xref: /openbmc/qemu/hw/pci/pci.c (revision dd873966)
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 #include "qemu/osdep.h"
25 #include "hw/hw.h"
26 #include "hw/pci/pci.h"
27 #include "hw/pci/pci_bridge.h"
28 #include "hw/pci/pci_bus.h"
29 #include "hw/pci/pci_host.h"
30 #include "monitor/monitor.h"
31 #include "net/net.h"
32 #include "sysemu/sysemu.h"
33 #include "hw/loader.h"
34 #include "qemu/error-report.h"
35 #include "qemu/range.h"
36 #include "qmp-commands.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 "qemu/cutils.h"
44 
45 //#define DEBUG_PCI
46 #ifdef DEBUG_PCI
47 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
48 #else
49 # define PCI_DPRINTF(format, ...)       do { } while (0)
50 #endif
51 
52 bool pci_available = true;
53 
54 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
55 static char *pcibus_get_dev_path(DeviceState *dev);
56 static char *pcibus_get_fw_dev_path(DeviceState *dev);
57 static void pcibus_reset(BusState *qbus);
58 
59 static Property pci_props[] = {
60     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
61     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
62     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
63     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
64                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
65     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
66                     QEMU_PCI_CAP_SERR_BITNR, true),
67     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
68                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
69     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
70                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
71     DEFINE_PROP_END_OF_LIST()
72 };
73 
74 static const VMStateDescription vmstate_pcibus = {
75     .name = "PCIBUS",
76     .version_id = 1,
77     .minimum_version_id = 1,
78     .fields = (VMStateField[]) {
79         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
80         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
81                              nirq, 0, vmstate_info_int32,
82                              int32_t),
83         VMSTATE_END_OF_LIST()
84     }
85 };
86 
87 static void pci_init_bus_master(PCIDevice *pci_dev)
88 {
89     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
90 
91     memory_region_init_alias(&pci_dev->bus_master_enable_region,
92                              OBJECT(pci_dev), "bus master",
93                              dma_as->root, 0, memory_region_size(dma_as->root));
94     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
95     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
96                                 &pci_dev->bus_master_enable_region);
97 }
98 
99 static void pcibus_machine_done(Notifier *notifier, void *data)
100 {
101     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
102     int i;
103 
104     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
105         if (bus->devices[i]) {
106             pci_init_bus_master(bus->devices[i]);
107         }
108     }
109 }
110 
111 static void pci_bus_realize(BusState *qbus, Error **errp)
112 {
113     PCIBus *bus = PCI_BUS(qbus);
114 
115     bus->machine_done.notify = pcibus_machine_done;
116     qemu_add_machine_init_done_notifier(&bus->machine_done);
117 
118     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
119 }
120 
121 static void pci_bus_unrealize(BusState *qbus, Error **errp)
122 {
123     PCIBus *bus = PCI_BUS(qbus);
124 
125     qemu_remove_machine_init_done_notifier(&bus->machine_done);
126 
127     vmstate_unregister(NULL, &vmstate_pcibus, bus);
128 }
129 
130 static bool pcibus_is_root(PCIBus *bus)
131 {
132     return !bus->parent_dev;
133 }
134 
135 static int pcibus_num(PCIBus *bus)
136 {
137     if (pcibus_is_root(bus)) {
138         return 0; /* pci host bridge */
139     }
140     return bus->parent_dev->config[PCI_SECONDARY_BUS];
141 }
142 
143 static uint16_t pcibus_numa_node(PCIBus *bus)
144 {
145     return NUMA_NODE_UNASSIGNED;
146 }
147 
148 static void pci_bus_class_init(ObjectClass *klass, void *data)
149 {
150     BusClass *k = BUS_CLASS(klass);
151     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
152 
153     k->print_dev = pcibus_dev_print;
154     k->get_dev_path = pcibus_get_dev_path;
155     k->get_fw_dev_path = pcibus_get_fw_dev_path;
156     k->realize = pci_bus_realize;
157     k->unrealize = pci_bus_unrealize;
158     k->reset = pcibus_reset;
159 
160     pbc->is_root = pcibus_is_root;
161     pbc->bus_num = pcibus_num;
162     pbc->numa_node = pcibus_numa_node;
163 }
164 
165 static const TypeInfo pci_bus_info = {
166     .name = TYPE_PCI_BUS,
167     .parent = TYPE_BUS,
168     .instance_size = sizeof(PCIBus),
169     .class_size = sizeof(PCIBusClass),
170     .class_init = pci_bus_class_init,
171 };
172 
173 static const TypeInfo pcie_interface_info = {
174     .name          = INTERFACE_PCIE_DEVICE,
175     .parent        = TYPE_INTERFACE,
176 };
177 
178 static const TypeInfo conventional_pci_interface_info = {
179     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
180     .parent        = TYPE_INTERFACE,
181 };
182 
183 static const TypeInfo pcie_bus_info = {
184     .name = TYPE_PCIE_BUS,
185     .parent = TYPE_PCI_BUS,
186 };
187 
188 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
189 static void pci_update_mappings(PCIDevice *d);
190 static void pci_irq_handler(void *opaque, int irq_num, int level);
191 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
192 static void pci_del_option_rom(PCIDevice *pdev);
193 
194 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
195 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
196 
197 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
198 
199 int pci_bar(PCIDevice *d, int reg)
200 {
201     uint8_t type;
202 
203     if (reg != PCI_ROM_SLOT)
204         return PCI_BASE_ADDRESS_0 + reg * 4;
205 
206     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
207     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
208 }
209 
210 static inline int pci_irq_state(PCIDevice *d, int irq_num)
211 {
212 	return (d->irq_state >> irq_num) & 0x1;
213 }
214 
215 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
216 {
217 	d->irq_state &= ~(0x1 << irq_num);
218 	d->irq_state |= level << irq_num;
219 }
220 
221 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
222 {
223     PCIBus *bus;
224     for (;;) {
225         bus = pci_dev->bus;
226         irq_num = bus->map_irq(pci_dev, irq_num);
227         if (bus->set_irq)
228             break;
229         pci_dev = bus->parent_dev;
230     }
231     bus->irq_count[irq_num] += change;
232     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
233 }
234 
235 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
236 {
237     assert(irq_num >= 0);
238     assert(irq_num < bus->nirq);
239     return !!bus->irq_count[irq_num];
240 }
241 
242 /* Update interrupt status bit in config space on interrupt
243  * state change. */
244 static void pci_update_irq_status(PCIDevice *dev)
245 {
246     if (dev->irq_state) {
247         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
248     } else {
249         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
250     }
251 }
252 
253 void pci_device_deassert_intx(PCIDevice *dev)
254 {
255     int i;
256     for (i = 0; i < PCI_NUM_PINS; ++i) {
257         pci_irq_handler(dev, i, 0);
258     }
259 }
260 
261 static void pci_do_device_reset(PCIDevice *dev)
262 {
263     int r;
264 
265     pci_device_deassert_intx(dev);
266     assert(dev->irq_state == 0);
267 
268     /* Clear all writable bits */
269     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
270                                  pci_get_word(dev->wmask + PCI_COMMAND) |
271                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
272     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
273                                  pci_get_word(dev->wmask + PCI_STATUS) |
274                                  pci_get_word(dev->w1cmask + PCI_STATUS));
275     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
276     dev->config[PCI_INTERRUPT_LINE] = 0x0;
277     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
278         PCIIORegion *region = &dev->io_regions[r];
279         if (!region->size) {
280             continue;
281         }
282 
283         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
284             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
285             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
286         } else {
287             pci_set_long(dev->config + pci_bar(dev, r), region->type);
288         }
289     }
290     pci_update_mappings(dev);
291 
292     msi_reset(dev);
293     msix_reset(dev);
294 }
295 
296 /*
297  * This function is called on #RST and FLR.
298  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
299  */
300 void pci_device_reset(PCIDevice *dev)
301 {
302     qdev_reset_all(&dev->qdev);
303     pci_do_device_reset(dev);
304 }
305 
306 /*
307  * Trigger pci bus reset under a given bus.
308  * Called via qbus_reset_all on RST# assert, after the devices
309  * have been reset qdev_reset_all-ed already.
