xref: /openbmc/qemu/hw/pci/pci.c (revision 932a8d1f)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu-common.h"
27 #include "qemu/datadir.h"
28 #include "qemu/units.h"
29 #include "hw/irq.h"
30 #include "hw/pci/pci.h"
31 #include "hw/pci/pci_bridge.h"
32 #include "hw/pci/pci_bus.h"
33 #include "hw/pci/pci_host.h"
34 #include "hw/qdev-properties.h"
35 #include "hw/qdev-properties-system.h"
36 #include "migration/qemu-file-types.h"
37 #include "migration/vmstate.h"
38 #include "monitor/monitor.h"
39 #include "net/net.h"
40 #include "sysemu/numa.h"
41 #include "sysemu/sysemu.h"
42 #include "hw/loader.h"
43 #include "qemu/error-report.h"
44 #include "qemu/range.h"
45 #include "trace.h"
46 #include "hw/pci/msi.h"
47 #include "hw/pci/msix.h"
48 #include "exec/address-spaces.h"
49 #include "hw/hotplug.h"
50 #include "hw/boards.h"
51 #include "qapi/error.h"
52 #include "qapi/qapi-commands-pci.h"
53 #include "qemu/cutils.h"
54 
55 //#define DEBUG_PCI
56 #ifdef DEBUG_PCI
57 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
58 #else
59 # define PCI_DPRINTF(format, ...)       do { } while (0)
60 #endif
61 
62 bool pci_available = true;
63 
64 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset(BusState *qbus);
68 
69 static Property pci_props[] = {
70     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
71     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
72     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
73     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
74     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
75                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
76     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
77                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
78     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
79                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
80     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
81                        failover_pair_id),
82     DEFINE_PROP_END_OF_LIST()
83 };
84 
85 static const VMStateDescription vmstate_pcibus = {
86     .name = "PCIBUS",
87     .version_id = 1,
88     .minimum_version_id = 1,
89     .fields = (VMStateField[]) {
90         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
91         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
92                              nirq, 0, vmstate_info_int32,
93                              int32_t),
94         VMSTATE_END_OF_LIST()
95     }
96 };
97 
98 static void pci_init_bus_master(PCIDevice *pci_dev)
99 {
100     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
101 
102     memory_region_init_alias(&pci_dev->bus_master_enable_region,
103                              OBJECT(pci_dev), "bus master",
104                              dma_as->root, 0, memory_region_size(dma_as->root));
105     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
106     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
107                                 &pci_dev->bus_master_enable_region);
108 }
109 
110 static void pcibus_machine_done(Notifier *notifier, void *data)
111 {
112     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
113     int i;
114 
115     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
116         if (bus->devices[i]) {
117             pci_init_bus_master(bus->devices[i]);
118         }
119     }
120 }
121 
122 static void pci_bus_realize(BusState *qbus, Error **errp)
123 {
124     PCIBus *bus = PCI_BUS(qbus);
125 
126     bus->machine_done.notify = pcibus_machine_done;
127     qemu_add_machine_init_done_notifier(&bus->machine_done);
128 
129     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
130 }
131 
132 static void pcie_bus_realize(BusState *qbus, Error **errp)
133 {
134     PCIBus *bus = PCI_BUS(qbus);
135 
136     pci_bus_realize(qbus, errp);
137 
138     /*
139      * A PCI-E bus can support extended config space if it's the root
140      * bus, or if the bus/bridge above it does as well
141      */
142     if (pci_bus_is_root(bus)) {
143         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
144     } else {
145         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
146 
147         if (pci_bus_allows_extended_config_space(parent_bus)) {
148             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
149         }
150     }
151 }
152 
153 static void pci_bus_unrealize(BusState *qbus)
154 {
155     PCIBus *bus = PCI_BUS(qbus);
156 
157     qemu_remove_machine_init_done_notifier(&bus->machine_done);
158 
159     vmstate_unregister(NULL, &vmstate_pcibus, bus);
160 }
161 
162 static int pcibus_num(PCIBus *bus)
163 {
164     if (pci_bus_is_root(bus)) {
165         return 0; /* pci host bridge */
166     }
167     return bus->parent_dev->config[PCI_SECONDARY_BUS];
168 }
169 
170 static uint16_t pcibus_numa_node(PCIBus *bus)
171 {
172     return NUMA_NODE_UNASSIGNED;
173 }
174 
175 static void pci_bus_class_init(ObjectClass *klass, void *data)
176 {
177     BusClass *k = BUS_CLASS(klass);
178     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
179 
180     k->print_dev = pcibus_dev_print;
181     k->get_dev_path = pcibus_get_dev_path;
182     k->get_fw_dev_path = pcibus_get_fw_dev_path;
183     k->realize = pci_bus_realize;
184     k->unrealize = pci_bus_unrealize;
185     k->reset = pcibus_reset;
186 
187     pbc->bus_num = pcibus_num;
188     pbc->numa_node = pcibus_numa_node;
189 }
190 
191 static const TypeInfo pci_bus_info = {
192     .name = TYPE_PCI_BUS,
193     .parent = TYPE_BUS,
194     .instance_size = sizeof(PCIBus),
195     .class_size = sizeof(PCIBusClass),
196     .class_init = pci_bus_class_init,
197 };
198 
199 static const TypeInfo pcie_interface_info = {
200     .name          = INTERFACE_PCIE_DEVICE,
201     .parent        = TYPE_INTERFACE,
202 };
203 
204 static const TypeInfo conventional_pci_interface_info = {
205     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
206     .parent        = TYPE_INTERFACE,
207 };
208 
209 static void pcie_bus_class_init(ObjectClass *klass, void *data)
210 {
211     BusClass *k = BUS_CLASS(klass);
212 
213     k->realize = pcie_bus_realize;
214 }
215 
216 static const TypeInfo pcie_bus_info = {
217     .name = TYPE_PCIE_BUS,
218     .parent = TYPE_PCI_BUS,
219     .class_init = pcie_bus_class_init,
220 };
221 
222 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
223 static void pci_update_mappings(PCIDevice *d);
224 static void pci_irq_handler(void *opaque, int irq_num, int level);
225 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
226 static void pci_del_option_rom(PCIDevice *pdev);
227 
228 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
229 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
230 
231 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
232 
233 int pci_bar(PCIDevice *d, int reg)
234 {
235     uint8_t type;
236 
237     if (reg != PCI_ROM_SLOT)
238         return PCI_BASE_ADDRESS_0 + reg * 4;
239 
240     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
241     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
242 }
243 
244 static inline int pci_irq_state(PCIDevice *d, int irq_num)
245 {
246         return (d->irq_state >> irq_num) & 0x1;
247 }
248 
249 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
250 {
251         d->irq_state &= ~(0x1 << irq_num);
252         d->irq_state |= level << irq_num;
253 }
254 
255 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
256 {
257     assert(irq_num >= 0);
258     assert(irq_num < bus->nirq);
259     bus->irq_count[irq_num] += change;
260     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
261 }
262 
263 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
264 {
265     PCIBus *bus;
266     for (;;) {
267         bus = pci_get_bus(pci_dev);
268         irq_num = bus->map_irq(pci_dev, irq_num);
269         if (bus->set_irq)
270             break;
271         pci_dev = bus->parent_dev;
272     }
273     pci_bus_change_irq_level(bus, irq_num, change);
274 }
275 
276 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
277 {
278     assert(irq_num >= 0);
279     assert(irq_num < bus->nirq);
280     return !!bus->irq_count[irq_num];
281 }
282 
283 /* Update interrupt status bit in config space on interrupt
284  * state change. */
285 static void pci_update_irq_status(PCIDevice *dev)
286 {
287     if (dev->irq_state) {
288         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
289     } else {
290         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
291     }
292 }
293 
294 void pci_device_deassert_intx(PCIDevice *dev)
295 {
296     int i;
297     for (i = 0; i < PCI_NUM_PINS; ++i) {
298         pci_irq_handler(dev, i, 0);
299     }
300 }
301 
302 static void pci_do_device_reset(PCIDevice *dev)
303 {
304     int r;
305 
306     pci_device_deassert_intx(dev);
307     assert(dev->irq_state == 0);
308 
309     /* Clear all writable bits */
310     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
311                                  pci_get_word(dev->wmask + PCI_COMMAND) |
312                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
313     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
314                                  pci_get_word(dev->wmask + PCI_STATUS) |
315                                  pci_get_word(dev->w1cmask + PCI_STATUS));
316     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
317     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
318                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
319                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
320     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
321     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
322         PCIIORegion *region = &dev->io_regions[r];
323         if (!region->size) {
324             continue;
325         }
326 
327         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
328             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
329             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
330         } else {
331             pci_set_long(dev->config + pci_bar(dev, r), region->type);
332         }
333     }
334     pci_update_mappings(dev);
335 
336     msi_reset(dev);
337     msix_reset(dev);
338 }
339 
340 /*
341  * This function is called on #RST and FLR.
342  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
343  */
344 void pci_device_reset(PCIDevice *dev)
345 {
346     qdev_reset_all(&dev->qdev);
347     pci_do_device_reset(dev);
348 }
349 
350 /*
351  * Trigger pci bus reset under a given bus.
352  * Called via qbus_reset_all on RST# assert, after the devices
353  * have been reset qdev_reset_all-ed already.
