
4144     if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
4145         vdev->status = vdev->status | VIRTIO_CONFIG_S_NEEDS_RESET;
4146         virtio_notify_config(vdev);
4147     }
4148 
4149     vdev->broken = true;
4150 }
4151 
4152 static void virtio_memory_listener_commit(MemoryListener *listener)
4153 {
4154     VirtIODevice *vdev = container_of(listener, VirtIODevice, listener);
4155     int i;
4156 
4157     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
4158         if (vdev->vq[i].vring.num == 0) {
4159             break;
4160         }
4161         virtio_init_region_cache(vdev, i);
4162     }
4163 }
4164 
4165 static void virtio_device_realize(DeviceState *dev, Error **errp)
4166 {
4167     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
4168     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
4169     Error *err = NULL;
4170 
4171     /* Devices should either use vmsd or the load/save methods */
4172     assert(!vdc->vmsd || !vdc->load);
4173 
4174     if (vdc->realize != NULL) {
4175         vdc->realize(dev, &err);
4176         if (err != NULL) {
4177             error_propagate(errp, err);
4178             return;
4179         }
4180     }
4181 
4182     virtio_bus_device_plugged(vdev, &err);
4183     if (err != NULL) {
4184         error_propagate(errp, err);
4185         vdc->unrealize(dev);
4186         return;
4187     }
4188 
4189     vdev->listener.commit = virtio_memory_listener_commit;
4190     vdev->listener.name = "virtio";
4191     memory_listener_register(&vdev->listener, vdev->dma_as);
4192     QTAILQ_INSERT_TAIL(&virtio_list, vdev, next);
4193 }
4194 
4195 static void virtio_device_unrealize(DeviceState *dev)
4196 {
4197     VirtIODevice *vdev = VIRTIO_DEVICE(dev);
4198     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(dev);
4199 
4200     memory_listener_unregister(&vdev->listener);
4201     virtio_bus_device_unplugged(vdev);
4202 
4203     if (vdc->unrealize != NULL) {
4204         vdc->unrealize(dev);
4205     }
4206 
4207     QTAILQ_REMOVE(&virtio_list, vdev, next);
4208     g_free(vdev->bus_name);
4209     vdev->bus_name = NULL;
4210 }
4211 
4212 static void virtio_device_free_virtqueues(VirtIODevice *vdev)
4213 {
4214     int i;
4215     if (!vdev->vq) {
4216         return;
4217     }
4218 
4219     for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
4220         if (vdev->vq[i].vring.num == 0) {
4221             break;
4222         }
4223         virtio_virtqueue_reset_region_cache(&vdev->vq[i]);
4224     }
4225     g_free(vdev->vq);
4226 }
4227 
4228 static void virtio_device_instance_finalize(Object *obj)
4229 {
4230     VirtIODevice *vdev = VIRTIO_DEVICE(obj);
4231 
4232     virtio_device_free_virtqueues(vdev);
4233 
4234     g_free(vdev->config);
4235     g_free(vdev->vector_queues);
4236 }
4237 
4238 static Property virtio_properties[] = {
4239     DEFINE_VIRTIO_COMMON_FEATURES(VirtIODevice, host_features),
4240     DEFINE_PROP_BOOL("use-started", VirtIODevice, use_started, true),
4241     DEFINE_PROP_BOOL("use-disabled-flag", VirtIODevice, use_disabled_flag, true),
4242     DEFINE_PROP_BOOL("x-disable-legacy-check", VirtIODevice,
4243                      disable_legacy_check, false),
4244     DEFINE_PROP_END_OF_LIST(),
4245 };
4246 
4247 static int virtio_device_start_ioeventfd_impl(VirtIODevice *vdev)
4248 {
4249     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
4250     int i, n, r, err;
4251 
4252     /*
4253      * Batch all the host notifiers in a single transaction to avoid
4254      * quadratic time complexity in address_space_update_ioeventfds().