310  */
311 static void pcibus_reset(BusState *qbus)
312 {
313     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
314     int i;
315 
316     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
317         if (bus->devices[i]) {
318             pci_do_device_reset(bus->devices[i]);
319         }
320     }
321 
322     for (i = 0; i < bus->nirq; i++) {
323         assert(bus->irq_count[i] == 0);
324     }
325 }
326 
327 static void pci_host_bus_register(DeviceState *host)
328 {
329     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
330 
331     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
332 }
333 
334 PCIBus *pci_find_primary_bus(void)
335 {
336     PCIBus *primary_bus = NULL;
337     PCIHostState *host;
338 
339     QLIST_FOREACH(host, &pci_host_bridges, next) {
340         if (primary_bus) {
341             /* We have multiple root buses, refuse to select a primary */
342             return NULL;
343         }
344         primary_bus = host->bus;
345     }
346 
347     return primary_bus;
348 }
349 
350 PCIBus *pci_device_root_bus(const PCIDevice *d)
351 {
352     PCIBus *bus = d->bus;
353 
354     while (!pci_bus_is_root(bus)) {
355         d = bus->parent_dev;
356         assert(d != NULL);
357 
358         bus = d->bus;
359     }
360 
361     return bus;
362 }
363 
364 const char *pci_root_bus_path(PCIDevice *dev)
365 {
366     PCIBus *rootbus = pci_device_root_bus(dev);
367     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
368     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
369 
370     assert(host_bridge->bus == rootbus);
371 
372     if (hc->root_bus_path) {
373         return (*hc->root_bus_path)(host_bridge, rootbus);
374     }
375 
376     return rootbus->qbus.name;
377 }
378 
379 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
380                          MemoryRegion *address_space_mem,
381                          MemoryRegion *address_space_io,
382                          uint8_t devfn_min)
383 {
384     assert(PCI_FUNC(devfn_min) == 0);
385     bus->devfn_min = devfn_min;
386     bus->slot_reserved_mask = 0x0;
387     bus->address_space_mem = address_space_mem;
388     bus->address_space_io = address_space_io;
389 
390     /* host bridge */
391     QLIST_INIT(&bus->child);
392 
393     pci_host_bus_register(parent);
394 }
395 
396 bool pci_bus_is_express(PCIBus *bus)
397 {
398     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
399 }
400 
401 bool pci_bus_is_root(PCIBus *bus)
402 {
403     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
404 }
405 
406 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
407                          const char *name,
408                          MemoryRegion *address_space_mem,
409                          MemoryRegion *address_space_io,
410                          uint8_t devfn_min, const char *typename)
411 {
412     qbus_create_inplace(bus, bus_size, typename, parent, name);
413     pci_bus_init(bus, parent, address_space_mem, address_space_io, devfn_min);
414 }
415 
416 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
417                     MemoryRegion *address_space_mem,
418                     MemoryRegion *address_space_io,
419                     uint8_t devfn_min, const char *typename)
420 {
421     PCIBus *bus;
422 
423     bus = PCI_BUS(qbus_create(typename, parent, name));
424     pci_bus_init(bus, parent, address_space_mem, address_space_io, devfn_min);
425     return bus;
426 }
427 
428 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
429                   void *irq_opaque, int nirq)
430 {
431     bus->set_irq = set_irq;
432     bus->map_irq = map_irq;
433     bus->irq_opaque = irq_opaque;
434     bus->nirq = nirq;
435     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
436 }
437 
438 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
439                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
440                          void *irq_opaque,
441                          MemoryRegion *address_space_mem,
442                          MemoryRegion *address_space_io,
443                          uint8_t devfn_min, int nirq, const char *typename)
444 {
445     PCIBus *bus;
446 
447     bus = pci_bus_new(parent, name, address_space_mem,
448                       address_space_io, devfn_min, typename);
449     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
450     return bus;
451 }
452 
453 int pci_bus_num(PCIBus *s)
454 {
455     return PCI_BUS_GET_CLASS(s)->bus_num(s);
456 }
457 
458 int pci_bus_numa_node(PCIBus *bus)
459 {
460     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
461 }
462 
463 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
464                                  VMStateField *field)
465 {
466     PCIDevice *s = container_of(pv, PCIDevice, config);
467     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
468     uint8_t *config;
469     int i;
470 
471     assert(size == pci_config_size(s));
472     config = g_malloc(size);
473 
474     qemu_get_buffer(f, config, size);
475     for (i = 0; i < size; ++i) {
476         if ((config[i] ^ s->config[i]) &
477             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
478             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
479                          "cmask: %x wmask: %x w1cmask:%x", __func__,
480                          i, config[i], s->config[i],
481                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
482             g_free(config);
483             return -EINVAL;
484         }
485     }
486     memcpy(s->config, config, size);
487 
488     pci_update_mappings(s);
489     if (pc->is_bridge) {
490         PCIBridge *b = PCI_BRIDGE(s);
491         pci_bridge_update_mappings(b);
492     }
493 
494     memory_region_set_enabled(&s->bus_master_enable_region,
495                               pci_get_word(s->config + PCI_COMMAND)
496                               & PCI_COMMAND_MASTER);
497 
498     g_free(config);
499     return 0;
500 }
501 
502 /* just put buffer */
503 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
504                                  VMStateField *field, QJSON *vmdesc)
505 {
506     const uint8_t **v = pv;
507     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
508     qemu_put_buffer(f, *v, size);
509 
510     return 0;
511 }
512 
513 static VMStateInfo vmstate_info_pci_config = {
514     .name = "pci config",
515     .get  = get_pci_config_device,
516     .put  = put_pci_config_device,
517 };
518 
519 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
520                              VMStateField *field)
521 {
522     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
523     uint32_t irq_state[PCI_NUM_PINS];
524     int i;
525     for (i = 0; i < PCI_NUM_PINS; ++i) {
526         irq_state[i] = qemu_get_be32(f);
527         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
528             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
529                     irq_state[i]);
530             return -EINVAL;
531         }
532     }
533 
534     for (i = 0; i < PCI_NUM_PINS; ++i) {
535         pci_set_irq_state(s, i, irq_state[i]);
536     }
537 
538     return 0;
539 }
540 
541 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
542                              VMStateField *field, QJSON *vmdesc)
543 {
544     int i;
545     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
546 
547     for (i = 0; i < PCI_NUM_PINS; ++i) {
548         qemu_put_be32(f, pci_irq_state(s, i));
549     }
550 
551     return 0;
552 }
553 
554 static VMStateInfo vmstate_info_pci_irq_state = {
555     .name = "pci irq state",
556     .get  = get_pci_irq_state,
557     .put  = put_pci_irq_state,
558 };
559 
560 static bool migrate_is_pcie(void *opaque, int version_id)
561 {
562     return pci_is_express((PCIDevice *)opaque);
563 }
564 
565 static bool migrate_is_not_pcie(void *opaque, int version_id)
566 {
567     return !pci_is_express((PCIDevice *)opaque);
568 }
569 
570 const VMStateDescription vmstate_pci_device = {
571     .name = "PCIDevice",
572     .version_id = 2,
573     .minimum_version_id = 1,
574     .fields = (VMStateField[]) {
575         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
576         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
577                                    migrate_is_not_pcie,
578                                    0, vmstate_info_pci_config,
579                                    PCI_CONFIG_SPACE_SIZE),
580         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
581                                    migrate_is_pcie,
582                                    0, vmstate_info_pci_config,
583                                    PCIE_CONFIG_SPACE_SIZE),
584         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
585 				   vmstate_info_pci_irq_state,
586 				   PCI_NUM_PINS * sizeof(int32_t)),
587         VMSTATE_END_OF_LIST()
588     }
589 };
590 
591 
592 void pci_device_save(PCIDevice *s, QEMUFile *f)
593 {
594     /* Clear interrupt status bit: it is implicit
595      * in irq_state which we are saving.
596      * This makes us compatible with old devices
597      * which never set or clear this bit. */
598     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
599     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
600     /* Restore the interrupt status bit. */
601     pci_update_irq_status(s);
602 }
603 
604 int pci_device_load(PCIDevice *s, QEMUFile *f)
605 {
606     int ret;
607     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
608     /* Restore the interrupt status bit. */
609     pci_update_irq_status(s);
610     return ret;
611 }
612 
613 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
614 {
615     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
616                  pci_default_sub_vendor_id);
617     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
618                  pci_default_sub_device_id);
619 }
620 
621 /*
622  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
623  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
624  */
625 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
626                              unsigned int *slotp, unsigned int *funcp)
627 {
628     const char *p;
629     char *e;
630     unsigned long val;
631     unsigned long dom = 0, bus = 0;
632     unsigned int slot = 0;
633     unsigned int func = 0;
634 
635     p = addr;
636     val = strtoul(p, &e, 16);
637     if (e == p)
638 	return -1;
639     if (*e == ':') {
640 	bus = val;
641 	p = e + 1;
642 	val = strtoul(p, &e, 16);
643 	if (e == p)
644 	    return -1;
645 	if (*e == ':') {
646 	    dom = bus;
647 	    bus = val;
648 	    p = e + 1;
649 	    val = strtoul(p, &e, 16);
650 	    if (e == p)
651 		return -1;
652 	}
653     }
654 
655     slot = val;
656 
657     if (funcp != NULL) {
658         if (*e != '.')
659             return -1;
660 
661         p = e + 1;
662         val = strtoul(p, &e, 16);
663         if (e == p)
664             return -1;
665 
666         func = val;
667     }
668 
669     /* if funcp == NULL func is 0 */
670     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
671 	return -1;
672 
673     if (*e)
674 	return -1;
675 
676     *domp = dom;
677     *busp = bus;
678     *slotp = slot;
679     if (funcp != NULL)
680         *funcp = func;
681     return 0;
682 }
683 
684 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
685                                  const char *devaddr)
686 {
687     int dom, bus;
688     unsigned slot;
689 
690     if (!root) {
691         fprintf(stderr, "No primary PCI bus\n");
692         return NULL;
693     }
694 
695     assert(!root->parent_dev);
696 
697     if (!devaddr) {
698         *devfnp = -1;
699         return pci_find_bus_nr(root, 0);
700     }
701 
702     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
703         return NULL;
704     }
705 
706     if (dom != 0) {
707         fprintf(stderr, "No support for non-zero PCI domains\n");
708         return NULL;
709     }
710 
711     *devfnp = PCI_DEVFN(slot, 0);
712     return pci_find_bus_nr(root, bus);
713 }
714 
715 static void pci_init_cmask(PCIDevice *dev)
716 {
717     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
718     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
719     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
720     dev->cmask[PCI_REVISION_ID] = 0xff;
721     dev->cmask[PCI_CLASS_PROG] = 0xff;
722     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
723     dev->cmask[PCI_HEADER_TYPE] = 0xff;
724     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
725 }
726 
727 static void pci_init_wmask(PCIDevice *dev)
728 {
729     int config_size = pci_config_size(dev);
730 
731     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
732     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
733     pci_set_word(dev->wmask + PCI_COMMAND,
734                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
735                  PCI_COMMAND_INTX_DISABLE);
736     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
737         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
738     }
739 
740     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
741            config_size - PCI_CONFIG_HEADER_SIZE);
742 }
743 
744 static void pci_init_w1cmask(PCIDevice *dev)
745 {
746     /*
747      * Note: It's okay to set w1cmask even for readonly bits as
748      * long as their value is hardwired to 0.
