xref: /openbmc/qemu/hw/pci/pci.c (revision 9dea2df8)
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 "hw/hw.h"
27 #include "hw/pci/pci.h"
28 #include "hw/pci/pci_bridge.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/pci/pci_host.h"
31 #include "monitor/monitor.h"
32 #include "net/net.h"
33 #include "sysemu/sysemu.h"
34 #include "hw/loader.h"
35 #include "qemu/error-report.h"
36 #include "qemu/range.h"
37 #include "trace.h"
38 #include "hw/pci/msi.h"
39 #include "hw/pci/msix.h"
40 #include "exec/address-spaces.h"
41 #include "hw/hotplug.h"
42 #include "hw/boards.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-misc.h"
45 #include "qemu/cutils.h"
46 
47 //#define DEBUG_PCI
48 #ifdef DEBUG_PCI
49 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
50 #else
51 # define PCI_DPRINTF(format, ...)       do { } while (0)
52 #endif
53 
54 bool pci_available = true;
55 
56 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
57 static char *pcibus_get_dev_path(DeviceState *dev);
58 static char *pcibus_get_fw_dev_path(DeviceState *dev);
59 static void pcibus_reset(BusState *qbus);
60 
61 static Property pci_props[] = {
62     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
63     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
64     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
65     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
66                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
67     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
68                     QEMU_PCI_CAP_SERR_BITNR, true),
69     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
70                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
71     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
72                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
73     DEFINE_PROP_END_OF_LIST()
74 };
75 
76 static const VMStateDescription vmstate_pcibus = {
77     .name = "PCIBUS",
78     .version_id = 1,
79     .minimum_version_id = 1,
80     .fields = (VMStateField[]) {
81         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
82         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
83                              nirq, 0, vmstate_info_int32,
84                              int32_t),
85         VMSTATE_END_OF_LIST()
86     }
87 };
88 
89 static void pci_init_bus_master(PCIDevice *pci_dev)
90 {
91     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
92 
93     memory_region_init_alias(&pci_dev->bus_master_enable_region,
94                              OBJECT(pci_dev), "bus master",
95                              dma_as->root, 0, memory_region_size(dma_as->root));
96     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
97     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
98                                 &pci_dev->bus_master_enable_region);
99 }
100 
101 static void pcibus_machine_done(Notifier *notifier, void *data)
102 {
103     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
104     int i;
105 
106     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
107         if (bus->devices[i]) {
108             pci_init_bus_master(bus->devices[i]);
109         }
110     }
111 }
112 
113 static void pci_bus_realize(BusState *qbus, Error **errp)
114 {
115     PCIBus *bus = PCI_BUS(qbus);
116 
117     bus->machine_done.notify = pcibus_machine_done;
118     qemu_add_machine_init_done_notifier(&bus->machine_done);
119 
120     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
121 }
122 
123 static void pci_bus_unrealize(BusState *qbus, Error **errp)
124 {
125     PCIBus *bus = PCI_BUS(qbus);
126 
127     qemu_remove_machine_init_done_notifier(&bus->machine_done);
128 
129     vmstate_unregister(NULL, &vmstate_pcibus, bus);
130 }
131 
132 static bool pcibus_is_root(PCIBus *bus)
133 {
134     return !bus->parent_dev;
135 }
136 
137 static int pcibus_num(PCIBus *bus)
138 {
139     if (pcibus_is_root(bus)) {
140         return 0; /* pci host bridge */
141     }
142     return bus->parent_dev->config[PCI_SECONDARY_BUS];
143 }
144 
145 static uint16_t pcibus_numa_node(PCIBus *bus)
146 {
147     return NUMA_NODE_UNASSIGNED;
148 }
149 
150 static void pci_bus_class_init(ObjectClass *klass, void *data)
151 {
152     BusClass *k = BUS_CLASS(klass);
153     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
154 
155     k->print_dev = pcibus_dev_print;
156     k->get_dev_path = pcibus_get_dev_path;
157     k->get_fw_dev_path = pcibus_get_fw_dev_path;
158     k->realize = pci_bus_realize;
159     k->unrealize = pci_bus_unrealize;
160     k->reset = pcibus_reset;
161 
162     pbc->is_root = pcibus_is_root;
163     pbc->bus_num = pcibus_num;
164     pbc->numa_node = pcibus_numa_node;
165 }
166 
167 static const TypeInfo pci_bus_info = {
168     .name = TYPE_PCI_BUS,
169     .parent = TYPE_BUS,
170     .instance_size = sizeof(PCIBus),
171     .class_size = sizeof(PCIBusClass),
172     .class_init = pci_bus_class_init,
173 };
174 
175 static const TypeInfo pcie_interface_info = {
176     .name          = INTERFACE_PCIE_DEVICE,
177     .parent        = TYPE_INTERFACE,
178 };
179 
180 static const TypeInfo conventional_pci_interface_info = {
181     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
182     .parent        = TYPE_INTERFACE,
183 };
184 
185 static const TypeInfo pcie_bus_info = {
186     .name = TYPE_PCIE_BUS,
187     .parent = TYPE_PCI_BUS,
188 };
189 
190 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
191 static void pci_update_mappings(PCIDevice *d);
192 static void pci_irq_handler(void *opaque, int irq_num, int level);
193 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
194 static void pci_del_option_rom(PCIDevice *pdev);
195 
196 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
197 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
198 
199 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
200 
201 int pci_bar(PCIDevice *d, int reg)
202 {
203     uint8_t type;
204 
205     if (reg != PCI_ROM_SLOT)
206         return PCI_BASE_ADDRESS_0 + reg * 4;
207 
208     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
209     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
210 }
211 
212 static inline int pci_irq_state(PCIDevice *d, int irq_num)
213 {
214         return (d->irq_state >> irq_num) & 0x1;
215 }
216 
217 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
218 {
219         d->irq_state &= ~(0x1 << irq_num);
220         d->irq_state |= level << irq_num;
221 }
222 
223 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
224 {
225     PCIBus *bus;
226     for (;;) {
227         bus = pci_get_bus(pci_dev);
228         irq_num = bus->map_irq(pci_dev, irq_num);
229         if (bus->set_irq)
230             break;
231         pci_dev = bus->parent_dev;
232     }
233     bus->irq_count[irq_num] += change;
234     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
235 }
236 
237 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
238 {
239     assert(irq_num >= 0);
240     assert(irq_num < bus->nirq);
241     return !!bus->irq_count[irq_num];
242 }
243 
244 /* Update interrupt status bit in config space on interrupt
245  * state change. */
246 static void pci_update_irq_status(PCIDevice *dev)
247 {
248     if (dev->irq_state) {
249         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
250     } else {
251         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
252     }
253 }
254 
255 void pci_device_deassert_intx(PCIDevice *dev)
256 {
257     int i;
258     for (i = 0; i < PCI_NUM_PINS; ++i) {
259         pci_irq_handler(dev, i, 0);
260     }
261 }
262 
263 static void pci_do_device_reset(PCIDevice *dev)
264 {
265     int r;
266 
267     pci_device_deassert_intx(dev);
268     assert(dev->irq_state == 0);
269 
270     /* Clear all writable bits */
271     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
272                                  pci_get_word(dev->wmask + PCI_COMMAND) |
273                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
274     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
275                                  pci_get_word(dev->wmask + PCI_STATUS) |
276                                  pci_get_word(dev->w1cmask + PCI_STATUS));
277     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
278     dev->config[PCI_INTERRUPT_LINE] = 0x0;
279     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
280         PCIIORegion *region = &dev->io_regions[r];
281         if (!region->size) {
282             continue;
283         }
284 
285         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
286             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
287             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
288         } else {
289             pci_set_long(dev->config + pci_bar(dev, r), region->type);
290         }
291     }
292     pci_update_mappings(dev);
293 
294     msi_reset(dev);
295     msix_reset(dev);
296 }
297 
298 /*
299  * This function is called on #RST and FLR.
300  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
301  */
302 void pci_device_reset(PCIDevice *dev)
303 {
304     qdev_reset_all(&dev->qdev);
305     pci_do_device_reset(dev);
306 }
307 
308 /*
309  * Trigger pci bus reset under a given bus.
310  * Called via qbus_reset_all on RST# assert, after the devices
311  * have been reset qdev_reset_all-ed already.