4255      */
4256     memory_region_transaction_begin();
4257     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
4258         VirtQueue *vq = &vdev->vq[n];
4259         if (!virtio_queue_get_num(vdev, n)) {
4260             continue;
4261         }
4262         r = virtio_bus_set_host_notifier(qbus, n, true);
4263         if (r < 0) {
4264             err = r;
4265             goto assign_error;
4266         }
4267         event_notifier_set_handler(&vq->host_notifier,
4268                                    virtio_queue_host_notifier_read);
4269     }
4270 
4271     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
4272         /* Kick right away to begin processing requests already in vring */
4273         VirtQueue *vq = &vdev->vq[n];
4274         if (!vq->vring.num) {
4275             continue;
4276         }
4277         event_notifier_set(&vq->host_notifier);
4278     }
4279     memory_region_transaction_commit();
4280     return 0;
4281 
4282 assign_error:
4283     i = n; /* save n for a second iteration after transaction is committed. */
4284     while (--n >= 0) {
4285         VirtQueue *vq = &vdev->vq[n];
4286         if (!virtio_queue_get_num(vdev, n)) {
4287             continue;
4288         }
4289 
4290         event_notifier_set_handler(&vq->host_notifier, NULL);
4291         r = virtio_bus_set_host_notifier(qbus, n, false);
4292         assert(r >= 0);
4293     }
4294     /*
4295      * The transaction expects the ioeventfds to be open when it
4296      * commits. Do it now, before the cleanup loop.
4297      */
4298     memory_region_transaction_commit();
4299 
4300     while (--i >= 0) {
4301         if (!virtio_queue_get_num(vdev, i)) {
4302             continue;
4303         }
4304         virtio_bus_cleanup_host_notifier(qbus, i);
4305     }
4306     return err;
4307 }
4308 
4309 int virtio_device_start_ioeventfd(VirtIODevice *vdev)
4310 {
4311     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
4312     VirtioBusState *vbus = VIRTIO_BUS(qbus);
4313 
4314     return virtio_bus_start_ioeventfd(vbus);
4315 }
4316 
4317 static void virtio_device_stop_ioeventfd_impl(VirtIODevice *vdev)
4318 {
4319     VirtioBusState *qbus = VIRTIO_BUS(qdev_get_parent_bus(DEVICE(vdev)));
4320     int n, r;
4321 
4322     /*
4323      * Batch all the host notifiers in a single transaction to avoid
4324      * quadratic time complexity in address_space_update_ioeventfds().
4325      */
4326     memory_region_transaction_begin();
4327     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
4328         VirtQueue *vq = &vdev->vq[n];
4329 
4330         if (!virtio_queue_get_num(vdev, n)) {
4331             continue;
4332         }
4333         event_notifier_set_handler(&vq->host_notifier, NULL);
4334         r = virtio_bus_set_host_notifier(qbus, n, false);
4335         assert(r >= 0);
4336     }
4337     /*
4338      * The transaction expects the ioeventfds to be open when it
4339      * commits. Do it now, before the cleanup loop.
4340      */
4341     memory_region_transaction_commit();
4342 
4343     for (n = 0; n < VIRTIO_QUEUE_MAX; n++) {
4344         if (!virtio_queue_get_num(vdev, n)) {
4345             continue;
4346         }
4347         virtio_bus_cleanup_host_notifier(qbus, n);
4348     }
4349 }
4350 
4351 int virtio_device_grab_ioeventfd(VirtIODevice *vdev)
4352 {
4353     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
4354     VirtioBusState *vbus = VIRTIO_BUS(qbus);
4355 
4356     return virtio_bus_grab_ioeventfd(vbus);
4357 }
4358 
4359 void virtio_device_release_ioeventfd(VirtIODevice *vdev)
4360 {
4361     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
4362     VirtioBusState *vbus = VIRTIO_BUS(qbus);
4363 
4364     virtio_bus_release_ioeventfd(vbus);
4365 }
4366 
4367 static void virtio_device_class_init(ObjectClass *klass, void *data)
4368 {
4369     /* Set the default value here. */
4370     VirtioDeviceClass *vdc = VIRTIO_DEVICE_CLASS(klass);