749      */
750     pci_set_word(dev->w1cmask + PCI_STATUS,
751                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
752                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
753                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
754 }
755 
756 static void pci_init_mask_bridge(PCIDevice *d)
757 {
758     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
759        PCI_SEC_LETENCY_TIMER */
760     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
761 
762     /* base and limit */
763     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
764     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
765     pci_set_word(d->wmask + PCI_MEMORY_BASE,
766                  PCI_MEMORY_RANGE_MASK & 0xffff);
767     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
768                  PCI_MEMORY_RANGE_MASK & 0xffff);
769     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
770                  PCI_PREF_RANGE_MASK & 0xffff);
771     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
772                  PCI_PREF_RANGE_MASK & 0xffff);
773 
774     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
775     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
776 
777     /* Supported memory and i/o types */
778     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
779     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
780     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
781                                PCI_PREF_RANGE_TYPE_64);
782     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
783                                PCI_PREF_RANGE_TYPE_64);
784 
785     /*
786      * TODO: Bridges default to 10-bit VGA decoding but we currently only
787      * implement 16-bit decoding (no alias support).
788      */
789     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
790                  PCI_BRIDGE_CTL_PARITY |
791                  PCI_BRIDGE_CTL_SERR |
792                  PCI_BRIDGE_CTL_ISA |
793                  PCI_BRIDGE_CTL_VGA |
794                  PCI_BRIDGE_CTL_VGA_16BIT |
795                  PCI_BRIDGE_CTL_MASTER_ABORT |
796                  PCI_BRIDGE_CTL_BUS_RESET |
797                  PCI_BRIDGE_CTL_FAST_BACK |
798                  PCI_BRIDGE_CTL_DISCARD |
799                  PCI_BRIDGE_CTL_SEC_DISCARD |
800                  PCI_BRIDGE_CTL_DISCARD_SERR);
801     /* Below does not do anything as we never set this bit, put here for
802      * completeness. */
803     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
804                  PCI_BRIDGE_CTL_DISCARD_STATUS);
805     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
806     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
807     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
808                                PCI_PREF_RANGE_TYPE_MASK);
809     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
810                                PCI_PREF_RANGE_TYPE_MASK);
811 }
812 
813 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
814 {
815     uint8_t slot = PCI_SLOT(dev->devfn);
816     uint8_t func;
817 
818     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
819         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
820     }
821 
822     /*
823      * multifunction bit is interpreted in two ways as follows.
824      *   - all functions must set the bit to 1.
825      *     Example: Intel X53
826      *   - function 0 must set the bit, but the rest function (> 0)
827      *     is allowed to leave the bit to 0.
828      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
829      *
830      * So OS (at least Linux) checks the bit of only function 0,
831      * and doesn't see the bit of function > 0.
832      *
833      * The below check allows both interpretation.
834      */
835     if (PCI_FUNC(dev->devfn)) {
836         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
837         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
838             /* function 0 should set multifunction bit */
839             error_setg(errp, "PCI: single function device can't be populated "
840                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
841             return;
842         }
843         return;
844     }
845 
846     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
847         return;
848     }
849     /* function 0 indicates single function, so function > 0 must be NULL */
850     for (func = 1; func < PCI_FUNC_MAX; ++func) {
851         if (bus->devices[PCI_DEVFN(slot, func)]) {
852             error_setg(errp, "PCI: %x.0 indicates single function, "
853                        "but %x.%x is already populated.",
854                        slot, slot, func);
855             return;
856         }
857     }
858 }
859 
860 static void pci_config_alloc(PCIDevice *pci_dev)
861 {
862     int config_size = pci_config_size(pci_dev);
863 
864     pci_dev->config = g_malloc0(config_size);
865     pci_dev->cmask = g_malloc0(config_size);
866     pci_dev->wmask = g_malloc0(config_size);
867     pci_dev->w1cmask = g_malloc0(config_size);
868     pci_dev->used = g_malloc0(config_size);
869 }
870 
871 static void pci_config_free(PCIDevice *pci_dev)
872 {
873     g_free(pci_dev->config);
874     g_free(pci_dev->cmask);
875     g_free(pci_dev->wmask);
876     g_free(pci_dev->w1cmask);
877     g_free(pci_dev->used);
878 }
879 
880 static void do_pci_unregister_device(PCIDevice *pci_dev)
881 {
882     pci_dev->bus->devices[pci_dev->devfn] = NULL;
883     pci_config_free(pci_dev);
884 
885     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
886         memory_region_del_subregion(&pci_dev->bus_master_container_region,
887                                     &pci_dev->bus_master_enable_region);
888     }
889     address_space_destroy(&pci_dev->bus_master_as);
890 }
891 
892 /* Extract PCIReqIDCache into BDF format */
893 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
894 {
895     uint8_t bus_n;
896     uint16_t result;
897 
898     switch (cache->type) {
899     case PCI_REQ_ID_BDF:
900         result = pci_get_bdf(cache->dev);
901         break;
902     case PCI_REQ_ID_SECONDARY_BUS:
903         bus_n = pci_bus_num(cache->dev->bus);
904         result = PCI_BUILD_BDF(bus_n, 0);
905         break;
906     default:
907         error_printf("Invalid PCI requester ID cache type: %d\n",
908                      cache->type);
909         exit(1);
910         break;
911     }
912 
913     return result;
914 }
915 
916 /* Parse bridges up to the root complex and return requester ID
917  * cache for specific device.  For full PCIe topology, the cache
918  * result would be exactly the same as getting BDF of the device.
919  * However, several tricks are required when system mixed up with
920  * legacy PCI devices and PCIe-to-PCI bridges.
921  *
922  * Here we cache the proxy device (and type) not requester ID since
923  * bus number might change from time to time.
924  */
925 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
926 {
927     PCIDevice *parent;
928     PCIReqIDCache cache = {
929         .dev = dev,
930         .type = PCI_REQ_ID_BDF,
931     };
932 
933     while (!pci_bus_is_root(dev->bus)) {
934         /* We are under PCI/PCIe bridges */
935         parent = dev->bus->parent_dev;
936         if (pci_is_express(parent)) {
937             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
938                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
939                  * override the requester ID using secondary bus
940                  * number of parent bridge with zeroed devfn
941                  * (pcie-to-pci bridge spec chap 2.3). */
942                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
943                 cache.dev = dev;
944             }
945         } else {
946             /* Legacy PCI, override requester ID with the bridge's
947              * BDF upstream.  When the root complex connects to
948              * legacy PCI devices (including buses), it can only
949              * obtain requester ID info from directly attached
950              * devices.  If devices are attached under bridges, only
951              * the requester ID of the bridge that is directly
952              * attached to the root complex can be recognized. */
953             cache.type = PCI_REQ_ID_BDF;
954             cache.dev = parent;
955         }
956         dev = parent;
957     }
958 
959     return cache;
960 }
961 
962 uint16_t pci_requester_id(PCIDevice *dev)
963 {
964     return pci_req_id_cache_extract(&dev->requester_id_cache);
965 }
966 
967 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
968 {
969     return !(bus->devices[devfn]);
970 }
971 
972 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
973 {
974     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
975 }
976 
977 /* -1 for devfn means auto assign */
978 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
979                                          const char *name, int devfn,
980                                          Error **errp)
981 {
982     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
983     PCIConfigReadFunc *config_read = pc->config_read;
984     PCIConfigWriteFunc *config_write = pc->config_write;
985     Error *local_err = NULL;
986     DeviceState *dev = DEVICE(pci_dev);
987 
988     pci_dev->bus = bus;
989     /* Only pci bridges can be attached to extra PCI root buses */
990     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
991         error_setg(errp,
992                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
993                     bus->parent_dev->name);
994         return NULL;
995     }
996 
997     if (devfn < 0) {
998         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
999             devfn += PCI_FUNC_MAX) {
1000             if (pci_bus_devfn_available(bus, devfn) &&
1001                    !pci_bus_devfn_reserved(bus, devfn)) {
1002                 goto found;
1003             }
1004         }
1005         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1006                    "or reserved", name);
1007         return NULL;
1008     found: ;
1009     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1010         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1011                    " reserved",
1012                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1013         return NULL;
1014     } else if (!pci_bus_devfn_available(bus, devfn)) {
1015         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1016                    " in use by %s",
1017                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1018                    bus->devices[devfn]->name);
1019         return NULL;
1020     } else if (dev->hotplugged &&
1021                pci_get_function_0(pci_dev)) {
1022         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1023                    " new func %s cannot be exposed to guest.",
1024                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1025                    pci_get_function_0(pci_dev)->name,
1026                    name);
1027 
1028        return NULL;
1029     }
1030 
1031     pci_dev->devfn = devfn;
1032     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1033     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1034 
1035     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1036                        "bus master container", UINT64_MAX);
