?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
qemu-iotests: Test qemu-img bitmap/commit exit code on error
[qemu.git] / hw / virtio / virtio-pci.c
blob247325c1933ca5cd503a40b8652cb44ec0ac41f3
1 /*
2 * Virtio PCI Bindings
3 *
4 * Copyright IBM, Corp. 2007
5 * Copyright (c) 2009 CodeSourcery
6 *
7 * Authors:
8 * Anthony Liguori <aliguori@us.ibm.com>
9 * Paul Brook <paul@codesourcery.com>
10 *
11 * This work is licensed under the terms of the GNU GPL, version 2. See
12 * the COPYING file in the top-level directory.
13 *
14 * Contributions after 2012-01-13 are licensed under the terms of the
15 * GNU GPL, version 2 or (at your option) any later version.
16 */
17
18 #include "qemu/osdep.h"
19
20 #include "exec/memop.h"
21 #include "standard-headers/linux/virtio_pci.h"
22 #include "standard-headers/linux/virtio_ids.h"
23 #include "hw/boards.h"
24 #include "hw/virtio/virtio.h"
25 #include "migration/qemu-file-types.h"
26 #include "hw/pci/pci.h"
27 #include "hw/pci/pci_bus.h"
28 #include "hw/qdev-properties.h"
29 #include "qapi/error.h"
30 #include "qemu/error-report.h"
31 #include "qemu/log.h"
32 #include "qemu/module.h"
33 #include "hw/pci/msi.h"
34 #include "hw/pci/msix.h"
35 #include "hw/loader.h"
36 #include "sysemu/kvm.h"
37 #include "hw/virtio/virtio-pci.h"
38 #include "qemu/range.h"
39 #include "hw/virtio/virtio-bus.h"
40 #include "qapi/visitor.h"
41 #include "sysemu/replay.h"
42 #include "trace.h"
43
44 #define VIRTIO_PCI_REGION_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_present(dev))
45
46 #undef VIRTIO_PCI_CONFIG
47
48 /* The remaining space is defined by each driver as the per-driver
49 * configuration space */
50 #define VIRTIO_PCI_CONFIG_SIZE(dev) VIRTIO_PCI_CONFIG_OFF(msix_enabled(dev))
51
52 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
53 VirtIOPCIProxy *dev);
54 static void virtio_pci_reset(DeviceState *qdev);
55
56 /* virtio device */
57 /* DeviceState to VirtIOPCIProxy. For use off data-path. TODO: use QOM. */
58 static inline VirtIOPCIProxy *to_virtio_pci_proxy(DeviceState *d)
59 {
60 return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
61 }
62
63 /* DeviceState to VirtIOPCIProxy. Note: used on datapath,
64 * be careful and test performance if you change this.
65 */
66 static inline VirtIOPCIProxy *to_virtio_pci_proxy_fast(DeviceState *d)
67 {
68 return container_of(d, VirtIOPCIProxy, pci_dev.qdev);
69 }
70
71 static void virtio_pci_notify(DeviceState *d, uint16_t vector)
72 {
73 VirtIOPCIProxy *proxy = to_virtio_pci_proxy_fast(d);
74
75 if (msix_enabled(&proxy->pci_dev)) {
76 if (vector != VIRTIO_NO_VECTOR) {
77 msix_notify(&proxy->pci_dev, vector);
78 }
79 } else {
80 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
81 pci_set_irq(&proxy->pci_dev, qatomic_read(&vdev->isr) & 1);
82 }
83 }
84
85 static void virtio_pci_save_config(DeviceState *d, QEMUFile *f)
86 {
87 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
88 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
89
90 pci_device_save(&proxy->pci_dev, f);
91 msix_save(&proxy->pci_dev, f);
92 if (msix_present(&proxy->pci_dev))
93 qemu_put_be16(f, vdev->config_vector);
94 }
95
96 static const VMStateDescription vmstate_virtio_pci_modern_queue_state = {
97 .name = "virtio_pci/modern_queue_state",
98 .version_id = 1,
99 .minimum_version_id = 1,
100 .fields = (VMStateField[]) {
101 VMSTATE_UINT16(num, VirtIOPCIQueue),
102 VMSTATE_UNUSED(1), /* enabled was stored as be16 */
103 VMSTATE_BOOL(enabled, VirtIOPCIQueue),
104 VMSTATE_UINT32_ARRAY(desc, VirtIOPCIQueue, 2),
105 VMSTATE_UINT32_ARRAY(avail, VirtIOPCIQueue, 2),
106 VMSTATE_UINT32_ARRAY(used, VirtIOPCIQueue, 2),
107 VMSTATE_END_OF_LIST()
108 }
109 };
110
111 static bool virtio_pci_modern_state_needed(void *opaque)
112 {
113 VirtIOPCIProxy *proxy = opaque;
114
115 return virtio_pci_modern(proxy);
116 }
117
118 static const VMStateDescription vmstate_virtio_pci_modern_state_sub = {
119 .name = "virtio_pci/modern_state",
120 .version_id = 1,
121 .minimum_version_id = 1,
122 .needed = &virtio_pci_modern_state_needed,
123 .fields = (VMStateField[]) {
124 VMSTATE_UINT32(dfselect, VirtIOPCIProxy),
125 VMSTATE_UINT32(gfselect, VirtIOPCIProxy),
126 VMSTATE_UINT32_ARRAY(guest_features, VirtIOPCIProxy, 2),
127 VMSTATE_STRUCT_ARRAY(vqs, VirtIOPCIProxy, VIRTIO_QUEUE_MAX, 0,
128 vmstate_virtio_pci_modern_queue_state,
129 VirtIOPCIQueue),
130 VMSTATE_END_OF_LIST()
131 }
132 };
133
134 static const VMStateDescription vmstate_virtio_pci = {
135 .name = "virtio_pci",
136 .version_id = 1,
137 .minimum_version_id = 1,
138 .fields = (VMStateField[]) {
139 VMSTATE_END_OF_LIST()
140 },
141 .subsections = (const VMStateDescription*[]) {
142 &vmstate_virtio_pci_modern_state_sub,
143 NULL
144 }
145 };
146
147 static bool virtio_pci_has_extra_state(DeviceState *d)
148 {
149 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
150
151 return proxy->flags & VIRTIO_PCI_FLAG_MIGRATE_EXTRA;
152 }
153
154 static void virtio_pci_save_extra_state(DeviceState *d, QEMUFile *f)
155 {
156 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
157
158 vmstate_save_state(f, &vmstate_virtio_pci, proxy, NULL);
159 }
160
161 static int virtio_pci_load_extra_state(DeviceState *d, QEMUFile *f)
162 {
163 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
164
165 return vmstate_load_state(f, &vmstate_virtio_pci, proxy, 1);
166 }
167
168 static void virtio_pci_save_queue(DeviceState *d, int n, QEMUFile *f)
169 {
170 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
171 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
172
173 if (msix_present(&proxy->pci_dev))
174 qemu_put_be16(f, virtio_queue_vector(vdev, n));
175 }
176
177 static int virtio_pci_load_config(DeviceState *d, QEMUFile *f)
178 {
179 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
180 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
181 uint16_t vector;
182
183 int ret;
184 ret = pci_device_load(&proxy->pci_dev, f);
185 if (ret) {
186 return ret;
187 }
188 msix_unuse_all_vectors(&proxy->pci_dev);
189 msix_load(&proxy->pci_dev, f);
190 if (msix_present(&proxy->pci_dev)) {
191 qemu_get_be16s(f, &vector);
192
193 if (vector != VIRTIO_NO_VECTOR && vector >= proxy->nvectors) {
194 return -EINVAL;
195 }
196 } else {
197 vector = VIRTIO_NO_VECTOR;
198 }
199 vdev->config_vector = vector;
200 if (vector != VIRTIO_NO_VECTOR) {
201 msix_vector_use(&proxy->pci_dev, vector);
202 }
203 return 0;
204 }
205
206 static int virtio_pci_load_queue(DeviceState *d, int n, QEMUFile *f)
207 {
208 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
209 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
210
211 uint16_t vector;
212 if (msix_present(&proxy->pci_dev)) {
213 qemu_get_be16s(f, &vector);
214 if (vector != VIRTIO_NO_VECTOR && vector >= proxy->nvectors) {
215 return -EINVAL;
216 }
217 } else {
218 vector = VIRTIO_NO_VECTOR;
219 }
220 virtio_queue_set_vector(vdev, n, vector);
221 if (vector != VIRTIO_NO_VECTOR) {
222 msix_vector_use(&proxy->pci_dev, vector);
223 }
224
225 return 0;
226 }
227
228 typedef struct VirtIOPCIIDInfo {
229 /* virtio id */
230 uint16_t vdev_id;
231 /* pci device id for the transitional device */
232 uint16_t trans_devid;
233 uint16_t class_id;
234 } VirtIOPCIIDInfo;
235
236 static const VirtIOPCIIDInfo virtio_pci_id_info[] = {
237 {
238 .vdev_id = VIRTIO_ID_CRYPTO,
239 .class_id = PCI_CLASS_OTHERS,
240 }, {
241 .vdev_id = VIRTIO_ID_FS,
242 .class_id = PCI_CLASS_STORAGE_OTHER,
243 }, {
244 .vdev_id = VIRTIO_ID_NET,
245 .trans_devid = PCI_DEVICE_ID_VIRTIO_NET,
246 .class_id = PCI_CLASS_NETWORK_ETHERNET,
247 }, {
248 .vdev_id = VIRTIO_ID_BLOCK,
249 .trans_devid = PCI_DEVICE_ID_VIRTIO_BLOCK,
250 .class_id = PCI_CLASS_STORAGE_SCSI,
251 }, {
252 .vdev_id = VIRTIO_ID_CONSOLE,
253 .trans_devid = PCI_DEVICE_ID_VIRTIO_CONSOLE,
254 .class_id = PCI_CLASS_COMMUNICATION_OTHER,
255 }, {
256 .vdev_id = VIRTIO_ID_SCSI,
257 .trans_devid = PCI_DEVICE_ID_VIRTIO_SCSI,
258 .class_id = PCI_CLASS_STORAGE_SCSI
259 }, {
260 .vdev_id = VIRTIO_ID_9P,
261 .trans_devid = PCI_DEVICE_ID_VIRTIO_9P,
262 .class_id = PCI_BASE_CLASS_NETWORK,
263 }, {
264 .vdev_id = VIRTIO_ID_BALLOON,
265 .trans_devid = PCI_DEVICE_ID_VIRTIO_BALLOON,
266 .class_id = PCI_CLASS_OTHERS,
267 }, {
268 .vdev_id = VIRTIO_ID_RNG,
269 .trans_devid = PCI_DEVICE_ID_VIRTIO_RNG,
270 .class_id = PCI_CLASS_OTHERS,
271 },
272 };
273
274 static const VirtIOPCIIDInfo *virtio_pci_get_id_info(uint16_t vdev_id)
275 {
276 const VirtIOPCIIDInfo *info = NULL;
277 int i;
278
279 for (i = 0; i < ARRAY_SIZE(virtio_pci_id_info); i++) {
280 if (virtio_pci_id_info[i].vdev_id == vdev_id) {
281 info = &virtio_pci_id_info[i];
282 break;
283 }
284 }
285
286 if (!info) {
287 /* The device id is invalid or not added to the id_info yet. */
288 error_report("Invalid virtio device(id %u)", vdev_id);
289 abort();
290 }
291
292 return info;
293 }
294
295 /*
296 * Get the Transitional Device ID for the specific device, return
297 * zero if the device is non-transitional.
298 */
299 uint16_t virtio_pci_get_trans_devid(uint16_t device_id)
300 {
301 return virtio_pci_get_id_info(device_id)->trans_devid;
302 }
303
304 /*
305 * Get the Class ID for the specific device.
306 */
307 uint16_t virtio_pci_get_class_id(uint16_t device_id)
308 {
309 return virtio_pci_get_id_info(device_id)->class_id;
310 }
311
312 static bool virtio_pci_ioeventfd_enabled(DeviceState *d)
313 {
314 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
315
316 return (proxy->flags & VIRTIO_PCI_FLAG_USE_IOEVENTFD) != 0;
317 }
318
319 #define QEMU_VIRTIO_PCI_QUEUE_MEM_MULT 0x1000
320
321 static inline int virtio_pci_queue_mem_mult(struct VirtIOPCIProxy *proxy)
322 {
323 return (proxy->flags & VIRTIO_PCI_FLAG_PAGE_PER_VQ) ?