312  */
313 static void pcibus_reset(BusState *qbus)
314 {
315     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
316     int i;
317 
318     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
319         if (bus->devices[i]) {
320             pci_do_device_reset(bus->devices[i]);
321         }
322     }
323 
324     for (i = 0; i < bus->nirq; i++) {
325         assert(bus->irq_count[i] == 0);
326     }
327 }
328 
329 static void pci_host_bus_register(DeviceState *host)
330 {
331     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
332 
333     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
334 }
335 
336 static void pci_host_bus_unregister(DeviceState *host)
337 {
338     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
339 
340     QLIST_REMOVE(host_bridge, next);
341 }
342 
343 PCIBus *pci_device_root_bus(const PCIDevice *d)
344 {
345     PCIBus *bus = pci_get_bus(d);
346 
347     while (!pci_bus_is_root(bus)) {
348         d = bus->parent_dev;
349         assert(d != NULL);
350 
351         bus = pci_get_bus(d);
352     }
353 
354     return bus;
355 }
356 
357 const char *pci_root_bus_path(PCIDevice *dev)
358 {
359     PCIBus *rootbus = pci_device_root_bus(dev);
360     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
361     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
362 
363     assert(host_bridge->bus == rootbus);
364 
365     if (hc->root_bus_path) {
366         return (*hc->root_bus_path)(host_bridge, rootbus);
367     }
368 
369     return rootbus->qbus.name;
370 }
371 
372 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
373                               MemoryRegion *address_space_mem,
374                               MemoryRegion *address_space_io,
375                               uint8_t devfn_min)
376 {
377     assert(PCI_FUNC(devfn_min) == 0);
378     bus->devfn_min = devfn_min;
379     bus->slot_reserved_mask = 0x0;
380     bus->address_space_mem = address_space_mem;
381     bus->address_space_io = address_space_io;
382 
383     /* host bridge */
384     QLIST_INIT(&bus->child);
385 
386     pci_host_bus_register(parent);
387 }
388 
389 static void pci_bus_uninit(PCIBus *bus)
390 {
391     pci_host_bus_unregister(BUS(bus)->parent);
392 }
393 
394 bool pci_bus_is_express(PCIBus *bus)
395 {
396     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
397 }
398 
399 bool pci_bus_is_root(PCIBus *bus)
400 {
401     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
402 }
403 
404 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
405                               const char *name,
406                               MemoryRegion *address_space_mem,
407                               MemoryRegion *address_space_io,
408                               uint8_t devfn_min, const char *typename)
409 {
410     qbus_create_inplace(bus, bus_size, typename, parent, name);
411     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
412                       devfn_min);
413 }
414 
415 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
416                          MemoryRegion *address_space_mem,
417                          MemoryRegion *address_space_io,
418                          uint8_t devfn_min, const char *typename)
419 {
420     PCIBus *bus;
421 
422     bus = PCI_BUS(qbus_create(typename, parent, name));
423     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
424                       devfn_min);
425     return bus;
426 }
427 
428 void pci_root_bus_cleanup(PCIBus *bus)
429 {
430     pci_bus_uninit(bus);
431     object_unparent(OBJECT(bus));
432 }
433 
434 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
435                   void *irq_opaque, int nirq)
436 {
437     bus->set_irq = set_irq;
438     bus->map_irq = map_irq;
439     bus->irq_opaque = irq_opaque;
440     bus->nirq = nirq;
441     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
442 }
443 
444 void pci_bus_irqs_cleanup(PCIBus *bus)
445 {
446     bus->set_irq = NULL;
447     bus->map_irq = NULL;
448     bus->irq_opaque = NULL;
449     bus->nirq = 0;
450     g_free(bus->irq_count);
451 }
452 
453 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
454                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
455                               void *irq_opaque,
456                               MemoryRegion *address_space_mem,
457                               MemoryRegion *address_space_io,
458                               uint8_t devfn_min, int nirq,
459                               const char *typename)
460 {
461     PCIBus *bus;
462 
463     bus = pci_root_bus_new(parent, name, address_space_mem,
464                            address_space_io, devfn_min, typename);
465     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
466     return bus;
467 }
468 
469 void pci_unregister_root_bus(PCIBus *bus)
470 {
471     pci_bus_irqs_cleanup(bus);
472     pci_root_bus_cleanup(bus);
473 }
474 
475 int pci_bus_num(PCIBus *s)
476 {
477     return PCI_BUS_GET_CLASS(s)->bus_num(s);
478 }
479 
480 int pci_bus_numa_node(PCIBus *bus)
481 {
482     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
483 }
484 
485 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
486                                  const VMStateField *field)
487 {
488     PCIDevice *s = container_of(pv, PCIDevice, config);
489     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
490     uint8_t *config;
491     int i;
492 
493     assert(size == pci_config_size(s));
494     config = g_malloc(size);
495 
496     qemu_get_buffer(f, config, size);
497     for (i = 0; i < size; ++i) {
498         if ((config[i] ^ s->config[i]) &
499             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
500             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
501                          "cmask: %x wmask: %x w1cmask:%x", __func__,
502                          i, config[i], s->config[i],
503                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
504             g_free(config);
505             return -EINVAL;
506         }
507     }
508     memcpy(s->config, config, size);
509 
510     pci_update_mappings(s);
511     if (pc->is_bridge) {
512         PCIBridge *b = PCI_BRIDGE(s);
513         pci_bridge_update_mappings(b);
514     }
515 
516     memory_region_set_enabled(&s->bus_master_enable_region,
517                               pci_get_word(s->config + PCI_COMMAND)
518                               & PCI_COMMAND_MASTER);
519 
520     g_free(config);
521     return 0;
522 }
523 
524 /* just put buffer */
525 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
526                                  const VMStateField *field, QJSON *vmdesc)
527 {
528     const uint8_t **v = pv;
529     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
530     qemu_put_buffer(f, *v, size);
531 
532     return 0;
533 }
534 
535 static VMStateInfo vmstate_info_pci_config = {
536     .name = "pci config",
537     .get  = get_pci_config_device,
538     .put  = put_pci_config_device,
539 };
540 
541 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
542                              const VMStateField *field)
543 {
544     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
545     uint32_t irq_state[PCI_NUM_PINS];
546     int i;
547     for (i = 0; i < PCI_NUM_PINS; ++i) {
548         irq_state[i] = qemu_get_be32(f);
549         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
550             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
551                     irq_state[i]);
552             return -EINVAL;
553         }
554     }
555 
556     for (i = 0; i < PCI_NUM_PINS; ++i) {
557         pci_set_irq_state(s, i, irq_state[i]);
558     }
559 
560     return 0;
561 }
562 
563 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
564                              const VMStateField *field, QJSON *vmdesc)
565 {
566     int i;
567     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
568 
569     for (i = 0; i < PCI_NUM_PINS; ++i) {
570         qemu_put_be32(f, pci_irq_state(s, i));
571     }
572 
573     return 0;
574 }
575 
576 static VMStateInfo vmstate_info_pci_irq_state = {
577     .name = "pci irq state",
578     .get  = get_pci_irq_state,
579     .put  = put_pci_irq_state,
580 };
581 
582 static bool migrate_is_pcie(void *opaque, int version_id)
583 {
584     return pci_is_express((PCIDevice *)opaque);
585 }
586 
587 static bool migrate_is_not_pcie(void *opaque, int version_id)
588 {
589     return !pci_is_express((PCIDevice *)opaque);
590 }
591 
592 const VMStateDescription vmstate_pci_device = {
593     .name = "PCIDevice",
594     .version_id = 2,
595     .minimum_version_id = 1,
596     .fields = (VMStateField[]) {
597         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
598         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
599                                    migrate_is_not_pcie,
600                                    0, vmstate_info_pci_config,
601                                    PCI_CONFIG_SPACE_SIZE),
602         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
603                                    migrate_is_pcie,
604                                    0, vmstate_info_pci_config,
605                                    PCIE_CONFIG_SPACE_SIZE),
606         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
607                                    vmstate_info_pci_irq_state,
608                                    PCI_NUM_PINS * sizeof(int32_t)),
609         VMSTATE_END_OF_LIST()
610     }
611 };
612 
613 
614 void pci_device_save(PCIDevice *s, QEMUFile *f)
615 {
616     /* Clear interrupt status bit: it is implicit
617      * in irq_state which we are saving.
618      * This makes us compatible with old devices
619      * which never set or clear this bit. */
620     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
621     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
622     /* Restore the interrupt status bit. */
623     pci_update_irq_status(s);
624 }
625 
626 int pci_device_load(PCIDevice *s, QEMUFile *f)
627 {
628     int ret;
629     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
630     /* Restore the interrupt status bit. */
631     pci_update_irq_status(s);
632     return ret;
633 }
634 
635 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
636 {
637     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
638                  pci_default_sub_vendor_id);
639     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
640                  pci_default_sub_device_id);
641 }
642 
643 /*
644  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
645  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
646  */
647 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
648                              unsigned int *slotp, unsigned int *funcp)
649 {
650     const char *p;
651     char *e;
652     unsigned long val;
653     unsigned long dom = 0, bus = 0;
654     unsigned int slot = 0;
655     unsigned int func = 0;
656 
657     p = addr;
658     val = strtoul(p, &e, 16);
659     if (e == p)
660         return -1;
661     if (*e == ':') {
662         bus = val;
663         p = e + 1;
664         val = strtoul(p, &e, 16);
665         if (e == p)
666             return -1;
667         if (*e == ':') {
668             dom = bus;
669             bus = val;
670             p = e + 1;
671             val = strtoul(p, &e, 16);
672             if (e == p)
673                 return -1;
674         }
675     }
676 
677     slot = val;
678 
679     if (funcp != NULL) {
680         if (*e != '.')
681             return -1;
682 
683         p = e + 1;
684         val = strtoul(p, &e, 16);
685         if (e == p)
686             return -1;
687 
688         func = val;
689     }
690 
691     /* if funcp == NULL func is 0 */
692     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
693         return -1;
694 
695     if (*e)
696         return -1;
697 
698     *domp = dom;
699     *busp = bus;
700     *slotp = slot;
701     if (funcp != NULL)
702         *funcp = func;
703     return 0;
704 }
705 
706 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
707                                  const char *devaddr)
708 {
709     int dom, bus;
710     unsigned slot;
711 
712     if (!root) {
713         fprintf(stderr, "No primary PCI bus\n");
714         return NULL;
715     }
716 
717     assert(!root->parent_dev);
718 
719     if (!devaddr) {
720         *devfnp = -1;
721         return pci_find_bus_nr(root, 0);
722     }
723 
724     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
725         return NULL;
726     }
727 
728     if (dom != 0) {
729         fprintf(stderr, "No support for non-zero PCI domains\n");
730         return NULL;
731     }
732 
733     *devfnp = PCI_DEVFN(slot, 0);
734     return pci_find_bus_nr(root, bus);
735 }
736 
737 static void pci_init_cmask(PCIDevice *dev)
738 {
739     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
740     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
741     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
742     dev->cmask[PCI_REVISION_ID] = 0xff;
743     dev->cmask[PCI_CLASS_PROG] = 0xff;
744     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
745     dev->cmask[PCI_HEADER_TYPE] = 0xff;
746     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
747 }
748 
749 static void pci_init_wmask(PCIDevice *dev)
750 {
751     int config_size = pci_config_size(dev);
752 
753     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
754     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
755     pci_set_word(dev->wmask + PCI_COMMAND,
756                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
757                  PCI_COMMAND_INTX_DISABLE);
758     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
759         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
760     }
761 
762     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
763            config_size - PCI_CONFIG_HEADER_SIZE);
764 }
765 
766 static void pci_init_w1cmask(PCIDevice *dev)
767 {
768     /*
769      * Note: It's okay to set w1cmask even for readonly bits as
770      * long as their value is hardwired to 0.
