xref: /openbmc/qemu/hw/pci/pci.c (revision 40d6ee94)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "hw/hw.h"
27 #include "hw/pci/pci.h"
28 #include "hw/pci/pci_bridge.h"
29 #include "hw/pci/pci_bus.h"
30 #include "hw/pci/pci_host.h"
31 #include "monitor/monitor.h"
32 #include "net/net.h"
33 #include "sysemu/sysemu.h"
34 #include "hw/loader.h"
35 #include "qemu/error-report.h"
36 #include "qemu/range.h"
37 #include "trace.h"
38 #include "hw/pci/msi.h"
39 #include "hw/pci/msix.h"
40 #include "exec/address-spaces.h"
41 #include "hw/hotplug.h"
42 #include "hw/boards.h"
43 #include "qapi/error.h"
44 #include "qapi/qapi-commands-misc.h"
45 #include "qemu/cutils.h"
46 
47 //#define DEBUG_PCI
48 #ifdef DEBUG_PCI
49 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
50 #else
51 # define PCI_DPRINTF(format, ...)       do { } while (0)
52 #endif
53 
54 bool pci_available = true;
55 
56 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
57 static char *pcibus_get_dev_path(DeviceState *dev);
58 static char *pcibus_get_fw_dev_path(DeviceState *dev);
59 static void pcibus_reset(BusState *qbus);
60 
61 static Property pci_props[] = {
62     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
63     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
64     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
65     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
66                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
67     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
68                     QEMU_PCI_CAP_SERR_BITNR, true),
69     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
70                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
71     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
72                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
73     DEFINE_PROP_END_OF_LIST()
74 };
75 
76 static const VMStateDescription vmstate_pcibus = {
77     .name = "PCIBUS",
78     .version_id = 1,
79     .minimum_version_id = 1,
80     .fields = (VMStateField[]) {
81         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
82         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
83                              nirq, 0, vmstate_info_int32,
84                              int32_t),
85         VMSTATE_END_OF_LIST()
86     }
87 };
88 
89 static void pci_init_bus_master(PCIDevice *pci_dev)
90 {
91     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
92 
93     memory_region_init_alias(&pci_dev->bus_master_enable_region,
94                              OBJECT(pci_dev), "bus master",
95                              dma_as->root, 0, memory_region_size(dma_as->root));
96     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
97     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
98                                 &pci_dev->bus_master_enable_region);
99 }
100 
101 static void pcibus_machine_done(Notifier *notifier, void *data)
102 {
103     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
104     int i;
105 
106     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
107         if (bus->devices[i]) {
108             pci_init_bus_master(bus->devices[i]);
109         }
110     }
111 }
112 
113 static void pci_bus_realize(BusState *qbus, Error **errp)
114 {
115     PCIBus *bus = PCI_BUS(qbus);
116 
117     bus->machine_done.notify = pcibus_machine_done;
118     qemu_add_machine_init_done_notifier(&bus->machine_done);
119 
120     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
121 }
122 
123 static void pci_bus_unrealize(BusState *qbus, Error **errp)
124 {
125     PCIBus *bus = PCI_BUS(qbus);
126 
127     qemu_remove_machine_init_done_notifier(&bus->machine_done);
128 
129     vmstate_unregister(NULL, &vmstate_pcibus, bus);
130 }
131 
132 static bool pcibus_is_root(PCIBus *bus)
133 {
134     return !bus->parent_dev;
135 }
136 
137 static int pcibus_num(PCIBus *bus)
138 {
139     if (pcibus_is_root(bus)) {
140         return 0; /* pci host bridge */
141     }
142     return bus->parent_dev->config[PCI_SECONDARY_BUS];
143 }
144 
145 static uint16_t pcibus_numa_node(PCIBus *bus)
146 {
147     return NUMA_NODE_UNASSIGNED;
148 }
149 
150 static void pci_bus_class_init(ObjectClass *klass, void *data)
151 {
152     BusClass *k = BUS_CLASS(klass);
153     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
154 
155     k->print_dev = pcibus_dev_print;
156     k->get_dev_path = pcibus_get_dev_path;
157     k->get_fw_dev_path = pcibus_get_fw_dev_path;
158     k->realize = pci_bus_realize;
159     k->unrealize = pci_bus_unrealize;
160     k->reset = pcibus_reset;
161 
162     pbc->is_root = pcibus_is_root;
163     pbc->bus_num = pcibus_num;