324 QEMU_VIRTIO_PCI_QUEUE_MEM_MULT : 4;
325 }
326
327 static int virtio_pci_ioeventfd_assign(DeviceState *d, EventNotifier *notifier,
328 int n, bool assign)
329 {
330 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
331 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
332 VirtQueue *vq = virtio_get_queue(vdev, n);
333 bool legacy = virtio_pci_legacy(proxy);
334 bool modern = virtio_pci_modern(proxy);
335 bool fast_mmio = kvm_ioeventfd_any_length_enabled();
336 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
337 MemoryRegion *modern_mr = &proxy->notify.mr;
338 MemoryRegion *modern_notify_mr = &proxy->notify_pio.mr;
339 MemoryRegion *legacy_mr = &proxy->bar;
340 hwaddr modern_addr = virtio_pci_queue_mem_mult(proxy) *
341 virtio_get_queue_index(vq);
342 hwaddr legacy_addr = VIRTIO_PCI_QUEUE_NOTIFY;
343
344 if (assign) {
345 if (modern) {
346 if (fast_mmio) {
347 memory_region_add_eventfd(modern_mr, modern_addr, 0,
348 false, n, notifier);
349 } else {
350 memory_region_add_eventfd(modern_mr, modern_addr, 2,
351 false, n, notifier);
352 }
353 if (modern_pio) {
354 memory_region_add_eventfd(modern_notify_mr, 0, 2,
355 true, n, notifier);
356 }
357 }
358 if (legacy) {
359 memory_region_add_eventfd(legacy_mr, legacy_addr, 2,
360 true, n, notifier);
361 }
362 } else {
363 if (modern) {
364 if (fast_mmio) {
365 memory_region_del_eventfd(modern_mr, modern_addr, 0,
366 false, n, notifier);
367 } else {
368 memory_region_del_eventfd(modern_mr, modern_addr, 2,
369 false, n, notifier);
370 }
371 if (modern_pio) {
372 memory_region_del_eventfd(modern_notify_mr, 0, 2,
373 true, n, notifier);
374 }
375 }
376 if (legacy) {
377 memory_region_del_eventfd(legacy_mr, legacy_addr, 2,
378 true, n, notifier);
379 }
380 }
381 return 0;
382 }
383
384 static void virtio_pci_start_ioeventfd(VirtIOPCIProxy *proxy)
385 {
386 virtio_bus_start_ioeventfd(&proxy->bus);
387 }
388
389 static void virtio_pci_stop_ioeventfd(VirtIOPCIProxy *proxy)
390 {
391 virtio_bus_stop_ioeventfd(&proxy->bus);
392 }
393
394 static void virtio_ioport_write(void *opaque, uint32_t addr, uint32_t val)
395 {
396 VirtIOPCIProxy *proxy = opaque;
397 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
398 uint16_t vector;
399 hwaddr pa;
400
401 switch (addr) {
402 case VIRTIO_PCI_GUEST_FEATURES:
403 /* Guest does not negotiate properly? We have to assume nothing. */
404 if (val & (1 << VIRTIO_F_BAD_FEATURE)) {
405 val = virtio_bus_get_vdev_bad_features(&proxy->bus);
406 }
407 virtio_set_features(vdev, val);
408 break;
409 case VIRTIO_PCI_QUEUE_PFN:
410 pa = (hwaddr)val << VIRTIO_PCI_QUEUE_ADDR_SHIFT;
411 if (pa == 0) {
412 virtio_pci_reset(DEVICE(proxy));
413 }
414 else
415 virtio_queue_set_addr(vdev, vdev->queue_sel, pa);
416 break;
417 case VIRTIO_PCI_QUEUE_SEL:
418 if (val < VIRTIO_QUEUE_MAX)
419 vdev->queue_sel = val;
420 break;
421 case VIRTIO_PCI_QUEUE_NOTIFY:
422 if (val < VIRTIO_QUEUE_MAX) {
423 virtio_queue_notify(vdev, val);
424 }
425 break;
426 case VIRTIO_PCI_STATUS:
427 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
428 virtio_pci_stop_ioeventfd(proxy);
429 }
430
431 virtio_set_status(vdev, val & 0xFF);
432
433 if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
434 virtio_pci_start_ioeventfd(proxy);
435 }
436
437 if (vdev->status == 0) {
438 virtio_pci_reset(DEVICE(proxy));
439 }
440
441 /* Linux before 2.6.34 drives the device without enabling
442 the PCI device bus master bit. Enable it automatically
443 for the guest. This is a PCI spec violation but so is
444 initiating DMA with bus master bit clear. */
445 if (val == (VIRTIO_CONFIG_S_ACKNOWLEDGE | VIRTIO_CONFIG_S_DRIVER)) {
446 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
447 proxy->pci_dev.config[PCI_COMMAND] |
448 PCI_COMMAND_MASTER, 1);
449 }
450 break;
451 case VIRTIO_MSI_CONFIG_VECTOR:
452 if (vdev->config_vector != VIRTIO_NO_VECTOR) {
453 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
454 }
455 /* Make it possible for guest to discover an error took place. */
456 if (val < proxy->nvectors) {
457 msix_vector_use(&proxy->pci_dev, val);
458 } else {
459 val = VIRTIO_NO_VECTOR;
460 }
461 vdev->config_vector = val;
462 break;
463 case VIRTIO_MSI_QUEUE_VECTOR:
464 vector = virtio_queue_vector(vdev, vdev->queue_sel);
465 if (vector != VIRTIO_NO_VECTOR) {
466 msix_vector_unuse(&proxy->pci_dev, vector);
467 }
468 /* Make it possible for guest to discover an error took place. */
469 if (val < proxy->nvectors) {
470 msix_vector_use(&proxy->pci_dev, val);
471 } else {
472 val = VIRTIO_NO_VECTOR;
473 }
474 virtio_queue_set_vector(vdev, vdev->queue_sel, val);
475 break;
476 default:
477 qemu_log_mask(LOG_GUEST_ERROR,
478 "%s: unexpected address 0x%x value 0x%x\n",
479 __func__, addr, val);
480 break;
481 }
482 }
483
484 static uint32_t virtio_ioport_read(VirtIOPCIProxy *proxy, uint32_t addr)
485 {
486 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
487 uint32_t ret = 0xFFFFFFFF;
488
489 switch (addr) {
490 case VIRTIO_PCI_HOST_FEATURES:
491 ret = vdev->host_features;
492 break;
493 case VIRTIO_PCI_GUEST_FEATURES:
494 ret = vdev->guest_features;
495 break;
496 case VIRTIO_PCI_QUEUE_PFN:
497 ret = virtio_queue_get_addr(vdev, vdev->queue_sel)
498 >> VIRTIO_PCI_QUEUE_ADDR_SHIFT;
499 break;
500 case VIRTIO_PCI_QUEUE_NUM:
501 ret = virtio_queue_get_num(vdev, vdev->queue_sel);
502 break;
503 case VIRTIO_PCI_QUEUE_SEL:
504 ret = vdev->queue_sel;
505 break;
506 case VIRTIO_PCI_STATUS:
507 ret = vdev->status;
508 break;
509 case VIRTIO_PCI_ISR:
510 /* reading from the ISR also clears it. */
511 ret = qatomic_xchg(&vdev->isr, 0);
512 pci_irq_deassert(&proxy->pci_dev);
513 break;
514 case VIRTIO_MSI_CONFIG_VECTOR:
515 ret = vdev->config_vector;
516 break;
517 case VIRTIO_MSI_QUEUE_VECTOR:
518 ret = virtio_queue_vector(vdev, vdev->queue_sel);
519 break;
520 default:
521 break;
522 }
523
524 return ret;
525 }
526
527 static uint64_t virtio_pci_config_read(void *opaque, hwaddr addr,
528 unsigned size)
529 {
530 VirtIOPCIProxy *proxy = opaque;
531 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
532 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
533 uint64_t val = 0;
534
535 if (vdev == NULL) {
536 return UINT64_MAX;
537 }
538
539 if (addr < config) {
540 return virtio_ioport_read(proxy, addr);
541 }
542 addr -= config;
543
544 switch (size) {
545 case 1:
546 val = virtio_config_readb(vdev, addr);
547 break;
548 case 2:
549 val = virtio_config_readw(vdev, addr);
550 if (virtio_is_big_endian(vdev)) {
551 val = bswap16(val);
552 }
553 break;
554 case 4:
555 val = virtio_config_readl(vdev, addr);
556 if (virtio_is_big_endian(vdev)) {
557 val = bswap32(val);
558 }
559 break;
560 }
561 return val;
562 }
563
564 static void virtio_pci_config_write(void *opaque, hwaddr addr,
565 uint64_t val, unsigned size)
566 {
567 VirtIOPCIProxy *proxy = opaque;
568 uint32_t config = VIRTIO_PCI_CONFIG_SIZE(&proxy->pci_dev);
569 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
570
571 if (vdev == NULL) {
572 return;
573 }
574
575 if (addr < config) {
576 virtio_ioport_write(proxy, addr, val);
577 return;
578 }
579 addr -= config;
580 /*
581 * Virtio-PCI is odd. Ioports are LE but config space is target native
582 * endian.
583 */
584 switch (size) {
585 case 1:
586 virtio_config_writeb(vdev, addr, val);
587 break;
588 case 2:
589 if (virtio_is_big_endian(vdev)) {
590 val = bswap16(val);
591 }
592 virtio_config_writew(vdev, addr, val);
593 break;
594 case 4:
595 if (virtio_is_big_endian(vdev)) {
596 val = bswap32(val);
597 }
598 virtio_config_writel(vdev, addr, val);
599 break;
600 }
601 }
602
603 static const MemoryRegionOps virtio_pci_config_ops = {
604 .read = virtio_pci_config_read,
605 .write = virtio_pci_config_write,
606 .impl = {
607 .min_access_size = 1,
608 .max_access_size = 4,
609 },
610 .endianness = DEVICE_LITTLE_ENDIAN,
611 };
612
613 static MemoryRegion *virtio_address_space_lookup(VirtIOPCIProxy *proxy,
614 hwaddr *off, int len)
615 {
616 int i;
617 VirtIOPCIRegion *reg;
618
619 for (i = 0; i < ARRAY_SIZE(proxy->regs); ++i) {
620 reg = &proxy->regs[i];
621 if (*off >= reg->offset &&
622 *off + len <= reg->offset + reg->size) {
623 *off -= reg->offset;
624 return &reg->mr;
625 }
626 }
627
628 return NULL;
629 }
630
631 /* Below are generic functions to do memcpy from/to an address space,
632 * without byteswaps, with input validation.
633 *
634 * As regular address_space_* APIs all do some kind of byteswap at least for
635 * some host/target combinations, we are forced to explicitly convert to a
636 * known-endianness integer value.
637 * It doesn't really matter which endian format to go through, so the code
638 * below selects the endian that causes the least amount of work on the given
639 * host.
640 *
641 * Note: host pointer must be aligned.
642 */
643 static
644 void virtio_address_space_write(VirtIOPCIProxy *proxy, hwaddr addr,
645 const uint8_t *buf, int len)
646 {
647 uint64_t val;
648 MemoryRegion *mr;
649
650 /* address_space_* APIs assume an aligned address.
651 * As address is under guest control, handle illegal values.
652 */
653 addr &= ~(len - 1);
654
655 mr = virtio_address_space_lookup(proxy, &addr, len);
656 if (!mr) {
657 return;
658 }
659
660 /* Make sure caller aligned buf properly */
661 assert(!(((uintptr_t)buf) & (len - 1)));
662
663 switch (len) {
664 case 1:
665 val = pci_get_byte(buf);
666 break;
667 case 2:
668 val = pci_get_word(buf);
669 break;
670 case 4:
671 val = pci_get_long(buf);
672 break;
673 default:
674 /* As length is under guest control, handle illegal values. */
675 return;
676 }
677 memory_region_dispatch_write(mr, addr, val, size_memop(len) | MO_LE,
678 MEMTXATTRS_UNSPECIFIED);
679 }
680
681 static void
682 virtio_address_space_read(VirtIOPCIProxy *proxy, hwaddr addr,
683 uint8_t *buf, int len)
684 {
685 uint64_t val;
686 MemoryRegion *mr;
687
688 /* address_space_* APIs assume an aligned address.