771      */
772     pci_set_word(dev->w1cmask + PCI_STATUS,
773                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
774                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
775                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
776 }
777 
778 static void pci_init_mask_bridge(PCIDevice *d)
779 {
780     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
781        PCI_SEC_LETENCY_TIMER */
782     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
783 
784     /* base and limit */
785     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
786     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
787     pci_set_word(d->wmask + PCI_MEMORY_BASE,
788                  PCI_MEMORY_RANGE_MASK & 0xffff);
789     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
790                  PCI_MEMORY_RANGE_MASK & 0xffff);
791     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
792                  PCI_PREF_RANGE_MASK & 0xffff);
793     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
794                  PCI_PREF_RANGE_MASK & 0xffff);
795 
796     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
797     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
798 
799     /* Supported memory and i/o types */
800     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
801     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
802     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
803                                PCI_PREF_RANGE_TYPE_64);
804     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
805                                PCI_PREF_RANGE_TYPE_64);
806 
807     /*
808      * TODO: Bridges default to 10-bit VGA decoding but we currently only
809      * implement 16-bit decoding (no alias support).
810      */
811     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
812                  PCI_BRIDGE_CTL_PARITY |
813                  PCI_BRIDGE_CTL_SERR |
814                  PCI_BRIDGE_CTL_ISA |
815                  PCI_BRIDGE_CTL_VGA |
816                  PCI_BRIDGE_CTL_VGA_16BIT |
817                  PCI_BRIDGE_CTL_MASTER_ABORT |
818                  PCI_BRIDGE_CTL_BUS_RESET |
819                  PCI_BRIDGE_CTL_FAST_BACK |
820                  PCI_BRIDGE_CTL_DISCARD |
821                  PCI_BRIDGE_CTL_SEC_DISCARD |
822                  PCI_BRIDGE_CTL_DISCARD_SERR);
823     /* Below does not do anything as we never set this bit, put here for
824      * completeness. */
825     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
826                  PCI_BRIDGE_CTL_DISCARD_STATUS);
827     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
828     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
829     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
830                                PCI_PREF_RANGE_TYPE_MASK);
831     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
832                                PCI_PREF_RANGE_TYPE_MASK);
833 }
834 
835 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
836 {
837     uint8_t slot = PCI_SLOT(dev->devfn);
838     uint8_t func;
839 
840     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
841         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
842     }
843 
844     /*
845      * multifunction bit is interpreted in two ways as follows.
846      *   - all functions must set the bit to 1.
847      *     Example: Intel X53
848      *   - function 0 must set the bit, but the rest function (> 0)
849      *     is allowed to leave the bit to 0.
850      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
851      *
852      * So OS (at least Linux) checks the bit of only function 0,
853      * and doesn't see the bit of function > 0.
854      *
855      * The below check allows both interpretation.
856      */
857     if (PCI_FUNC(dev->devfn)) {
858         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
859         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
860             /* function 0 should set multifunction bit */
861             error_setg(errp, "PCI: single function device can't be populated "
862                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
863             return;
864         }
865         return;
866     }
867 
868     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
869         return;
870     }
871     /* function 0 indicates single function, so function > 0 must be NULL */
872     for (func = 1; func < PCI_FUNC_MAX; ++func) {
873         if (bus->devices[PCI_DEVFN(slot, func)]) {
874             error_setg(errp, "PCI: %x.0 indicates single function, "
875                        "but %x.%x is already populated.",
876                        slot, slot, func);
877             return;
878         }
879     }
880 }
881 
882 static void pci_config_alloc(PCIDevice *pci_dev)
883 {
884     int config_size = pci_config_size(pci_dev);
885 
886     pci_dev->config = g_malloc0(config_size);
887     pci_dev->cmask = g_malloc0(config_size);
888     pci_dev->wmask = g_malloc0(config_size);
889     pci_dev->w1cmask = g_malloc0(config_size);
890     pci_dev->used = g_malloc0(config_size);
891 }
892 
893 static void pci_config_free(PCIDevice *pci_dev)
894 {
895     g_free(pci_dev->config);
896     g_free(pci_dev->cmask);
897     g_free(pci_dev->wmask);
898     g_free(pci_dev->w1cmask);
899     g_free(pci_dev->used);
900 }
901 
902 static void do_pci_unregister_device(PCIDevice *pci_dev)
903 {
904     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
905     pci_config_free(pci_dev);
906 
907     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
908         memory_region_del_subregion(&pci_dev->bus_master_container_region,
909                                     &pci_dev->bus_master_enable_region);
910     }
911     address_space_destroy(&pci_dev->bus_master_as);
912 }
913 
914 /* Extract PCIReqIDCache into BDF format */
915 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
916 {
917     uint8_t bus_n;
918     uint16_t result;
919 
920     switch (cache->type) {
921     case PCI_REQ_ID_BDF:
922         result = pci_get_bdf(cache->dev);
923         break;
924     case PCI_REQ_ID_SECONDARY_BUS:
925         bus_n = pci_dev_bus_num(cache->dev);
926         result = PCI_BUILD_BDF(bus_n, 0);
927         break;
928     default:
929         error_printf("Invalid PCI requester ID cache type: %d\n",
930                      cache->type);
931         exit(1);
932         break;
933     }
934 
935     return result;
936 }
937 
938 /* Parse bridges up to the root complex and return requester ID
939  * cache for specific device.  For full PCIe topology, the cache
940  * result would be exactly the same as getting BDF of the device.
941  * However, several tricks are required when system mixed up with
942  * legacy PCI devices and PCIe-to-PCI bridges.
943  *
944  * Here we cache the proxy device (and type) not requester ID since
945  * bus number might change from time to time.
946  */
947 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
948 {
949     PCIDevice *parent;
950     PCIReqIDCache cache = {
951         .dev = dev,
952         .type = PCI_REQ_ID_BDF,
953     };
954 
955     while (!pci_bus_is_root(pci_get_bus(dev))) {
956         /* We are under PCI/PCIe bridges */
957         parent = pci_get_bus(dev)->parent_dev;
958         if (pci_is_express(parent)) {
959             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
960                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
961                  * override the requester ID using secondary bus
962                  * number of parent bridge with zeroed devfn
963                  * (pcie-to-pci bridge spec chap 2.3). */
964                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
965                 cache.dev = dev;
966             }
967         } else {
968             /* Legacy PCI, override requester ID with the bridge's
969              * BDF upstream.  When the root complex connects to
970              * legacy PCI devices (including buses), it can only
971              * obtain requester ID info from directly attached
972              * devices.  If devices are attached under bridges, only
973              * the requester ID of the bridge that is directly
974              * attached to the root complex can be recognized. */
975             cache.type = PCI_REQ_ID_BDF;
976             cache.dev = parent;
977         }
978         dev = parent;
979     }
980 
981     return cache;
982 }
983 
984 uint16_t pci_requester_id(PCIDevice *dev)
985 {
986     return pci_req_id_cache_extract(&dev->requester_id_cache);
987 }
988 
989 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
990 {
991     return !(bus->devices[devfn]);
992 }
993 
994 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
995 {
996     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
997 }
998 
999 /* -1 for devfn means auto assign */
1000 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1001                                          const char *name, int devfn,
1002                                          Error **errp)
1003 {
1004     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1005     PCIConfigReadFunc *config_read = pc->config_read;
1006     PCIConfigWriteFunc *config_write = pc->config_write;
1007     Error *local_err = NULL;
1008     DeviceState *dev = DEVICE(pci_dev);
1009     PCIBus *bus = pci_get_bus(pci_dev);
1010 
1011     /* Only pci bridges can be attached to extra PCI root buses */
1012     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1013         error_setg(errp,
1014                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1015                     bus->parent_dev->name);
1016         return NULL;
1017     }
1018 
1019     if (devfn < 0) {
1020         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1021             devfn += PCI_FUNC_MAX) {
1022             if (pci_bus_devfn_available(bus, devfn) &&
1023                    !pci_bus_devfn_reserved(bus, devfn)) {
1024                 goto found;
1025             }
1026         }
1027         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1028                    "or reserved", name);
1029         return NULL;
1030     found: ;
1031     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1032         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1033                    " reserved",
1034                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1035         return NULL;
1036     } else if (!pci_bus_devfn_available(bus, devfn)) {
1037         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1038                    " in use by %s",
1039                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1040                    bus->devices[devfn]->name);
1041         return NULL;
1042     } else if (dev->hotplugged &&
1043                pci_get_function_0(pci_dev)) {
1044         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1045                    " new func %s cannot be exposed to guest.",
1046                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1047                    pci_get_function_0(pci_dev)->name,
1048                    name);
1049 
1050        return NULL;
1051     }
1052 
1053     pci_dev->devfn = devfn;
1054     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1055     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1056 