354  */
355 static void pcibus_reset(BusState *qbus)
356 {
357     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
358     int i;
359 
360     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
361         if (bus->devices[i]) {
362             pci_do_device_reset(bus->devices[i]);
363         }
364     }
365 
366     for (i = 0; i < bus->nirq; i++) {
367         assert(bus->irq_count[i] == 0);
368     }
369 }
370 
371 static void pci_host_bus_register(DeviceState *host)
372 {
373     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
374 
375     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
376 }
377 
378 static void pci_host_bus_unregister(DeviceState *host)
379 {
380     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
381 
382     QLIST_REMOVE(host_bridge, next);
383 }
384 
385 PCIBus *pci_device_root_bus(const PCIDevice *d)
386 {
387     PCIBus *bus = pci_get_bus(d);
388 
389     while (!pci_bus_is_root(bus)) {
390         d = bus->parent_dev;
391         assert(d != NULL);
392 
393         bus = pci_get_bus(d);
394     }
395 
396     return bus;
397 }
398 
399 const char *pci_root_bus_path(PCIDevice *dev)
400 {
401     PCIBus *rootbus = pci_device_root_bus(dev);
402     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
403     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
404 
405     assert(host_bridge->bus == rootbus);
406 
407     if (hc->root_bus_path) {
408         return (*hc->root_bus_path)(host_bridge, rootbus);
409     }
410 
411     return rootbus->qbus.name;
412 }
413 
414 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
415                               MemoryRegion *address_space_mem,
416                               MemoryRegion *address_space_io,
417                               uint8_t devfn_min)
418 {
419     assert(PCI_FUNC(devfn_min) == 0);
420     bus->devfn_min = devfn_min;
421     bus->slot_reserved_mask = 0x0;
422     bus->address_space_mem = address_space_mem;
423     bus->address_space_io = address_space_io;
424     bus->flags |= PCI_BUS_IS_ROOT;
425 
426     /* host bridge */
427     QLIST_INIT(&bus->child);
428 
429     pci_host_bus_register(parent);
430 }
431 
432 static void pci_bus_uninit(PCIBus *bus)
433 {
434     pci_host_bus_unregister(BUS(bus)->parent);
435 }
436 
437 bool pci_bus_is_express(PCIBus *bus)
438 {
439     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
440 }
441 
442 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
443                               const char *name,
444                               MemoryRegion *address_space_mem,
445                               MemoryRegion *address_space_io,
446                               uint8_t devfn_min, const char *typename)
447 {
448     qbus_create_inplace(bus, bus_size, typename, parent, name);
449     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
450                       devfn_min);
451 }
452 
453 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
454                          MemoryRegion *address_space_mem,
455                          MemoryRegion *address_space_io,
456                          uint8_t devfn_min, const char *typename)
457 {
458     PCIBus *bus;
459 
460     bus = PCI_BUS(qbus_create(typename, parent, name));
461     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
462                       devfn_min);
463     return bus;
464 }
465 
466 void pci_root_bus_cleanup(PCIBus *bus)
467 {
468     pci_bus_uninit(bus);
469     /* the caller of the unplug hotplug handler will delete this device */
470     qbus_unrealize(BUS(bus));
471 }
472 
473 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
474                   void *irq_opaque, int nirq)
475 {
476     bus->set_irq = set_irq;
477     bus->map_irq = map_irq;
478     bus->irq_opaque = irq_opaque;
479     bus->nirq = nirq;
480     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
481 }
482 
483 void pci_bus_irqs_cleanup(PCIBus *bus)
484 {
485     bus->set_irq = NULL;
486     bus->map_irq = NULL;
487     bus->irq_opaque = NULL;
488     bus->nirq = 0;
489     g_free(bus->irq_count);
490 }
491 
492 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
493                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
494                               void *irq_opaque,
495                               MemoryRegion *address_space_mem,
496                               MemoryRegion *address_space_io,
497                               uint8_t devfn_min, int nirq,
498                               const char *typename)
499 {
500     PCIBus *bus;
501 
502     bus = pci_root_bus_new(parent, name, address_space_mem,
503                            address_space_io, devfn_min, typename);
504     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
505     return bus;
506 }
507 
508 void pci_unregister_root_bus(PCIBus *bus)
509 {
510     pci_bus_irqs_cleanup(bus);
511     pci_root_bus_cleanup(bus);
512 }
513 
514 int pci_bus_num(PCIBus *s)
515 {
516     return PCI_BUS_GET_CLASS(s)->bus_num(s);
517 }
518 
519 int pci_bus_numa_node(PCIBus *bus)
520 {
521     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
522 }
523 
524 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
525                                  const VMStateField *field)
526 {
527     PCIDevice *s = container_of(pv, PCIDevice, config);
528     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
529     uint8_t *config;
530     int i;
531 
532     assert(size == pci_config_size(s));
533     config = g_malloc(size);
534 
535     qemu_get_buffer(f, config, size);
536     for (i = 0; i < size; ++i) {
537         if ((config[i] ^ s->config[i]) &
538             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
539             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
540                          "cmask: %x wmask: %x w1cmask:%x", __func__,
541                          i, config[i], s->config[i],
542                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
543             g_free(config);
544             return -EINVAL;
545         }
546     }
547     memcpy(s->config, config, size);
548 
549     pci_update_mappings(s);
550     if (pc->is_bridge) {
551         PCIBridge *b = PCI_BRIDGE(s);
552         pci_bridge_update_mappings(b);
553     }
554 
555     memory_region_set_enabled(&s->bus_master_enable_region,
556                               pci_get_word(s->config + PCI_COMMAND)
557                               & PCI_COMMAND_MASTER);
558 
559     g_free(config);
560     return 0;
561 }
562 
563 /* just put buffer */
564 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
565                                  const VMStateField *field, JSONWriter *vmdesc)
566 {
567     const uint8_t **v = pv;
568     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
569     qemu_put_buffer(f, *v, size);
570 
571     return 0;
572 }
573 
574 static VMStateInfo vmstate_info_pci_config = {
575     .name = "pci config",
576     .get  = get_pci_config_device,
577     .put  = put_pci_config_device,
578 };
579 
580 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
581                              const VMStateField *field)
582 {
583     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
584     uint32_t irq_state[PCI_NUM_PINS];
585     int i;
586     for (i = 0; i < PCI_NUM_PINS; ++i) {
587         irq_state[i] = qemu_get_be32(f);
588         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
589             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
590                     irq_state[i]);
591             return -EINVAL;
592         }
593     }
594 
595     for (i = 0; i < PCI_NUM_PINS; ++i) {
596         pci_set_irq_state(s, i, irq_state[i]);
597     }
598 
599     return 0;
600 }
601 
602 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
603                              const VMStateField *field, JSONWriter *vmdesc)
604 {
605     int i;
606     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
607 
608     for (i = 0; i < PCI_NUM_PINS; ++i) {
609         qemu_put_be32(f, pci_irq_state(s, i));
610     }
611 
612     return 0;
613 }
614 
615 static VMStateInfo vmstate_info_pci_irq_state = {
616     .name = "pci irq state",
617     .get  = get_pci_irq_state,
618     .put  = put_pci_irq_state,
619 };
620 
621 static bool migrate_is_pcie(void *opaque, int version_id)
622 {
623     return pci_is_express((PCIDevice *)opaque);
624 }
625 
626 static bool migrate_is_not_pcie(void *opaque, int version_id)
627 {
628     return !pci_is_express((PCIDevice *)opaque);
629 }
630 
631 const VMStateDescription vmstate_pci_device = {
632     .name = "PCIDevice",
633     .version_id = 2,
634     .minimum_version_id = 1,
635     .fields = (VMStateField[]) {
636         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
637         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
638                                    migrate_is_not_pcie,
639                                    0, vmstate_info_pci_config,
640                                    PCI_CONFIG_SPACE_SIZE),
641         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
642                                    migrate_is_pcie,
643                                    0, vmstate_info_pci_config,
644                                    PCIE_CONFIG_SPACE_SIZE),
645         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
646                                    vmstate_info_pci_irq_state,
647                                    PCI_NUM_PINS * sizeof(int32_t)),
648         VMSTATE_END_OF_LIST()
649     }
650 };
651 
652 
653 void pci_device_save(PCIDevice *s, QEMUFile *f)
654 {
655     /* Clear interrupt status bit: it is implicit
656      * in irq_state which we are saving.
657      * This makes us compatible with old devices
658      * which never set or clear this bit. */
659     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
660     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
661     /* Restore the interrupt status bit. */
662     pci_update_irq_status(s);
663 }
664 
665 int pci_device_load(PCIDevice *s, QEMUFile *f)
666 {
667     int ret;
668     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
669     /* Restore the interrupt status bit. */
670     pci_update_irq_status(s);
671     return ret;
672 }
673 
674 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
675 {
676     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
677                  pci_default_sub_vendor_id);
678     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
679                  pci_default_sub_device_id);
680 }
681 
682 /*
683  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
684  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
685  */
686 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
687                              unsigned int *slotp, unsigned int *funcp)
688 {
689     const char *p;
690     char *e;
691     unsigned long val;
692     unsigned long dom = 0, bus = 0;
693     unsigned int slot = 0;
694     unsigned int func = 0;
695 
696     p = addr;
697     val = strtoul(p, &e, 16);
698     if (e == p)
699         return -1;
700     if (*e == ':') {
701         bus = val;
702         p = e + 1;
703         val = strtoul(p, &e, 16);
704         if (e == p)
705             return -1;
706         if (*e == ':') {
707             dom = bus;
708             bus = val;
709             p = e + 1;
710             val = strtoul(p, &e, 16);
711             if (e == p)
712                 return -1;
713         }
714     }
715 
716     slot = val;
717 
718     if (funcp != NULL) {
719         if (*e != '.')
720             return -1;
721 
722         p = e + 1;
723         val = strtoul(p, &e, 16);
724         if (e == p)
725             return -1;
726 
727         func = val;
728     }
729 
730     /* if funcp == NULL func is 0 */
731     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
732         return -1;
733 
734     if (*e)
735         return -1;
736 
737     *domp = dom;
738     *busp = bus;
739     *slotp = slot;
740     if (funcp != NULL)
741         *funcp = func;
742     return 0;
743 }
744 
745 static void pci_init_cmask(PCIDevice *dev)
746 {
747     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
748     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
749     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
750     dev->cmask[PCI_REVISION_ID] = 0xff;
751     dev->cmask[PCI_CLASS_PROG] = 0xff;
752     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
753     dev->cmask[PCI_HEADER_TYPE] = 0xff;
754     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
755 }
756 
757 static void pci_init_wmask(PCIDevice *dev)
758 {
759     int config_size = pci_config_size(dev);
760 
761     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
762     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
763     pci_set_word(dev->wmask + PCI_COMMAND,
764                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
765                  PCI_COMMAND_INTX_DISABLE);
766     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
767 
768     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
769            config_size - PCI_CONFIG_HEADER_SIZE);
770 }
771 
772 static void pci_init_w1cmask(PCIDevice *dev)
773 {
774     /*
775      * Note: It's okay to set w1cmask even for readonly bits as
776      * long as their value is hardwired to 0.