4371     DeviceClass *dc = DEVICE_CLASS(klass);
4372 
4373     dc->realize = virtio_device_realize;
4374     dc->unrealize = virtio_device_unrealize;
4375     dc->bus_type = TYPE_VIRTIO_BUS;
4376     device_class_set_props(dc, virtio_properties);
4377     vdc->start_ioeventfd = virtio_device_start_ioeventfd_impl;
4378     vdc->stop_ioeventfd = virtio_device_stop_ioeventfd_impl;
4379 
4380     vdc->legacy_features |= VIRTIO_LEGACY_FEATURES;
4381 
4382     QTAILQ_INIT(&virtio_list);
4383 }
4384 
4385 bool virtio_device_ioeventfd_enabled(VirtIODevice *vdev)
4386 {
4387     BusState *qbus = qdev_get_parent_bus(DEVICE(vdev));
4388     VirtioBusState *vbus = VIRTIO_BUS(qbus);
4389 
4390     return virtio_bus_ioeventfd_enabled(vbus);
4391 }
4392 
4393 VirtioInfoList *qmp_x_query_virtio(Error **errp)
4394 {
4395     VirtioInfoList *list = NULL;
4396     VirtioInfoList *node;
4397     VirtIODevice *vdev;
4398 
4399     QTAILQ_FOREACH(vdev, &virtio_list, next) {
4400         DeviceState *dev = DEVICE(vdev);
4401         Error *err = NULL;
4402         QObject *obj = qmp_qom_get(dev->canonical_path, "realized", &err);
4403 
4404         if (err == NULL) {
4405             GString *is_realized = qobject_to_json_pretty(obj, true);
4406             /* virtio device is NOT realized, remove it from list */
4407             if (!strncmp(is_realized->str, "false", 4)) {
4408                 QTAILQ_REMOVE(&virtio_list, vdev, next);
4409             } else {
4410                 node = g_new0(VirtioInfoList, 1);
4411                 node->value = g_new(VirtioInfo, 1);
4412                 node->value->path = g_strdup(dev->canonical_path);
4413                 node->value->name = g_strdup(vdev->name);
4414                 QAPI_LIST_PREPEND(list, node->value);
4415             }
4416            g_string_free(is_realized, true);
4417         }
4418         qobject_unref(obj);
4419     }
4420 
4421     return list;
4422 }
4423 
4424 static VirtIODevice *virtio_device_find(const char *path)
4425 {
4426     VirtIODevice *vdev;
4427 
4428     QTAILQ_FOREACH(vdev, &virtio_list, next) {
4429         DeviceState *dev = DEVICE(vdev);
4430 
4431         if (strcmp(dev->canonical_path, path) != 0) {
4432             continue;
4433         }
4434 
4435         Error *err = NULL;
4436         QObject *obj = qmp_qom_get(dev->canonical_path, "realized", &err);
4437         if (err == NULL) {
4438             GString *is_realized = qobject_to_json_pretty(obj, true);
4439             /* virtio device is NOT realized, remove it from list */
4440             if (!strncmp(is_realized->str, "false", 4)) {
4441                 g_string_free(is_realized, true);
4442                 qobject_unref(obj);
4443                 QTAILQ_REMOVE(&virtio_list, vdev, next);
4444                 return NULL;
4445             }
4446             g_string_free(is_realized, true);
4447         } else {
4448             /* virtio device doesn't exist in QOM tree */
4449             QTAILQ_REMOVE(&virtio_list, vdev, next);
4450             qobject_unref(obj);
4451             return NULL;
4452         }
4453         /* device exists in QOM tree & is realized */
4454         qobject_unref(obj);
4455         return vdev;
4456     }
4457     return NULL;
4458 }
4459 
4460 #define CONVERT_FEATURES(type, map, is_status, bitmap)   \
4461     ({                                                   \
4462         type *list = NULL;                               \
4463         type *node;                                      \
4464         for (i = 0; map[i].virtio_bit != -1; i++) {      \
4465             if (is_status) {                             \
4466                 bit = map[i].virtio_bit;                 \
4467             }                                            \
4468             else {                                       \
4469                 bit = 1ULL << map[i].virtio_bit;         \
4470             }                                            \