1057     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1058                        "bus master container", UINT64_MAX);
1059     address_space_init(&pci_dev->bus_master_as,
1060                        &pci_dev->bus_master_container_region, pci_dev->name);
1061 
1062     if (qdev_hotplug) {
1063         pci_init_bus_master(pci_dev);
1064     }
1065     pci_dev->irq_state = 0;
1066     pci_config_alloc(pci_dev);
1067 
1068     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1069     pci_config_set_device_id(pci_dev->config, pc->device_id);
1070     pci_config_set_revision(pci_dev->config, pc->revision);
1071     pci_config_set_class(pci_dev->config, pc->class_id);
1072 
1073     if (!pc->is_bridge) {
1074         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1075             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1076                          pc->subsystem_vendor_id);
1077             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1078                          pc->subsystem_id);
1079         } else {
1080             pci_set_default_subsystem_id(pci_dev);
1081         }
1082     } else {
1083         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1084         assert(!pc->subsystem_vendor_id);
1085         assert(!pc->subsystem_id);
1086     }
1087     pci_init_cmask(pci_dev);
1088     pci_init_wmask(pci_dev);
1089     pci_init_w1cmask(pci_dev);
1090     if (pc->is_bridge) {
1091         pci_init_mask_bridge(pci_dev);
1092     }
1093     pci_init_multifunction(bus, pci_dev, &local_err);
1094     if (local_err) {
1095         error_propagate(errp, local_err);
1096         do_pci_unregister_device(pci_dev);
1097         return NULL;
1098     }
1099 
1100     if (!config_read)
1101         config_read = pci_default_read_config;
1102     if (!config_write)
1103         config_write = pci_default_write_config;
1104     pci_dev->config_read = config_read;
1105     pci_dev->config_write = config_write;
1106     bus->devices[devfn] = pci_dev;
1107     pci_dev->version_id = 2; /* Current pci device vmstate version */
1108     return pci_dev;
1109 }
1110 
1111 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1112 {
1113     PCIIORegion *r;
1114     int i;
1115 
1116     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1117         r = &pci_dev->io_regions[i];
1118         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1119             continue;
1120         memory_region_del_subregion(r->address_space, r->memory);
1121     }
1122 
1123     pci_unregister_vga(pci_dev);
1124 }
1125 
1126 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1127 {
1128     PCIDevice *pci_dev = PCI_DEVICE(dev);
1129     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1130 
1131     pci_unregister_io_regions(pci_dev);
1132     pci_del_option_rom(pci_dev);
1133 
1134     if (pc->exit) {
1135         pc->exit(pci_dev);
1136     }
1137 
1138     pci_device_deassert_intx(pci_dev);
1139     do_pci_unregister_device(pci_dev);
1140 }
1141 
1142 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1143                       uint8_t type, MemoryRegion *memory)
1144 {
1145     PCIIORegion *r;
1146     uint32_t addr; /* offset in pci config space */
1147     uint64_t wmask;
1148     pcibus_t size = memory_region_size(memory);
1149 
1150     assert(region_num >= 0);
1151     assert(region_num < PCI_NUM_REGIONS);
1152     if (size & (size-1)) {
1153         error_report("ERROR: PCI region size must be pow2 "
1154                     "type=0x%x, size=0x%"FMT_PCIBUS"", type, size);
1155         exit(1);
1156     }
1157 
1158     r = &pci_dev->io_regions[region_num];
1159     r->addr = PCI_BAR_UNMAPPED;
1160     r->size = size;
1161     r->type = type;
1162     r->memory = memory;
1163     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1164                         ? pci_get_bus(pci_dev)->address_space_io
1165                         : pci_get_bus(pci_dev)->address_space_mem;
1166 
1167     wmask = ~(size - 1);
1168     if (region_num == PCI_ROM_SLOT) {
1169         /* ROM enable bit is writable */
1170         wmask |= PCI_ROM_ADDRESS_ENABLE;
1171     }
1172 
1173     addr = pci_bar(pci_dev, region_num);
1174     pci_set_long(pci_dev->config + addr, type);
1175 
1176     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1177         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1178         pci_set_quad(pci_dev->wmask + addr, wmask);
1179         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1180     } else {
1181         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1182         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1183     }
1184 }
1185 
1186 static void pci_update_vga(PCIDevice *pci_dev)
1187 {
1188     uint16_t cmd;
1189 
1190     if (!pci_dev->has_vga) {
1191         return;
1192     }
1193 
1194     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1195 
1196     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1197                               cmd & PCI_COMMAND_MEMORY);
1198     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1199                               cmd & PCI_COMMAND_IO);
1200     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1201                               cmd & PCI_COMMAND_IO);
1202 }
1203 
1204 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1205                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1206 {
1207     PCIBus *bus = pci_get_bus(pci_dev);
1208 
1209     assert(!pci_dev->has_vga);
1210 
1211     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1212     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1213     memory_region_add_subregion_overlap(bus->address_space_mem,
1214                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1215 
1216     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1217     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1218     memory_region_add_subregion_overlap(bus->address_space_io,
1219                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1220 
1221     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1222     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1223     memory_region_add_subregion_overlap(bus->address_space_io,
1224                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1225     pci_dev->has_vga = true;
1226 
1227     pci_update_vga(pci_dev);
1228 }
1229 
1230 void pci_unregister_vga(PCIDevice *pci_dev)
1231 {
1232     PCIBus *bus = pci_get_bus(pci_dev);
1233 
1234     if (!pci_dev->has_vga) {
1235         return;
1236     }
1237 
1238     memory_region_del_subregion(bus->address_space_mem,
1239                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1240     memory_region_del_subregion(bus->address_space_io,
1241                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1242     memory_region_del_subregion(bus->address_space_io,
1243                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1244     pci_dev->has_vga = false;
1245 }
1246 
1247 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1248 {
1249     return pci_dev->io_regions[region_num].addr;
1250 }
1251 
1252 static pcibus_t pci_bar_address(PCIDevice *d,
1253                                 int reg, uint8_t type, pcibus_t size)
1254 {
1255     pcibus_t new_addr, last_addr;
1256     int bar = pci_bar(d, reg);
1257     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1258     Object *machine = qdev_get_machine();
1259     ObjectClass *oc = object_get_class(machine);
1260     MachineClass *mc = MACHINE_CLASS(oc);
1261     bool allow_0_address = mc->pci_allow_0_address;
1262 
1263     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1264         if (!(cmd & PCI_COMMAND_IO)) {
1265             return PCI_BAR_UNMAPPED;
1266         }
1267         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1268         last_addr = new_addr + size - 1;
1269         /* Check if 32 bit BAR wraps around explicitly.
1270          * TODO: make priorities correct and remove this work around.
1271          */
1272         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1273             (!allow_0_address && new_addr == 0)) {
1274             return PCI_BAR_UNMAPPED;
1275         }
1276         return new_addr;
1277     }
1278 
1279     if (!(cmd & PCI_COMMAND_MEMORY)) {
1280         return PCI_BAR_UNMAPPED;
1281     }
1282     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1283         new_addr = pci_get_quad(d->config + bar);
1284     } else {
1285         new_addr = pci_get_long(d->config + bar);
1286     }
1287     /* the ROM slot has a specific enable bit */
1288     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1289         return PCI_BAR_UNMAPPED;
1290     }
1291     new_addr &= ~(size - 1);
1292     last_addr = new_addr + size - 1;
1293     /* NOTE: we do not support wrapping */
1294     /* XXX: as we cannot support really dynamic
1295        mappings, we handle specific values as invalid
1296        mappings. */
1297     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1298         (!allow_0_address && new_addr == 0)) {
1299         return PCI_BAR_UNMAPPED;
1300     }
1301 
1302     /* Now pcibus_t is 64bit.
1303      * Check if 32 bit BAR wraps around explicitly.
1304      * Without this, PC ide doesn't work well.
1305      * TODO: remove this work around.
1306      */
1307     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1308         return PCI_BAR_UNMAPPED;
1309     }
1310 
1311     /*
1312      * OS is allowed to set BAR beyond its addressable
1313      * bits. For example, 32 bit OS can set 64bit bar
1314      * to >4G. Check it. TODO: we might need to support
1315      * it in the future for e.g. PAE.
1316      */
1317     if (last_addr >= HWADDR_MAX) {
1318         return PCI_BAR_UNMAPPED;
1319     }
1320 
1321     return new_addr;
1322 }
1323 
1324 static void pci_update_mappings(PCIDevice *d)
1325 {
1326     PCIIORegion *r;
1327     int i;
1328     pcibus_t new_addr;
1329 
1330     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1331         r = &d->io_regions[i];
1332 
1333         /* this region isn't registered */
1334         if (!r->size)
1335             continue;
1336 
1337         new_addr = pci_bar_address(d, i, r->type, r->size);
1338 
1339         /* This bar isn't changed */
1340         if (new_addr == r->addr)
1341             continue;
1342 
1343         /* now do the real mapping */
1344         if (r->addr != PCI_BAR_UNMAPPED) {
1345             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1346                                           PCI_SLOT(d->devfn),
1347                                           PCI_FUNC(d->devfn),
1348                                           i, r->addr, r->size);
1349             memory_region_del_subregion(r->address_space, r->memory);
1350         }
1351         r->addr = new_addr;
1352         if (r->addr != PCI_BAR_UNMAPPED) {
1353             trace_pci_update_mappings_add(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_add_subregion_overlap(r->address_space,
1358                                                 r->addr, r->memory, 1);
1359         }
1360     }
1361 
1362     pci_update_vga(d);
1363 }
1364 
1365 static inline int pci_irq_disabled(PCIDevice *d)
1366 {
1367     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1368 }
1369 
1370 /* Called after interrupt disabled field update in config space,
1371  * assert/deassert interrupts if necessary.