777      */
778     pci_set_word(dev->w1cmask + PCI_STATUS,
779                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
780                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
781                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
782 }
783 
784 static void pci_init_mask_bridge(PCIDevice *d)
785 {
786     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
787        PCI_SEC_LETENCY_TIMER */
788     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
789 
790     /* base and limit */
791     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
792     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
793     pci_set_word(d->wmask + PCI_MEMORY_BASE,
794                  PCI_MEMORY_RANGE_MASK & 0xffff);
795     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
796                  PCI_MEMORY_RANGE_MASK & 0xffff);
797     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
798                  PCI_PREF_RANGE_MASK & 0xffff);
799     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
800                  PCI_PREF_RANGE_MASK & 0xffff);
801 
802     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
803     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
804 
805     /* Supported memory and i/o types */
806     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
807     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
808     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
809                                PCI_PREF_RANGE_TYPE_64);
810     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
811                                PCI_PREF_RANGE_TYPE_64);
812 
813     /*
814      * TODO: Bridges default to 10-bit VGA decoding but we currently only
815      * implement 16-bit decoding (no alias support).
816      */
817     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
818                  PCI_BRIDGE_CTL_PARITY |
819                  PCI_BRIDGE_CTL_SERR |
820                  PCI_BRIDGE_CTL_ISA |
821                  PCI_BRIDGE_CTL_VGA |
822                  PCI_BRIDGE_CTL_VGA_16BIT |
823                  PCI_BRIDGE_CTL_MASTER_ABORT |
824                  PCI_BRIDGE_CTL_BUS_RESET |
825                  PCI_BRIDGE_CTL_FAST_BACK |
826                  PCI_BRIDGE_CTL_DISCARD |
827                  PCI_BRIDGE_CTL_SEC_DISCARD |
828                  PCI_BRIDGE_CTL_DISCARD_SERR);
829     /* Below does not do anything as we never set this bit, put here for
830      * completeness. */
831     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
832                  PCI_BRIDGE_CTL_DISCARD_STATUS);
833     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
834     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
835     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
836                                PCI_PREF_RANGE_TYPE_MASK);
837     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
838                                PCI_PREF_RANGE_TYPE_MASK);
839 }
840 
841 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
842 {
843     uint8_t slot = PCI_SLOT(dev->devfn);
844     uint8_t func;
845 
846     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
847         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
848     }
849 
850     /*
851      * multifunction bit is interpreted in two ways as follows.
852      *   - all functions must set the bit to 1.
853      *     Example: Intel X53
854      *   - function 0 must set the bit, but the rest function (> 0)
855      *     is allowed to leave the bit to 0.
856      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
857      *
858      * So OS (at least Linux) checks the bit of only function 0,
859      * and doesn't see the bit of function > 0.
860      *
861      * The below check allows both interpretation.
862      */
863     if (PCI_FUNC(dev->devfn)) {
864         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
865         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
866             /* function 0 should set multifunction bit */
867             error_setg(errp, "PCI: single function device can't be populated "
868                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
869             return;
870         }
871         return;
872     }
873 
874     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
875         return;
876     }
877     /* function 0 indicates single function, so function > 0 must be NULL */
878     for (func = 1; func < PCI_FUNC_MAX; ++func) {
879         if (bus->devices[PCI_DEVFN(slot, func)]) {
880             error_setg(errp, "PCI: %x.0 indicates single function, "
881                        "but %x.%x is already populated.",
882                        slot, slot, func);
883             return;
884         }
885     }
886 }
887 
888 static void pci_config_alloc(PCIDevice *pci_dev)
889 {
890     int config_size = pci_config_size(pci_dev);
891 
892     pci_dev->config = g_malloc0(config_size);
893     pci_dev->cmask = g_malloc0(config_size);
894     pci_dev->wmask = g_malloc0(config_size);
895     pci_dev->w1cmask = g_malloc0(config_size);
896     pci_dev->used = g_malloc0(config_size);
897 }
898 
899 static void pci_config_free(PCIDevice *pci_dev)
900 {
901     g_free(pci_dev->config);
902     g_free(pci_dev->cmask);
903     g_free(pci_dev->wmask);
904     g_free(pci_dev->w1cmask);
905     g_free(pci_dev->used);
906 }
907 
908 static void do_pci_unregister_device(PCIDevice *pci_dev)
909 {
910     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
911     pci_config_free(pci_dev);
912 
913     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
914         memory_region_del_subregion(&pci_dev->bus_master_container_region,
915                                     &pci_dev->bus_master_enable_region);
916     }
917     address_space_destroy(&pci_dev->bus_master_as);
918 }
919 
920 /* Extract PCIReqIDCache into BDF format */
921 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
922 {
923     uint8_t bus_n;
924     uint16_t result;
925 
926     switch (cache->type) {
927     case PCI_REQ_ID_BDF:
928         result = pci_get_bdf(cache->dev);
929         break;
930     case PCI_REQ_ID_SECONDARY_BUS:
931         bus_n = pci_dev_bus_num(cache->dev);
932         result = PCI_BUILD_BDF(bus_n, 0);
933         break;
934     default:
935         error_report("Invalid PCI requester ID cache type: %d",
936                      cache->type);
937         exit(1);
938         break;
939     }
940 
941     return result;
942 }
943 
944 /* Parse bridges up to the root complex and return requester ID
945  * cache for specific device.  For full PCIe topology, the cache
946  * result would be exactly the same as getting BDF of the device.
947  * However, several tricks are required when system mixed up with
948  * legacy PCI devices and PCIe-to-PCI bridges.
949  *
950  * Here we cache the proxy device (and type) not requester ID since
951  * bus number might change from time to time.
952  */
953 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
954 {
955     PCIDevice *parent;
956     PCIReqIDCache cache = {
957         .dev = dev,
958         .type = PCI_REQ_ID_BDF,
959     };
960 
961     while (!pci_bus_is_root(pci_get_bus(dev))) {
962         /* We are under PCI/PCIe bridges */
963         parent = pci_get_bus(dev)->parent_dev;
964         if (pci_is_express(parent)) {
965             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
966                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
967                  * override the requester ID using secondary bus
968                  * number of parent bridge with zeroed devfn
969                  * (pcie-to-pci bridge spec chap 2.3). */
970                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
971                 cache.dev = dev;
972             }
973         } else {
974             /* Legacy PCI, override requester ID with the bridge's
975              * BDF upstream.  When the root complex connects to
976              * legacy PCI devices (including buses), it can only
977              * obtain requester ID info from directly attached
978              * devices.  If devices are attached under bridges, only
979              * the requester ID of the bridge that is directly
980              * attached to the root complex can be recognized. */
981             cache.type = PCI_REQ_ID_BDF;
982             cache.dev = parent;
983         }
984         dev = parent;
985     }
986 
987     return cache;
988 }
989 
990 uint16_t pci_requester_id(PCIDevice *dev)
991 {
992     return pci_req_id_cache_extract(&dev->requester_id_cache);
993 }
994 
995 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
996 {
997     return !(bus->devices[devfn]);
998 }
999 
1000 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1001 {
1002     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1003 }
1004 
1005 /* -1 for devfn means auto assign */
1006 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1007                                          const char *name, int devfn,
1008                                          Error **errp)
1009 {
1010     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1011     PCIConfigReadFunc *config_read = pc->config_read;
1012     PCIConfigWriteFunc *config_write = pc->config_write;
1013     Error *local_err = NULL;
1014     DeviceState *dev = DEVICE(pci_dev);
1015     PCIBus *bus = pci_get_bus(pci_dev);
1016 
1017     /* Only pci bridges can be attached to extra PCI root buses */
1018     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1019         error_setg(errp,
1020                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1021                     bus->parent_dev->name);
1022         return NULL;
1023     }
1024 
1025     if (devfn < 0) {
1026         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1027             devfn += PCI_FUNC_MAX) {
1028             if (pci_bus_devfn_available(bus, devfn) &&
1029                    !pci_bus_devfn_reserved(bus, devfn)) {
1030                 goto found;
1031             }
1032         }
1033         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1034                    "or reserved", name);
1035         return NULL;
1036     found: ;
1037     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1038         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1039                    " reserved",
1040                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1041         return NULL;
1042     } else if (!pci_bus_devfn_available(bus, devfn)) {
1043         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1044                    " in use by %s",
1045                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1046                    bus->devices[devfn]->name);
1047         return NULL;
1048     } else if (dev->hotplugged &&
1049                pci_get_function_0(pci_dev)) {
1050         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1051                    " new func %s cannot be exposed to guest.",
1052                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1053                    pci_get_function_0(pci_dev)->name,
1054                    name);
1055 
1056        return NULL;
1057     }
1058 
1059     pci_dev->devfn = devfn;
1060     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1061     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1062 
1063     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1064                        "bus master container", UINT64_MAX);
1065     address_space_init(&pci_dev->bus_master_as,