4471             if ((bitmap & bit) == 0) {                   \
4472                 continue;                                \
4473             }                                            \
4474             node = g_new0(type, 1);                      \
4475             node->value = g_strdup(map[i].feature_desc); \
4476             node->next = list;                           \
4477             list = node;                                 \
4478             bitmap ^= bit;                               \
4479         }                                                \
4480         list;                                            \
4481     })
4482 
4483 static VirtioDeviceStatus *qmp_decode_status(uint8_t bitmap)
4484 {
4485     VirtioDeviceStatus *status;
4486     uint8_t bit;
4487     int i;
4488 
4489     status = g_new0(VirtioDeviceStatus, 1);
4490     status->statuses = CONVERT_FEATURES(strList, virtio_config_status_map,
4491                                         1, bitmap);
4492     status->has_unknown_statuses = bitmap != 0;
4493     if (status->has_unknown_statuses) {
4494         status->unknown_statuses = bitmap;
4495     }
4496 
4497     return status;
4498 }
4499 
4500 static VhostDeviceProtocols *qmp_decode_protocols(uint64_t bitmap)
4501 {
4502     VhostDeviceProtocols *vhu_protocols;
4503     uint64_t bit;
4504     int i;
4505 
4506     vhu_protocols = g_new0(VhostDeviceProtocols, 1);
4507     vhu_protocols->protocols =
4508                     CONVERT_FEATURES(strList,
4509                                      vhost_user_protocol_map, 0, bitmap);
4510     vhu_protocols->has_unknown_protocols = bitmap != 0;
4511     if (vhu_protocols->has_unknown_protocols) {
4512         vhu_protocols->unknown_protocols = bitmap;
4513     }
4514 
4515     return vhu_protocols;
4516 }
4517 
4518 static VirtioDeviceFeatures *qmp_decode_features(uint16_t device_id,
4519                                                  uint64_t bitmap)
4520 {
4521     VirtioDeviceFeatures *features;
4522     uint64_t bit;
4523     int i;
4524 
4525     features = g_new0(VirtioDeviceFeatures, 1);
4526     features->has_dev_features = true;
4527 
4528     /* transport features */
4529     features->transports = CONVERT_FEATURES(strList, virtio_transport_map, 0,
4530                                             bitmap);
4531 
4532     /* device features */
4533     switch (device_id) {
4534 #ifdef CONFIG_VIRTIO_SERIAL
4535     case VIRTIO_ID_CONSOLE:
4536         features->dev_features =
4537             CONVERT_FEATURES(strList, virtio_serial_feature_map, 0, bitmap);
4538         break;
4539 #endif
4540 #ifdef CONFIG_VIRTIO_BLK
4541     case VIRTIO_ID_BLOCK:
4542         features->dev_features =
4543             CONVERT_FEATURES(strList, virtio_blk_feature_map, 0, bitmap);
4544         break;
4545 #endif
4546 #ifdef CONFIG_VIRTIO_GPU
4547     case VIRTIO_ID_GPU:
4548         features->dev_features =
4549             CONVERT_FEATURES(strList, virtio_gpu_feature_map, 0, bitmap);
4550         break;
4551 #endif
4552 #ifdef CONFIG_VIRTIO_NET
4553     case VIRTIO_ID_NET:
4554         features->dev_features =
4555             CONVERT_FEATURES(strList, virtio_net_feature_map, 0, bitmap);
4556         break;
4557 #endif
4558 #ifdef CONFIG_VIRTIO_SCSI
4559     case VIRTIO_ID_SCSI:
4560         features->dev_features =
4561             CONVERT_FEATURES(strList, virtio_scsi_feature_map, 0, bitmap);
4562         break;
4563 #endif
4564 #ifdef CONFIG_VIRTIO_BALLOON
4565     case VIRTIO_ID_BALLOON:
4566         features->dev_features =
4567             CONVERT_FEATURES(strList, virtio_balloon_feature_map, 0, bitmap);
4568         break;
4569 #endif
4570 #ifdef CONFIG_VIRTIO_IOMMU
4571     case VIRTIO_ID_IOMMU:
4572         features->dev_features =
4573             CONVERT_FEATURES(strList, virtio_iommu_feature_map, 0, bitmap);
4574         break;
4575 #endif
4576 #ifdef CONFIG_VIRTIO_INPUT
4577     case VIRTIO_ID_INPUT:
4578         features->dev_features =