1372  * Gets original interrupt disable bit value (before update). */
1373 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1374 {
1375     int i, disabled = pci_irq_disabled(d);
1376     if (disabled == was_irq_disabled)
1377         return;
1378     for (i = 0; i < PCI_NUM_PINS; ++i) {
1379         int state = pci_irq_state(d, i);
1380         pci_change_irq_level(d, i, disabled ? -state : state);
1381     }
1382 }
1383 
1384 uint32_t pci_default_read_config(PCIDevice *d,
1385                                  uint32_t address, int len)
1386 {
1387     uint32_t val = 0;
1388 
1389     if (pci_is_express_downstream_port(d) &&
1390         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1391         pcie_sync_bridge_lnk(d);
1392     }
1393     memcpy(&val, d->config + address, len);
1394     return le32_to_cpu(val);
1395 }
1396 
1397 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1398 {
1399     int i, was_irq_disabled = pci_irq_disabled(d);
1400     uint32_t val = val_in;
1401 
1402     for (i = 0; i < l; val >>= 8, ++i) {
1403         uint8_t wmask = d->wmask[addr + i];
1404         uint8_t w1cmask = d->w1cmask[addr + i];
1405         assert(!(wmask & w1cmask));
1406         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1407         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1408     }
1409     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1410         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1411         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1412         range_covers_byte(addr, l, PCI_COMMAND))
1413         pci_update_mappings(d);
1414 
1415     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1416         pci_update_irq_disabled(d, was_irq_disabled);
1417         memory_region_set_enabled(&d->bus_master_enable_region,
1418                                   pci_get_word(d->config + PCI_COMMAND)
1419                                     & PCI_COMMAND_MASTER);
1420     }
1421 
1422     msi_write_config(d, addr, val_in, l);
1423     msix_write_config(d, addr, val_in, l);
1424 }
1425 
1426 /***********************************************************/
1427 /* generic PCI irq support */
1428 
1429 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1430 static void pci_irq_handler(void *opaque, int irq_num, int level)
1431 {
1432     PCIDevice *pci_dev = opaque;
1433     int change;
1434 
1435     change = level - pci_irq_state(pci_dev, irq_num);
1436     if (!change)
1437         return;
1438 
1439     pci_set_irq_state(pci_dev, irq_num, level);
1440     pci_update_irq_status(pci_dev);
1441     if (pci_irq_disabled(pci_dev))
1442         return;
1443     pci_change_irq_level(pci_dev, irq_num, change);
1444 }
1445 
1446 static inline int pci_intx(PCIDevice *pci_dev)
1447 {
1448     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1449 }
1450 
1451 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1452 {
1453     int intx = pci_intx(pci_dev);
1454 
1455     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1456 }
1457 
1458 void pci_set_irq(PCIDevice *pci_dev, int level)
1459 {
1460     int intx = pci_intx(pci_dev);
1461     pci_irq_handler(pci_dev, intx, level);
1462 }
1463 
1464 /* Special hooks used by device assignment */
1465 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1466 {
1467     assert(pci_bus_is_root(bus));
1468     bus->route_intx_to_irq = route_intx_to_irq;
1469 }
1470 
1471 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1472 {
1473     PCIBus *bus;
1474 
1475     do {
1476         bus = pci_get_bus(dev);
1477         pin = bus->map_irq(dev, pin);
1478         dev = bus->parent_dev;
1479     } while (dev);
1480 
1481     if (!bus->route_intx_to_irq) {
1482         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1483                      object_get_typename(OBJECT(bus->qbus.parent)));
1484         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1485     }
1486 
1487     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1488 }
1489 
1490 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1491 {
1492     return old->mode != new->mode || old->irq != new->irq;
1493 }
1494 
1495 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1496 {
1497     PCIDevice *dev;
1498     PCIBus *sec;
1499     int i;
1500 
1501     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1502         dev = bus->devices[i];
1503         if (dev && dev->intx_routing_notifier) {
1504             dev->intx_routing_notifier(dev);
1505         }
1506     }
1507 
1508     QLIST_FOREACH(sec, &bus->child, sibling) {
1509         pci_bus_fire_intx_routing_notifier(sec);
1510     }
1511 }
1512 
1513 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1514                                           PCIINTxRoutingNotifier notifier)
1515 {
1516     dev->intx_routing_notifier = notifier;
1517 }
1518 
1519 /*
1520  * PCI-to-PCI bridge specification
1521  * 9.1: Interrupt routing. Table 9-1
1522  *
1523  * the PCI Express Base Specification, Revision 2.1
1524  * 2.2.8.1: INTx interrutp signaling - Rules
1525  *          the Implementation Note
1526  *          Table 2-20
1527  */
1528 /*
1529  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1530  * 0-origin unlike PCI interrupt pin register.
1531  */
1532 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1533 {
1534     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1535 }
1536 
1537 /***********************************************************/
1538 /* monitor info on PCI */
1539 
1540 typedef struct {
1541     uint16_t class;
1542     const char *desc;
1543     const char *fw_name;
1544     uint16_t fw_ign_bits;
1545 } pci_class_desc;
1546 
1547 static const pci_class_desc pci_class_descriptions[] =
1548 {
1549     { 0x0001, "VGA controller", "display"},
1550     { 0x0100, "SCSI controller", "scsi"},
1551     { 0x0101, "IDE controller", "ide"},
1552     { 0x0102, "Floppy controller", "fdc"},
1553     { 0x0103, "IPI controller", "ipi"},
1554     { 0x0104, "RAID controller", "raid"},
1555     { 0x0106, "SATA controller"},
1556     { 0x0107, "SAS controller"},
1557     { 0x0180, "Storage controller"},
1558     { 0x0200, "Ethernet controller", "ethernet"},
1559     { 0x0201, "Token Ring controller", "token-ring"},
1560     { 0x0202, "FDDI controller", "fddi"},
1561     { 0x0203, "ATM controller", "atm"},
1562     { 0x0280, "Network controller"},
1563     { 0x0300, "VGA controller", "display", 0x00ff},
1564     { 0x0301, "XGA controller"},
1565     { 0x0302, "3D controller"},
1566     { 0x0380, "Display controller"},
1567     { 0x0400, "Video controller", "video"},
1568     { 0x0401, "Audio controller", "sound"},
1569     { 0x0402, "Phone"},
1570     { 0x0403, "Audio controller", "sound"},
1571     { 0x0480, "Multimedia controller"},
1572     { 0x0500, "RAM controller", "memory"},
1573     { 0x0501, "Flash controller", "flash"},
1574     { 0x0580, "Memory controller"},
1575     { 0x0600, "Host bridge", "host"},
1576     { 0x0601, "ISA bridge", "isa"},
1577     { 0x0602, "EISA bridge", "eisa"},
1578     { 0x0603, "MC bridge", "mca"},
1579     { 0x0604, "PCI bridge", "pci-bridge"},
1580     { 0x0605, "PCMCIA bridge", "pcmcia"},
1581     { 0x0606, "NUBUS bridge", "nubus"},
1582     { 0x0607, "CARDBUS bridge", "cardbus"},
1583     { 0x0608, "RACEWAY bridge"},
1584     { 0x0680, "Bridge"},
1585     { 0x0700, "Serial port", "serial"},
1586     { 0x0701, "Parallel port", "parallel"},
1587     { 0x0800, "Interrupt controller", "interrupt-controller"},
1588     { 0x0801, "DMA controller", "dma-controller"},
1589     { 0x0802, "Timer", "timer"},
1590     { 0x0803, "RTC", "rtc"},
1591     { 0x0900, "Keyboard", "keyboard"},
1592     { 0x0901, "Pen", "pen"},
1593     { 0x0902, "Mouse", "mouse"},
1594     { 0x0A00, "Dock station", "dock", 0x00ff},
1595     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1596     { 0x0c00, "Fireware contorller", "fireware"},
1597     { 0x0c01, "Access bus controller", "access-bus"},
1598     { 0x0c02, "SSA controller", "ssa"},
1599     { 0x0c03, "USB controller", "usb"},
1600     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1601     { 0x0c05, "SMBus"},
1602     { 0, NULL}
1603 };
1604 
1605 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1606                                                   void (*fn)(PCIBus *b,
1607                                                              PCIDevice *d,
1608                                                              void *opaque),
1609                                                   void *opaque)
1610 {
1611     PCIDevice *d;
1612     int devfn;
1613 
1614     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1615         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1616         if (d) {
1617             fn(bus, d, opaque);
1618         }
1619     }
1620 }
1621 
1622 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1623                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1624                          void *opaque)
1625 {
1626     bus = pci_find_bus_nr(bus, bus_num);
1627 
1628     if (bus) {
1629         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1630     }
1631 }
1632 
1633 static void pci_for_each_device_under_bus(PCIBus *bus,
1634                                           void (*fn)(PCIBus *b, PCIDevice *d,
1635                                                      void *opaque),
1636                                           void *opaque)
1637 {
1638     PCIDevice *d;
1639     int devfn;
1640 
1641     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1642         d = bus->devices[devfn];
1643         if (d) {
1644             fn(bus, d, opaque);
1645         }
1646     }
1647 }
1648 
1649 void pci_for_each_device(PCIBus *bus, int bus_num,
1650                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1651                          void *opaque)
1652 {
1653     bus = pci_find_bus_nr(bus, bus_num);
1654 
1655     if (bus) {
1656         pci_for_each_device_under_bus(bus, fn, opaque);
1657     }
1658 }
1659 
1660 static const pci_class_desc *get_class_desc(int class)
1661 {
1662     const pci_class_desc *desc;
1663 
1664     desc = pci_class_descriptions;
1665     while (desc->desc && class != desc->class) {
1666         desc++;
1667     }
1668 
1669     return desc;
1670 }
1671 
1672 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1673 
1674 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1675 {
1676     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1677     int i;
1678 
1679     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1680         const PCIIORegion *r = &dev->io_regions[i];
1681         PciMemoryRegionList *region;
1682 
1683         if (!r->size) {
1684             continue;
1685         }
1686 
1687         region = g_malloc0(sizeof(*region));
1688         region->value = g_malloc0(sizeof(*region->value));
1689 
1690         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1691             region->value->type = g_strdup("io");