689 * As address is under guest control, handle illegal values.
690 */
691 addr &= ~(len - 1);
692
693 mr = virtio_address_space_lookup(proxy, &addr, len);
694 if (!mr) {
695 return;
696 }
697
698 /* Make sure caller aligned buf properly */
699 assert(!(((uintptr_t)buf) & (len - 1)));
700
701 memory_region_dispatch_read(mr, addr, &val, size_memop(len) | MO_LE,
702 MEMTXATTRS_UNSPECIFIED);
703 switch (len) {
704 case 1:
705 pci_set_byte(buf, val);
706 break;
707 case 2:
708 pci_set_word(buf, val);
709 break;
710 case 4:
711 pci_set_long(buf, val);
712 break;
713 default:
714 /* As length is under guest control, handle illegal values. */
715 break;
716 }
717 }
718
719 static void virtio_write_config(PCIDevice *pci_dev, uint32_t address,
720 uint32_t val, int len)
721 {
722 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
723 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
724 struct virtio_pci_cfg_cap *cfg;
725
726 pci_default_write_config(pci_dev, address, val, len);
727
728 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
729 pcie_cap_flr_write_config(pci_dev, address, val, len);
730 }
731
732 if (range_covers_byte(address, len, PCI_COMMAND)) {
733 if (!(pci_dev->config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
734 virtio_set_disabled(vdev, true);
735 virtio_pci_stop_ioeventfd(proxy);
736 virtio_set_status(vdev, vdev->status & ~VIRTIO_CONFIG_S_DRIVER_OK);
737 } else {
738 virtio_set_disabled(vdev, false);
739 }
740 }
741
742 if (proxy->config_cap &&
743 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
744 pci_cfg_data),
745 sizeof cfg->pci_cfg_data)) {
746 uint32_t off;
747 uint32_t len;
748
749 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
750 off = le32_to_cpu(cfg->cap.offset);
751 len = le32_to_cpu(cfg->cap.length);
752
753 if (len == 1 || len == 2 || len == 4) {
754 assert(len <= sizeof cfg->pci_cfg_data);
755 virtio_address_space_write(proxy, off, cfg->pci_cfg_data, len);
756 }
757 }
758 }
759
760 static uint32_t virtio_read_config(PCIDevice *pci_dev,
761 uint32_t address, int len)
762 {
763 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
764 struct virtio_pci_cfg_cap *cfg;
765
766 if (proxy->config_cap &&
767 ranges_overlap(address, len, proxy->config_cap + offsetof(struct virtio_pci_cfg_cap,
768 pci_cfg_data),
769 sizeof cfg->pci_cfg_data)) {
770 uint32_t off;
771 uint32_t len;
772
773 cfg = (void *)(proxy->pci_dev.config + proxy->config_cap);
774 off = le32_to_cpu(cfg->cap.offset);
775 len = le32_to_cpu(cfg->cap.length);
776
777 if (len == 1 || len == 2 || len == 4) {
778 assert(len <= sizeof cfg->pci_cfg_data);
779 virtio_address_space_read(proxy, off, cfg->pci_cfg_data, len);
780 }
781 }
782
783 return pci_default_read_config(pci_dev, address, len);
784 }
785
786 static int kvm_virtio_pci_vq_vector_use(VirtIOPCIProxy *proxy,
787 unsigned int vector)
788 {
789 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
790 int ret;
791
792 if (irqfd->users == 0) {
793 KVMRouteChange c = kvm_irqchip_begin_route_changes(kvm_state);
794 ret = kvm_irqchip_add_msi_route(&c, vector, &proxy->pci_dev);
795 if (ret < 0) {
796 return ret;
797 }
798 kvm_irqchip_commit_route_changes(&c);
799 irqfd->virq = ret;
800 }
801 irqfd->users++;
802 return 0;
803 }
804
805 static void kvm_virtio_pci_vq_vector_release(VirtIOPCIProxy *proxy,
806 unsigned int vector)
807 {
808 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
809 if (--irqfd->users == 0) {
810 kvm_irqchip_release_virq(kvm_state, irqfd->virq);
811 }
812 }
813
814 static int kvm_virtio_pci_irqfd_use(VirtIOPCIProxy *proxy,
815 EventNotifier *n,
816 unsigned int vector)
817 {
818 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
819 return kvm_irqchip_add_irqfd_notifier_gsi(kvm_state, n, NULL, irqfd->virq);
820 }
821
822 static void kvm_virtio_pci_irqfd_release(VirtIOPCIProxy *proxy,
823 EventNotifier *n ,
824 unsigned int vector)
825 {
826 VirtIOIRQFD *irqfd = &proxy->vector_irqfd[vector];
827 int ret;
828
829 ret = kvm_irqchip_remove_irqfd_notifier_gsi(kvm_state, n, irqfd->virq);
830 assert(ret == 0);
831 }
832 static int virtio_pci_get_notifier(VirtIOPCIProxy *proxy, int queue_no,
833 EventNotifier **n, unsigned int *vector)
834 {
835 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
836 VirtQueue *vq;
837
838 if (queue_no == VIRTIO_CONFIG_IRQ_IDX) {
839 *n = virtio_config_get_guest_notifier(vdev);
840 *vector = vdev->config_vector;
841 } else {
842 if (!virtio_queue_get_num(vdev, queue_no)) {
843 return -1;
844 }
845 *vector = virtio_queue_vector(vdev, queue_no);
846 vq = virtio_get_queue(vdev, queue_no);
847 *n = virtio_queue_get_guest_notifier(vq);
848 }
849 return 0;
850 }
851
852 static int kvm_virtio_pci_vector_use_one(VirtIOPCIProxy *proxy, int queue_no)
853 {
854 unsigned int vector;
855 int ret;
856 EventNotifier *n;
857 PCIDevice *dev = &proxy->pci_dev;
858 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
859 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
860
861 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector);
862 if (ret < 0) {
863 return ret;
864 }
865 if (vector >= msix_nr_vectors_allocated(dev)) {
866 return 0;
867 }
868 ret = kvm_virtio_pci_vq_vector_use(proxy, vector);
869 if (ret < 0) {
870 goto undo;
871 }
872 /*
873 * If guest supports masking, set up irqfd now.
874 * Otherwise, delay until unmasked in the frontend.
875 */
876 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
877 ret = kvm_virtio_pci_irqfd_use(proxy, n, vector);
878 if (ret < 0) {
879 kvm_virtio_pci_vq_vector_release(proxy, vector);
880 goto undo;
881 }
882 }
883
884 return 0;
885 undo:
886
887 vector = virtio_queue_vector(vdev, queue_no);
888 if (vector >= msix_nr_vectors_allocated(dev)) {
889 return ret;
890 }
891 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
892 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector);
893 if (ret < 0) {
894 return ret;
895 }
896 kvm_virtio_pci_irqfd_release(proxy, n, vector);
897 }
898 return ret;
899 }
900 static int kvm_virtio_pci_vector_vq_use(VirtIOPCIProxy *proxy, int nvqs)
901 {
902 int queue_no;
903 int ret = 0;
904 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
905
906 for (queue_no = 0; queue_no < nvqs; queue_no++) {
907 if (!virtio_queue_get_num(vdev, queue_no)) {
908 return -1;
909 }
910 ret = kvm_virtio_pci_vector_use_one(proxy, queue_no);
911 }
912 return ret;
913 }
914
915 static int kvm_virtio_pci_vector_config_use(VirtIOPCIProxy *proxy)
916 {
917 return kvm_virtio_pci_vector_use_one(proxy, VIRTIO_CONFIG_IRQ_IDX);
918 }
919
920 static void kvm_virtio_pci_vector_release_one(VirtIOPCIProxy *proxy,
921 int queue_no)
922 {
923 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
924 unsigned int vector;
925 EventNotifier *n;
926 int ret;
927 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
928 PCIDevice *dev = &proxy->pci_dev;
929
930 ret = virtio_pci_get_notifier(proxy, queue_no, &n, &vector);
931 if (ret < 0) {
932 return;
933 }
934 if (vector >= msix_nr_vectors_allocated(dev)) {
935 return;
936 }
937 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
938 kvm_virtio_pci_irqfd_release(proxy, n, vector);
939 }
940 kvm_virtio_pci_vq_vector_release(proxy, vector);
941 }
942
943 static void kvm_virtio_pci_vector_vq_release(VirtIOPCIProxy *proxy, int nvqs)
944 {
945 int queue_no;
946 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
947
948 for (queue_no = 0; queue_no < nvqs; queue_no++) {
949 if (!virtio_queue_get_num(vdev, queue_no)) {
950 break;
951 }
952 kvm_virtio_pci_vector_release_one(proxy, queue_no);
953 }
954 }
955
956 static void kvm_virtio_pci_vector_config_release(VirtIOPCIProxy *proxy)
957 {
958 kvm_virtio_pci_vector_release_one(proxy, VIRTIO_CONFIG_IRQ_IDX);
959 }
960
961 static int virtio_pci_one_vector_unmask(VirtIOPCIProxy *proxy,
962 unsigned int queue_no,
963 unsigned int vector,
964 MSIMessage msg,
965 EventNotifier *n)
966 {
967 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
968 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
969 VirtIOIRQFD *irqfd;
970 int ret = 0;
971
972 if (proxy->vector_irqfd) {
973 irqfd = &proxy->vector_irqfd[vector];
974 if (irqfd->msg.data != msg.data || irqfd->msg.address != msg.address) {
975 ret = kvm_irqchip_update_msi_route(kvm_state, irqfd->virq, msg,
976 &proxy->pci_dev);
977 if (ret < 0) {
978 return ret;
979 }
980 kvm_irqchip_commit_routes(kvm_state);
981 }
982 }
983
984 /* If guest supports masking, irqfd is already setup, unmask it.
985 * Otherwise, set it up now.
986 */
987 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
988 k->guest_notifier_mask(vdev, queue_no, false);
989 /* Test after unmasking to avoid losing events. */
990 if (k->guest_notifier_pending &&
991 k->guest_notifier_pending(vdev, queue_no)) {
992 event_notifier_set(n);
993 }
994 } else {
995 ret = kvm_virtio_pci_irqfd_use(proxy, n, vector);
996 }
997 return ret;
998 }
999
1000 static void virtio_pci_one_vector_mask(VirtIOPCIProxy *proxy,
1001 unsigned int queue_no,
1002 unsigned int vector,
1003 EventNotifier *n)
1004 {
1005 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1006 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1007
1008 /* If guest supports masking, keep irqfd but mask it.
1009 * Otherwise, clean it up now.