4579             CONVERT_FEATURES(strList, virtio_input_feature_map, 0, bitmap);
4580         break;
4581 #endif
4582 #ifdef CONFIG_VHOST_USER_FS
4583     case VIRTIO_ID_FS:
4584         features->dev_features =
4585             CONVERT_FEATURES(strList, virtio_fs_feature_map, 0, bitmap);
4586         break;
4587 #endif
4588 #ifdef CONFIG_VHOST_VSOCK
4589     case VIRTIO_ID_VSOCK:
4590         features->dev_features =
4591             CONVERT_FEATURES(strList, virtio_vsock_feature_map, 0, bitmap);
4592         break;
4593 #endif
4594 #ifdef CONFIG_VIRTIO_CRYPTO
4595     case VIRTIO_ID_CRYPTO:
4596         features->dev_features =
4597             CONVERT_FEATURES(strList, virtio_crypto_feature_map, 0, bitmap);
4598         break;
4599 #endif
4600 #ifdef CONFIG_VIRTIO_MEM
4601     case VIRTIO_ID_MEM:
4602         features->dev_features =
4603             CONVERT_FEATURES(strList, virtio_mem_feature_map, 0, bitmap);
4604         break;
4605 #endif
4606 #ifdef CONFIG_VIRTIO_I2C_ADAPTER
4607     case VIRTIO_ID_I2C_ADAPTER:
4608         features->dev_features =
4609             CONVERT_FEATURES(strList, virtio_i2c_feature_map, 0, bitmap);
4610         break;
4611 #endif
4612 #ifdef CONFIG_VIRTIO_RNG
4613     case VIRTIO_ID_RNG:
4614         features->dev_features =
4615             CONVERT_FEATURES(strList, virtio_rng_feature_map, 0, bitmap);
4616         break;
4617 #endif
4618     /* No features */
4619     case VIRTIO_ID_9P:
4620     case VIRTIO_ID_PMEM:
4621     case VIRTIO_ID_IOMEM:
4622     case VIRTIO_ID_RPMSG:
4623     case VIRTIO_ID_CLOCK:
4624     case VIRTIO_ID_MAC80211_WLAN:
4625     case VIRTIO_ID_MAC80211_HWSIM:
4626     case VIRTIO_ID_RPROC_SERIAL:
4627     case VIRTIO_ID_MEMORY_BALLOON:
4628     case VIRTIO_ID_CAIF:
4629     case VIRTIO_ID_SIGNAL_DIST:
4630     case VIRTIO_ID_PSTORE:
4631     case VIRTIO_ID_SOUND:
4632     case VIRTIO_ID_BT:
4633     case VIRTIO_ID_RPMB:
4634     case VIRTIO_ID_VIDEO_ENCODER:
4635     case VIRTIO_ID_VIDEO_DECODER:
4636     case VIRTIO_ID_SCMI:
4637     case VIRTIO_ID_NITRO_SEC_MOD:
4638     case VIRTIO_ID_WATCHDOG:
4639     case VIRTIO_ID_CAN:
4640     case VIRTIO_ID_DMABUF:
4641     case VIRTIO_ID_PARAM_SERV:
4642     case VIRTIO_ID_AUDIO_POLICY:
4643     case VIRTIO_ID_GPIO:
4644         break;
4645     default:
4646         g_assert_not_reached();
4647     }
4648 
4649     features->has_unknown_dev_features = bitmap != 0;
4650     if (features->has_unknown_dev_features) {
4651         features->unknown_dev_features = bitmap;
4652     }
4653 
4654     return features;
4655 }
4656 
4657 VirtioStatus *qmp_x_query_virtio_status(const char *path, Error **errp)
4658 {
4659     VirtIODevice *vdev;
4660     VirtioStatus *status;
4661 
4662     vdev = virtio_device_find(path);
4663     if (vdev == NULL) {
4664         error_setg(errp, "Path %s is not a VirtIODevice", path);
4665         return NULL;
4666     }
4667 
4668     status = g_new0(VirtioStatus, 1);
4669     status->name = g_strdup(vdev->name);
4670     status->device_id = vdev->device_id;
4671     status->vhost_started = vdev->vhost_started;
4672     status->guest_features = qmp_decode_features(vdev->device_id,
4673                                                  vdev->guest_features);
4674     status->host_features = qmp_decode_features(vdev->device_id,
4675                                                 vdev->host_features);
4676     status->backend_features = qmp_decode_features(vdev->device_id,
4677                                                    vdev->backend_features);
4678 
4679     switch (vdev->device_endian) {
4680     case VIRTIO_DEVICE_ENDIAN_LITTLE:
4681         status->device_endian = g_strdup("little");
4682         break;
4683     case VIRTIO_DEVICE_ENDIAN_BIG:
4684         status->device_endian = g_strdup("big");
4685         break;
4686     default:
4687         status->device_endian = g_strdup("unknown");
4688         break;
4689     }
4690 