1066                        &pci_dev->bus_master_container_region, pci_dev->name);
1067 
1068     if (phase_check(PHASE_MACHINE_READY)) {
1069         pci_init_bus_master(pci_dev);
1070     }
1071     pci_dev->irq_state = 0;
1072     pci_config_alloc(pci_dev);
1073 
1074     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1075     pci_config_set_device_id(pci_dev->config, pc->device_id);
1076     pci_config_set_revision(pci_dev->config, pc->revision);
1077     pci_config_set_class(pci_dev->config, pc->class_id);
1078 
1079     if (!pc->is_bridge) {
1080         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1081             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1082                          pc->subsystem_vendor_id);
1083             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1084                          pc->subsystem_id);
1085         } else {
1086             pci_set_default_subsystem_id(pci_dev);
1087         }
1088     } else {
1089         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1090         assert(!pc->subsystem_vendor_id);
1091         assert(!pc->subsystem_id);
1092     }
1093     pci_init_cmask(pci_dev);
1094     pci_init_wmask(pci_dev);
1095     pci_init_w1cmask(pci_dev);
1096     if (pc->is_bridge) {
1097         pci_init_mask_bridge(pci_dev);
1098     }
1099     pci_init_multifunction(bus, pci_dev, &local_err);
1100     if (local_err) {
1101         error_propagate(errp, local_err);
1102         do_pci_unregister_device(pci_dev);
1103         return NULL;
1104     }
1105 
1106     if (!config_read)
1107         config_read = pci_default_read_config;
1108     if (!config_write)
1109         config_write = pci_default_write_config;
1110     pci_dev->config_read = config_read;
1111     pci_dev->config_write = config_write;
1112     bus->devices[devfn] = pci_dev;
1113     pci_dev->version_id = 2; /* Current pci device vmstate version */
1114     return pci_dev;
1115 }
1116 
1117 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1118 {
1119     PCIIORegion *r;
1120     int i;
1121 
1122     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1123         r = &pci_dev->io_regions[i];
1124         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1125             continue;
1126         memory_region_del_subregion(r->address_space, r->memory);
1127     }
1128 
1129     pci_unregister_vga(pci_dev);
1130 }
1131 
1132 static void pci_qdev_unrealize(DeviceState *dev)
1133 {
1134     PCIDevice *pci_dev = PCI_DEVICE(dev);
1135     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1136 
1137     pci_unregister_io_regions(pci_dev);
1138     pci_del_option_rom(pci_dev);
1139 
1140     if (pc->exit) {
1141         pc->exit(pci_dev);
1142     }
1143 
1144     pci_device_deassert_intx(pci_dev);
1145     do_pci_unregister_device(pci_dev);
1146 }
1147 
1148 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1149                       uint8_t type, MemoryRegion *memory)
1150 {
1151     PCIIORegion *r;
1152     uint32_t addr; /* offset in pci config space */
1153     uint64_t wmask;
1154     pcibus_t size = memory_region_size(memory);
1155     uint8_t hdr_type;
1156 
1157     assert(region_num >= 0);
1158     assert(region_num < PCI_NUM_REGIONS);
1159     assert(is_power_of_2(size));
1160 
1161     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1162     hdr_type =
1163         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1164     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1165 
1166     r = &pci_dev->io_regions[region_num];
1167     r->addr = PCI_BAR_UNMAPPED;
1168     r->size = size;
1169     r->type = type;
1170     r->memory = memory;
1171     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1172                         ? pci_get_bus(pci_dev)->address_space_io
1173                         : pci_get_bus(pci_dev)->address_space_mem;
1174 
1175     wmask = ~(size - 1);
1176     if (region_num == PCI_ROM_SLOT) {
1177         /* ROM enable bit is writable */
1178         wmask |= PCI_ROM_ADDRESS_ENABLE;
1179     }
1180 
1181     addr = pci_bar(pci_dev, region_num);
1182     pci_set_long(pci_dev->config + addr, type);
1183 
1184     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1185         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1186         pci_set_quad(pci_dev->wmask + addr, wmask);
1187         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1188     } else {
1189         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1190         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1191     }
1192 }
1193 
1194 static void pci_update_vga(PCIDevice *pci_dev)
1195 {
1196     uint16_t cmd;
1197 
1198     if (!pci_dev->has_vga) {
1199         return;
1200     }
1201 
1202     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1203 
1204     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1205                               cmd & PCI_COMMAND_MEMORY);
1206     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1207                               cmd & PCI_COMMAND_IO);
1208     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1209                               cmd & PCI_COMMAND_IO);
1210 }
1211 
1212 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1213                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1214 {
1215     PCIBus *bus = pci_get_bus(pci_dev);
1216 
1217     assert(!pci_dev->has_vga);
1218 
1219     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1220     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1221     memory_region_add_subregion_overlap(bus->address_space_mem,
1222                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1223 
1224     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1225     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1226     memory_region_add_subregion_overlap(bus->address_space_io,
1227                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1228 
1229     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1230     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1231     memory_region_add_subregion_overlap(bus->address_space_io,
1232                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1233     pci_dev->has_vga = true;
1234 
1235     pci_update_vga(pci_dev);
1236 }
1237 
1238 void pci_unregister_vga(PCIDevice *pci_dev)
1239 {
1240     PCIBus *bus = pci_get_bus(pci_dev);
1241 
1242     if (!pci_dev->has_vga) {
1243         return;
1244     }
1245 
1246     memory_region_del_subregion(bus->address_space_mem,
1247                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1248     memory_region_del_subregion(bus->address_space_io,
1249                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1250     memory_region_del_subregion(bus->address_space_io,
1251                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1252     pci_dev->has_vga = false;
1253 }
1254 
1255 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1256 {
1257     return pci_dev->io_regions[region_num].addr;
1258 }
1259 
1260 static pcibus_t pci_bar_address(PCIDevice *d,
1261                                 int reg, uint8_t type, pcibus_t size)
1262 {
1263     pcibus_t new_addr, last_addr;
1264     int bar = pci_bar(d, reg);
1265     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1266     Object *machine = qdev_get_machine();
1267     ObjectClass *oc = object_get_class(machine);
1268     MachineClass *mc = MACHINE_CLASS(oc);
1269     bool allow_0_address = mc->pci_allow_0_address;
1270 
1271     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1272         if (!(cmd & PCI_COMMAND_IO)) {
1273             return PCI_BAR_UNMAPPED;
1274         }
1275         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1276         last_addr = new_addr + size - 1;
1277         /* Check if 32 bit BAR wraps around explicitly.
1278          * TODO: make priorities correct and remove this work around.
1279          */
1280         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1281             (!allow_0_address && new_addr == 0)) {
1282             return PCI_BAR_UNMAPPED;
1283         }
1284         return new_addr;
1285     }
1286 
1287     if (!(cmd & PCI_COMMAND_MEMORY)) {
1288         return PCI_BAR_UNMAPPED;
1289     }
1290     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1291         new_addr = pci_get_quad(d->config + bar);
1292     } else {
1293         new_addr = pci_get_long(d->config + bar);
1294     }
1295     /* the ROM slot has a specific enable bit */
1296     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1297         return PCI_BAR_UNMAPPED;
1298     }
1299     new_addr &= ~(size - 1);
1300     last_addr = new_addr + size - 1;
1301     /* NOTE: we do not support wrapping */
1302     /* XXX: as we cannot support really dynamic
1303        mappings, we handle specific values as invalid
1304        mappings. */
1305     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1306         (!allow_0_address && new_addr == 0)) {
1307         return PCI_BAR_UNMAPPED;
1308     }
1309 
1310     /* Now pcibus_t is 64bit.
1311      * Check if 32 bit BAR wraps around explicitly.
1312      * Without this, PC ide doesn't work well.
1313      * TODO: remove this work around.
1314      */
1315     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1316         return PCI_BAR_UNMAPPED;
1317     }
1318 
1319     /*
1320      * OS is allowed to set BAR beyond its addressable
1321      * bits. For example, 32 bit OS can set 64bit bar
1322      * to >4G. Check it. TODO: we might need to support
1323      * it in the future for e.g. PAE.
1324      */
1325     if (last_addr >= HWADDR_MAX) {
1326         return PCI_BAR_UNMAPPED;
1327     }
1328 
1329     return new_addr;
1330 }
1331 
1332 static void pci_update_mappings(PCIDevice *d)
1333 {
1334     PCIIORegion *r;
1335     int i;
1336     pcibus_t new_addr;
1337 
1338     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1339         r = &d->io_regions[i];
1340 
1341         /* this region isn't registered */
1342         if (!r->size)
1343             continue;
1344 
1345         new_addr = pci_bar_address(d, i, r->type, r->size);
1346 
1347         /* This bar isn't changed */
1348         if (new_addr == r->addr)
1349             continue;
1350 
1351         /* now do the real mapping */
1352         if (r->addr != PCI_BAR_UNMAPPED) {
1353             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1354                                           PCI_SLOT(d->devfn),
1355                                           PCI_FUNC(d->devfn),
1356                                           i, r->addr, r->size);
1357             memory_region_del_subregion(r->address_space, r->memory);
1358         }
1359         r->addr = new_addr;
1360         if (r->addr != PCI_BAR_UNMAPPED) {
1361             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1362                                           PCI_SLOT(d->devfn),
1363                                           PCI_FUNC(d->devfn),
1364                                           i, r->addr, r->size);
1365             memory_region_add_subregion_overlap(r->address_space,
1366                                                 r->addr, r->memory, 1);
1367         }
1368     }
1369 
1370     pci_update_vga(d);
1371 }
1372 
1373 static inline int pci_irq_disabled(PCIDevice *d)
1374 {
1375     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1376 }
1377 
1378 /* Called after interrupt disabled field update in config space,
1379  * assert/deassert interrupts if necessary.