1692         } else {
1693             region->value->type = g_strdup("memory");
1694             region->value->has_prefetch = true;
1695             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1696             region->value->has_mem_type_64 = true;
1697             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1698         }
1699 
1700         region->value->bar = i;
1701         region->value->address = r->addr;
1702         region->value->size = r->size;
1703 
1704         /* XXX: waiting for the qapi to support GSList */
1705         if (!cur_item) {
1706             head = cur_item = region;
1707         } else {
1708             cur_item->next = region;
1709             cur_item = region;
1710         }
1711     }
1712 
1713     return head;
1714 }
1715 
1716 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1717                                            int bus_num)
1718 {
1719     PciBridgeInfo *info;
1720     PciMemoryRange *range;
1721 
1722     info = g_new0(PciBridgeInfo, 1);
1723 
1724     info->bus = g_new0(PciBusInfo, 1);
1725     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1726     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1727     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1728 
1729     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1730     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1731     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1732 
1733     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1734     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1735     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1736 
1737     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1738     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1739     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1740 
1741     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1742         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1743         if (child_bus) {
1744             info->has_devices = true;
1745             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1746         }
1747     }
1748 
1749     return info;
1750 }
1751 
1752 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1753                                            int bus_num)
1754 {
1755     const pci_class_desc *desc;
1756     PciDeviceInfo *info;
1757     uint8_t type;
1758     int class;
1759 
1760     info = g_new0(PciDeviceInfo, 1);
1761     info->bus = bus_num;
1762     info->slot = PCI_SLOT(dev->devfn);
1763     info->function = PCI_FUNC(dev->devfn);
1764 
1765     info->class_info = g_new0(PciDeviceClass, 1);
1766     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1767     info->class_info->q_class = class;
1768     desc = get_class_desc(class);
1769     if (desc->desc) {
1770         info->class_info->has_desc = true;
1771         info->class_info->desc = g_strdup(desc->desc);
1772     }
1773 
1774     info->id = g_new0(PciDeviceId, 1);
1775     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1776     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1777     info->regions = qmp_query_pci_regions(dev);
1778     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1779 
1780     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1781         info->has_irq = true;
1782         info->irq = dev->config[PCI_INTERRUPT_LINE];
1783     }
1784 
1785     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1786     if (type == PCI_HEADER_TYPE_BRIDGE) {
1787         info->has_pci_bridge = true;
1788         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1789     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1790         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1791         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1792         info->id->subsystem_vendor =
1793             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1794     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1795         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1796         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1797         info->id->subsystem_vendor =
1798             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1799     }
1800 
1801     return info;
1802 }
1803 
1804 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1805 {
1806     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1807     PCIDevice *dev;
1808     int devfn;
1809 
1810     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1811         dev = bus->devices[devfn];
1812         if (dev) {
1813             info = g_malloc0(sizeof(*info));
1814             info->value = qmp_query_pci_device(dev, bus, bus_num);
1815 
1816             /* XXX: waiting for the qapi to support GSList */
1817             if (!cur_item) {
1818                 head = cur_item = info;
1819             } else {
1820                 cur_item->next = info;
1821                 cur_item = info;
1822             }
1823         }
1824     }
1825 
1826     return head;
1827 }
1828 
1829 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1830 {
1831     PciInfo *info = NULL;
1832 
1833     bus = pci_find_bus_nr(bus, bus_num);
1834     if (bus) {
1835         info = g_malloc0(sizeof(*info));
1836         info->bus = bus_num;
1837         info->devices = qmp_query_pci_devices(bus, bus_num);
1838     }
1839 
1840     return info;
1841 }
1842 
1843 PciInfoList *qmp_query_pci(Error **errp)
1844 {
1845     PciInfoList *info, *head = NULL, *cur_item = NULL;
1846     PCIHostState *host_bridge;
1847 
1848     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1849         info = g_malloc0(sizeof(*info));
1850         info->value = qmp_query_pci_bus(host_bridge->bus,
1851                                         pci_bus_num(host_bridge->bus));
1852 
1853         /* XXX: waiting for the qapi to support GSList */
1854         if (!cur_item) {
1855             head = cur_item = info;
1856         } else {
1857             cur_item->next = info;
1858             cur_item = info;
1859         }
1860     }
1861 
1862     return head;
1863 }
1864 
1865 /* Initialize a PCI NIC.  */
1866 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1867                                const char *default_model,
1868                                const char *default_devaddr)
1869 {
1870     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1871     GSList *list;
1872     GPtrArray *pci_nic_models;
1873     PCIBus *bus;
1874     PCIDevice *pci_dev;
1875     DeviceState *dev;
1876     int devfn;
1877     int i;
1878 
1879     if (nd->model && !strcmp(nd->model, "virtio")) {
1880         g_free(nd->model);
1881         nd->model = g_strdup("virtio-net-pci");
1882     }
1883 
1884     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1885     pci_nic_models = g_ptr_array_new();
1886     while (list) {
1887         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1888                                              TYPE_DEVICE);
1889         GSList *next;
1890         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1891             dc->user_creatable) {
1892             const char *name = object_class_get_name(list->data);
1893             g_ptr_array_add(pci_nic_models, (gpointer)name);
1894         }
1895         next = list->next;
1896         g_slist_free_1(list);
1897         list = next;
1898     }
1899     g_ptr_array_add(pci_nic_models, NULL);
1900 
1901     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1902         exit(0);
1903     }
1904 
1905     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1906                             default_model);
1907     if (i < 0) {
1908         exit(1);
1909     }
1910 
1911     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1912     if (!bus) {
1913         error_report("Invalid PCI device address %s for device %s",
1914                      devaddr, nd->model);
1915         exit(1);
1916     }
1917 
1918     pci_dev = pci_create(bus, devfn, nd->model);
1919     dev = &pci_dev->qdev;
1920     qdev_set_nic_properties(dev, nd);
1921     qdev_init_nofail(dev);
1922     g_ptr_array_free(pci_nic_models, true);
1923     return pci_dev;
1924 }
1925 
1926 PCIDevice *pci_vga_init(PCIBus *bus)
1927 {
1928     switch (vga_interface_type) {
1929     case VGA_CIRRUS:
1930         return pci_create_simple(bus, -1, "cirrus-vga");
1931     case VGA_QXL:
1932         return pci_create_simple(bus, -1, "qxl-vga");
1933     case VGA_STD:
1934         return pci_create_simple(bus, -1, "VGA");
1935     case VGA_VMWARE:
1936         return pci_create_simple(bus, -1, "vmware-svga");
1937     case VGA_VIRTIO:
1938         return pci_create_simple(bus, -1, "virtio-vga");
1939     case VGA_NONE:
1940     default: /* Other non-PCI types. Checking for unsupported types is already
1941                 done in vl.c. */
1942         return NULL;
1943     }
1944 }
1945 
1946 /* Whether a given bus number is in range of the secondary
1947  * bus of the given bridge device. */
1948 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1949 {
1950     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1951              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1952         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1953         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1954 }
1955 
1956 /* Whether a given bus number is in a range of a root bus */
1957 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1958 {
1959     int i;
1960 
1961     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1962         PCIDevice *dev = bus->devices[i];
1963 
1964         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1965             if (pci_secondary_bus_in_range(dev, bus_num)) {
1966                 return true;
1967             }
1968         }
1969     }
1970 
1971     return false;
1972 }
1973 
1974 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1975 {
1976     PCIBus *sec;
1977 
1978     if (!bus) {
1979         return NULL;
1980     }
1981 
1982     if (pci_bus_num(bus) == bus_num) {
1983         return bus;
1984     }
1985 
1986     /* Consider all bus numbers in range for the host pci bridge. */
1987     if (!pci_bus_is_root(bus) &&
1988         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1989         return NULL;
1990     }
1991 
1992     /* try child bus */
1993     for (; bus; bus = sec) {
1994         QLIST_FOREACH(sec, &bus->child, sibling) {
1995             if (pci_bus_num(sec) == bus_num) {
1996                 return sec;
1997             }
1998             /* PXB buses assumed to be children of bus 0 */
1999             if (pci_bus_is_root(sec)) {
2000                 if (pci_root_bus_in_range(sec, bus_num)) {
2001                     break;
2002                 }
2003             } else {
2004                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2005                     break;
2006                 }
2007             }
2008         }