4691     status->num_vqs = virtio_get_num_queues(vdev);
4692     status->status = qmp_decode_status(vdev->status);
4693     status->isr = vdev->isr;
4694     status->queue_sel = vdev->queue_sel;
4695     status->vm_running = vdev->vm_running;
4696     status->broken = vdev->broken;
4697     status->disabled = vdev->disabled;
4698     status->use_started = vdev->use_started;
4699     status->started = vdev->started;
4700     status->start_on_kick = vdev->start_on_kick;
4701     status->disable_legacy_check = vdev->disable_legacy_check;
4702     status->bus_name = g_strdup(vdev->bus_name);
4703     status->use_guest_notifier_mask = vdev->use_guest_notifier_mask;
4704 
4705     if (vdev->vhost_started) {
4706         VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
4707         struct vhost_dev *hdev = vdc->get_vhost(vdev);
4708 
4709         status->vhost_dev = g_new0(VhostStatus, 1);
4710         status->vhost_dev->n_mem_sections = hdev->n_mem_sections;
4711         status->vhost_dev->n_tmp_sections = hdev->n_tmp_sections;
4712         status->vhost_dev->nvqs = hdev->nvqs;
4713         status->vhost_dev->vq_index = hdev->vq_index;
4714         status->vhost_dev->features =
4715             qmp_decode_features(vdev->device_id, hdev->features);
4716         status->vhost_dev->acked_features =
4717             qmp_decode_features(vdev->device_id, hdev->acked_features);
4718         status->vhost_dev->backend_features =
4719             qmp_decode_features(vdev->device_id, hdev->backend_features);
4720         status->vhost_dev->protocol_features =
4721             qmp_decode_protocols(hdev->protocol_features);
4722         status->vhost_dev->max_queues = hdev->max_queues;
4723         status->vhost_dev->backend_cap = hdev->backend_cap;
4724         status->vhost_dev->log_enabled = hdev->log_enabled;
4725         status->vhost_dev->log_size = hdev->log_size;
4726     }
4727 
4728     return status;
4729 }
4730 
4731 VirtVhostQueueStatus *qmp_x_query_virtio_vhost_queue_status(const char *path,
4732                                                             uint16_t queue,
4733                                                             Error **errp)
4734 {
4735     VirtIODevice *vdev;
4736     VirtVhostQueueStatus *status;
4737 
4738     vdev = virtio_device_find(path);
4739     if (vdev == NULL) {
4740         error_setg(errp, "Path %s is not a VirtIODevice", path);
4741         return NULL;
4742     }
4743 
4744     if (!vdev->vhost_started) {
4745         error_setg(errp, "Error: vhost device has not started yet");
4746         return NULL;
4747     }
4748 
4749     VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
4750     struct vhost_dev *hdev = vdc->get_vhost(vdev);
4751 
4752     if (queue < hdev->vq_index || queue >= hdev->vq_index + hdev->nvqs) {
4753         error_setg(errp, "Invalid vhost virtqueue number %d", queue);
4754         return NULL;
4755     }
4756 
4757     status = g_new0(VirtVhostQueueStatus, 1);
4758     status->name = g_strdup(vdev->name);
4759     status->kick = hdev->vqs[queue].kick;
4760     status->call = hdev->vqs[queue].call;
4761     status->desc = (uintptr_t)hdev->vqs[queue].desc;
4762     status->avail = (uintptr_t)hdev->vqs[queue].avail;
4763     status->used = (uintptr_t)hdev->vqs[queue].used;
4764     status->num = hdev->vqs[queue].num;
4765     status->desc_phys = hdev->vqs[queue].desc_phys;
4766     status->desc_size = hdev->vqs[queue].desc_size;
4767     status->avail_phys = hdev->vqs[queue].avail_phys;
4768     status->avail_size = hdev->vqs[queue].avail_size;
4769     status->used_phys = hdev->vqs[queue].used_phys;
4770     status->used_size = hdev->vqs[queue].used_size;
4771 
4772     return status;
4773 }
4774 
4775 VirtQueueStatus *qmp_x_query_virtio_queue_status(const char *path,
4776                                                  uint16_t queue,
4777                                                  Error **errp)
4778 {
4779     VirtIODevice *vdev;
4780     VirtQueueStatus *status;