1010 */
1011 if (vdev->use_guest_notifier_mask && k->guest_notifier_mask) {
1012 k->guest_notifier_mask(vdev, queue_no, true);
1013 } else {
1014 kvm_virtio_pci_irqfd_release(proxy, n, vector);
1015 }
1016 }
1017
1018 static int virtio_pci_vector_unmask(PCIDevice *dev, unsigned vector,
1019 MSIMessage msg)
1020 {
1021 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
1022 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1023 VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
1024 EventNotifier *n;
1025 int ret, index, unmasked = 0;
1026
1027 while (vq) {
1028 index = virtio_get_queue_index(vq);
1029 if (!virtio_queue_get_num(vdev, index)) {
1030 break;
1031 }
1032 if (index < proxy->nvqs_with_notifiers) {
1033 n = virtio_queue_get_guest_notifier(vq);
1034 ret = virtio_pci_one_vector_unmask(proxy, index, vector, msg, n);
1035 if (ret < 0) {
1036 goto undo;
1037 }
1038 ++unmasked;
1039 }
1040 vq = virtio_vector_next_queue(vq);
1041 }
1042 /* unmask config intr */
1043 if (vector == vdev->config_vector) {
1044 n = virtio_config_get_guest_notifier(vdev);
1045 ret = virtio_pci_one_vector_unmask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector,
1046 msg, n);
1047 if (ret < 0) {
1048 goto undo_config;
1049 }
1050 }
1051 return 0;
1052 undo_config:
1053 n = virtio_config_get_guest_notifier(vdev);
1054 virtio_pci_one_vector_mask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector, n);
1055 undo:
1056 vq = virtio_vector_first_queue(vdev, vector);
1057 while (vq && unmasked >= 0) {
1058 index = virtio_get_queue_index(vq);
1059 if (index < proxy->nvqs_with_notifiers) {
1060 n = virtio_queue_get_guest_notifier(vq);
1061 virtio_pci_one_vector_mask(proxy, index, vector, n);
1062 --unmasked;
1063 }
1064 vq = virtio_vector_next_queue(vq);
1065 }
1066 return ret;
1067 }
1068
1069 static void virtio_pci_vector_mask(PCIDevice *dev, unsigned vector)
1070 {
1071 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
1072 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1073 VirtQueue *vq = virtio_vector_first_queue(vdev, vector);
1074 EventNotifier *n;
1075 int index;
1076
1077 while (vq) {
1078 index = virtio_get_queue_index(vq);
1079 n = virtio_queue_get_guest_notifier(vq);
1080 if (!virtio_queue_get_num(vdev, index)) {
1081 break;
1082 }
1083 if (index < proxy->nvqs_with_notifiers) {
1084 virtio_pci_one_vector_mask(proxy, index, vector, n);
1085 }
1086 vq = virtio_vector_next_queue(vq);
1087 }
1088
1089 if (vector == vdev->config_vector) {
1090 n = virtio_config_get_guest_notifier(vdev);
1091 virtio_pci_one_vector_mask(proxy, VIRTIO_CONFIG_IRQ_IDX, vector, n);
1092 }
1093 }
1094
1095 static void virtio_pci_vector_poll(PCIDevice *dev,
1096 unsigned int vector_start,
1097 unsigned int vector_end)
1098 {
1099 VirtIOPCIProxy *proxy = container_of(dev, VirtIOPCIProxy, pci_dev);
1100 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1101 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1102 int queue_no;
1103 unsigned int vector;
1104 EventNotifier *notifier;
1105 int ret;
1106
1107 for (queue_no = 0; queue_no < proxy->nvqs_with_notifiers; queue_no++) {
1108 ret = virtio_pci_get_notifier(proxy, queue_no, &notifier, &vector);
1109 if (ret < 0) {
1110 break;
1111 }
1112 if (vector < vector_start || vector >= vector_end ||
1113 !msix_is_masked(dev, vector)) {
1114 continue;
1115 }
1116 if (k->guest_notifier_pending) {
1117 if (k->guest_notifier_pending(vdev, queue_no)) {
1118 msix_set_pending(dev, vector);
1119 }
1120 } else if (event_notifier_test_and_clear(notifier)) {
1121 msix_set_pending(dev, vector);
1122 }
1123 }
1124 /* poll the config intr */
1125 ret = virtio_pci_get_notifier(proxy, VIRTIO_CONFIG_IRQ_IDX, &notifier,
1126 &vector);
1127 if (ret < 0) {
1128 return;
1129 }
1130 if (vector < vector_start || vector >= vector_end ||
1131 !msix_is_masked(dev, vector)) {
1132 return;
1133 }
1134 if (k->guest_notifier_pending) {
1135 if (k->guest_notifier_pending(vdev, VIRTIO_CONFIG_IRQ_IDX)) {
1136 msix_set_pending(dev, vector);
1137 }
1138 } else if (event_notifier_test_and_clear(notifier)) {
1139 msix_set_pending(dev, vector);
1140 }
1141 }
1142
1143 void virtio_pci_set_guest_notifier_fd_handler(VirtIODevice *vdev, VirtQueue *vq,
1144 int n, bool assign,
1145 bool with_irqfd)
1146 {
1147 if (n == VIRTIO_CONFIG_IRQ_IDX) {
1148 virtio_config_set_guest_notifier_fd_handler(vdev, assign, with_irqfd);
1149 } else {
1150 virtio_queue_set_guest_notifier_fd_handler(vq, assign, with_irqfd);
1151 }
1152 }
1153
1154 static int virtio_pci_set_guest_notifier(DeviceState *d, int n, bool assign,
1155 bool with_irqfd)
1156 {
1157 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1158 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1159 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1160 VirtQueue *vq = NULL;
1161 EventNotifier *notifier = NULL;
1162
1163 if (n == VIRTIO_CONFIG_IRQ_IDX) {
1164 notifier = virtio_config_get_guest_notifier(vdev);
1165 } else {
1166 vq = virtio_get_queue(vdev, n);
1167 notifier = virtio_queue_get_guest_notifier(vq);
1168 }
1169
1170 if (assign) {
1171 int r = event_notifier_init(notifier, 0);
1172 if (r < 0) {
1173 return r;
1174 }
1175 virtio_pci_set_guest_notifier_fd_handler(vdev, vq, n, true, with_irqfd);
1176 } else {
1177 virtio_pci_set_guest_notifier_fd_handler(vdev, vq, n, false,
1178 with_irqfd);
1179 event_notifier_cleanup(notifier);
1180 }
1181
1182 if (!msix_enabled(&proxy->pci_dev) &&
1183 vdev->use_guest_notifier_mask &&
1184 vdc->guest_notifier_mask) {
1185 vdc->guest_notifier_mask(vdev, n, !assign);
1186 }
1187
1188 return 0;
1189 }
1190
1191 static bool virtio_pci_query_guest_notifiers(DeviceState *d)
1192 {
1193 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1194 return msix_enabled(&proxy->pci_dev);
1195 }
1196
1197 static int virtio_pci_set_guest_notifiers(DeviceState *d, int nvqs, bool assign)
1198 {
1199 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1200 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1201 VirtioDeviceClass *k = VIRTIO_DEVICE_GET_CLASS(vdev);
1202 int r, n;
1203 bool with_irqfd = msix_enabled(&proxy->pci_dev) &&
1204 kvm_msi_via_irqfd_enabled();
1205
1206 nvqs = MIN(nvqs, VIRTIO_QUEUE_MAX);
1207
1208 /*
1209 * When deassigning, pass a consistent nvqs value to avoid leaking
1210 * notifiers. But first check we've actually been configured, exit
1211 * early if we haven't.
1212 */
1213 if (!assign && !proxy->nvqs_with_notifiers) {
1214 return 0;
1215 }
1216 assert(assign || nvqs == proxy->nvqs_with_notifiers);
1217
1218 proxy->nvqs_with_notifiers = nvqs;
1219
1220 /* Must unset vector notifier while guest notifier is still assigned */
1221 if ((proxy->vector_irqfd ||
1222 (vdev->use_guest_notifier_mask && k->guest_notifier_mask)) &&
1223 !assign) {
1224 msix_unset_vector_notifiers(&proxy->pci_dev);
1225 if (proxy->vector_irqfd) {
1226 kvm_virtio_pci_vector_vq_release(proxy, nvqs);
1227 kvm_virtio_pci_vector_config_release(proxy);
1228 g_free(proxy->vector_irqfd);
1229 proxy->vector_irqfd = NULL;
1230 }
1231 }
1232
1233 for (n = 0; n < nvqs; n++) {
1234 if (!virtio_queue_get_num(vdev, n)) {
1235 break;
1236 }
1237
1238 r = virtio_pci_set_guest_notifier(d, n, assign, with_irqfd);
1239 if (r < 0) {
1240 goto assign_error;
1241 }
1242 }
1243 r = virtio_pci_set_guest_notifier(d, VIRTIO_CONFIG_IRQ_IDX, assign,
1244 with_irqfd);
1245 if (r < 0) {
1246 goto config_assign_error;
1247 }
1248 /* Must set vector notifier after guest notifier has been assigned */
1249 if ((with_irqfd ||
1250 (vdev->use_guest_notifier_mask && k->guest_notifier_mask)) &&
1251 assign) {
1252 if (with_irqfd) {
1253 proxy->vector_irqfd =
1254 g_malloc0(sizeof(*proxy->vector_irqfd) *
1255 msix_nr_vectors_allocated(&proxy->pci_dev));
1256 r = kvm_virtio_pci_vector_vq_use(proxy, nvqs);
1257 if (r < 0) {
1258 goto config_assign_error;
1259 }
1260 r = kvm_virtio_pci_vector_config_use(proxy);
1261 if (r < 0) {
1262 goto config_error;
1263 }
1264 }
1265
1266 r = msix_set_vector_notifiers(&proxy->pci_dev, virtio_pci_vector_unmask,
1267 virtio_pci_vector_mask,
1268 virtio_pci_vector_poll);
1269 if (r < 0) {
1270 goto notifiers_error;
1271 }
1272 }
1273
1274 return 0;
1275
1276 notifiers_error:
1277 if (with_irqfd) {
1278 assert(assign);
1279 kvm_virtio_pci_vector_vq_release(proxy, nvqs);
1280 }
1281 config_error:
1282 if (with_irqfd) {
1283 kvm_virtio_pci_vector_config_release(proxy);
1284 }
1285 config_assign_error:
1286 virtio_pci_set_guest_notifier(d, VIRTIO_CONFIG_IRQ_IDX, !assign,
1287 with_irqfd);
1288 assign_error:
1289 /* We get here on assignment failure. Recover by undoing for VQs 0 .. n. */
1290 assert(assign);
1291 while (--n >= 0) {
1292 virtio_pci_set_guest_notifier(d, n, !assign, with_irqfd);
1293 }
1294 g_free(proxy->vector_irqfd);
1295 proxy->vector_irqfd = NULL;
1296 return r;
1297 }
1298
1299 static int virtio_pci_set_host_notifier_mr(DeviceState *d, int n,
1300 MemoryRegion *mr, bool assign)
1301 {
1302 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1303 int offset;
1304
1305 if (n >= VIRTIO_QUEUE_MAX || !virtio_pci_modern(proxy) ||
1306 virtio_pci_queue_mem_mult(proxy) != memory_region_size(mr)) {
1307 return -1;
1308 }
1309
1310 if (assign) {
1311 offset = virtio_pci_queue_mem_mult(proxy) * n;
1312 memory_region_add_subregion_overlap(&proxy->notify.mr, offset, mr, 1);
1313 } else {
1314 memory_region_del_subregion(&proxy->notify.mr, mr);
1315 }
1316
1317 return 0;
1318 }
1319
1320 static void virtio_pci_vmstate_change(DeviceState *d, bool running)
1321 {
1322 VirtIOPCIProxy *proxy = to_virtio_pci_proxy(d);
1323 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1324
1325 if (running) {
1326 /* Old QEMU versions did not set bus master enable on status write.
1327 * Detect DRIVER set and enable it.
1328 */
1329 if ((proxy->flags & VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION) &&
1330 (vdev->status & VIRTIO_CONFIG_S_DRIVER) &&
1331 !(proxy->pci_dev.config[PCI_COMMAND] & PCI_COMMAND_MASTER)) {
1332 pci_default_write_config(&proxy->pci_dev, PCI_COMMAND,
1333 proxy->pci_dev.config[PCI_COMMAND] |
1334 PCI_COMMAND_MASTER, 1);
1335 }
1336 virtio_pci_start_ioeventfd(proxy);
1337 } else {
1338 virtio_pci_stop_ioeventfd(proxy);
1339 }
1340 }
1341
1342 /*
1343 * virtio-pci: This is the PCIDevice which has a virtio-pci-bus.