1380  * Gets original interrupt disable bit value (before update). */
1381 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1382 {
1383     int i, disabled = pci_irq_disabled(d);
1384     if (disabled == was_irq_disabled)
1385         return;
1386     for (i = 0; i < PCI_NUM_PINS; ++i) {
1387         int state = pci_irq_state(d, i);
1388         pci_change_irq_level(d, i, disabled ? -state : state);
1389     }
1390 }
1391 
1392 uint32_t pci_default_read_config(PCIDevice *d,
1393                                  uint32_t address, int len)
1394 {
1395     uint32_t val = 0;
1396 
1397     assert(address + len <= pci_config_size(d));
1398 
1399     if (pci_is_express_downstream_port(d) &&
1400         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1401         pcie_sync_bridge_lnk(d);
1402     }
1403     memcpy(&val, d->config + address, len);
1404     return le32_to_cpu(val);
1405 }
1406 
1407 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1408 {
1409     int i, was_irq_disabled = pci_irq_disabled(d);
1410     uint32_t val = val_in;
1411 
1412     assert(addr + l <= pci_config_size(d));
1413 
1414     for (i = 0; i < l; val >>= 8, ++i) {
1415         uint8_t wmask = d->wmask[addr + i];
1416         uint8_t w1cmask = d->w1cmask[addr + i];
1417         assert(!(wmask & w1cmask));
1418         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1419         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1420     }
1421     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1422         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1423         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1424         range_covers_byte(addr, l, PCI_COMMAND))
1425         pci_update_mappings(d);
1426 
1427     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1428         pci_update_irq_disabled(d, was_irq_disabled);
1429         memory_region_set_enabled(&d->bus_master_enable_region,
1430                                   pci_get_word(d->config + PCI_COMMAND)
1431                                     & PCI_COMMAND_MASTER);
1432     }
1433 
1434     msi_write_config(d, addr, val_in, l);
1435     msix_write_config(d, addr, val_in, l);
1436 }
1437 
1438 /***********************************************************/
1439 /* generic PCI irq support */
1440 
1441 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1442 static void pci_irq_handler(void *opaque, int irq_num, int level)
1443 {
1444     PCIDevice *pci_dev = opaque;
1445     int change;
1446 
1447     change = level - pci_irq_state(pci_dev, irq_num);
1448     if (!change)
1449         return;
1450 
1451     pci_set_irq_state(pci_dev, irq_num, level);
1452     pci_update_irq_status(pci_dev);
1453     if (pci_irq_disabled(pci_dev))
1454         return;
1455     pci_change_irq_level(pci_dev, irq_num, change);
1456 }
1457 
1458 static inline int pci_intx(PCIDevice *pci_dev)
1459 {
1460     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1461 }
1462 
1463 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1464 {
1465     int intx = pci_intx(pci_dev);
1466 
1467     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1468 }
1469 
1470 void pci_set_irq(PCIDevice *pci_dev, int level)
1471 {
1472     int intx = pci_intx(pci_dev);
1473     pci_irq_handler(pci_dev, intx, level);
1474 }
1475 
1476 /* Special hooks used by device assignment */
1477 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1478 {
1479     assert(pci_bus_is_root(bus));
1480     bus->route_intx_to_irq = route_intx_to_irq;
1481 }
1482 
1483 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1484 {
1485     PCIBus *bus;
1486 
1487     do {
1488         bus = pci_get_bus(dev);
1489         pin = bus->map_irq(dev, pin);
1490         dev = bus->parent_dev;
1491     } while (dev);
1492 
1493     if (!bus->route_intx_to_irq) {
1494         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1495                      object_get_typename(OBJECT(bus->qbus.parent)));
1496         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1497     }
1498 
1499     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1500 }
1501 
1502 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1503 {
1504     return old->mode != new->mode || old->irq != new->irq;
1505 }
1506 
1507 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1508 {
1509     PCIDevice *dev;
1510     PCIBus *sec;
1511     int i;
1512 
1513     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1514         dev = bus->devices[i];
1515         if (dev && dev->intx_routing_notifier) {
1516             dev->intx_routing_notifier(dev);
1517         }
1518     }
1519 
1520     QLIST_FOREACH(sec, &bus->child, sibling) {
1521         pci_bus_fire_intx_routing_notifier(sec);
1522     }
1523 }
1524 
1525 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1526                                           PCIINTxRoutingNotifier notifier)
1527 {
1528     dev->intx_routing_notifier = notifier;
1529 }
1530 
1531 /*
1532  * PCI-to-PCI bridge specification
1533  * 9.1: Interrupt routing. Table 9-1
1534  *
1535  * the PCI Express Base Specification, Revision 2.1
1536  * 2.2.8.1: INTx interrutp signaling - Rules
1537  *          the Implementation Note
1538  *          Table 2-20
1539  */
1540 /*
1541  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1542  * 0-origin unlike PCI interrupt pin register.
1543  */
1544 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1545 {
1546     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1547 }
1548 
1549 /***********************************************************/
1550 /* monitor info on PCI */
1551 
1552 typedef struct {
1553     uint16_t class;
1554     const char *desc;
1555     const char *fw_name;
1556     uint16_t fw_ign_bits;
1557 } pci_class_desc;
1558 
1559 static const pci_class_desc pci_class_descriptions[] =
1560 {
1561     { 0x0001, "VGA controller", "display"},
1562     { 0x0100, "SCSI controller", "scsi"},
1563     { 0x0101, "IDE controller", "ide"},
1564     { 0x0102, "Floppy controller", "fdc"},
1565     { 0x0103, "IPI controller", "ipi"},
1566     { 0x0104, "RAID controller", "raid"},
1567     { 0x0106, "SATA controller"},
1568     { 0x0107, "SAS controller"},
1569     { 0x0180, "Storage controller"},
1570     { 0x0200, "Ethernet controller", "ethernet"},
1571     { 0x0201, "Token Ring controller", "token-ring"},
1572     { 0x0202, "FDDI controller", "fddi"},
1573     { 0x0203, "ATM controller", "atm"},
1574     { 0x0280, "Network controller"},
1575     { 0x0300, "VGA controller", "display", 0x00ff},
1576     { 0x0301, "XGA controller"},
1577     { 0x0302, "3D controller"},
1578     { 0x0380, "Display controller"},
1579     { 0x0400, "Video controller", "video"},
1580     { 0x0401, "Audio controller", "sound"},
1581     { 0x0402, "Phone"},
1582     { 0x0403, "Audio controller", "sound"},
1583     { 0x0480, "Multimedia controller"},
1584     { 0x0500, "RAM controller", "memory"},
1585     { 0x0501, "Flash controller", "flash"},
1586     { 0x0580, "Memory controller"},
1587     { 0x0600, "Host bridge", "host"},
1588     { 0x0601, "ISA bridge", "isa"},
1589     { 0x0602, "EISA bridge", "eisa"},
1590     { 0x0603, "MC bridge", "mca"},
1591     { 0x0604, "PCI bridge", "pci-bridge"},
1592     { 0x0605, "PCMCIA bridge", "pcmcia"},
1593     { 0x0606, "NUBUS bridge", "nubus"},
1594     { 0x0607, "CARDBUS bridge", "cardbus"},
1595     { 0x0608, "RACEWAY bridge"},
1596     { 0x0680, "Bridge"},
1597     { 0x0700, "Serial port", "serial"},
1598     { 0x0701, "Parallel port", "parallel"},
1599     { 0x0800, "Interrupt controller", "interrupt-controller"},
1600     { 0x0801, "DMA controller", "dma-controller"},
1601     { 0x0802, "Timer", "timer"},
1602     { 0x0803, "RTC", "rtc"},
1603     { 0x0900, "Keyboard", "keyboard"},
1604     { 0x0901, "Pen", "pen"},
1605     { 0x0902, "Mouse", "mouse"},
1606     { 0x0A00, "Dock station", "dock", 0x00ff},
1607     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1608     { 0x0c00, "Fireware contorller", "fireware"},
1609     { 0x0c01, "Access bus controller", "access-bus"},
1610     { 0x0c02, "SSA controller", "ssa"},
1611     { 0x0c03, "USB controller", "usb"},
1612     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1613     { 0x0c05, "SMBus"},
1614     { 0, NULL}
1615 };
1616 
1617 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1618                                                   void (*fn)(PCIBus *b,
1619                                                              PCIDevice *d,
1620                                                              void *opaque),
1621                                                   void *opaque)
1622 {
1623     PCIDevice *d;
1624     int devfn;
1625 
1626     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1627         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1628         if (d) {
1629             fn(bus, d, opaque);
1630         }
1631     }
1632 }
1633 
1634 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1635                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1636                          void *opaque)
1637 {
1638     bus = pci_find_bus_nr(bus, bus_num);
1639 
1640     if (bus) {
1641         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1642     }
1643 }
1644 
1645 static void pci_for_each_device_under_bus(PCIBus *bus,
1646                                           void (*fn)(PCIBus *b, PCIDevice *d,
1647                                                      void *opaque),
1648                                           void *opaque)
1649 {
1650     PCIDevice *d;
1651     int devfn;
1652 
1653     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1654         d = bus->devices[devfn];
1655         if (d) {
1656             fn(bus, d, opaque);
1657         }
1658     }
1659 }
1660 
1661 void pci_for_each_device(PCIBus *bus, int bus_num,
1662                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1663                          void *opaque)
1664 {
1665     bus = pci_find_bus_nr(bus, bus_num);
1666 
1667     if (bus) {
1668         pci_for_each_device_under_bus(bus, fn, opaque);
1669     }
1670 }
1671 
1672 static const pci_class_desc *get_class_desc(int class)
1673 {
1674     const pci_class_desc *desc;
1675 
1676     desc = pci_class_descriptions;
1677     while (desc->desc && class != desc->class) {
1678         desc++;
1679     }
1680 
1681     return desc;
1682 }
1683 
1684 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1685 
1686 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1687 {
1688     PciMemoryRegionList *head = NULL, **tail = &head;
1689     int i;
1690 
1691     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1692         const PCIIORegion *r = &dev->io_regions[i];
1693         PciMemoryRegion *region;
1694 
1695         if (!r->size) {
1696             continue;
1697         }
1698 
1699         region = g_malloc0(sizeof(*region));
1700 
1701         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1702             region->type = g_strdup("io");
1703         } else {
1704             region->type = g_strdup("memory");
1705             region->has_prefetch = true;
1706             region->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1707             region->has_mem_type_64 = true;
1708             region->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1709         }
1710 
1711         region->bar = i;
1712         region->address = r->addr;
1713         region->size = r->size;
1714 
1715         QAPI_LIST_APPEND(tail, region);
1716     }
1717 
1718     return head;
1719 }
1720 
1721 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1722                                            int bus_num)
1723 {
1724     PciBridgeInfo *info;
1725     PciMemoryRange *range;
1726 
1727     info = g_new0(PciBridgeInfo, 1);
1728 
1729     info->bus = g_new0(PciBusInfo, 1);