4781 
4782     vdev = virtio_device_find(path);
4783     if (vdev == NULL) {
4784         error_setg(errp, "Path %s is not a VirtIODevice", path);
4785         return NULL;
4786     }
4787 
4788     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
4789         error_setg(errp, "Invalid virtqueue number %d", queue);
4790         return NULL;
4791     }
4792 
4793     status = g_new0(VirtQueueStatus, 1);
4794     status->name = g_strdup(vdev->name);
4795     status->queue_index = vdev->vq[queue].queue_index;
4796     status->inuse = vdev->vq[queue].inuse;
4797     status->vring_num = vdev->vq[queue].vring.num;
4798     status->vring_num_default = vdev->vq[queue].vring.num_default;
4799     status->vring_align = vdev->vq[queue].vring.align;
4800     status->vring_desc = vdev->vq[queue].vring.desc;
4801     status->vring_avail = vdev->vq[queue].vring.avail;
4802     status->vring_used = vdev->vq[queue].vring.used;
4803     status->used_idx = vdev->vq[queue].used_idx;
4804     status->signalled_used = vdev->vq[queue].signalled_used;
4805     status->signalled_used_valid = vdev->vq[queue].signalled_used_valid;
4806 
4807     if (vdev->vhost_started) {
4808         VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
4809         struct vhost_dev *hdev = vdc->get_vhost(vdev);
4810 
4811         /* check if vq index exists for vhost as well  */
4812         if (queue >= hdev->vq_index && queue < hdev->vq_index + hdev->nvqs) {
4813             status->has_last_avail_idx = true;
4814 
4815             int vhost_vq_index =
4816                 hdev->vhost_ops->vhost_get_vq_index(hdev, queue);
4817             struct vhost_vring_state state = {
4818                 .index = vhost_vq_index,
4819             };
4820 
4821             status->last_avail_idx =
4822                 hdev->vhost_ops->vhost_get_vring_base(hdev, &state);
4823         }
4824     } else {
4825         status->has_shadow_avail_idx = true;
4826         status->has_last_avail_idx = true;
4827         status->last_avail_idx = vdev->vq[queue].last_avail_idx;
4828         status->shadow_avail_idx = vdev->vq[queue].shadow_avail_idx;
4829     }
4830 
4831     return status;
4832 }
4833 
4834 static strList *qmp_decode_vring_desc_flags(uint16_t flags)
4835 {
4836     strList *list = NULL;
4837     strList *node;
4838     int i;
4839 
4840     struct {
4841         uint16_t flag;
4842         const char *value;
4843     } map[] = {
4844         { VRING_DESC_F_NEXT, "next" },
4845         { VRING_DESC_F_WRITE, "write" },
4846         { VRING_DESC_F_INDIRECT, "indirect" },
4847         { 1 << VRING_PACKED_DESC_F_AVAIL, "avail" },
4848         { 1 << VRING_PACKED_DESC_F_USED, "used" },
4849         { 0, "" }
4850     };
4851 
4852     for (i = 0; map[i].flag; i++) {
4853         if ((map[i].flag & flags) == 0) {
4854             continue;
4855         }
4856         node = g_malloc0(sizeof(strList));
4857         node->value = g_strdup(map[i].value);
4858         node->next = list;
4859         list = node;
4860     }
4861 
4862     return list;
4863 }
4864 
4865 VirtioQueueElement *qmp_x_query_virtio_queue_element(const char *path,
4866                                                      uint16_t queue,
4867                                                      bool has_index,
4868                                                      uint16_t index,
4869                                                      Error **errp)
4870 {
4871     VirtIODevice *vdev;
4872     VirtQueue *vq;
4873     VirtioQueueElement *element = NULL;
4874 
4875     vdev = virtio_device_find(path);
4876     if (vdev == NULL) {
4877         error_setg(errp, "Path %s is not a VirtIO device", path);
4878         return NULL;
4879     }
4880 
4881     if (queue >= VIRTIO_QUEUE_MAX || !virtio_queue_get_num(vdev, queue)) {
4882         error_setg(errp, "Invalid virtqueue number %d", queue);
4883         return NULL;
4884     }
4885     vq = &vdev->vq[queue];
4886 