1344 */
1345
1346 static int virtio_pci_query_nvectors(DeviceState *d)
1347 {
1348 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1349
1350 return proxy->nvectors;
1351 }
1352
1353 static AddressSpace *virtio_pci_get_dma_as(DeviceState *d)
1354 {
1355 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1356 PCIDevice *dev = &proxy->pci_dev;
1357
1358 return pci_get_address_space(dev);
1359 }
1360
1361 static bool virtio_pci_iommu_enabled(DeviceState *d)
1362 {
1363 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1364 PCIDevice *dev = &proxy->pci_dev;
1365 AddressSpace *dma_as = pci_device_iommu_address_space(dev);
1366
1367 if (dma_as == &address_space_memory) {
1368 return false;
1369 }
1370
1371 return true;
1372 }
1373
1374 static bool virtio_pci_queue_enabled(DeviceState *d, int n)
1375 {
1376 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1377 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1378
1379 if (virtio_vdev_has_feature(vdev, VIRTIO_F_VERSION_1)) {
1380 return proxy->vqs[n].enabled;
1381 }
1382
1383 return virtio_queue_enabled_legacy(vdev, n);
1384 }
1385
1386 static int virtio_pci_add_mem_cap(VirtIOPCIProxy *proxy,
1387 struct virtio_pci_cap *cap)
1388 {
1389 PCIDevice *dev = &proxy->pci_dev;
1390 int offset;
1391
1392 offset = pci_add_capability(dev, PCI_CAP_ID_VNDR, 0,
1393 cap->cap_len, &error_abort);
1394
1395 assert(cap->cap_len >= sizeof *cap);
1396 memcpy(dev->config + offset + PCI_CAP_FLAGS, &cap->cap_len,
1397 cap->cap_len - PCI_CAP_FLAGS);
1398
1399 return offset;
1400 }
1401
1402 static uint64_t virtio_pci_common_read(void *opaque, hwaddr addr,
1403 unsigned size)
1404 {
1405 VirtIOPCIProxy *proxy = opaque;
1406 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1407 uint32_t val = 0;
1408 int i;
1409
1410 if (vdev == NULL) {
1411 return UINT64_MAX;
1412 }
1413
1414 switch (addr) {
1415 case VIRTIO_PCI_COMMON_DFSELECT:
1416 val = proxy->dfselect;
1417 break;
1418 case VIRTIO_PCI_COMMON_DF:
1419 if (proxy->dfselect <= 1) {
1420 VirtioDeviceClass *vdc = VIRTIO_DEVICE_GET_CLASS(vdev);
1421
1422 val = (vdev->host_features & ~vdc->legacy_features) >>
1423 (32 * proxy->dfselect);
1424 }
1425 break;
1426 case VIRTIO_PCI_COMMON_GFSELECT:
1427 val = proxy->gfselect;
1428 break;
1429 case VIRTIO_PCI_COMMON_GF:
1430 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1431 val = proxy->guest_features[proxy->gfselect];
1432 }
1433 break;
1434 case VIRTIO_PCI_COMMON_MSIX:
1435 val = vdev->config_vector;
1436 break;
1437 case VIRTIO_PCI_COMMON_NUMQ:
1438 for (i = 0; i < VIRTIO_QUEUE_MAX; ++i) {
1439 if (virtio_queue_get_num(vdev, i)) {
1440 val = i + 1;
1441 }
1442 }
1443 break;
1444 case VIRTIO_PCI_COMMON_STATUS:
1445 val = vdev->status;
1446 break;
1447 case VIRTIO_PCI_COMMON_CFGGENERATION:
1448 val = vdev->generation;
1449 break;
1450 case VIRTIO_PCI_COMMON_Q_SELECT:
1451 val = vdev->queue_sel;
1452 break;
1453 case VIRTIO_PCI_COMMON_Q_SIZE:
1454 val = virtio_queue_get_num(vdev, vdev->queue_sel);
1455 break;
1456 case VIRTIO_PCI_COMMON_Q_MSIX:
1457 val = virtio_queue_vector(vdev, vdev->queue_sel);
1458 break;
1459 case VIRTIO_PCI_COMMON_Q_ENABLE:
1460 val = proxy->vqs[vdev->queue_sel].enabled;
1461 break;
1462 case VIRTIO_PCI_COMMON_Q_NOFF:
1463 /* Simply map queues in order */
1464 val = vdev->queue_sel;
1465 break;
1466 case VIRTIO_PCI_COMMON_Q_DESCLO:
1467 val = proxy->vqs[vdev->queue_sel].desc[0];
1468 break;
1469 case VIRTIO_PCI_COMMON_Q_DESCHI:
1470 val = proxy->vqs[vdev->queue_sel].desc[1];
1471 break;
1472 case VIRTIO_PCI_COMMON_Q_AVAILLO:
1473 val = proxy->vqs[vdev->queue_sel].avail[0];
1474 break;
1475 case VIRTIO_PCI_COMMON_Q_AVAILHI:
1476 val = proxy->vqs[vdev->queue_sel].avail[1];
1477 break;
1478 case VIRTIO_PCI_COMMON_Q_USEDLO:
1479 val = proxy->vqs[vdev->queue_sel].used[0];
1480 break;
1481 case VIRTIO_PCI_COMMON_Q_USEDHI:
1482 val = proxy->vqs[vdev->queue_sel].used[1];
1483 break;
1484 case VIRTIO_PCI_COMMON_Q_RESET:
1485 val = proxy->vqs[vdev->queue_sel].reset;
1486 break;
1487 default:
1488 val = 0;
1489 }
1490
1491 return val;
1492 }
1493
1494 static void virtio_pci_common_write(void *opaque, hwaddr addr,
1495 uint64_t val, unsigned size)
1496 {
1497 VirtIOPCIProxy *proxy = opaque;
1498 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1499 uint16_t vector;
1500
1501 if (vdev == NULL) {
1502 return;
1503 }
1504
1505 switch (addr) {
1506 case VIRTIO_PCI_COMMON_DFSELECT:
1507 proxy->dfselect = val;
1508 break;
1509 case VIRTIO_PCI_COMMON_GFSELECT:
1510 proxy->gfselect = val;
1511 break;
1512 case VIRTIO_PCI_COMMON_GF:
1513 if (proxy->gfselect < ARRAY_SIZE(proxy->guest_features)) {
1514 proxy->guest_features[proxy->gfselect] = val;
1515 virtio_set_features(vdev,
1516 (((uint64_t)proxy->guest_features[1]) << 32) |
1517 proxy->guest_features[0]);
1518 }
1519 break;
1520 case VIRTIO_PCI_COMMON_MSIX:
1521 if (vdev->config_vector != VIRTIO_NO_VECTOR) {
1522 msix_vector_unuse(&proxy->pci_dev, vdev->config_vector);
1523 }
1524 /* Make it possible for guest to discover an error took place. */
1525 if (val < proxy->nvectors) {
1526 msix_vector_use(&proxy->pci_dev, val);
1527 } else {
1528 val = VIRTIO_NO_VECTOR;
1529 }
1530 vdev->config_vector = val;
1531 break;
1532 case VIRTIO_PCI_COMMON_STATUS:
1533 if (!(val & VIRTIO_CONFIG_S_DRIVER_OK)) {
1534 virtio_pci_stop_ioeventfd(proxy);
1535 }
1536
1537 virtio_set_status(vdev, val & 0xFF);
1538
1539 if (val & VIRTIO_CONFIG_S_DRIVER_OK) {
1540 virtio_pci_start_ioeventfd(proxy);
1541 }
1542
1543 if (vdev->status == 0) {
1544 virtio_pci_reset(DEVICE(proxy));
1545 }
1546
1547 break;
1548 case VIRTIO_PCI_COMMON_Q_SELECT:
1549 if (val < VIRTIO_QUEUE_MAX) {
1550 vdev->queue_sel = val;
1551 }
1552 break;
1553 case VIRTIO_PCI_COMMON_Q_SIZE:
1554 proxy->vqs[vdev->queue_sel].num = val;
1555 virtio_queue_set_num(vdev, vdev->queue_sel,
1556 proxy->vqs[vdev->queue_sel].num);
1557 break;
1558 case VIRTIO_PCI_COMMON_Q_MSIX:
1559 vector = virtio_queue_vector(vdev, vdev->queue_sel);
1560 if (vector != VIRTIO_NO_VECTOR) {
1561 msix_vector_unuse(&proxy->pci_dev, vector);
1562 }
1563 /* Make it possible for guest to discover an error took place. */
1564 if (val < proxy->nvectors) {
1565 msix_vector_use(&proxy->pci_dev, val);
1566 } else {
1567 val = VIRTIO_NO_VECTOR;
1568 }
1569 virtio_queue_set_vector(vdev, vdev->queue_sel, val);
1570 break;
1571 case VIRTIO_PCI_COMMON_Q_ENABLE:
1572 if (val == 1) {
1573 virtio_queue_set_num(vdev, vdev->queue_sel,
1574 proxy->vqs[vdev->queue_sel].num);
1575 virtio_queue_set_rings(vdev, vdev->queue_sel,
1576 ((uint64_t)proxy->vqs[vdev->queue_sel].desc[1]) << 32 |
1577 proxy->vqs[vdev->queue_sel].desc[0],
1578 ((uint64_t)proxy->vqs[vdev->queue_sel].avail[1]) << 32 |
1579 proxy->vqs[vdev->queue_sel].avail[0],
1580 ((uint64_t)proxy->vqs[vdev->queue_sel].used[1]) << 32 |
1581 proxy->vqs[vdev->queue_sel].used[0]);
1582 proxy->vqs[vdev->queue_sel].enabled = 1;
1583 proxy->vqs[vdev->queue_sel].reset = 0;
1584 virtio_queue_enable(vdev, vdev->queue_sel);
1585 } else {
1586 virtio_error(vdev, "wrong value for queue_enable %"PRIx64, val);
1587 }
1588 break;
1589 case VIRTIO_PCI_COMMON_Q_DESCLO:
1590 proxy->vqs[vdev->queue_sel].desc[0] = val;
1591 break;
1592 case VIRTIO_PCI_COMMON_Q_DESCHI:
1593 proxy->vqs[vdev->queue_sel].desc[1] = val;
1594 break;
1595 case VIRTIO_PCI_COMMON_Q_AVAILLO:
1596 proxy->vqs[vdev->queue_sel].avail[0] = val;
1597 break;
1598 case VIRTIO_PCI_COMMON_Q_AVAILHI:
1599 proxy->vqs[vdev->queue_sel].avail[1] = val;
1600 break;
1601 case VIRTIO_PCI_COMMON_Q_USEDLO:
1602 proxy->vqs[vdev->queue_sel].used[0] = val;
1603 break;
1604 case VIRTIO_PCI_COMMON_Q_USEDHI:
1605 proxy->vqs[vdev->queue_sel].used[1] = val;
1606 break;
1607 case VIRTIO_PCI_COMMON_Q_RESET:
1608 if (val == 1) {
1609 proxy->vqs[vdev->queue_sel].reset = 1;
1610
1611 virtio_queue_reset(vdev, vdev->queue_sel);
1612
1613 proxy->vqs[vdev->queue_sel].reset = 0;
1614 proxy->vqs[vdev->queue_sel].enabled = 0;
1615 }
1616 break;
1617 default:
1618 break;
1619 }
1620 }
1621
1622
1623 static uint64_t virtio_pci_notify_read(void *opaque, hwaddr addr,
1624 unsigned size)
1625 {
1626 VirtIOPCIProxy *proxy = opaque;
1627 if (virtio_bus_get_device(&proxy->bus) == NULL) {
1628 return UINT64_MAX;
1629 }
1630
1631 return 0;
1632 }
1633
1634 static void virtio_pci_notify_write(void *opaque, hwaddr addr,
1635 uint64_t val, unsigned size)
1636 {
1637 VirtIOPCIProxy *proxy = opaque;
1638 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1639