1730     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1731     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1732     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1733 
1734     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1735     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1736     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1737 
1738     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1739     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1740     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1741 
1742     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1743     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1744     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1745 
1746     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1747         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1748         if (child_bus) {
1749             info->has_devices = true;
1750             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1751         }
1752     }
1753 
1754     return info;
1755 }
1756 
1757 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1758                                            int bus_num)
1759 {
1760     const pci_class_desc *desc;
1761     PciDeviceInfo *info;
1762     uint8_t type;
1763     int class;
1764 
1765     info = g_new0(PciDeviceInfo, 1);
1766     info->bus = bus_num;
1767     info->slot = PCI_SLOT(dev->devfn);
1768     info->function = PCI_FUNC(dev->devfn);
1769 
1770     info->class_info = g_new0(PciDeviceClass, 1);
1771     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1772     info->class_info->q_class = class;
1773     desc = get_class_desc(class);
1774     if (desc->desc) {
1775         info->class_info->has_desc = true;
1776         info->class_info->desc = g_strdup(desc->desc);
1777     }
1778 
1779     info->id = g_new0(PciDeviceId, 1);
1780     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1781     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1782     info->regions = qmp_query_pci_regions(dev);
1783     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1784 
1785     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1786     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1787         info->has_irq = true;
1788         info->irq = dev->config[PCI_INTERRUPT_LINE];
1789     }
1790 
1791     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1792     if (type == PCI_HEADER_TYPE_BRIDGE) {
1793         info->has_pci_bridge = true;
1794         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1795     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1796         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1797         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1798         info->id->subsystem_vendor =
1799             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1800     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1801         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1802         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1803         info->id->subsystem_vendor =
1804             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1805     }
1806 
1807     return info;
1808 }
1809 
1810 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1811 {
1812     PciDeviceInfoList *head = NULL, **tail = &head;
1813     PCIDevice *dev;
1814     int devfn;
1815 
1816     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1817         dev = bus->devices[devfn];
1818         if (dev) {
1819             QAPI_LIST_APPEND(tail, qmp_query_pci_device(dev, bus, bus_num));
1820         }
1821     }
1822 
1823     return head;
1824 }
1825 
1826 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1827 {
1828     PciInfo *info = NULL;
1829 
1830     bus = pci_find_bus_nr(bus, bus_num);
1831     if (bus) {
1832         info = g_malloc0(sizeof(*info));
1833         info->bus = bus_num;
1834         info->devices = qmp_query_pci_devices(bus, bus_num);
1835     }
1836 
1837     return info;
1838 }
1839 
1840 PciInfoList *qmp_query_pci(Error **errp)
1841 {
1842     PciInfoList *head = NULL, **tail = &head;
1843     PCIHostState *host_bridge;
1844 
1845     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1846         QAPI_LIST_APPEND(tail,
1847                          qmp_query_pci_bus(host_bridge->bus,
1848                                            pci_bus_num(host_bridge->bus)));
1849     }
1850 
1851     return head;
1852 }
1853 
1854 /* Initialize a PCI NIC.  */
1855 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1856                                const char *default_model,
1857                                const char *default_devaddr)
1858 {
1859     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1860     GSList *list;
1861     GPtrArray *pci_nic_models;
1862     PCIBus *bus;
1863     PCIDevice *pci_dev;
1864     DeviceState *dev;
1865     int devfn;
1866     int i;
1867     int dom, busnr;
1868     unsigned slot;
1869 
1870     if (nd->model && !strcmp(nd->model, "virtio")) {
1871         g_free(nd->model);
1872         nd->model = g_strdup("virtio-net-pci");
1873     }
1874 
1875     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1876     pci_nic_models = g_ptr_array_new();
1877     while (list) {
1878         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1879                                              TYPE_DEVICE);
1880         GSList *next;
1881         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1882             dc->user_creatable) {
1883             const char *name = object_class_get_name(list->data);
1884             /*
1885              * A network device might also be something else than a NIC, see
1886              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1887              * property, too. Unfortunately, some devices like virtio-net only
1888              * create this property during instance_init, so we have to create
1889              * a temporary instance here to be able to check it.
1890              */
1891             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1892             if (object_property_find(obj, "netdev")) {
1893                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1894             }
1895             object_unref(obj);
1896         }
1897         next = list->next;
1898         g_slist_free_1(list);
1899         list = next;
1900     }
1901     g_ptr_array_add(pci_nic_models, NULL);
1902 
1903     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1904         exit(0);
1905     }
1906 
1907     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1908                             default_model);
1909     if (i < 0) {
1910         exit(1);
1911     }
1912 
1913     if (!rootbus) {
1914         error_report("No primary PCI bus");
1915         exit(1);
1916     }
1917 
1918     assert(!rootbus->parent_dev);
1919 
1920     if (!devaddr) {
1921         devfn = -1;
1922         busnr = 0;
1923     } else {
1924         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1925             error_report("Invalid PCI device address %s for device %s",
1926                          devaddr, nd->model);
1927             exit(1);
1928         }
1929 
1930         if (dom != 0) {
1931             error_report("No support for non-zero PCI domains");
1932             exit(1);
1933         }
1934 
1935         devfn = PCI_DEVFN(slot, 0);
1936     }
1937 
1938     bus = pci_find_bus_nr(rootbus, busnr);
1939     if (!bus) {
1940         error_report("Invalid PCI device address %s for device %s",
1941                      devaddr, nd->model);
1942         exit(1);
1943     }
1944 
1945     pci_dev = pci_new(devfn, nd->model);
1946     dev = &pci_dev->qdev;
1947     qdev_set_nic_properties(dev, nd);
1948     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1949     g_ptr_array_free(pci_nic_models, true);
1950     return pci_dev;
1951 }
1952 
1953 PCIDevice *pci_vga_init(PCIBus *bus)
1954 {
1955     switch (vga_interface_type) {
1956     case VGA_CIRRUS:
1957         return pci_create_simple(bus, -1, "cirrus-vga");
1958     case VGA_QXL:
1959         return pci_create_simple(bus, -1, "qxl-vga");
1960     case VGA_STD:
1961         return pci_create_simple(bus, -1, "VGA");
1962     case VGA_VMWARE:
1963         return pci_create_simple(bus, -1, "vmware-svga");
1964     case VGA_VIRTIO:
1965         return pci_create_simple(bus, -1, "virtio-vga");
1966     case VGA_NONE:
1967     default: /* Other non-PCI types. Checking for unsupported types is already
1968                 done in vl.c. */
1969         return NULL;
1970     }
1971 }
1972 
1973 /* Whether a given bus number is in range of the secondary
1974  * bus of the given bridge device. */
1975 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1976 {
1977     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1978              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1979         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1980         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1981 }
1982 
1983 /* Whether a given bus number is in a range of a root bus */
1984 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1985 {
1986     int i;
1987 
1988     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1989         PCIDevice *dev = bus->devices[i];
1990 
1991         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1992             if (pci_secondary_bus_in_range(dev, bus_num)) {
1993                 return true;
1994             }
1995         }
1996     }
1997 
1998     return false;
1999 }
2000 
2001 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2002 {
2003     PCIBus *sec;
2004 
2005     if (!bus) {
2006         return NULL;
2007     }
2008 
2009     if (pci_bus_num(bus) == bus_num) {
2010         return bus;
2011     }
2012 
2013     /* Consider all bus numbers in range for the host pci bridge. */
2014     if (!pci_bus_is_root(bus) &&
2015         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2016         return NULL;
2017     }
2018 
2019     /* try child bus */
2020     for (; bus; bus = sec) {
2021         QLIST_FOREACH(sec, &bus->child, sibling) {
2022             if (pci_bus_num(sec) == bus_num) {
2023                 return sec;
2024             }
2025             /* PXB buses assumed to be children of bus 0 */
2026             if (pci_bus_is_root(sec)) {
2027                 if (pci_root_bus_in_range(sec, bus_num)) {
2028                     break;
2029                 }
2030             } else {
2031                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2032                     break;
2033                 }
2034             }
2035         }
2036     }
2037 
2038     return NULL;
2039 }
2040 
2041 void pci_for_each_bus_depth_first(PCIBus *bus,
2042                                   void *(*begin)(PCIBus *bus, void *parent_state),
2043                                   void (*end)(PCIBus *bus, void *state),
2044                                   void *parent_state)
2045 {
2046     PCIBus *sec;
2047     void *state;
2048 
2049     if (!bus) {
2050         return;
2051     }
2052 
2053     if (begin) {
2054         state = begin(bus, parent_state);
2055     } else {
2056         state = parent_state;
2057     }
2058 
2059     QLIST_FOREACH(sec, &bus->child, sibling) {
2060         pci_for_each_bus_depth_first(sec, begin, end, state);
2061     }
2062 
2063     if (end) {
2064         end(bus, state);
2065     }
2066 }
2067 
2068 
2069 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2070 {
2071     bus = pci_find_bus_nr(bus, bus_num);
2072 
2073     if (!bus)
2074         return NULL;
2075 
2076     return bus->devices[devfn];
2077 }
2078 
2079 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2080 {
2081     PCIDevice *pci_dev = (PCIDevice *)qdev;