1037     address_space_init(&pci_dev->bus_master_as,
1038                        &pci_dev->bus_master_container_region, pci_dev->name);
1039 
1040     if (qdev_hotplug) {
1041         pci_init_bus_master(pci_dev);
1042     }
1043     pci_dev->irq_state = 0;
1044     pci_config_alloc(pci_dev);
1045 
1046     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1047     pci_config_set_device_id(pci_dev->config, pc->device_id);
1048     pci_config_set_revision(pci_dev->config, pc->revision);
1049     pci_config_set_class(pci_dev->config, pc->class_id);
1050 
1051     if (!pc->is_bridge) {
1052         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1053             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1054                          pc->subsystem_vendor_id);
1055             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1056                          pc->subsystem_id);
1057         } else {
1058             pci_set_default_subsystem_id(pci_dev);
1059         }
1060     } else {
1061         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1062         assert(!pc->subsystem_vendor_id);
1063         assert(!pc->subsystem_id);
1064     }
1065     pci_init_cmask(pci_dev);
1066     pci_init_wmask(pci_dev);
1067     pci_init_w1cmask(pci_dev);
1068     if (pc->is_bridge) {
1069         pci_init_mask_bridge(pci_dev);
1070     }
1071     pci_init_multifunction(bus, pci_dev, &local_err);
1072     if (local_err) {
1073         error_propagate(errp, local_err);
1074         do_pci_unregister_device(pci_dev);
1075         return NULL;
1076     }
1077 
1078     if (!config_read)
1079         config_read = pci_default_read_config;
1080     if (!config_write)
1081         config_write = pci_default_write_config;
1082     pci_dev->config_read = config_read;
1083     pci_dev->config_write = config_write;
1084     bus->devices[devfn] = pci_dev;
1085     pci_dev->version_id = 2; /* Current pci device vmstate version */
1086     return pci_dev;
1087 }
1088 
1089 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1090 {
1091     PCIIORegion *r;
1092     int i;
1093 
1094     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1095         r = &pci_dev->io_regions[i];
1096         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1097             continue;
1098         memory_region_del_subregion(r->address_space, r->memory);
1099     }
1100 
1101     pci_unregister_vga(pci_dev);
1102 }
1103 
1104 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1105 {
1106     PCIDevice *pci_dev = PCI_DEVICE(dev);
1107     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1108 
1109     pci_unregister_io_regions(pci_dev);
1110     pci_del_option_rom(pci_dev);
1111 
1112     if (pc->exit) {
1113         pc->exit(pci_dev);
1114     }
1115 
1116     pci_device_deassert_intx(pci_dev);
1117     do_pci_unregister_device(pci_dev);
1118 }
1119 
1120 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1121                       uint8_t type, MemoryRegion *memory)
1122 {
1123     PCIIORegion *r;
1124     uint32_t addr; /* offset in pci config space */
1125     uint64_t wmask;
1126     pcibus_t size = memory_region_size(memory);
1127 
1128     assert(region_num >= 0);
1129     assert(region_num < PCI_NUM_REGIONS);
1130     if (size & (size-1)) {
1131         fprintf(stderr, "ERROR: PCI region size must be pow2 "
1132                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
1133         exit(1);
1134     }
1135 
1136     r = &pci_dev->io_regions[region_num];
1137     r->addr = PCI_BAR_UNMAPPED;
1138     r->size = size;
1139     r->type = type;
1140     r->memory = memory;
1141     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1142                         ? pci_dev->bus->address_space_io
1143                         : pci_dev->bus->address_space_mem;
1144 
1145     wmask = ~(size - 1);
1146     if (region_num == PCI_ROM_SLOT) {
1147         /* ROM enable bit is writable */
1148         wmask |= PCI_ROM_ADDRESS_ENABLE;
1149     }
1150 
1151     addr = pci_bar(pci_dev, region_num);
1152     pci_set_long(pci_dev->config + addr, type);
1153 
1154     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1155         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1156         pci_set_quad(pci_dev->wmask + addr, wmask);
1157         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1158     } else {
1159         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1160         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1161     }
1162 }
1163 
1164 static void pci_update_vga(PCIDevice *pci_dev)
1165 {
1166     uint16_t cmd;
1167 
1168     if (!pci_dev->has_vga) {
1169         return;
1170     }
1171 
1172     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1173 
1174     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1175                               cmd & PCI_COMMAND_MEMORY);
1176     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1177                               cmd & PCI_COMMAND_IO);
1178     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1179                               cmd & PCI_COMMAND_IO);
1180 }
1181 
1182 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1183                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1184 {
1185     assert(!pci_dev->has_vga);
1186 
1187     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1188     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1189     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
1190                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1191 
1192     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1193     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1194     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1195                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1196 
1197     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1198     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1199     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1200                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1201     pci_dev->has_vga = true;
1202 
1203     pci_update_vga(pci_dev);
1204 }
1205 
1206 void pci_unregister_vga(PCIDevice *pci_dev)
1207 {
1208     if (!pci_dev->has_vga) {
1209         return;
1210     }
1211 
1212     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1213                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1214     memory_region_del_subregion(pci_dev->bus->address_space_io,
1215                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1216     memory_region_del_subregion(pci_dev->bus->address_space_io,
1217                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1218     pci_dev->has_vga = false;
1219 }
1220 
1221 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1222 {
1223     return pci_dev->io_regions[region_num].addr;
1224 }
1225 
1226 static pcibus_t pci_bar_address(PCIDevice *d,
1227 				int reg, uint8_t type, pcibus_t size)
1228 {
1229     pcibus_t new_addr, last_addr;
1230     int bar = pci_bar(d, reg);
1231     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1232     Object *machine = qdev_get_machine();
1233     ObjectClass *oc = object_get_class(machine);
1234     MachineClass *mc = MACHINE_CLASS(oc);
1235     bool allow_0_address = mc->pci_allow_0_address;
1236 
1237     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1238         if (!(cmd & PCI_COMMAND_IO)) {
1239             return PCI_BAR_UNMAPPED;
1240         }
1241         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1242         last_addr = new_addr + size - 1;
1243         /* Check if 32 bit BAR wraps around explicitly.
1244          * TODO: make priorities correct and remove this work around.
1245          */
1246         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1247             (!allow_0_address && new_addr == 0)) {
1248             return PCI_BAR_UNMAPPED;
1249         }
1250         return new_addr;
1251     }
1252 
1253     if (!(cmd & PCI_COMMAND_MEMORY)) {
1254         return PCI_BAR_UNMAPPED;
1255     }
1256     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1257         new_addr = pci_get_quad(d->config + bar);
1258     } else {
1259         new_addr = pci_get_long(d->config + bar);
1260     }
1261     /* the ROM slot has a specific enable bit */
1262     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1263         return PCI_BAR_UNMAPPED;
1264     }
1265     new_addr &= ~(size - 1);
1266     last_addr = new_addr + size - 1;
1267     /* NOTE: we do not support wrapping */
1268     /* XXX: as we cannot support really dynamic
1269        mappings, we handle specific values as invalid
1270        mappings. */
1271     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1272         (!allow_0_address && new_addr == 0)) {
1273         return PCI_BAR_UNMAPPED;
1274     }
1275 
1276     /* Now pcibus_t is 64bit.
1277      * Check if 32 bit BAR wraps around explicitly.
1278      * Without this, PC ide doesn't work well.
1279      * TODO: remove this work around.
1280      */
1281     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1282         return PCI_BAR_UNMAPPED;
1283     }
1284 
1285     /*
1286      * OS is allowed to set BAR beyond its addressable
1287      * bits. For example, 32 bit OS can set 64bit bar
1288      * to >4G. Check it. TODO: we might need to support
1289      * it in the future for e.g. PAE.
1290      */
1291     if (last_addr >= HWADDR_MAX) {
1292         return PCI_BAR_UNMAPPED;
1293     }
1294 
1295     return new_addr;
1296 }
1297 
1298 static void pci_update_mappings(PCIDevice *d)
1299 {
1300     PCIIORegion *r;
1301     int i;
1302     pcibus_t new_addr;
1303 
1304     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1305         r = &d->io_regions[i];
1306 
1307         /* this region isn't registered */
1308         if (!r->size)
1309             continue;
1310 
1311         new_addr = pci_bar_address(d, i, r->type, r->size);
1312 
1313         /* This bar isn't changed */
1314         if (new_addr == r->addr)
1315             continue;
1316 
1317         /* now do the real mapping */
1318         if (r->addr != PCI_BAR_UNMAPPED) {
1319             trace_pci_update_mappings_del(d, pci_bus_num(d->bus),
1320                                           PCI_SLOT(d->devfn),
1321                                           PCI_FUNC(d->devfn),
1322                                           i, r->addr, r->size);
1323             memory_region_del_subregion(r->address_space, r->memory);
1324         }
1325         r->addr = new_addr;
1326         if (r->addr != PCI_BAR_UNMAPPED) {
1327             trace_pci_update_mappings_add(d, pci_bus_num(d->bus),
1328                                           PCI_SLOT(d->devfn),
1329                                           PCI_FUNC(d->devfn),
1330                                           i, r->addr, r->size);
1331             memory_region_add_subregion_overlap(r->address_space,
1332                                                 r->addr, r->memory, 1);
1333         }
1334     }
1335 
1336     pci_update_vga(d);
1337 }
1338 
1339 static inline int pci_irq_disabled(PCIDevice *d)
1340 {
1341     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1342 }
1343 
1344 /* Called after interrupt disabled field update in config space,
1345  * assert/deassert interrupts if necessary.