1640 unsigned queue = addr / virtio_pci_queue_mem_mult(proxy);
1641
1642 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1643 trace_virtio_pci_notify_write(addr, val, size);
1644 virtio_queue_notify(vdev, queue);
1645 }
1646 }
1647
1648 static void virtio_pci_notify_write_pio(void *opaque, hwaddr addr,
1649 uint64_t val, unsigned size)
1650 {
1651 VirtIOPCIProxy *proxy = opaque;
1652 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1653
1654 unsigned queue = val;
1655
1656 if (vdev != NULL && queue < VIRTIO_QUEUE_MAX) {
1657 trace_virtio_pci_notify_write_pio(addr, val, size);
1658 virtio_queue_notify(vdev, queue);
1659 }
1660 }
1661
1662 static uint64_t virtio_pci_isr_read(void *opaque, hwaddr addr,
1663 unsigned size)
1664 {
1665 VirtIOPCIProxy *proxy = opaque;
1666 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1667 uint64_t val;
1668
1669 if (vdev == NULL) {
1670 return UINT64_MAX;
1671 }
1672
1673 val = qatomic_xchg(&vdev->isr, 0);
1674 pci_irq_deassert(&proxy->pci_dev);
1675 return val;
1676 }
1677
1678 static void virtio_pci_isr_write(void *opaque, hwaddr addr,
1679 uint64_t val, unsigned size)
1680 {
1681 }
1682
1683 static uint64_t virtio_pci_device_read(void *opaque, hwaddr addr,
1684 unsigned size)
1685 {
1686 VirtIOPCIProxy *proxy = opaque;
1687 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1688 uint64_t val;
1689
1690 if (vdev == NULL) {
1691 return UINT64_MAX;
1692 }
1693
1694 switch (size) {
1695 case 1:
1696 val = virtio_config_modern_readb(vdev, addr);
1697 break;
1698 case 2:
1699 val = virtio_config_modern_readw(vdev, addr);
1700 break;
1701 case 4:
1702 val = virtio_config_modern_readl(vdev, addr);
1703 break;
1704 default:
1705 val = 0;
1706 break;
1707 }
1708 return val;
1709 }
1710
1711 static void virtio_pci_device_write(void *opaque, hwaddr addr,
1712 uint64_t val, unsigned size)
1713 {
1714 VirtIOPCIProxy *proxy = opaque;
1715 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1716
1717 if (vdev == NULL) {
1718 return;
1719 }
1720
1721 switch (size) {
1722 case 1:
1723 virtio_config_modern_writeb(vdev, addr, val);
1724 break;
1725 case 2:
1726 virtio_config_modern_writew(vdev, addr, val);
1727 break;
1728 case 4:
1729 virtio_config_modern_writel(vdev, addr, val);
1730 break;
1731 }
1732 }
1733
1734 static void virtio_pci_modern_regions_init(VirtIOPCIProxy *proxy,
1735 const char *vdev_name)
1736 {
1737 static const MemoryRegionOps common_ops = {
1738 .read = virtio_pci_common_read,
1739 .write = virtio_pci_common_write,
1740 .impl = {
1741 .min_access_size = 1,
1742 .max_access_size = 4,
1743 },
1744 .endianness = DEVICE_LITTLE_ENDIAN,
1745 };
1746 static const MemoryRegionOps isr_ops = {
1747 .read = virtio_pci_isr_read,
1748 .write = virtio_pci_isr_write,
1749 .impl = {
1750 .min_access_size = 1,
1751 .max_access_size = 4,
1752 },
1753 .endianness = DEVICE_LITTLE_ENDIAN,
1754 };
1755 static const MemoryRegionOps device_ops = {
1756 .read = virtio_pci_device_read,
1757 .write = virtio_pci_device_write,
1758 .impl = {
1759 .min_access_size = 1,
1760 .max_access_size = 4,
1761 },
1762 .endianness = DEVICE_LITTLE_ENDIAN,
1763 };
1764 static const MemoryRegionOps notify_ops = {
1765 .read = virtio_pci_notify_read,
1766 .write = virtio_pci_notify_write,
1767 .impl = {
1768 .min_access_size = 1,
1769 .max_access_size = 4,
1770 },
1771 .endianness = DEVICE_LITTLE_ENDIAN,
1772 };
1773 static const MemoryRegionOps notify_pio_ops = {
1774 .read = virtio_pci_notify_read,
1775 .write = virtio_pci_notify_write_pio,
1776 .impl = {
1777 .min_access_size = 1,
1778 .max_access_size = 4,
1779 },
1780 .endianness = DEVICE_LITTLE_ENDIAN,
1781 };
1782 g_autoptr(GString) name = g_string_new(NULL);
1783
1784 g_string_printf(name, "virtio-pci-common-%s", vdev_name);
1785 memory_region_init_io(&proxy->common.mr, OBJECT(proxy),
1786 &common_ops,
1787 proxy,
1788 name->str,
1789 proxy->common.size);
1790
1791 g_string_printf(name, "virtio-pci-isr-%s", vdev_name);
1792 memory_region_init_io(&proxy->isr.mr, OBJECT(proxy),
1793 &isr_ops,
1794 proxy,
1795 name->str,
1796 proxy->isr.size);
1797
1798 g_string_printf(name, "virtio-pci-device-%s", vdev_name);
1799 memory_region_init_io(&proxy->device.mr, OBJECT(proxy),
1800 &device_ops,
1801 proxy,
1802 name->str,
1803 proxy->device.size);
1804
1805 g_string_printf(name, "virtio-pci-notify-%s", vdev_name);
1806 memory_region_init_io(&proxy->notify.mr, OBJECT(proxy),
1807 &notify_ops,
1808 proxy,
1809 name->str,
1810 proxy->notify.size);
1811
1812 g_string_printf(name, "virtio-pci-notify-pio-%s", vdev_name);
1813 memory_region_init_io(&proxy->notify_pio.mr, OBJECT(proxy),
1814 &notify_pio_ops,
1815 proxy,
1816 name->str,
1817 proxy->notify_pio.size);
1818 }
1819
1820 static void virtio_pci_modern_region_map(VirtIOPCIProxy *proxy,
1821 VirtIOPCIRegion *region,
1822 struct virtio_pci_cap *cap,
1823 MemoryRegion *mr,
1824 uint8_t bar)
1825 {
1826 memory_region_add_subregion(mr, region->offset, &region->mr);
1827
1828 cap->cfg_type = region->type;
1829 cap->bar = bar;
1830 cap->offset = cpu_to_le32(region->offset);
1831 cap->length = cpu_to_le32(region->size);
1832 virtio_pci_add_mem_cap(proxy, cap);
1833
1834 }
1835
1836 static void virtio_pci_modern_mem_region_map(VirtIOPCIProxy *proxy,
1837 VirtIOPCIRegion *region,
1838 struct virtio_pci_cap *cap)
1839 {
1840 virtio_pci_modern_region_map(proxy, region, cap,
1841 &proxy->modern_bar, proxy->modern_mem_bar_idx);
1842 }
1843
1844 static void virtio_pci_modern_io_region_map(VirtIOPCIProxy *proxy,
1845 VirtIOPCIRegion *region,
1846 struct virtio_pci_cap *cap)
1847 {
1848 virtio_pci_modern_region_map(proxy, region, cap,
1849 &proxy->io_bar, proxy->modern_io_bar_idx);
1850 }
1851
1852 static void virtio_pci_modern_mem_region_unmap(VirtIOPCIProxy *proxy,
1853 VirtIOPCIRegion *region)
1854 {
1855 memory_region_del_subregion(&proxy->modern_bar,
1856 &region->mr);
1857 }
1858
1859 static void virtio_pci_modern_io_region_unmap(VirtIOPCIProxy *proxy,
1860 VirtIOPCIRegion *region)
1861 {
1862 memory_region_del_subregion(&proxy->io_bar,
1863 &region->mr);
1864 }
1865
1866 static void virtio_pci_pre_plugged(DeviceState *d, Error **errp)
1867 {
1868 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1869 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1870
1871 if (virtio_pci_modern(proxy)) {
1872 virtio_add_feature(&vdev->host_features, VIRTIO_F_VERSION_1);
1873 }
1874
1875 virtio_add_feature(&vdev->host_features, VIRTIO_F_BAD_FEATURE);
1876 }
1877
1878 /* This is called by virtio-bus just after the device is plugged. */
1879 static void virtio_pci_device_plugged(DeviceState *d, Error **errp)
1880 {
1881 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
1882 VirtioBusState *bus = &proxy->bus;
1883 bool legacy = virtio_pci_legacy(proxy);
1884 bool modern;
1885 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
1886 uint8_t *config;
1887 uint32_t size;
1888 VirtIODevice *vdev = virtio_bus_get_device(&proxy->bus);
1889
1890 /*
1891 * Virtio capabilities present without
1892 * VIRTIO_F_VERSION_1 confuses guests
1893 */
1894 if (!proxy->ignore_backend_features &&
1895 !virtio_has_feature(vdev->host_features, VIRTIO_F_VERSION_1)) {
1896 virtio_pci_disable_modern(proxy);
1897
1898 if (!legacy) {
1899 error_setg(errp, "Device doesn't support modern mode, and legacy"
1900 " mode is disabled");
1901 error_append_hint(errp, "Set disable-legacy to off\n");
1902
1903 return;
1904 }
1905 }
1906
1907 modern = virtio_pci_modern(proxy);
1908
1909 config = proxy->pci_dev.config;
1910 if (proxy->class_code) {
1911 pci_config_set_class(config, proxy->class_code);
1912 }
1913
1914 if (legacy) {
1915 if (!virtio_legacy_allowed(vdev)) {
1916 /*
1917 * To avoid migration issues, we allow legacy mode when legacy
1918 * check is disabled in the old machine types (< 5.1).
1919 */
1920 if (virtio_legacy_check_disabled(vdev)) {
1921 warn_report("device is modern-only, but for backward "
1922 "compatibility legacy is allowed");
1923 } else {
1924 error_setg(errp,
1925 "device is modern-only, use disable-legacy=on");
1926 return;
1927 }
1928 }
1929 if (virtio_host_has_feature(vdev, VIRTIO_F_IOMMU_PLATFORM)) {
1930 error_setg(errp, "VIRTIO_F_IOMMU_PLATFORM was supported by"
1931 " neither legacy nor transitional device");
1932 return;
1933 }
1934 /*
1935 * Legacy and transitional devices use specific subsystem IDs.
1936 * Note that the subsystem vendor ID (config + PCI_SUBSYSTEM_VENDOR_ID)
1937 * is set to PCI_SUBVENDOR_ID_REDHAT_QUMRANET by default.