2082     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2083     ObjectClass *klass = OBJECT_CLASS(pc);
2084     Error *local_err = NULL;
2085     bool is_default_rom;
2086     uint16_t class_id;
2087 
2088     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2089         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2090         return;
2091     }
2092 
2093     /* initialize cap_present for pci_is_express() and pci_config_size(),
2094      * Note that hybrid PCIs are not set automatically and need to manage
2095      * QEMU_PCI_CAP_EXPRESS manually */
2096     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2097        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2098         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2099     }
2100 
2101     pci_dev = do_pci_register_device(pci_dev,
2102                                      object_get_typename(OBJECT(qdev)),
2103                                      pci_dev->devfn, errp);
2104     if (pci_dev == NULL)
2105         return;
2106 
2107     if (pc->realize) {
2108         pc->realize(pci_dev, &local_err);
2109         if (local_err) {
2110             error_propagate(errp, local_err);
2111             do_pci_unregister_device(pci_dev);
2112             return;
2113         }
2114     }
2115 
2116     if (pci_dev->failover_pair_id) {
2117         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2118             error_setg(errp, "failover primary device must be on "
2119                              "PCIExpress bus");
2120             pci_qdev_unrealize(DEVICE(pci_dev));
2121             return;
2122         }
2123         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2124         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2125             error_setg(errp, "failover primary device is not an "
2126                              "Ethernet device");
2127             pci_qdev_unrealize(DEVICE(pci_dev));
2128             return;
2129         }
2130         if (!(pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2131             && (PCI_FUNC(pci_dev->devfn) == 0)) {
2132             qdev->allow_unplug_during_migration = true;
2133         } else {
2134             error_setg(errp, "failover: primary device must be in its own "
2135                               "PCI slot");
2136             pci_qdev_unrealize(DEVICE(pci_dev));
2137             return;
2138         }
2139         qdev->allow_unplug_during_migration = true;
2140     }
2141 
2142     /* rom loading */
2143     is_default_rom = false;
2144     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2145         pci_dev->romfile = g_strdup(pc->romfile);
2146         is_default_rom = true;
2147     }
2148 
2149     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2150     if (local_err) {
2151         error_propagate(errp, local_err);
2152         pci_qdev_unrealize(DEVICE(pci_dev));
2153         return;
2154     }
2155 }
2156 
2157 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2158                                  const char *name)
2159 {
2160     DeviceState *dev;
2161 
2162     dev = qdev_new(name);
2163     qdev_prop_set_int32(dev, "addr", devfn);
2164     qdev_prop_set_bit(dev, "multifunction", multifunction);
2165     return PCI_DEVICE(dev);
2166 }
2167 
2168 PCIDevice *pci_new(int devfn, const char *name)
2169 {
2170     return pci_new_multifunction(devfn, false, name);
2171 }
2172 
2173 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2174 {
2175     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2176 }
2177 
2178 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2179                                            bool multifunction,
2180                                            const char *name)
2181 {
2182     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2183     pci_realize_and_unref(dev, bus, &error_fatal);
2184     return dev;
2185 }
2186 
2187 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2188 {
2189     return pci_create_simple_multifunction(bus, devfn, false, name);
2190 }
2191 
2192 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2193 {
2194     int offset = PCI_CONFIG_HEADER_SIZE;
2195     int i;
2196     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2197         if (pdev->used[i])
2198             offset = i + 1;
2199         else if (i - offset + 1 == size)
2200             return offset;
2201     }
2202     return 0;
2203 }
2204 
2205 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2206                                         uint8_t *prev_p)
2207 {
2208     uint8_t next, prev;
2209 
2210     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2211         return 0;
2212 
2213     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2214          prev = next + PCI_CAP_LIST_NEXT)
2215         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2216             break;
2217 
2218     if (prev_p)
2219         *prev_p = prev;
2220     return next;
2221 }
2222 
2223 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2224 {
2225     uint8_t next, prev, found = 0;
2226 
2227     if (!(pdev->used[offset])) {
2228         return 0;
2229     }
2230 
2231     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2232 
2233     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2234          prev = next + PCI_CAP_LIST_NEXT) {
2235         if (next <= offset && next > found) {
2236             found = next;
2237         }
2238     }
2239     return found;
2240 }
2241 
2242 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2243    This is needed for an option rom which is used for more than one device. */
2244 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2245 {
2246     uint16_t vendor_id;
2247     uint16_t device_id;
2248     uint16_t rom_vendor_id;
2249     uint16_t rom_device_id;
2250     uint16_t rom_magic;
2251     uint16_t pcir_offset;
2252     uint8_t checksum;
2253 
2254     /* Words in rom data are little endian (like in PCI configuration),
2255        so they can be read / written with pci_get_word / pci_set_word. */
2256 
2257     /* Only a valid rom will be patched. */
2258     rom_magic = pci_get_word(ptr);
2259     if (rom_magic != 0xaa55) {
2260         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2261         return;
2262     }
2263     pcir_offset = pci_get_word(ptr + 0x18);
2264     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2265         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2266         return;
2267     }
2268 
2269     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2270     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2271     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2272     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2273 
2274     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2275                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2276 
2277     checksum = ptr[6];
2278 
2279     if (vendor_id != rom_vendor_id) {
2280         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2281         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2282         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2283         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2284         ptr[6] = checksum;
2285         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2286     }
2287 
2288     if (device_id != rom_device_id) {
2289         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2290         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2291         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2292         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2293         ptr[6] = checksum;
2294         pci_set_word(ptr + pcir_offset + 6, device_id);
2295     }
2296 }
2297 
2298 /* Add an option rom for the device */
2299 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2300                                Error **errp)
2301 {
2302     int64_t size;
2303     char *path;
2304     void *ptr;
2305     char name[32];
2306     const VMStateDescription *vmsd;
2307 
2308     if (!pdev->romfile)
2309         return;
2310     if (strlen(pdev->romfile) == 0)
2311         return;
2312 
2313     if (!pdev->rom_bar) {
2314         /*
2315          * Load rom via fw_cfg instead of creating a rom bar,
2316          * for 0.11 compatibility.
2317          */
2318         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2319 
2320         /*
2321          * Hot-plugged devices can't use the option ROM
2322          * if the rom bar is disabled.
2323          */
2324         if (DEVICE(pdev)->hotplugged) {
2325             error_setg(errp, "Hot-plugged device without ROM bar"
2326                        " can't have an option ROM");
2327             return;
2328         }
2329 
2330         if (class == 0x0300) {
2331             rom_add_vga(pdev->romfile);
2332         } else {
2333             rom_add_option(pdev->romfile, -1);
2334         }
2335         return;
2336     }
2337 
2338     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2339     if (path == NULL) {
2340         path = g_strdup(pdev->romfile);
2341     }
2342 
2343     size = get_image_size(path);
2344     if (size < 0) {
2345         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2346         g_free(path);
2347         return;
2348     } else if (size == 0) {
2349         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2350         g_free(path);
2351         return;
2352     } else if (size > 2 * GiB) {
2353         error_setg(errp, "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2354                    pdev->romfile);
2355         g_free(path);
2356         return;
2357     }
2358     if (pdev->romsize != -1) {
2359         if (size > pdev->romsize) {
2360             error_setg(errp, "romfile \"%s\" (%u bytes) is too large for ROM size %u",
2361                        pdev->romfile, (uint32_t)size, pdev->romsize);
2362             g_free(path);
2363             return;
2364         }
2365     } else {
2366         pdev->romsize = pow2ceil(size);
2367     }
2368 
2369     vmsd = qdev_get_vmsd(DEVICE(pdev));
2370 
2371     if (vmsd) {
2372         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2373     } else {
2374         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2375     }
2376     pdev->has_rom = true;
2377     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize, &error_fatal);
2378     ptr = memory_region_get_ram_ptr(&pdev->rom);
2379     if (load_image_size(path, ptr, size) < 0) {
2380         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2381         g_free(path);
2382         return;
2383     }
2384     g_free(path);
2385 
2386     if (is_default_rom) {
2387         /* Only the default rom images will be patched (if needed). */
2388         pci_patch_ids(pdev, ptr, size);
2389     }
2390 
2391     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2392 }
2393 
2394 static void pci_del_option_rom(PCIDevice *pdev)
2395 {
2396     if (!pdev->has_rom)
2397         return;
2398 
2399     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2400     pdev->has_rom = false;
2401 }
2402 
2403 /*
2404  * On success, pci_add_capability() returns a positive value
2405  * that the offset of the pci capability.
2406  * On failure, it sets an error and returns a negative error
2407  * code.
2408  */
2409 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2410                        uint8_t offset, uint8_t size,
2411                        Error **errp)
2412 {
2413     uint8_t *config;
2414     int i, overlapping_cap;
2415 
2416     if (!offset) {
2417         offset = pci_find_space(pdev, size);
2418         /* out of PCI config space is programming error */
2419         assert(offset);
2420     } else {
2421         /* Verify that capabilities don't overlap.  Note: device assignment
2422          * depends on this check to verify that the device is not broken.