1346  * Gets original interrupt disable bit value (before update). */
1347 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1348 {
1349     int i, disabled = pci_irq_disabled(d);
1350     if (disabled == was_irq_disabled)
1351         return;
1352     for (i = 0; i < PCI_NUM_PINS; ++i) {
1353         int state = pci_irq_state(d, i);
1354         pci_change_irq_level(d, i, disabled ? -state : state);
1355     }
1356 }
1357 
1358 uint32_t pci_default_read_config(PCIDevice *d,
1359                                  uint32_t address, int len)
1360 {
1361     uint32_t val = 0;
1362 
1363     memcpy(&val, d->config + address, len);
1364     return le32_to_cpu(val);
1365 }
1366 
1367 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1368 {
1369     int i, was_irq_disabled = pci_irq_disabled(d);
1370     uint32_t val = val_in;
1371 
1372     for (i = 0; i < l; val >>= 8, ++i) {
1373         uint8_t wmask = d->wmask[addr + i];
1374         uint8_t w1cmask = d->w1cmask[addr + i];
1375         assert(!(wmask & w1cmask));
1376         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1377         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1378     }
1379     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1380         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1381         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1382         range_covers_byte(addr, l, PCI_COMMAND))
1383         pci_update_mappings(d);
1384 
1385     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1386         pci_update_irq_disabled(d, was_irq_disabled);
1387         memory_region_set_enabled(&d->bus_master_enable_region,
1388                                   pci_get_word(d->config + PCI_COMMAND)
1389                                     & PCI_COMMAND_MASTER);
1390     }
1391 
1392     msi_write_config(d, addr, val_in, l);
1393     msix_write_config(d, addr, val_in, l);
1394 }
1395 
1396 /***********************************************************/
1397 /* generic PCI irq support */
1398 
1399 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1400 static void pci_irq_handler(void *opaque, int irq_num, int level)
1401 {
1402     PCIDevice *pci_dev = opaque;
1403     int change;
1404 
1405     change = level - pci_irq_state(pci_dev, irq_num);
1406     if (!change)
1407         return;
1408 
1409     pci_set_irq_state(pci_dev, irq_num, level);
1410     pci_update_irq_status(pci_dev);
1411     if (pci_irq_disabled(pci_dev))
1412         return;
1413     pci_change_irq_level(pci_dev, irq_num, change);
1414 }
1415 
1416 static inline int pci_intx(PCIDevice *pci_dev)
1417 {
1418     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1419 }
1420 
1421 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1422 {
1423     int intx = pci_intx(pci_dev);
1424 
1425     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1426 }
1427 
1428 void pci_set_irq(PCIDevice *pci_dev, int level)
1429 {
1430     int intx = pci_intx(pci_dev);
1431     pci_irq_handler(pci_dev, intx, level);
1432 }
1433 
1434 /* Special hooks used by device assignment */
1435 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1436 {
1437     assert(pci_bus_is_root(bus));
1438     bus->route_intx_to_irq = route_intx_to_irq;
1439 }
1440 
1441 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1442 {
1443     PCIBus *bus;
1444 
1445     do {
1446          bus = dev->bus;
1447          pin = bus->map_irq(dev, pin);
1448          dev = bus->parent_dev;
1449     } while (dev);
1450 
1451     if (!bus->route_intx_to_irq) {
1452         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1453                      object_get_typename(OBJECT(bus->qbus.parent)));
1454         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1455     }
1456 
1457     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1458 }
1459 
1460 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1461 {
1462     return old->mode != new->mode || old->irq != new->irq;
1463 }
1464 
1465 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1466 {
1467     PCIDevice *dev;
1468     PCIBus *sec;
1469     int i;
1470 
1471     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1472         dev = bus->devices[i];
1473         if (dev && dev->intx_routing_notifier) {
1474             dev->intx_routing_notifier(dev);
1475         }
1476     }
1477 
1478     QLIST_FOREACH(sec, &bus->child, sibling) {
1479         pci_bus_fire_intx_routing_notifier(sec);
1480     }
1481 }
1482 
1483 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1484                                           PCIINTxRoutingNotifier notifier)
1485 {
1486     dev->intx_routing_notifier = notifier;
1487 }
1488 
1489 /*
1490  * PCI-to-PCI bridge specification
1491  * 9.1: Interrupt routing. Table 9-1
1492  *
1493  * the PCI Express Base Specification, Revision 2.1
1494  * 2.2.8.1: INTx interrutp signaling - Rules
1495  *          the Implementation Note
1496  *          Table 2-20
1497  */
1498 /*
1499  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1500  * 0-origin unlike PCI interrupt pin register.
1501  */
1502 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1503 {
1504     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1505 }
1506 
1507 /***********************************************************/
1508 /* monitor info on PCI */
1509 
1510 typedef struct {
1511     uint16_t class;
1512     const char *desc;
1513     const char *fw_name;
1514     uint16_t fw_ign_bits;
1515 } pci_class_desc;
1516 
1517 static const pci_class_desc pci_class_descriptions[] =
1518 {
1519     { 0x0001, "VGA controller", "display"},
1520     { 0x0100, "SCSI controller", "scsi"},
1521     { 0x0101, "IDE controller", "ide"},
1522     { 0x0102, "Floppy controller", "fdc"},
1523     { 0x0103, "IPI controller", "ipi"},
1524     { 0x0104, "RAID controller", "raid"},
1525     { 0x0106, "SATA controller"},
1526     { 0x0107, "SAS controller"},
1527     { 0x0180, "Storage controller"},
1528     { 0x0200, "Ethernet controller", "ethernet"},
1529     { 0x0201, "Token Ring controller", "token-ring"},
1530     { 0x0202, "FDDI controller", "fddi"},
1531     { 0x0203, "ATM controller", "atm"},
1532     { 0x0280, "Network controller"},
1533     { 0x0300, "VGA controller", "display", 0x00ff},
1534     { 0x0301, "XGA controller"},
1535     { 0x0302, "3D controller"},
1536     { 0x0380, "Display controller"},
1537     { 0x0400, "Video controller", "video"},
1538     { 0x0401, "Audio controller", "sound"},
1539     { 0x0402, "Phone"},
1540     { 0x0403, "Audio controller", "sound"},
1541     { 0x0480, "Multimedia controller"},
1542     { 0x0500, "RAM controller", "memory"},
1543     { 0x0501, "Flash controller", "flash"},
1544     { 0x0580, "Memory controller"},
1545     { 0x0600, "Host bridge", "host"},
1546     { 0x0601, "ISA bridge", "isa"},
1547     { 0x0602, "EISA bridge", "eisa"},
1548     { 0x0603, "MC bridge", "mca"},
1549     { 0x0604, "PCI bridge", "pci-bridge"},
1550     { 0x0605, "PCMCIA bridge", "pcmcia"},
1551     { 0x0606, "NUBUS bridge", "nubus"},
1552     { 0x0607, "CARDBUS bridge", "cardbus"},
1553     { 0x0608, "RACEWAY bridge"},
1554     { 0x0680, "Bridge"},
1555     { 0x0700, "Serial port", "serial"},
1556     { 0x0701, "Parallel port", "parallel"},
1557     { 0x0800, "Interrupt controller", "interrupt-controller"},
1558     { 0x0801, "DMA controller", "dma-controller"},
1559     { 0x0802, "Timer", "timer"},
1560     { 0x0803, "RTC", "rtc"},
1561     { 0x0900, "Keyboard", "keyboard"},
1562     { 0x0901, "Pen", "pen"},
1563     { 0x0902, "Mouse", "mouse"},
1564     { 0x0A00, "Dock station", "dock", 0x00ff},
1565     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1566     { 0x0c00, "Fireware contorller", "fireware"},
1567     { 0x0c01, "Access bus controller", "access-bus"},
1568     { 0x0c02, "SSA controller", "ssa"},
1569     { 0x0c03, "USB controller", "usb"},
1570     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1571     { 0x0c05, "SMBus"},
1572     { 0, NULL}
1573 };
1574 
1575 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1576                                                   void (*fn)(PCIBus *b,
1577                                                              PCIDevice *d,
1578                                                              void *opaque),
1579                                                   void *opaque)
1580 {
1581     PCIDevice *d;
1582     int devfn;
1583 
1584     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1585         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1586         if (d) {
1587             fn(bus, d, opaque);
1588         }
1589     }
1590 }
1591 
1592 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1593                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1594                          void *opaque)
1595 {
1596     bus = pci_find_bus_nr(bus, bus_num);
1597 
1598     if (bus) {
1599         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1600     }
1601 }
1602 
1603 static void pci_for_each_device_under_bus(PCIBus *bus,
1604                                           void (*fn)(PCIBus *b, PCIDevice *d,
1605                                                      void *opaque),
1606                                           void *opaque)
1607 {
1608     PCIDevice *d;
1609     int devfn;
1610 
1611     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1612         d = bus->devices[devfn];
1613         if (d) {
1614             fn(bus, d, opaque);
1615         }
1616     }
1617 }
1618 
1619 void pci_for_each_device(PCIBus *bus, int bus_num,
1620                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1621                          void *opaque)
1622 {
1623     bus = pci_find_bus_nr(bus, bus_num);
1624 
1625     if (bus) {
1626         pci_for_each_device_under_bus(bus, fn, opaque);
1627     }
1628 }
1629 
1630 static const pci_class_desc *get_class_desc(int class)
1631 {
1632     const pci_class_desc *desc;
1633 
1634     desc = pci_class_descriptions;
1635     while (desc->desc && class != desc->class) {
1636         desc++;
1637     }
1638 
1639     return desc;
1640 }
1641 
1642 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1643 
1644 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1645 {
1646     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1647     int i;
1648 
1649     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1650         const PCIIORegion *r = &dev->io_regions[i];
1651         PciMemoryRegionList *region;
1652 
1653         if (!r->size) {
1654             continue;