2009     }
2010 
2011     return NULL;
2012 }
2013 
2014 void pci_for_each_bus_depth_first(PCIBus *bus,
2015                                   void *(*begin)(PCIBus *bus, void *parent_state),
2016                                   void (*end)(PCIBus *bus, void *state),
2017                                   void *parent_state)
2018 {
2019     PCIBus *sec;
2020     void *state;
2021 
2022     if (!bus) {
2023         return;
2024     }
2025 
2026     if (begin) {
2027         state = begin(bus, parent_state);
2028     } else {
2029         state = parent_state;
2030     }
2031 
2032     QLIST_FOREACH(sec, &bus->child, sibling) {
2033         pci_for_each_bus_depth_first(sec, begin, end, state);
2034     }
2035 
2036     if (end) {
2037         end(bus, state);
2038     }
2039 }
2040 
2041 
2042 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2043 {
2044     bus = pci_find_bus_nr(bus, bus_num);
2045 
2046     if (!bus)
2047         return NULL;
2048 
2049     return bus->devices[devfn];
2050 }
2051 
2052 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2053 {
2054     PCIDevice *pci_dev = (PCIDevice *)qdev;
2055     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2056     ObjectClass *klass = OBJECT_CLASS(pc);
2057     Error *local_err = NULL;
2058     bool is_default_rom;
2059 
2060     /* initialize cap_present for pci_is_express() and pci_config_size(),
2061      * Note that hybrid PCIs are not set automatically and need to manage
2062      * QEMU_PCI_CAP_EXPRESS manually */
2063     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2064        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2065         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2066     }
2067 
2068     pci_dev = do_pci_register_device(pci_dev,
2069                                      object_get_typename(OBJECT(qdev)),
2070                                      pci_dev->devfn, errp);
2071     if (pci_dev == NULL)
2072         return;
2073 
2074     if (pc->realize) {
2075         pc->realize(pci_dev, &local_err);
2076         if (local_err) {
2077             error_propagate(errp, local_err);
2078             do_pci_unregister_device(pci_dev);
2079             return;
2080         }
2081     }
2082 
2083     /* rom loading */
2084     is_default_rom = false;
2085     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2086         pci_dev->romfile = g_strdup(pc->romfile);
2087         is_default_rom = true;
2088     }
2089 
2090     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2091     if (local_err) {
2092         error_propagate(errp, local_err);
2093         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2094         return;
2095     }
2096 }
2097 
2098 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2099                                     const char *name)
2100 {
2101     DeviceState *dev;
2102 
2103     dev = qdev_create(&bus->qbus, name);
2104     qdev_prop_set_int32(dev, "addr", devfn);
2105     qdev_prop_set_bit(dev, "multifunction", multifunction);
2106     return PCI_DEVICE(dev);
2107 }
2108 
2109 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2110                                            bool multifunction,
2111                                            const char *name)
2112 {
2113     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2114     qdev_init_nofail(&dev->qdev);
2115     return dev;
2116 }
2117 
2118 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2119 {
2120     return pci_create_multifunction(bus, devfn, false, name);
2121 }
2122 
2123 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2124 {
2125     return pci_create_simple_multifunction(bus, devfn, false, name);
2126 }
2127 
2128 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2129 {
2130     int offset = PCI_CONFIG_HEADER_SIZE;
2131     int i;
2132     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2133         if (pdev->used[i])
2134             offset = i + 1;
2135         else if (i - offset + 1 == size)
2136             return offset;
2137     }
2138     return 0;
2139 }
2140 
2141 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2142                                         uint8_t *prev_p)
2143 {
2144     uint8_t next, prev;
2145 
2146     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2147         return 0;
2148 
2149     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2150          prev = next + PCI_CAP_LIST_NEXT)
2151         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2152             break;
2153 
2154     if (prev_p)
2155         *prev_p = prev;
2156     return next;
2157 }
2158 
2159 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2160 {
2161     uint8_t next, prev, found = 0;
2162 
2163     if (!(pdev->used[offset])) {
2164         return 0;
2165     }
2166 
2167     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2168 
2169     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2170          prev = next + PCI_CAP_LIST_NEXT) {
2171         if (next <= offset && next > found) {
2172             found = next;
2173         }
2174     }
2175     return found;
2176 }
2177 
2178 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2179    This is needed for an option rom which is used for more than one device. */
2180 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2181 {
2182     uint16_t vendor_id;
2183     uint16_t device_id;
2184     uint16_t rom_vendor_id;
2185     uint16_t rom_device_id;
2186     uint16_t rom_magic;
2187     uint16_t pcir_offset;
2188     uint8_t checksum;
2189 
2190     /* Words in rom data are little endian (like in PCI configuration),
2191        so they can be read / written with pci_get_word / pci_set_word. */
2192 
2193     /* Only a valid rom will be patched. */
2194     rom_magic = pci_get_word(ptr);
2195     if (rom_magic != 0xaa55) {
2196         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2197         return;
2198     }
2199     pcir_offset = pci_get_word(ptr + 0x18);
2200     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2201         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2202         return;
2203     }
2204 
2205     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2206     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2207     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2208     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2209 
2210     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2211                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2212 
2213     checksum = ptr[6];
2214 
2215     if (vendor_id != rom_vendor_id) {
2216         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2217         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2218         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2219         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2220         ptr[6] = checksum;
2221         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2222     }
2223 
2224     if (device_id != rom_device_id) {
2225         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2226         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2227         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2228         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2229         ptr[6] = checksum;
2230         pci_set_word(ptr + pcir_offset + 6, device_id);
2231     }
2232 }
2233 
2234 /* Add an option rom for the device */
2235 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2236                                Error **errp)
2237 {
2238     int size;
2239     char *path;
2240     void *ptr;
2241     char name[32];
2242     const VMStateDescription *vmsd;
2243 
2244     if (!pdev->romfile)
2245         return;
2246     if (strlen(pdev->romfile) == 0)
2247         return;
2248 
2249     if (!pdev->rom_bar) {
2250         /*
2251          * Load rom via fw_cfg instead of creating a rom bar,
2252          * for 0.11 compatibility.
2253          */
2254         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2255 
2256         /*
2257          * Hot-plugged devices can't use the option ROM
2258          * if the rom bar is disabled.
2259          */
2260         if (DEVICE(pdev)->hotplugged) {
2261             error_setg(errp, "Hot-plugged device without ROM bar"
2262                        " can't have an option ROM");
2263             return;
2264         }
2265 
2266         if (class == 0x0300) {
2267             rom_add_vga(pdev->romfile);
2268         } else {
2269             rom_add_option(pdev->romfile, -1);
2270         }
2271         return;
2272     }
2273 
2274     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2275     if (path == NULL) {
2276         path = g_strdup(pdev->romfile);
2277     }
2278 
2279     size = get_image_size(path);
2280     if (size < 0) {
2281         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2282         g_free(path);
2283         return;
2284     } else if (size == 0) {
2285         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2286         g_free(path);
2287         return;
2288     }
2289     size = pow2ceil(size);
2290 
2291     vmsd = qdev_get_vmsd(DEVICE(pdev));
2292 
2293     if (vmsd) {
2294         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2295     } else {
2296         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2297     }
2298     pdev->has_rom = true;
2299     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2300     ptr = memory_region_get_ram_ptr(&pdev->rom);
2301     if (load_image_size(path, ptr, size) < 0) {
2302         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2303         g_free(path);
2304         return;
2305     }
2306     g_free(path);
2307 
2308     if (is_default_rom) {
2309         /* Only the default rom images will be patched (if needed). */
2310         pci_patch_ids(pdev, ptr, size);
2311     }
2312 
2313     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2314 }
2315 
2316 static void pci_del_option_rom(PCIDevice *pdev)
2317 {
2318     if (!pdev->has_rom)
2319         return;
2320 
2321     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2322     pdev->has_rom = false;
2323 }
2324 
2325 /*
2326  * On success, pci_add_capability() returns a positive value
2327  * that the offset of the pci capability.
2328  * On failure, it sets an error and returns a negative error
2329  * code.
2330  */
2331 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2332                        uint8_t offset, uint8_t size,
2333                        Error **errp)
2334 {
2335     uint8_t *config;
2336     int i, overlapping_cap;
2337 
2338     if (!offset) {
2339         offset = pci_find_space(pdev, size);
2340         /* out of PCI config space is programming error */
2341         assert(offset);
2342     } else {
2343         /* Verify that capabilities don't overlap.  Note: device assignment
2344          * depends on this check to verify that the device is not broken.