4887     if (virtio_vdev_has_feature(vdev, VIRTIO_F_RING_PACKED)) {
4888         error_setg(errp, "Packed ring not supported");
4889         return NULL;
4890     } else {
4891         unsigned int head, i, max;
4892         VRingMemoryRegionCaches *caches;
4893         MemoryRegionCache indirect_desc_cache = MEMORY_REGION_CACHE_INVALID;
4894         MemoryRegionCache *desc_cache;
4895         VRingDesc desc;
4896         VirtioRingDescList *list = NULL;
4897         VirtioRingDescList *node;
4898         int rc; int ndescs;
4899 
4900         RCU_READ_LOCK_GUARD();
4901 
4902         max = vq->vring.num;
4903 
4904         if (!has_index) {
4905             head = vring_avail_ring(vq, vq->last_avail_idx % vq->vring.num);
4906         } else {
4907             head = vring_avail_ring(vq, index % vq->vring.num);
4908         }
4909         i = head;
4910 
4911         caches = vring_get_region_caches(vq);
4912         if (!caches) {
4913             error_setg(errp, "Region caches not initialized");
4914             return NULL;
4915         }
4916         if (caches->desc.len < max * sizeof(VRingDesc)) {
4917             error_setg(errp, "Cannot map descriptor ring");
4918             return NULL;
4919         }
4920 
4921         desc_cache = &caches->desc;
4922         vring_split_desc_read(vdev, &desc, desc_cache, i);
4923         if (desc.flags & VRING_DESC_F_INDIRECT) {
4924             int64_t len;
4925             len = address_space_cache_init(&indirect_desc_cache, vdev->dma_as,
4926                                            desc.addr, desc.len, false);
4927             desc_cache = &indirect_desc_cache;
4928             if (len < desc.len) {
4929                 error_setg(errp, "Cannot map indirect buffer");
4930                 goto done;
4931             }
4932 
4933             max = desc.len / sizeof(VRingDesc);
4934             i = 0;
4935             vring_split_desc_read(vdev, &desc, desc_cache, i);
4936         }
4937 
4938         element = g_new0(VirtioQueueElement, 1);
4939         element->avail = g_new0(VirtioRingAvail, 1);
4940         element->used = g_new0(VirtioRingUsed, 1);
4941         element->name = g_strdup(vdev->name);
4942         element->index = head;
4943         element->avail->flags = vring_avail_flags(vq);
4944         element->avail->idx = vring_avail_idx(vq);
4945         element->avail->ring = head;
4946         element->used->flags = vring_used_flags(vq);
4947         element->used->idx = vring_used_idx(vq);
4948         ndescs = 0;
4949 
4950         do {
4951             /* A buggy driver may produce an infinite loop */
4952             if (ndescs >= max) {
4953                 break;
4954             }
4955             node = g_new0(VirtioRingDescList, 1);
4956             node->value = g_new0(VirtioRingDesc, 1);
4957             node->value->addr = desc.addr;
4958             node->value->len = desc.len;
4959             node->value->flags = qmp_decode_vring_desc_flags(desc.flags);
4960             node->next = list;
4961             list = node;
4962 
4963             ndescs++;
4964             rc = virtqueue_split_read_next_desc(vdev, &desc, desc_cache,
4965                                                 max, &i);
4966         } while (rc == VIRTQUEUE_READ_DESC_MORE);
4967         element->descs = list;
4968 done:
4969         address_space_cache_destroy(&indirect_desc_cache);
4970     }
4971 
4972     return element;
4973 }
4974 
4975 static const TypeInfo virtio_device_info = {
4976     .name = TYPE_VIRTIO_DEVICE,
4977     .parent = TYPE_DEVICE,
4978     .instance_size = sizeof(VirtIODevice),
4979     .class_init = virtio_device_class_init,
4980     .instance_finalize = virtio_device_instance_finalize,
4981     .abstract = true,
4982     .class_size = sizeof(VirtioDeviceClass),
4983 };
4984 
4985 static void virtio_register_types(void)
4986 {
4987     type_register_static(&virtio_device_info);
4988 }
4989 
4990 type_init(virtio_register_types)
4991