1655         }
1656 
1657         region = g_malloc0(sizeof(*region));
1658         region->value = g_malloc0(sizeof(*region->value));
1659 
1660         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1661             region->value->type = g_strdup("io");
1662         } else {
1663             region->value->type = g_strdup("memory");
1664             region->value->has_prefetch = true;
1665             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1666             region->value->has_mem_type_64 = true;
1667             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1668         }
1669 
1670         region->value->bar = i;
1671         region->value->address = r->addr;
1672         region->value->size = r->size;
1673 
1674         /* XXX: waiting for the qapi to support GSList */
1675         if (!cur_item) {
1676             head = cur_item = region;
1677         } else {
1678             cur_item->next = region;
1679             cur_item = region;
1680         }
1681     }
1682 
1683     return head;
1684 }
1685 
1686 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1687                                            int bus_num)
1688 {
1689     PciBridgeInfo *info;
1690     PciMemoryRange *range;
1691 
1692     info = g_new0(PciBridgeInfo, 1);
1693 
1694     info->bus = g_new0(PciBusInfo, 1);
1695     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1696     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1697     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1698 
1699     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1700     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1701     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1702 
1703     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1704     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1705     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1706 
1707     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1708     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1709     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1710 
1711     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1712         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1713         if (child_bus) {
1714             info->has_devices = true;
1715             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1716         }
1717     }
1718 
1719     return info;
1720 }
1721 
1722 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1723                                            int bus_num)
1724 {
1725     const pci_class_desc *desc;
1726     PciDeviceInfo *info;
1727     uint8_t type;
1728     int class;
1729 
1730     info = g_new0(PciDeviceInfo, 1);
1731     info->bus = bus_num;
1732     info->slot = PCI_SLOT(dev->devfn);
1733     info->function = PCI_FUNC(dev->devfn);
1734 
1735     info->class_info = g_new0(PciDeviceClass, 1);
1736     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1737     info->class_info->q_class = class;
1738     desc = get_class_desc(class);
1739     if (desc->desc) {
1740         info->class_info->has_desc = true;
1741         info->class_info->desc = g_strdup(desc->desc);
1742     }
1743 
1744     info->id = g_new0(PciDeviceId, 1);
1745     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1746     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1747     info->regions = qmp_query_pci_regions(dev);
1748     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1749 
1750     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1751         info->has_irq = true;
1752         info->irq = dev->config[PCI_INTERRUPT_LINE];
1753     }
1754 
1755     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1756     if (type == PCI_HEADER_TYPE_BRIDGE) {
1757         info->has_pci_bridge = true;
1758         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1759     }
1760 
1761     return info;
1762 }
1763 
1764 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1765 {
1766     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1767     PCIDevice *dev;
1768     int devfn;
1769 
1770     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1771         dev = bus->devices[devfn];
1772         if (dev) {
1773             info = g_malloc0(sizeof(*info));
1774             info->value = qmp_query_pci_device(dev, bus, bus_num);
1775 
1776             /* XXX: waiting for the qapi to support GSList */
1777             if (!cur_item) {
1778                 head = cur_item = info;
1779             } else {
1780                 cur_item->next = info;
1781                 cur_item = info;
1782             }
1783         }
1784     }
1785 
1786     return head;
1787 }
1788 
1789 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1790 {
1791     PciInfo *info = NULL;
1792 
1793     bus = pci_find_bus_nr(bus, bus_num);
1794     if (bus) {
1795         info = g_malloc0(sizeof(*info));
1796         info->bus = bus_num;
1797         info->devices = qmp_query_pci_devices(bus, bus_num);
1798     }
1799 
1800     return info;
1801 }
1802 
1803 PciInfoList *qmp_query_pci(Error **errp)
1804 {
1805     PciInfoList *info, *head = NULL, *cur_item = NULL;
1806     PCIHostState *host_bridge;
1807 
1808     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1809         info = g_malloc0(sizeof(*info));
1810         info->value = qmp_query_pci_bus(host_bridge->bus,
1811                                         pci_bus_num(host_bridge->bus));
1812 
1813         /* XXX: waiting for the qapi to support GSList */
1814         if (!cur_item) {
1815             head = cur_item = info;
1816         } else {
1817             cur_item->next = info;
1818             cur_item = info;
1819         }
1820     }
1821 
1822     return head;
1823 }
1824 
1825 static const char * const pci_nic_models[] = {
1826     "ne2k_pci",
1827     "i82551",
1828     "i82557b",
1829     "i82559er",
1830     "rtl8139",
1831     "e1000",
1832     "pcnet",
1833     "virtio",
1834     "sungem",
1835     NULL
1836 };
1837 
1838 static const char * const pci_nic_names[] = {
1839     "ne2k_pci",
1840     "i82551",
1841     "i82557b",
1842     "i82559er",
1843     "rtl8139",
1844     "e1000",
1845     "pcnet",
1846     "virtio-net-pci",
1847     "sungem",
1848     NULL
1849 };
1850 
1851 /* Initialize a PCI NIC.  */
1852 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1853                                const char *default_model,
1854                                const char *default_devaddr)
1855 {
1856     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1857     PCIBus *bus;
1858     PCIDevice *pci_dev;
1859     DeviceState *dev;
1860     int devfn;
1861     int i;
1862 
1863     if (qemu_show_nic_models(nd->model, pci_nic_models)) {
1864         exit(0);
1865     }
1866 
1867     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1868     if (i < 0) {
1869         exit(1);
1870     }
1871 
1872     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1873     if (!bus) {
1874         error_report("Invalid PCI device address %s for device %s",
1875                      devaddr, pci_nic_names[i]);
1876         exit(1);
1877     }
1878 
1879     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1880     dev = &pci_dev->qdev;
1881     qdev_set_nic_properties(dev, nd);
1882     qdev_init_nofail(dev);
1883 
1884     return pci_dev;
1885 }
1886 
1887 PCIDevice *pci_vga_init(PCIBus *bus)
1888 {
1889     switch (vga_interface_type) {
1890     case VGA_CIRRUS:
1891         return pci_create_simple(bus, -1, "cirrus-vga");
1892     case VGA_QXL:
1893         return pci_create_simple(bus, -1, "qxl-vga");
1894     case VGA_STD:
1895         return pci_create_simple(bus, -1, "VGA");
1896     case VGA_VMWARE:
1897         return pci_create_simple(bus, -1, "vmware-svga");
1898     case VGA_VIRTIO:
1899         return pci_create_simple(bus, -1, "virtio-vga");
1900     case VGA_NONE:
1901     default: /* Other non-PCI types. Checking for unsupported types is already
1902                 done in vl.c. */
1903         return NULL;
1904     }
1905 }
1906 
1907 /* Whether a given bus number is in range of the secondary
1908  * bus of the given bridge device. */
1909 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1910 {
1911     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1912              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1913         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1914         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1915 }
1916 
1917 /* Whether a given bus number is in a range of a root bus */
1918 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1919 {
1920     int i;
1921 
1922     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1923         PCIDevice *dev = bus->devices[i];
1924 
1925         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1926             if (pci_secondary_bus_in_range(dev, bus_num)) {
1927                 return true;
1928             }
1929         }
1930     }
1931 
1932     return false;
1933 }
1934 
1935 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1936 {
1937     PCIBus *sec;
1938 
1939     if (!bus) {
1940         return NULL;
1941     }
1942 
1943     if (pci_bus_num(bus) == bus_num) {
1944         return bus;
1945     }
1946 
1947     /* Consider all bus numbers in range for the host pci bridge. */
1948     if (!pci_bus_is_root(bus) &&
1949         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1950         return NULL;
1951     }
1952 
1953     /* try child bus */
1954     for (; bus; bus = sec) {
1955         QLIST_FOREACH(sec, &bus->child, sibling) {
1956             if (pci_bus_num(sec) == bus_num) {
1957                 return sec;
1958             }
1959             /* PXB buses assumed to be children of bus 0 */
1960             if (pci_bus_is_root(sec)) {
1961                 if (pci_root_bus_in_range(sec, bus_num)) {
1962                     break;
1963                 }
1964             } else {
1965                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1966                     break;
1967                 }
1968             }
1969         }
1970     }
1971 
1972     return NULL;
1973 }
1974 
1975 void pci_for_each_bus_depth_first(PCIBus *bus,
1976                                   void *(*begin)(PCIBus *bus, void *parent_state),
1977                                   void (*end)(PCIBus *bus, void *state),
1978                                   void *parent_state)
1979 {
1980     PCIBus *sec;
1981     void *state;
1982 
1983     if (!bus) {
1984         return;