2345          * Should never trigger for emulated devices, but it's helpful
2346          * for debugging these. */
2347         for (i = offset; i < offset + size; i++) {
2348             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2349             if (overlapping_cap) {
2350                 error_setg(errp, "%s:%02x:%02x.%x "
2351                            "Attempt to add PCI capability %x at offset "
2352                            "%x overlaps existing capability %x at offset %x",
2353                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2354                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2355                            cap_id, offset, overlapping_cap, i);
2356                 return -EINVAL;
2357             }
2358         }
2359     }
2360 
2361     config = pdev->config + offset;
2362     config[PCI_CAP_LIST_ID] = cap_id;
2363     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2364     pdev->config[PCI_CAPABILITY_LIST] = offset;
2365     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2366     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2367     /* Make capability read-only by default */
2368     memset(pdev->wmask + offset, 0, size);
2369     /* Check capability by default */
2370     memset(pdev->cmask + offset, 0xFF, size);
2371     return offset;
2372 }
2373 
2374 /* Unlink capability from the pci config space. */
2375 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2376 {
2377     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2378     if (!offset)
2379         return;
2380     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2381     /* Make capability writable again */
2382     memset(pdev->wmask + offset, 0xff, size);
2383     memset(pdev->w1cmask + offset, 0, size);
2384     /* Clear cmask as device-specific registers can't be checked */
2385     memset(pdev->cmask + offset, 0, size);
2386     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2387 
2388     if (!pdev->config[PCI_CAPABILITY_LIST])
2389         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2390 }
2391 
2392 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2393 {
2394     return pci_find_capability_list(pdev, cap_id, NULL);
2395 }
2396 
2397 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2398 {
2399     PCIDevice *d = (PCIDevice *)dev;
2400     const pci_class_desc *desc;
2401     char ctxt[64];
2402     PCIIORegion *r;
2403     int i, class;
2404 
2405     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2406     desc = pci_class_descriptions;
2407     while (desc->desc && class != desc->class)
2408         desc++;
2409     if (desc->desc) {
2410         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2411     } else {
2412         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2413     }
2414 
2415     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2416                    "pci id %04x:%04x (sub %04x:%04x)\n",
2417                    indent, "", ctxt, pci_dev_bus_num(d),
2418                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2419                    pci_get_word(d->config + PCI_VENDOR_ID),
2420                    pci_get_word(d->config + PCI_DEVICE_ID),
2421                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2422                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2423     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2424         r = &d->io_regions[i];
2425         if (!r->size)
2426             continue;
2427         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2428                        " [0x%"FMT_PCIBUS"]\n",
2429                        indent, "",
2430                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2431                        r->addr, r->addr + r->size - 1);
2432     }
2433 }
2434 
2435 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2436 {
2437     PCIDevice *d = (PCIDevice *)dev;
2438     const char *name = NULL;
2439     const pci_class_desc *desc =  pci_class_descriptions;
2440     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2441 
2442     while (desc->desc &&
2443           (class & ~desc->fw_ign_bits) !=
2444           (desc->class & ~desc->fw_ign_bits)) {
2445         desc++;
2446     }
2447 
2448     if (desc->desc) {
2449         name = desc->fw_name;
2450     }
2451 
2452     if (name) {
2453         pstrcpy(buf, len, name);
2454     } else {
2455         snprintf(buf, len, "pci%04x,%04x",
2456                  pci_get_word(d->config + PCI_VENDOR_ID),
2457                  pci_get_word(d->config + PCI_DEVICE_ID));
2458     }
2459 
2460     return buf;
2461 }
2462 
2463 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2464 {
2465     PCIDevice *d = (PCIDevice *)dev;
2466     char path[50], name[33];
2467     int off;
2468 
2469     off = snprintf(path, sizeof(path), "%s@%x",
2470                    pci_dev_fw_name(dev, name, sizeof name),
2471                    PCI_SLOT(d->devfn));
2472     if (PCI_FUNC(d->devfn))
2473         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2474     return g_strdup(path);
2475 }
2476 
2477 static char *pcibus_get_dev_path(DeviceState *dev)
2478 {
2479     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2480     PCIDevice *t;
2481     int slot_depth;
2482     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2483      * 00 is added here to make this format compatible with
2484      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2485      * Slot.Function list specifies the slot and function numbers for all
2486      * devices on the path from root to the specific device. */
2487     const char *root_bus_path;
2488     int root_bus_len;
2489     char slot[] = ":SS.F";
2490     int slot_len = sizeof slot - 1 /* For '\0' */;
2491     int path_len;
2492     char *path, *p;
2493     int s;
2494 
2495     root_bus_path = pci_root_bus_path(d);
2496     root_bus_len = strlen(root_bus_path);
2497 
2498     /* Calculate # of slots on path between device and root. */;
2499     slot_depth = 0;
2500     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2501         ++slot_depth;
2502     }
2503 
2504     path_len = root_bus_len + slot_len * slot_depth;
2505 
2506     /* Allocate memory, fill in the terminating null byte. */
2507     path = g_malloc(path_len + 1 /* For '\0' */);
2508     path[path_len] = '\0';
2509 
2510     memcpy(path, root_bus_path, root_bus_len);
2511 
2512     /* Fill in slot numbers. We walk up from device to root, so need to print
2513      * them in the reverse order, last to first. */
2514     p = path + path_len;
2515     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2516         p -= slot_len;
2517         s = snprintf(slot, sizeof slot, ":%02x.%x",
2518                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2519         assert(s == slot_len);
2520         memcpy(p, slot, slot_len);
2521     }
2522 
2523     return path;
2524 }
2525 
2526 static int pci_qdev_find_recursive(PCIBus *bus,
2527                                    const char *id, PCIDevice **pdev)
2528 {
2529     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2530     if (!qdev) {
2531         return -ENODEV;
2532     }
2533 
2534     /* roughly check if given qdev is pci device */
2535     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2536         *pdev = PCI_DEVICE(qdev);
2537         return 0;
2538     }
2539     return -EINVAL;
2540 }
2541 
2542 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2543 {
2544     PCIHostState *host_bridge;
2545     int rc = -ENODEV;
2546 
2547     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2548         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2549         if (!tmp) {
2550             rc = 0;
2551             break;
2552         }
2553         if (tmp != -ENODEV) {
2554             rc = tmp;
2555         }
2556     }
2557 
2558     return rc;
2559 }
2560 
2561 MemoryRegion *pci_address_space(PCIDevice *dev)
2562 {
2563     return pci_get_bus(dev)->address_space_mem;
2564 }
2565 
2566 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2567 {
2568     return pci_get_bus(dev)->address_space_io;
2569 }
2570 
2571 static void pci_device_class_init(ObjectClass *klass, void *data)
2572 {
2573     DeviceClass *k = DEVICE_CLASS(klass);
2574 
2575     k->realize = pci_qdev_realize;
2576     k->unrealize = pci_qdev_unrealize;
2577     k->bus_type = TYPE_PCI_BUS;
2578     k->props = pci_props;
2579 }
2580 
2581 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2582 {
2583     if (!object_class_is_abstract(klass)) {
2584         ObjectClass *conventional =
2585             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2586         ObjectClass *pcie =
2587             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2588         assert(conventional || pcie);
2589     }
2590 }
2591 
2592 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2593 {
2594     PCIBus *bus = pci_get_bus(dev);
2595     PCIBus *iommu_bus = bus;
2596 
2597     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2598         iommu_bus = pci_get_bus(iommu_bus->parent_dev);
2599     }
2600     if (iommu_bus && iommu_bus->iommu_fn) {
2601         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2602     }
2603     return &address_space_memory;
2604 }
2605 
2606 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2607 {
2608     bus->iommu_fn = fn;
2609     bus->iommu_opaque = opaque;
2610 }
2611 
2612 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2613 {
2614     Range *range = opaque;
2615     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2616     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2617     int i;
2618 
2619     if (!(cmd & PCI_COMMAND_MEMORY)) {
2620         return;
2621     }
2622 
2623     if (pc->is_bridge) {
2624         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2625         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2626 
2627         base = MAX(base, 0x1ULL << 32);
2628 
2629         if (limit >= base) {
2630             Range pref_range;
2631             range_set_bounds(&pref_range, base, limit);
2632             range_extend(range, &pref_range);
2633         }
2634     }
2635     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2636         PCIIORegion *r = &dev->io_regions[i];
2637         pcibus_t lob, upb;
2638         Range region_range;
2639 
2640         if (!r->size ||
2641             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2642             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2643             continue;
2644         }
2645 
2646         lob = pci_bar_address(dev, i, r->type, r->size);
2647         upb = lob + r->size - 1;
2648         if (lob == PCI_BAR_UNMAPPED) {
2649             continue;
2650         }
2651 
2652         lob = MAX(lob, 0x1ULL << 32);
2653 
2654         if (upb >= lob) {
2655             range_set_bounds(&region_range, lob, upb);
2656             range_extend(range, &region_range);
2657         }
2658     }
2659 }
2660 
2661 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2662 {
2663     range_make_empty(range);
2664     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2665 }
2666 
2667 static bool pcie_has_upstream_port(PCIDevice *dev)
2668 {
2669     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2670 
2671     /* Device associated with an upstream port.
2672      * As there are several types of these, it's easier to check the
2673      * parent device: upstream ports are always connected to
2674      * root or downstream ports.
2675      */
2676     return parent_dev &&
2677         pci_is_express(parent_dev) &&
2678         parent_dev->exp.exp_cap &&
2679         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2680          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2681 }
2682 
2683 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2684 {
2685     PCIBus *bus = pci_get_bus(pci_dev);
2686 
2687     if(pcie_has_upstream_port(pci_dev)) {
2688         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2689         return bus->devices[0];
2690     } else {
2691         /* Other bus types might support multiple devices at slots 0-31 */
2692         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2693     }
2694 }
2695 
2696 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2697 {
2698     MSIMessage msg;
2699     if (msix_enabled(dev)) {
2700         msg = msix_get_message(dev, vector);
2701     } else if (msi_enabled(dev)) {
2702         msg = msi_get_message(dev, vector);
2703     } else {
2704         /* Should never happen */
2705         error_report("%s: unknown interrupt type", __func__);
2706         abort();
2707     }
2708     return msg;
2709 }
2710 
2711 static const TypeInfo pci_device_type_info = {
2712     .name = TYPE_PCI_DEVICE,
2713     .parent = TYPE_DEVICE,
2714     .instance_size = sizeof(PCIDevice),
2715     .abstract = true,
2716     .class_size = sizeof(PCIDeviceClass),
2717     .class_init = pci_device_class_init,
2718     .class_base_init = pci_device_class_base_init,
2719 };
2720 
2721 static void pci_register_types(void)
2722 {
2723     type_register_static(&pci_bus_info);
2724     type_register_static(&pcie_bus_info);
2725     type_register_static(&conventional_pci_interface_info);
2726     type_register_static(&pcie_interface_info);
2727     type_register_static(&pci_device_type_info);
2728 }
2729 
2730 type_init(pci_register_types)
2731