2423          * Should never trigger for emulated devices, but it's helpful
2424          * for debugging these. */
2425         for (i = offset; i < offset + size; i++) {
2426             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2427             if (overlapping_cap) {
2428                 error_setg(errp, "%s:%02x:%02x.%x "
2429                            "Attempt to add PCI capability %x at offset "
2430                            "%x overlaps existing capability %x at offset %x",
2431                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2432                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2433                            cap_id, offset, overlapping_cap, i);
2434                 return -EINVAL;
2435             }
2436         }
2437     }
2438 
2439     config = pdev->config + offset;
2440     config[PCI_CAP_LIST_ID] = cap_id;
2441     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2442     pdev->config[PCI_CAPABILITY_LIST] = offset;
2443     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2444     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2445     /* Make capability read-only by default */
2446     memset(pdev->wmask + offset, 0, size);
2447     /* Check capability by default */
2448     memset(pdev->cmask + offset, 0xFF, size);
2449     return offset;
2450 }
2451 
2452 /* Unlink capability from the pci config space. */
2453 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2454 {
2455     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2456     if (!offset)
2457         return;
2458     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2459     /* Make capability writable again */
2460     memset(pdev->wmask + offset, 0xff, size);
2461     memset(pdev->w1cmask + offset, 0, size);
2462     /* Clear cmask as device-specific registers can't be checked */
2463     memset(pdev->cmask + offset, 0, size);
2464     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2465 
2466     if (!pdev->config[PCI_CAPABILITY_LIST])
2467         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2468 }
2469 
2470 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2471 {
2472     return pci_find_capability_list(pdev, cap_id, NULL);
2473 }
2474 
2475 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2476 {
2477     PCIDevice *d = (PCIDevice *)dev;
2478     const pci_class_desc *desc;
2479     char ctxt[64];
2480     PCIIORegion *r;
2481     int i, class;
2482 
2483     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2484     desc = pci_class_descriptions;
2485     while (desc->desc && class != desc->class)
2486         desc++;
2487     if (desc->desc) {
2488         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2489     } else {
2490         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2491     }
2492 
2493     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2494                    "pci id %04x:%04x (sub %04x:%04x)\n",
2495                    indent, "", ctxt, pci_dev_bus_num(d),
2496                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2497                    pci_get_word(d->config + PCI_VENDOR_ID),
2498                    pci_get_word(d->config + PCI_DEVICE_ID),
2499                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2500                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2501     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2502         r = &d->io_regions[i];
2503         if (!r->size)
2504             continue;
2505         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2506                        " [0x%"FMT_PCIBUS"]\n",
2507                        indent, "",
2508                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2509                        r->addr, r->addr + r->size - 1);
2510     }
2511 }
2512 
2513 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2514 {
2515     PCIDevice *d = (PCIDevice *)dev;
2516     const char *name = NULL;
2517     const pci_class_desc *desc =  pci_class_descriptions;
2518     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2519 
2520     while (desc->desc &&
2521           (class & ~desc->fw_ign_bits) !=
2522           (desc->class & ~desc->fw_ign_bits)) {
2523         desc++;
2524     }
2525 
2526     if (desc->desc) {
2527         name = desc->fw_name;
2528     }
2529 
2530     if (name) {
2531         pstrcpy(buf, len, name);
2532     } else {
2533         snprintf(buf, len, "pci%04x,%04x",
2534                  pci_get_word(d->config + PCI_VENDOR_ID),
2535                  pci_get_word(d->config + PCI_DEVICE_ID));
2536     }
2537 
2538     return buf;
2539 }
2540 
2541 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2542 {
2543     PCIDevice *d = (PCIDevice *)dev;
2544     char path[50], name[33];
2545     int off;
2546 
2547     off = snprintf(path, sizeof(path), "%s@%x",
2548                    pci_dev_fw_name(dev, name, sizeof name),
2549                    PCI_SLOT(d->devfn));
2550     if (PCI_FUNC(d->devfn))
2551         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2552     return g_strdup(path);
2553 }
2554 
2555 static char *pcibus_get_dev_path(DeviceState *dev)
2556 {
2557     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2558     PCIDevice *t;
2559     int slot_depth;
2560     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2561      * 00 is added here to make this format compatible with
2562      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2563      * Slot.Function list specifies the slot and function numbers for all
2564      * devices on the path from root to the specific device. */
2565     const char *root_bus_path;
2566     int root_bus_len;
2567     char slot[] = ":SS.F";
2568     int slot_len = sizeof slot - 1 /* For '\0' */;
2569     int path_len;
2570     char *path, *p;
2571     int s;
2572 
2573     root_bus_path = pci_root_bus_path(d);
2574     root_bus_len = strlen(root_bus_path);
2575 
2576     /* Calculate # of slots on path between device and root. */;
2577     slot_depth = 0;
2578     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2579         ++slot_depth;
2580     }
2581 
2582     path_len = root_bus_len + slot_len * slot_depth;
2583 
2584     /* Allocate memory, fill in the terminating null byte. */
2585     path = g_malloc(path_len + 1 /* For '\0' */);
2586     path[path_len] = '\0';
2587 
2588     memcpy(path, root_bus_path, root_bus_len);
2589 
2590     /* Fill in slot numbers. We walk up from device to root, so need to print
2591      * them in the reverse order, last to first. */
2592     p = path + path_len;
2593     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2594         p -= slot_len;
2595         s = snprintf(slot, sizeof slot, ":%02x.%x",
2596                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2597         assert(s == slot_len);
2598         memcpy(p, slot, slot_len);
2599     }
2600 
2601     return path;
2602 }
2603 
2604 static int pci_qdev_find_recursive(PCIBus *bus,
2605                                    const char *id, PCIDevice **pdev)
2606 {
2607     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2608     if (!qdev) {
2609         return -ENODEV;
2610     }
2611 
2612     /* roughly check if given qdev is pci device */
2613     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2614         *pdev = PCI_DEVICE(qdev);
2615         return 0;
2616     }
2617     return -EINVAL;
2618 }
2619 
2620 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2621 {
2622     PCIHostState *host_bridge;
2623     int rc = -ENODEV;
2624 
2625     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2626         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2627         if (!tmp) {
2628             rc = 0;
2629             break;
2630         }
2631         if (tmp != -ENODEV) {
2632             rc = tmp;
2633         }
2634     }
2635 
2636     return rc;
2637 }
2638 
2639 MemoryRegion *pci_address_space(PCIDevice *dev)
2640 {
2641     return pci_get_bus(dev)->address_space_mem;
2642 }
2643 
2644 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2645 {
2646     return pci_get_bus(dev)->address_space_io;
2647 }
2648 
2649 static void pci_device_class_init(ObjectClass *klass, void *data)
2650 {
2651     DeviceClass *k = DEVICE_CLASS(klass);
2652 
2653     k->realize = pci_qdev_realize;
2654     k->unrealize = pci_qdev_unrealize;
2655     k->bus_type = TYPE_PCI_BUS;
2656     device_class_set_props(k, pci_props);
2657 }
2658 
2659 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2660 {
2661     if (!object_class_is_abstract(klass)) {
2662         ObjectClass *conventional =
2663             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2664         ObjectClass *pcie =
2665             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2666         assert(conventional || pcie);
2667     }
2668 }
2669 
2670 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2671 {
2672     PCIBus *bus = pci_get_bus(dev);
2673     PCIBus *iommu_bus = bus;
2674     uint8_t devfn = dev->devfn;
2675 
2676     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2677         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2678 
2679         /*
2680          * The requester ID of the provided device may be aliased, as seen from
2681          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2682          * requester ID to provide a unique AddressSpace for devices, but
2683          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2684          * to-PCI bridge creates and accepts transactions on behalf of down-
2685          * stream devices.  When doing so, all downstream devices are masked
2686          * (aliased) behind a single requester ID.  The requester ID used
2687          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2688          * bridges, with a PCIe capability indicating such, follow the
2689          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2690          * where the bridge uses the seconary bus as the bridge portion of the
2691          * requester ID and devfn of 00.0.  For other bridges, typically those
2692          * found on the root complex such as the dmi-to-pci-bridge, we follow
2693          * the convention of typical bare-metal hardware, which uses the
2694          * requester ID of the bridge itself.  There are device specific
2695          * exceptions to these rules, but these are the defaults that the
2696          * Linux kernel uses when determining DMA aliases itself and believed
2697          * to be true for the bare metal equivalents of the devices emulated
2698          * in QEMU.
2699          */
2700         if (!pci_bus_is_express(iommu_bus)) {
2701             PCIDevice *parent = iommu_bus->parent_dev;
2702 
2703             if (pci_is_express(parent) &&
2704                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2705                 devfn = PCI_DEVFN(0, 0);
2706                 bus = iommu_bus;
2707             } else {
2708                 devfn = parent->devfn;
2709                 bus = parent_bus;
2710             }
2711         }
2712 
2713         iommu_bus = parent_bus;
2714     }
2715     if (iommu_bus && iommu_bus->iommu_fn) {
2716         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2717     }
2718     return &address_space_memory;
2719 }
2720 
2721 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2722 {
2723     bus->iommu_fn = fn;
2724     bus->iommu_opaque = opaque;
2725 }
2726 
2727 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2728 {
2729     Range *range = opaque;
2730     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2731     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2732     int i;
2733 
2734     if (!(cmd & PCI_COMMAND_MEMORY)) {
2735         return;
2736     }
2737 
2738     if (pc->is_bridge) {
2739         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2740         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2741 
2742         base = MAX(base, 0x1ULL << 32);
2743 
2744         if (limit >= base) {
2745             Range pref_range;
2746             range_set_bounds(&pref_range, base, limit);
2747             range_extend(range, &pref_range);
2748         }
2749     }
2750     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2751         PCIIORegion *r = &dev->io_regions[i];
2752         pcibus_t lob, upb;
2753         Range region_range;
2754 
2755         if (!r->size ||
2756             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2757             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2758             continue;
2759         }
2760 
2761         lob = pci_bar_address(dev, i, r->type, r->size);
2762         upb = lob + r->size - 1;
2763         if (lob == PCI_BAR_UNMAPPED) {
2764             continue;
2765         }
2766 
2767         lob = MAX(lob, 0x1ULL << 32);
2768 
2769         if (upb >= lob) {
2770             range_set_bounds(&region_range, lob, upb);
2771             range_extend(range, &region_range);
2772         }
2773     }
2774 }
2775 
2776 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2777 {
2778     range_make_empty(range);
2779     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2780 }
2781 
2782 static bool pcie_has_upstream_port(PCIDevice *dev)
2783 {
2784     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2785 
2786     /* Device associated with an upstream port.
2787      * As there are several types of these, it's easier to check the
2788      * parent device: upstream ports are always connected to
2789      * root or downstream ports.
2790      */
2791     return parent_dev &&
2792         pci_is_express(parent_dev) &&
2793         parent_dev->exp.exp_cap &&
2794         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2795          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2796 }
2797 
2798 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2799 {
2800     PCIBus *bus = pci_get_bus(pci_dev);
2801 
2802     if(pcie_has_upstream_port(pci_dev)) {
2803         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2804         return bus->devices[0];
2805     } else {
2806         /* Other bus types might support multiple devices at slots 0-31 */
2807         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2808     }
2809 }
2810 
2811 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2812 {
2813     MSIMessage msg;
2814     if (msix_enabled(dev)) {
2815         msg = msix_get_message(dev, vector);
2816     } else if (msi_enabled(dev)) {
2817         msg = msi_get_message(dev, vector);
2818     } else {
2819         /* Should never happen */
2820         error_report("%s: unknown interrupt type", __func__);
2821         abort();
2822     }
2823     return msg;
2824 }
2825 
2826 static const TypeInfo pci_device_type_info = {
2827     .name = TYPE_PCI_DEVICE,
2828     .parent = TYPE_DEVICE,
2829     .instance_size = sizeof(PCIDevice),
2830     .abstract = true,
2831     .class_size = sizeof(PCIDeviceClass),
2832     .class_init = pci_device_class_init,
2833     .class_base_init = pci_device_class_base_init,
2834 };
2835 
2836 static void pci_register_types(void)
2837 {
2838     type_register_static(&pci_bus_info);
2839     type_register_static(&pcie_bus_info);
2840     type_register_static(&conventional_pci_interface_info);
2841     type_register_static(&pcie_interface_info);
2842     type_register_static(&pci_device_type_info);
2843 }
2844 
2845 type_init(pci_register_types)
2846