1985     }
1986 
1987     if (begin) {
1988         state = begin(bus, parent_state);
1989     } else {
1990         state = parent_state;
1991     }
1992 
1993     QLIST_FOREACH(sec, &bus->child, sibling) {
1994         pci_for_each_bus_depth_first(sec, begin, end, state);
1995     }
1996 
1997     if (end) {
1998         end(bus, state);
1999     }
2000 }
2001 
2002 
2003 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2004 {
2005     bus = pci_find_bus_nr(bus, bus_num);
2006 
2007     if (!bus)
2008         return NULL;
2009 
2010     return bus->devices[devfn];
2011 }
2012 
2013 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2014 {
2015     PCIDevice *pci_dev = (PCIDevice *)qdev;
2016     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2017     Error *local_err = NULL;
2018     PCIBus *bus;
2019     bool is_default_rom;
2020 
2021     /* initialize cap_present for pci_is_express() and pci_config_size() */
2022     if (pc->is_express) {
2023         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2024     }
2025 
2026     bus = PCI_BUS(qdev_get_parent_bus(qdev));
2027     pci_dev = do_pci_register_device(pci_dev, bus,
2028                                      object_get_typename(OBJECT(qdev)),
2029                                      pci_dev->devfn, errp);
2030     if (pci_dev == NULL)
2031         return;
2032 
2033     if (pc->realize) {
2034         pc->realize(pci_dev, &local_err);
2035         if (local_err) {
2036             error_propagate(errp, local_err);
2037             do_pci_unregister_device(pci_dev);
2038             return;
2039         }
2040     }
2041 
2042     /* rom loading */
2043     is_default_rom = false;
2044     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2045         pci_dev->romfile = g_strdup(pc->romfile);
2046         is_default_rom = true;
2047     }
2048 
2049     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2050     if (local_err) {
2051         error_propagate(errp, local_err);
2052         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2053         return;
2054     }
2055 }
2056 
2057 static void pci_default_realize(PCIDevice *dev, Error **errp)
2058 {
2059     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2060 
2061     if (pc->init) {
2062         if (pc->init(dev) < 0) {
2063             error_setg(errp, "Device initialization failed");
2064             return;
2065         }
2066     }
2067 }
2068 
2069 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2070                                     const char *name)
2071 {
2072     DeviceState *dev;
2073 
2074     dev = qdev_create(&bus->qbus, name);
2075     qdev_prop_set_int32(dev, "addr", devfn);
2076     qdev_prop_set_bit(dev, "multifunction", multifunction);
2077     return PCI_DEVICE(dev);
2078 }
2079 
2080 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2081                                            bool multifunction,
2082                                            const char *name)
2083 {
2084     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2085     qdev_init_nofail(&dev->qdev);
2086     return dev;
2087 }
2088 
2089 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2090 {
2091     return pci_create_multifunction(bus, devfn, false, name);
2092 }
2093 
2094 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2095 {
2096     return pci_create_simple_multifunction(bus, devfn, false, name);
2097 }
2098 
2099 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2100 {
2101     int offset = PCI_CONFIG_HEADER_SIZE;
2102     int i;
2103     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2104         if (pdev->used[i])
2105             offset = i + 1;
2106         else if (i - offset + 1 == size)
2107             return offset;
2108     }
2109     return 0;
2110 }
2111 
2112 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2113                                         uint8_t *prev_p)
2114 {
2115     uint8_t next, prev;
2116 
2117     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2118         return 0;
2119 
2120     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2121          prev = next + PCI_CAP_LIST_NEXT)
2122         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2123             break;
2124 
2125     if (prev_p)
2126         *prev_p = prev;
2127     return next;
2128 }
2129 
2130 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2131 {
2132     uint8_t next, prev, found = 0;
2133 
2134     if (!(pdev->used[offset])) {
2135         return 0;
2136     }
2137 
2138     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2139 
2140     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2141          prev = next + PCI_CAP_LIST_NEXT) {
2142         if (next <= offset && next > found) {
2143             found = next;
2144         }
2145     }
2146     return found;
2147 }
2148 
2149 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2150    This is needed for an option rom which is used for more than one device. */
2151 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2152 {
2153     uint16_t vendor_id;
2154     uint16_t device_id;
2155     uint16_t rom_vendor_id;
2156     uint16_t rom_device_id;
2157     uint16_t rom_magic;
2158     uint16_t pcir_offset;
2159     uint8_t checksum;
2160 
2161     /* Words in rom data are little endian (like in PCI configuration),
2162        so they can be read / written with pci_get_word / pci_set_word. */
2163 
2164     /* Only a valid rom will be patched. */
2165     rom_magic = pci_get_word(ptr);
2166     if (rom_magic != 0xaa55) {
2167         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2168         return;
2169     }
2170     pcir_offset = pci_get_word(ptr + 0x18);
2171     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2172         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2173         return;
2174     }
2175 
2176     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2177     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2178     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2179     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2180 
2181     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2182                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2183 
2184     checksum = ptr[6];
2185 
2186     if (vendor_id != rom_vendor_id) {
2187         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2188         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2189         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2190         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2191         ptr[6] = checksum;
2192         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2193     }
2194 
2195     if (device_id != rom_device_id) {
2196         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2197         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2198         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2199         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2200         ptr[6] = checksum;
2201         pci_set_word(ptr + pcir_offset + 6, device_id);
2202     }
2203 }
2204 
2205 /* Add an option rom for the device */
2206 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2207                                Error **errp)
2208 {
2209     int size;
2210     char *path;
2211     void *ptr;
2212     char name[32];
2213     const VMStateDescription *vmsd;
2214 
2215     if (!pdev->romfile)
2216         return;
2217     if (strlen(pdev->romfile) == 0)
2218         return;
2219 
2220     if (!pdev->rom_bar) {
2221         /*
2222          * Load rom via fw_cfg instead of creating a rom bar,
2223          * for 0.11 compatibility.
2224          */
2225         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2226 
2227         /*
2228          * Hot-plugged devices can't use the option ROM
2229          * if the rom bar is disabled.
2230          */
2231         if (DEVICE(pdev)->hotplugged) {
2232             error_setg(errp, "Hot-plugged device without ROM bar"
2233                        " can't have an option ROM");
2234             return;
2235         }
2236 
2237         if (class == 0x0300) {
2238             rom_add_vga(pdev->romfile);
2239         } else {
2240             rom_add_option(pdev->romfile, -1);
2241         }
2242         return;
2243     }
2244 
2245     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2246     if (path == NULL) {
2247         path = g_strdup(pdev->romfile);
2248     }
2249 
2250     size = get_image_size(path);
2251     if (size < 0) {
2252         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2253         g_free(path);
2254         return;
2255     } else if (size == 0) {
2256         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2257         g_free(path);
2258         return;
2259     }
2260     size = pow2ceil(size);
2261 
2262     vmsd = qdev_get_vmsd(DEVICE(pdev));
2263 
2264     if (vmsd) {
2265         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2266     } else {
2267         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2268     }
2269     pdev->has_rom = true;
2270     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2271     ptr = memory_region_get_ram_ptr(&pdev->rom);
2272     load_image(path, ptr);
2273     g_free(path);
2274 
2275     if (is_default_rom) {
2276         /* Only the default rom images will be patched (if needed). */
2277         pci_patch_ids(pdev, ptr, size);
2278     }
2279 
2280     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2281 }
2282 
2283 static void pci_del_option_rom(PCIDevice *pdev)
2284 {
2285     if (!pdev->has_rom)
2286         return;
2287 
2288     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2289     pdev->has_rom = false;
2290 }
2291 
2292 /*
2293  * On success, pci_add_capability() returns a positive value
2294  * that the offset of the pci capability.
2295  * On failure, it sets an error and returns a negative error
2296  * code.
2297  */
2298 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2299                        uint8_t offset, uint8_t size,
2300                        Error **errp)
2301 {
2302     uint8_t *config;
2303     int i, overlapping_cap;
2304 
2305     if (!offset) {
2306         offset = pci_find_space(pdev, size);
2307         /* out of PCI config space is programming error */
2308         assert(offset);
2309     } else {
2310         /* Verify that capabilities don't overlap.  Note: device assignment
2311          * depends on this check to verify that the device is not broken.