1938 */
1939 pci_set_word(config + PCI_SUBSYSTEM_ID, virtio_bus_get_vdev_id(bus));
1940 if (proxy->trans_devid) {
1941 pci_config_set_device_id(config, proxy->trans_devid);
1942 }
1943 } else {
1944 /* pure virtio-1.0 */
1945 pci_set_word(config + PCI_VENDOR_ID,
1946 PCI_VENDOR_ID_REDHAT_QUMRANET);
1947 pci_set_word(config + PCI_DEVICE_ID,
1948 PCI_DEVICE_ID_VIRTIO_10_BASE + virtio_bus_get_vdev_id(bus));
1949 pci_config_set_revision(config, 1);
1950 }
1951 config[PCI_INTERRUPT_PIN] = 1;
1952
1953
1954 if (modern) {
1955 struct virtio_pci_cap cap = {
1956 .cap_len = sizeof cap,
1957 };
1958 struct virtio_pci_notify_cap notify = {
1959 .cap.cap_len = sizeof notify,
1960 .notify_off_multiplier =
1961 cpu_to_le32(virtio_pci_queue_mem_mult(proxy)),
1962 };
1963 struct virtio_pci_cfg_cap cfg = {
1964 .cap.cap_len = sizeof cfg,
1965 .cap.cfg_type = VIRTIO_PCI_CAP_PCI_CFG,
1966 };
1967 struct virtio_pci_notify_cap notify_pio = {
1968 .cap.cap_len = sizeof notify,
1969 .notify_off_multiplier = cpu_to_le32(0x0),
1970 };
1971
1972 struct virtio_pci_cfg_cap *cfg_mask;
1973
1974 virtio_pci_modern_regions_init(proxy, vdev->name);
1975
1976 virtio_pci_modern_mem_region_map(proxy, &proxy->common, &cap);
1977 virtio_pci_modern_mem_region_map(proxy, &proxy->isr, &cap);
1978 virtio_pci_modern_mem_region_map(proxy, &proxy->device, &cap);
1979 virtio_pci_modern_mem_region_map(proxy, &proxy->notify, &notify.cap);
1980
1981 if (modern_pio) {
1982 memory_region_init(&proxy->io_bar, OBJECT(proxy),
1983 "virtio-pci-io", 0x4);
1984
1985 pci_register_bar(&proxy->pci_dev, proxy->modern_io_bar_idx,
1986 PCI_BASE_ADDRESS_SPACE_IO, &proxy->io_bar);
1987
1988 virtio_pci_modern_io_region_map(proxy, &proxy->notify_pio,
1989 &notify_pio.cap);
1990 }
1991
1992 pci_register_bar(&proxy->pci_dev, proxy->modern_mem_bar_idx,
1993 PCI_BASE_ADDRESS_SPACE_MEMORY |
1994 PCI_BASE_ADDRESS_MEM_PREFETCH |
1995 PCI_BASE_ADDRESS_MEM_TYPE_64,
1996 &proxy->modern_bar);
1997
1998 proxy->config_cap = virtio_pci_add_mem_cap(proxy, &cfg.cap);
1999 cfg_mask = (void *)(proxy->pci_dev.wmask + proxy->config_cap);
2000 pci_set_byte(&cfg_mask->cap.bar, ~0x0);
2001 pci_set_long((uint8_t *)&cfg_mask->cap.offset, ~0x0);
2002 pci_set_long((uint8_t *)&cfg_mask->cap.length, ~0x0);
2003 pci_set_long(cfg_mask->pci_cfg_data, ~0x0);
2004 }
2005
2006 if (proxy->nvectors) {
2007 int err = msix_init_exclusive_bar(&proxy->pci_dev, proxy->nvectors,
2008 proxy->msix_bar_idx, NULL);
2009 if (err) {
2010 /* Notice when a system that supports MSIx can't initialize it */
2011 if (err != -ENOTSUP) {
2012 warn_report("unable to init msix vectors to %" PRIu32,
2013 proxy->nvectors);
2014 }
2015 proxy->nvectors = 0;
2016 }
2017 }
2018
2019 proxy->pci_dev.config_write = virtio_write_config;
2020 proxy->pci_dev.config_read = virtio_read_config;
2021
2022 if (legacy) {
2023 size = VIRTIO_PCI_REGION_SIZE(&proxy->pci_dev)
2024 + virtio_bus_get_vdev_config_len(bus);
2025 size = pow2ceil(size);
2026
2027 memory_region_init_io(&proxy->bar, OBJECT(proxy),
2028 &virtio_pci_config_ops,
2029 proxy, "virtio-pci", size);
2030
2031 pci_register_bar(&proxy->pci_dev, proxy->legacy_io_bar_idx,
2032 PCI_BASE_ADDRESS_SPACE_IO, &proxy->bar);
2033 }
2034 }
2035
2036 static void virtio_pci_device_unplugged(DeviceState *d)
2037 {
2038 VirtIOPCIProxy *proxy = VIRTIO_PCI(d);
2039 bool modern = virtio_pci_modern(proxy);
2040 bool modern_pio = proxy->flags & VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY;
2041
2042 virtio_pci_stop_ioeventfd(proxy);
2043
2044 if (modern) {
2045 virtio_pci_modern_mem_region_unmap(proxy, &proxy->common);
2046 virtio_pci_modern_mem_region_unmap(proxy, &proxy->isr);
2047 virtio_pci_modern_mem_region_unmap(proxy, &proxy->device);
2048 virtio_pci_modern_mem_region_unmap(proxy, &proxy->notify);
2049 if (modern_pio) {
2050 virtio_pci_modern_io_region_unmap(proxy, &proxy->notify_pio);
2051 }
2052 }
2053 }
2054
2055 static void virtio_pci_realize(PCIDevice *pci_dev, Error **errp)
2056 {
2057 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
2058 VirtioPCIClass *k = VIRTIO_PCI_GET_CLASS(pci_dev);
2059 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
2060 !pci_bus_is_root(pci_get_bus(pci_dev));
2061
2062 if (kvm_enabled() && !kvm_has_many_ioeventfds()) {
2063 proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
2064 }
2065
2066 /* fd-based ioevents can't be synchronized in record/replay */
2067 if (replay_mode != REPLAY_MODE_NONE) {
2068 proxy->flags &= ~VIRTIO_PCI_FLAG_USE_IOEVENTFD;
2069 }
2070
2071 /*
2072 * virtio pci bar layout used by default.
2073 * subclasses can re-arrange things if needed.
2074 *
2075 * region 0 -- virtio legacy io bar
2076 * region 1 -- msi-x bar
2077 * region 2 -- virtio modern io bar (off by default)
2078 * region 4+5 -- virtio modern memory (64bit) bar
2079 *
2080 */
2081 proxy->legacy_io_bar_idx = 0;
2082 proxy->msix_bar_idx = 1;
2083 proxy->modern_io_bar_idx = 2;
2084 proxy->modern_mem_bar_idx = 4;
2085
2086 proxy->common.offset = 0x0;
2087 proxy->common.size = 0x1000;
2088 proxy->common.type = VIRTIO_PCI_CAP_COMMON_CFG;
2089
2090 proxy->isr.offset = 0x1000;
2091 proxy->isr.size = 0x1000;
2092 proxy->isr.type = VIRTIO_PCI_CAP_ISR_CFG;
2093
2094 proxy->device.offset = 0x2000;
2095 proxy->device.size = 0x1000;
2096 proxy->device.type = VIRTIO_PCI_CAP_DEVICE_CFG;
2097
2098 proxy->notify.offset = 0x3000;
2099 proxy->notify.size = virtio_pci_queue_mem_mult(proxy) * VIRTIO_QUEUE_MAX;
2100 proxy->notify.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
2101
2102 proxy->notify_pio.offset = 0x0;
2103 proxy->notify_pio.size = 0x4;
2104 proxy->notify_pio.type = VIRTIO_PCI_CAP_NOTIFY_CFG;
2105
2106 /* subclasses can enforce modern, so do this unconditionally */
2107 memory_region_init(&proxy->modern_bar, OBJECT(proxy), "virtio-pci",
2108 /* PCI BAR regions must be powers of 2 */
2109 pow2ceil(proxy->notify.offset + proxy->notify.size));
2110
2111 if (proxy->disable_legacy == ON_OFF_AUTO_AUTO) {
2112 proxy->disable_legacy = pcie_port ? ON_OFF_AUTO_ON : ON_OFF_AUTO_OFF;
2113 }
2114
2115 if (!virtio_pci_modern(proxy) && !virtio_pci_legacy(proxy)) {
2116 error_setg(errp, "device cannot work as neither modern nor legacy mode"
2117 " is enabled");
2118 error_append_hint(errp, "Set either disable-modern or disable-legacy"
2119 " to off\n");
2120 return;
2121 }
2122
2123 if (pcie_port && pci_is_express(pci_dev)) {
2124 int pos;
2125 uint16_t last_pcie_cap_offset = PCI_CONFIG_SPACE_SIZE;
2126
2127 pos = pcie_endpoint_cap_init(pci_dev, 0);
2128 assert(pos > 0);
2129
2130 pos = pci_add_capability(pci_dev, PCI_CAP_ID_PM, 0,
2131 PCI_PM_SIZEOF, errp);
2132 if (pos < 0) {
2133 return;
2134 }
2135
2136 pci_dev->exp.pm_cap = pos;
2137
2138 /*
2139 * Indicates that this function complies with revision 1.2 of the
2140 * PCI Power Management Interface Specification.
2141 */
2142 pci_set_word(pci_dev->config + pos + PCI_PM_PMC, 0x3);
2143
2144 if (proxy->flags & VIRTIO_PCI_FLAG_AER) {
2145 pcie_aer_init(pci_dev, PCI_ERR_VER, last_pcie_cap_offset,
2146 PCI_ERR_SIZEOF, NULL);
2147 last_pcie_cap_offset += PCI_ERR_SIZEOF;
2148 }
2149
2150 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_DEVERR) {
2151 /* Init error enabling flags */
2152 pcie_cap_deverr_init(pci_dev);
2153 }
2154
2155 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_LNKCTL) {
2156 /* Init Link Control Register */
2157 pcie_cap_lnkctl_init(pci_dev);
2158 }
2159
2160 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_PM) {
2161 /* Init Power Management Control Register */
2162 pci_set_word(pci_dev->wmask + pos + PCI_PM_CTRL,
2163 PCI_PM_CTRL_STATE_MASK);
2164 }
2165
2166 if (proxy->flags & VIRTIO_PCI_FLAG_ATS) {
2167 pcie_ats_init(pci_dev, last_pcie_cap_offset,
2168 proxy->flags & VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED);
2169 last_pcie_cap_offset += PCI_EXT_CAP_ATS_SIZEOF;
2170 }
2171
2172 if (proxy->flags & VIRTIO_PCI_FLAG_INIT_FLR) {
2173 /* Set Function Level Reset capability bit */
2174 pcie_cap_flr_init(pci_dev);
2175 }
2176 } else {
2177 /*
2178 * make future invocations of pci_is_express() return false
2179 * and pci_config_size() return PCI_CONFIG_SPACE_SIZE.