2312          * Should never trigger for emulated devices, but it's helpful
2313          * for debugging these. */
2314         for (i = offset; i < offset + size; i++) {
2315             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2316             if (overlapping_cap) {
2317                 error_setg(errp, "%s:%02x:%02x.%x "
2318                            "Attempt to add PCI capability %x at offset "
2319                            "%x overlaps existing capability %x at offset %x",
2320                            pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2321                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2322                            cap_id, offset, overlapping_cap, i);
2323                 return -EINVAL;
2324             }
2325         }
2326     }
2327 
2328     config = pdev->config + offset;
2329     config[PCI_CAP_LIST_ID] = cap_id;
2330     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2331     pdev->config[PCI_CAPABILITY_LIST] = offset;
2332     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2333     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2334     /* Make capability read-only by default */
2335     memset(pdev->wmask + offset, 0, size);
2336     /* Check capability by default */
2337     memset(pdev->cmask + offset, 0xFF, size);
2338     return offset;
2339 }
2340 
2341 /* Unlink capability from the pci config space. */
2342 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2343 {
2344     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2345     if (!offset)
2346         return;
2347     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2348     /* Make capability writable again */
2349     memset(pdev->wmask + offset, 0xff, size);
2350     memset(pdev->w1cmask + offset, 0, size);
2351     /* Clear cmask as device-specific registers can't be checked */
2352     memset(pdev->cmask + offset, 0, size);
2353     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2354 
2355     if (!pdev->config[PCI_CAPABILITY_LIST])
2356         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2357 }
2358 
2359 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2360 {
2361     return pci_find_capability_list(pdev, cap_id, NULL);
2362 }
2363 
2364 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2365 {
2366     PCIDevice *d = (PCIDevice *)dev;
2367     const pci_class_desc *desc;
2368     char ctxt[64];
2369     PCIIORegion *r;
2370     int i, class;
2371 
2372     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2373     desc = pci_class_descriptions;
2374     while (desc->desc && class != desc->class)
2375         desc++;
2376     if (desc->desc) {
2377         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2378     } else {
2379         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2380     }
2381 
2382     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2383                    "pci id %04x:%04x (sub %04x:%04x)\n",
2384                    indent, "", ctxt, pci_bus_num(d->bus),
2385                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2386                    pci_get_word(d->config + PCI_VENDOR_ID),
2387                    pci_get_word(d->config + PCI_DEVICE_ID),
2388                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2389                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2390     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2391         r = &d->io_regions[i];
2392         if (!r->size)
2393             continue;
2394         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2395                        " [0x%"FMT_PCIBUS"]\n",
2396                        indent, "",
2397                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2398                        r->addr, r->addr + r->size - 1);
2399     }
2400 }
2401 
2402 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2403 {
2404     PCIDevice *d = (PCIDevice *)dev;
2405     const char *name = NULL;
2406     const pci_class_desc *desc =  pci_class_descriptions;
2407     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2408 
2409     while (desc->desc &&
2410           (class & ~desc->fw_ign_bits) !=
2411           (desc->class & ~desc->fw_ign_bits)) {
2412         desc++;
2413     }
2414 
2415     if (desc->desc) {
2416         name = desc->fw_name;
2417     }
2418 
2419     if (name) {
2420         pstrcpy(buf, len, name);
2421     } else {
2422         snprintf(buf, len, "pci%04x,%04x",
2423                  pci_get_word(d->config + PCI_VENDOR_ID),
2424                  pci_get_word(d->config + PCI_DEVICE_ID));
2425     }
2426 
2427     return buf;
2428 }
2429 
2430 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2431 {
2432     PCIDevice *d = (PCIDevice *)dev;
2433     char path[50], name[33];
2434     int off;
2435 
2436     off = snprintf(path, sizeof(path), "%s@%x",
2437                    pci_dev_fw_name(dev, name, sizeof name),
2438                    PCI_SLOT(d->devfn));
2439     if (PCI_FUNC(d->devfn))
2440         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2441     return g_strdup(path);
2442 }
2443 
2444 static char *pcibus_get_dev_path(DeviceState *dev)
2445 {
2446     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2447     PCIDevice *t;
2448     int slot_depth;
2449     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2450      * 00 is added here to make this format compatible with
2451      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2452      * Slot.Function list specifies the slot and function numbers for all
2453      * devices on the path from root to the specific device. */
2454     const char *root_bus_path;
2455     int root_bus_len;
2456     char slot[] = ":SS.F";
2457     int slot_len = sizeof slot - 1 /* For '\0' */;
2458     int path_len;
2459     char *path, *p;
2460     int s;
2461 
2462     root_bus_path = pci_root_bus_path(d);
2463     root_bus_len = strlen(root_bus_path);
2464 
2465     /* Calculate # of slots on path between device and root. */;
2466     slot_depth = 0;
2467     for (t = d; t; t = t->bus->parent_dev) {
2468         ++slot_depth;
2469     }
2470 
2471     path_len = root_bus_len + slot_len * slot_depth;
2472 
2473     /* Allocate memory, fill in the terminating null byte. */
2474     path = g_malloc(path_len + 1 /* For '\0' */);
2475     path[path_len] = '\0';
2476 
2477     memcpy(path, root_bus_path, root_bus_len);
2478 
2479     /* Fill in slot numbers. We walk up from device to root, so need to print
2480      * them in the reverse order, last to first. */
2481     p = path + path_len;
2482     for (t = d; t; t = t->bus->parent_dev) {
2483         p -= slot_len;
2484         s = snprintf(slot, sizeof slot, ":%02x.%x",
2485                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2486         assert(s == slot_len);
2487         memcpy(p, slot, slot_len);
2488     }
2489 
2490     return path;
2491 }
2492 
2493 static int pci_qdev_find_recursive(PCIBus *bus,
2494                                    const char *id, PCIDevice **pdev)
2495 {
2496     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2497     if (!qdev) {
2498         return -ENODEV;
2499     }
2500 
2501     /* roughly check if given qdev is pci device */
2502     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2503         *pdev = PCI_DEVICE(qdev);
2504         return 0;
2505     }
2506     return -EINVAL;
2507 }
2508 
2509 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2510 {
2511     PCIHostState *host_bridge;
2512     int rc = -ENODEV;
2513 
2514     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2515         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2516         if (!tmp) {
2517             rc = 0;
2518             break;
2519         }
2520         if (tmp != -ENODEV) {
2521             rc = tmp;
2522         }
2523     }
2524 
2525     return rc;
2526 }
2527 
2528 MemoryRegion *pci_address_space(PCIDevice *dev)
2529 {
2530     return dev->bus->address_space_mem;
2531 }
2532 
2533 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2534 {
2535     return dev->bus->address_space_io;
2536 }
2537 
2538 static void pci_device_class_init(ObjectClass *klass, void *data)
2539 {
2540     DeviceClass *k = DEVICE_CLASS(klass);
2541     PCIDeviceClass *pc = PCI_DEVICE_CLASS(klass);
2542 
2543     k->realize = pci_qdev_realize;
2544     k->unrealize = pci_qdev_unrealize;
2545     k->bus_type = TYPE_PCI_BUS;
2546     k->props = pci_props;
2547     pc->realize = pci_default_realize;
2548 }
2549 
2550 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2551 {
2552     if (!object_class_is_abstract(klass)) {
2553         ObjectClass *conventional =
2554             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2555         ObjectClass *pcie =
2556             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2557         assert(conventional || pcie);
2558     }
2559 }
2560 
2561 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2562 {
2563     PCIBus *bus = PCI_BUS(dev->bus);
2564     PCIBus *iommu_bus = bus;
2565 
2566     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2567         iommu_bus = PCI_BUS(iommu_bus->parent_dev->bus);
2568     }
2569     if (iommu_bus && iommu_bus->iommu_fn) {
2570         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2571     }
2572     return &address_space_memory;
2573 }
2574 
2575 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2576 {
2577     bus->iommu_fn = fn;
2578     bus->iommu_opaque = opaque;
2579 }
2580 
2581 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2582 {
2583     Range *range = opaque;
2584     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2585     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2586     int i;
2587 
2588     if (!(cmd & PCI_COMMAND_MEMORY)) {
2589         return;
2590     }
2591 
2592     if (pc->is_bridge) {
2593         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2594         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2595 
2596         base = MAX(base, 0x1ULL << 32);
2597 
2598         if (limit >= base) {
2599             Range pref_range;
2600             range_set_bounds(&pref_range, base, limit);
2601             range_extend(range, &pref_range);
2602         }
2603     }
2604     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2605         PCIIORegion *r = &dev->io_regions[i];
2606         pcibus_t lob, upb;
2607         Range region_range;
2608 
2609         if (!r->size ||
2610             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2611             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2612             continue;
2613         }
2614 
2615         lob = pci_bar_address(dev, i, r->type, r->size);
2616         upb = lob + r->size - 1;
2617         if (lob == PCI_BAR_UNMAPPED) {
2618             continue;
2619         }
2620 
2621         lob = MAX(lob, 0x1ULL << 32);
2622 
2623         if (upb >= lob) {
2624             range_set_bounds(&region_range, lob, upb);
2625             range_extend(range, &region_range);
2626         }
2627     }
2628 }
2629 
2630 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2631 {
2632     range_make_empty(range);
2633     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2634 }
2635 
2636 static bool pcie_has_upstream_port(PCIDevice *dev)
2637 {
2638     PCIDevice *parent_dev = pci_bridge_get_device(dev->bus);
2639 
2640     /* Device associated with an upstream port.
2641      * As there are several types of these, it's easier to check the
2642      * parent device: upstream ports are always connected to
2643      * root or downstream ports.
2644      */
2645     return parent_dev &&
2646         pci_is_express(parent_dev) &&
2647         parent_dev->exp.exp_cap &&
2648         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2649          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2650 }
2651 
2652 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2653 {
2654     if(pcie_has_upstream_port(pci_dev)) {
2655         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2656         return pci_dev->bus->devices[0];
2657     } else {
2658         /* Other bus types might support multiple devices at slots 0-31 */
2659         return pci_dev->bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2660     }
2661 }
2662 
2663 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2664 {
2665     MSIMessage msg;
2666     if (msix_enabled(dev)) {
2667         msg = msix_get_message(dev, vector);
2668     } else if (msi_enabled(dev)) {
2669         msg = msi_get_message(dev, vector);
2670     } else {
2671         /* Should never happen */
2672         error_report("%s: unknown interrupt type", __func__);
2673         abort();
2674     }
2675     return msg;
2676 }
2677 
2678 static const TypeInfo pci_device_type_info = {
2679     .name = TYPE_PCI_DEVICE,
2680     .parent = TYPE_DEVICE,
2681     .instance_size = sizeof(PCIDevice),
2682     .abstract = true,
2683     .class_size = sizeof(PCIDeviceClass),
2684     .class_init = pci_device_class_init,
2685     .class_base_init = pci_device_class_base_init,
2686 };
2687 
2688 static void pci_register_types(void)
2689 {
2690     type_register_static(&pci_bus_info);
2691     type_register_static(&pcie_bus_info);
2692     type_register_static(&conventional_pci_interface_info);
2693     type_register_static(&pcie_interface_info);
2694     type_register_static(&pci_device_type_info);
2695 }
2696 
2697 type_init(pci_register_types)
2698