2180 */
2181 pci_dev->cap_present &= ~QEMU_PCI_CAP_EXPRESS;
2182 }
2183
2184 virtio_pci_bus_new(&proxy->bus, sizeof(proxy->bus), proxy);
2185 if (k->realize) {
2186 k->realize(proxy, errp);
2187 }
2188 }
2189
2190 static void virtio_pci_exit(PCIDevice *pci_dev)
2191 {
2192 VirtIOPCIProxy *proxy = VIRTIO_PCI(pci_dev);
2193 bool pcie_port = pci_bus_is_express(pci_get_bus(pci_dev)) &&
2194 !pci_bus_is_root(pci_get_bus(pci_dev));
2195
2196 msix_uninit_exclusive_bar(pci_dev);
2197 if (proxy->flags & VIRTIO_PCI_FLAG_AER && pcie_port &&
2198 pci_is_express(pci_dev)) {
2199 pcie_aer_exit(pci_dev);
2200 }
2201 }
2202
2203 static void virtio_pci_reset(DeviceState *qdev)
2204 {
2205 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
2206 VirtioBusState *bus = VIRTIO_BUS(&proxy->bus);
2207 int i;
2208
2209 virtio_bus_reset(bus);
2210 msix_unuse_all_vectors(&proxy->pci_dev);
2211
2212 for (i = 0; i < VIRTIO_QUEUE_MAX; i++) {
2213 proxy->vqs[i].enabled = 0;
2214 proxy->vqs[i].reset = 0;
2215 proxy->vqs[i].num = 0;
2216 proxy->vqs[i].desc[0] = proxy->vqs[i].desc[1] = 0;
2217 proxy->vqs[i].avail[0] = proxy->vqs[i].avail[1] = 0;
2218 proxy->vqs[i].used[0] = proxy->vqs[i].used[1] = 0;
2219 }
2220 }
2221
2222 static void virtio_pci_bus_reset_hold(Object *obj)
2223 {
2224 PCIDevice *dev = PCI_DEVICE(obj);
2225 DeviceState *qdev = DEVICE(obj);
2226
2227 virtio_pci_reset(qdev);
2228
2229 if (pci_is_express(dev)) {
2230 pcie_cap_deverr_reset(dev);
2231 pcie_cap_lnkctl_reset(dev);
2232
2233 pci_set_word(dev->config + dev->exp.pm_cap + PCI_PM_CTRL, 0);
2234 }
2235 }
2236
2237 static Property virtio_pci_properties[] = {
2238 DEFINE_PROP_BIT("virtio-pci-bus-master-bug-migration", VirtIOPCIProxy, flags,
2239 VIRTIO_PCI_FLAG_BUS_MASTER_BUG_MIGRATION_BIT, false),
2240 DEFINE_PROP_BIT("migrate-extra", VirtIOPCIProxy, flags,
2241 VIRTIO_PCI_FLAG_MIGRATE_EXTRA_BIT, true),
2242 DEFINE_PROP_BIT("modern-pio-notify", VirtIOPCIProxy, flags,
2243 VIRTIO_PCI_FLAG_MODERN_PIO_NOTIFY_BIT, false),
2244 DEFINE_PROP_BIT("x-disable-pcie", VirtIOPCIProxy, flags,
2245 VIRTIO_PCI_FLAG_DISABLE_PCIE_BIT, false),
2246 DEFINE_PROP_BIT("page-per-vq", VirtIOPCIProxy, flags,
2247 VIRTIO_PCI_FLAG_PAGE_PER_VQ_BIT, false),
2248 DEFINE_PROP_BOOL("x-ignore-backend-features", VirtIOPCIProxy,
2249 ignore_backend_features, false),
2250 DEFINE_PROP_BIT("ats", VirtIOPCIProxy, flags,
2251 VIRTIO_PCI_FLAG_ATS_BIT, false),
2252 DEFINE_PROP_BIT("x-ats-page-aligned", VirtIOPCIProxy, flags,
2253 VIRTIO_PCI_FLAG_ATS_PAGE_ALIGNED_BIT, true),
2254 DEFINE_PROP_BIT("x-pcie-deverr-init", VirtIOPCIProxy, flags,
2255 VIRTIO_PCI_FLAG_INIT_DEVERR_BIT, true),
2256 DEFINE_PROP_BIT("x-pcie-lnkctl-init", VirtIOPCIProxy, flags,
2257 VIRTIO_PCI_FLAG_INIT_LNKCTL_BIT, true),
2258 DEFINE_PROP_BIT("x-pcie-pm-init", VirtIOPCIProxy, flags,
2259 VIRTIO_PCI_FLAG_INIT_PM_BIT, true),
2260 DEFINE_PROP_BIT("x-pcie-flr-init", VirtIOPCIProxy, flags,
2261 VIRTIO_PCI_FLAG_INIT_FLR_BIT, true),
2262 DEFINE_PROP_BIT("aer", VirtIOPCIProxy, flags,
2263 VIRTIO_PCI_FLAG_AER_BIT, false),
2264 DEFINE_PROP_END_OF_LIST(),
2265 };
2266
2267 static void virtio_pci_dc_realize(DeviceState *qdev, Error **errp)
2268 {
2269 VirtioPCIClass *vpciklass = VIRTIO_PCI_GET_CLASS(qdev);
2270 VirtIOPCIProxy *proxy = VIRTIO_PCI(qdev);
2271 PCIDevice *pci_dev = &proxy->pci_dev;
2272
2273 if (!(proxy->flags & VIRTIO_PCI_FLAG_DISABLE_PCIE) &&
2274 virtio_pci_modern(proxy)) {
2275 pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2276 }
2277
2278 vpciklass->parent_dc_realize(qdev, errp);
2279 }
2280
2281 static void virtio_pci_class_init(ObjectClass *klass, void *data)
2282 {
2283 DeviceClass *dc = DEVICE_CLASS(klass);
2284 PCIDeviceClass *k = PCI_DEVICE_CLASS(klass);
2285 VirtioPCIClass *vpciklass = VIRTIO_PCI_CLASS(klass);
2286 ResettableClass *rc = RESETTABLE_CLASS(klass);
2287
2288 device_class_set_props(dc, virtio_pci_properties);
2289 k->realize = virtio_pci_realize;
2290 k->exit = virtio_pci_exit;
2291 k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET;
2292 k->revision = VIRTIO_PCI_ABI_VERSION;
2293 k->class_id = PCI_CLASS_OTHERS;
2294 device_class_set_parent_realize(dc, virtio_pci_dc_realize,
2295 &vpciklass->parent_dc_realize);
2296 rc->phases.hold = virtio_pci_bus_reset_hold;
2297 }
2298
2299 static const TypeInfo virtio_pci_info = {
2300 .name = TYPE_VIRTIO_PCI,
2301 .parent = TYPE_PCI_DEVICE,
2302 .instance_size = sizeof(VirtIOPCIProxy),
2303 .class_init = virtio_pci_class_init,
2304 .class_size = sizeof(VirtioPCIClass),
2305 .abstract = true,
2306 };
2307
2308 static Property virtio_pci_generic_properties[] = {
2309 DEFINE_PROP_ON_OFF_AUTO("disable-legacy", VirtIOPCIProxy, disable_legacy,
2310 ON_OFF_AUTO_AUTO),
2311 DEFINE_PROP_BOOL("disable-modern", VirtIOPCIProxy, disable_modern, false),
2312 DEFINE_PROP_END_OF_LIST(),
2313 };
2314
2315 static void virtio_pci_base_class_init(ObjectClass *klass, void *data)
2316 {
2317 const VirtioPCIDeviceTypeInfo *t = data;
2318 if (t->class_init) {
2319 t->class_init(klass, NULL);
2320 }
2321 }
2322
2323 static void virtio_pci_generic_class_init(ObjectClass *klass, void *data)
2324 {
2325 DeviceClass *dc = DEVICE_CLASS(klass);
2326
2327 device_class_set_props(dc, virtio_pci_generic_properties);
2328 }
2329
2330 static void virtio_pci_transitional_instance_init(Object *obj)
2331 {
2332 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2333
2334 proxy->disable_legacy = ON_OFF_AUTO_OFF;
2335 proxy->disable_modern = false;
2336 }
2337
2338 static void virtio_pci_non_transitional_instance_init(Object *obj)
2339 {
2340 VirtIOPCIProxy *proxy = VIRTIO_PCI(obj);
2341
2342 proxy->disable_legacy = ON_OFF_AUTO_ON;
2343 proxy->disable_modern = false;
2344 }
2345
2346 void virtio_pci_types_register(const VirtioPCIDeviceTypeInfo *t)
2347 {
2348 char *base_name = NULL;
2349 TypeInfo base_type_info = {
2350 .name = t->base_name,
2351 .parent = t->parent ? t->parent : TYPE_VIRTIO_PCI,
2352 .instance_size = t->instance_size,
2353 .instance_init = t->instance_init,
2354 .class_size = t->class_size,
2355 .abstract = true,
2356 .interfaces = t->interfaces,
2357 };
2358 TypeInfo generic_type_info = {
2359 .name = t->generic_name,
2360 .parent = base_type_info.name,
2361 .class_init = virtio_pci_generic_class_init,
2362 .interfaces = (InterfaceInfo[]) {
2363 { INTERFACE_PCIE_DEVICE },
2364 { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2365 { }
2366 },
2367 };
2368
2369 if (!base_type_info.name) {
2370 /* No base type -> register a single generic device type */
2371 /* use intermediate %s-base-type to add generic device props */
2372 base_name = g_strdup_printf("%s-base-type", t->generic_name);
2373 base_type_info.name = base_name;
2374 base_type_info.class_init = virtio_pci_generic_class_init;
2375
2376 generic_type_info.parent = base_name;
2377 generic_type_info.class_init = virtio_pci_base_class_init;
2378 generic_type_info.class_data = (void *)t;
2379
2380 assert(!t->non_transitional_name);
2381 assert(!t->transitional_name);
2382 } else {
2383 base_type_info.class_init = virtio_pci_base_class_init;
2384 base_type_info.class_data = (void *)t;
2385 }
2386
2387 type_register(&base_type_info);
2388 if (generic_type_info.name) {
2389 type_register(&generic_type_info);
2390 }
2391
2392 if (t->non_transitional_name) {
2393 const TypeInfo non_transitional_type_info = {
2394 .name = t->non_transitional_name,
2395 .parent = base_type_info.name,
2396 .instance_init = virtio_pci_non_transitional_instance_init,
2397 .interfaces = (InterfaceInfo[]) {
2398 { INTERFACE_PCIE_DEVICE },
2399 { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2400 { }
2401 },
2402 };
2403 type_register(&non_transitional_type_info);
2404 }
2405
2406 if (t->transitional_name) {
2407 const TypeInfo transitional_type_info = {
2408 .name = t->transitional_name,
2409 .parent = base_type_info.name,
2410 .instance_init = virtio_pci_transitional_instance_init,
2411 .interfaces = (InterfaceInfo[]) {
2412 /*
2413 * Transitional virtio devices work only as Conventional PCI
2414 * devices because they require PIO ports.
2415 */
2416 { INTERFACE_CONVENTIONAL_PCI_DEVICE },
2417 { }
2418 },
2419 };
2420 type_register(&transitional_type_info);
2421 }
2422 g_free(base_name);
2423 }
2424
2425 unsigned virtio_pci_optimal_num_queues(unsigned fixed_queues)
2426 {
2427 /*
2428 * 1:1 vq to vCPU mapping is ideal because the same vCPU that submitted
2429 * virtqueue buffers can handle their completion. When a different vCPU
2430 * handles completion it may need to IPI the vCPU that submitted the
2431 * request and this adds overhead.
2432 *
2433 * Virtqueues consume guest RAM and MSI-X vectors. This is wasteful in
2434 * guests with very many vCPUs and a device that is only used by a few
2435 * vCPUs. Unfortunately optimizing that case requires manual pinning inside
2436 * the guest, so those users might as well manually set the number of
2437 * queues. There is no upper limit that can be applied automatically and
2438 * doing so arbitrarily would result in a sudden performance drop once the
2439 * threshold number of vCPUs is exceeded.
2440 */
2441 unsigned num_queues = current_machine->smp.cpus;
2442
2443 /*
2444 * The maximum number of MSI-X vectors is PCI_MSIX_FLAGS_QSIZE + 1, but the
2445 * config change interrupt and the fixed virtqueues must be taken into
2446 * account too.
2447 */
2448 num_queues = MIN(num_queues, PCI_MSIX_FLAGS_QSIZE - fixed_queues);
2449
2450 /*
2451 * There is a limit to how many virtqueues a device can have.
2452 */
2453 return MIN(num_queues, VIRTIO_QUEUE_MAX - fixed_queues);
2454 }
2455
2456 /* virtio-pci-bus */
2457
2458 static void virtio_pci_bus_new(VirtioBusState *bus, size_t bus_size,
2459 VirtIOPCIProxy *dev)
2460 {
2461 DeviceState *qdev = DEVICE(dev);
2462 char virtio_bus_name[] = "virtio-bus";
2463
2464 qbus_init(bus, bus_size, TYPE_VIRTIO_PCI_BUS, qdev, virtio_bus_name);
2465 }
2466
2467 static void virtio_pci_bus_class_init(ObjectClass *klass, void *data)
2468 {
2469 BusClass *bus_class = BUS_CLASS(klass);
2470 VirtioBusClass *k = VIRTIO_BUS_CLASS(klass);
2471 bus_class->max_dev = 1;
2472 k->notify = virtio_pci_notify;
2473 k->save_config = virtio_pci_save_config;
2474 k->load_config = virtio_pci_load_config;
2475 k->save_queue = virtio_pci_save_queue;
2476 k->load_queue = virtio_pci_load_queue;
2477 k->save_extra_state = virtio_pci_save_extra_state;
2478 k->load_extra_state = virtio_pci_load_extra_state;
2479 k->has_extra_state = virtio_pci_has_extra_state;
2480 k->query_guest_notifiers = virtio_pci_query_guest_notifiers;
2481 k->set_guest_notifiers = virtio_pci_set_guest_notifiers;
2482 k->set_host_notifier_mr = virtio_pci_set_host_notifier_mr;
2483 k->vmstate_change = virtio_pci_vmstate_change;
2484 k->pre_plugged = virtio_pci_pre_plugged;
2485 k->device_plugged = virtio_pci_device_plugged;
2486 k->device_unplugged = virtio_pci_device_unplugged;
2487 k->query_nvectors = virtio_pci_query_nvectors;
2488 k->ioeventfd_enabled = virtio_pci_ioeventfd_enabled;
2489 k->ioeventfd_assign = virtio_pci_ioeventfd_assign;
2490 k->get_dma_as = virtio_pci_get_dma_as;
2491 k->iommu_enabled = virtio_pci_iommu_enabled;
2492 k->queue_enabled = virtio_pci_queue_enabled;
2493 }
2494
2495 static const TypeInfo virtio_pci_bus_info = {
2496 .name = TYPE_VIRTIO_PCI_BUS,
2497 .parent = TYPE_VIRTIO_BUS,
2498 .instance_size = sizeof(VirtioPCIBusState),
2499 .class_size = sizeof(VirtioPCIBusClass),
2500 .class_init = virtio_pci_bus_class_init,
2501 };
2502
2503 static void virtio_pci_register_types(void)
2504 {
2505 /* Base types: */
2506 type_register_static(&virtio_pci_bus_info);
2507 type_register_static(&virtio_pci_info);
2508 }
2509
2510 type_init(virtio_pci_register_types)
2511
QEmu