164     pbc->numa_node = pcibus_numa_node;
165 }
166 
167 static const TypeInfo pci_bus_info = {
168     .name = TYPE_PCI_BUS,
169     .parent = TYPE_BUS,
170     .instance_size = sizeof(PCIBus),
171     .class_size = sizeof(PCIBusClass),
172     .class_init = pci_bus_class_init,
173 };
174 
175 static const TypeInfo pcie_interface_info = {
176     .name          = INTERFACE_PCIE_DEVICE,
177     .parent        = TYPE_INTERFACE,
178 };
179 
180 static const TypeInfo conventional_pci_interface_info = {
181     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
182     .parent        = TYPE_INTERFACE,
183 };
184 
185 static const TypeInfo pcie_bus_info = {
186     .name = TYPE_PCIE_BUS,
187     .parent = TYPE_PCI_BUS,
188 };
189 
190 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
191 static void pci_update_mappings(PCIDevice *d);
192 static void pci_irq_handler(void *opaque, int irq_num, int level);
193 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
194 static void pci_del_option_rom(PCIDevice *pdev);
195 
196 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
197 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
198 
199 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
200 
201 int pci_bar(PCIDevice *d, int reg)
202 {
203     uint8_t type;
204 
205     if (reg != PCI_ROM_SLOT)
206         return PCI_BASE_ADDRESS_0 + reg * 4;
207 
208     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
209     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
210 }
211 
212 static inline int pci_irq_state(PCIDevice *d, int irq_num)
213 {
214         return (d->irq_state >> irq_num) & 0x1;
215 }
216 
217 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
218 {
219         d->irq_state &= ~(0x1 << irq_num);
220         d->irq_state |= level << irq_num;
221 }
222 
223 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
224 {
225     PCIBus *bus;
226     for (;;) {
227         bus = pci_get_bus(pci_dev);
228         irq_num = bus->map_irq(pci_dev, irq_num);
229         if (bus->set_irq)
230             break;
231         pci_dev = bus->parent_dev;
232     }
233     bus->irq_count[irq_num] += change;
234     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
235 }
236 
237 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
238 {
239     assert(irq_num >= 0);
240     assert(irq_num < bus->nirq);
241     return !!bus->irq_count[irq_num];
242 }
243 
244 /* Update interrupt status bit in config space on interrupt
245  * state change. */
246 static void pci_update_irq_status(PCIDevice *dev)
247 {
248     if (dev->irq_state) {
249         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
250     } else {
251         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
252     }
253 }
254 
255 void pci_device_deassert_intx(PCIDevice *dev)
256 {
257     int i;
258     for (i = 0; i < PCI_NUM_PINS; ++i) {
259         pci_irq_handler(dev, i, 0);
260     }
261 }
262 
263 static void pci_do_device_reset(PCIDevice *dev)
264 {
265     int r;
266 
267     pci_device_deassert_intx(dev);
268     assert(dev->irq_state == 0);
269 
270     /* Clear all writable bits */
271     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
272                                  pci_get_word(dev->wmask + PCI_COMMAND) |
273                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
274     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
275                                  pci_get_word(dev->wmask + PCI_STATUS) |
276                                  pci_get_word(dev->w1cmask + PCI_STATUS));
277     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
278     dev->config[PCI_INTERRUPT_LINE] = 0x0;
279     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
280         PCIIORegion *region = &dev->io_regions[r];
281         if (!region->size) {
282             continue;
283         }
284 
285         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
286             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
287             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
288         } else {
289             pci_set_long(dev->config + pci_bar(dev, r), region->type);
290         }
291     }
292     pci_update_mappings(dev);
293 
294     msi_reset(dev);
295     msix_reset(dev);
296 }
297 
298 /*
299  * This function is called on #RST and FLR.
300  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
301  */
302 void pci_device_reset(PCIDevice *dev)
303 {
304     qdev_reset_all(&dev->qdev);
305     pci_do_device_reset(dev);
306 }
307 
308 /*
309  * Trigger pci bus reset under a given bus.
310  * Called via qbus_reset_all on RST# assert, after the devices
311  * have been reset qdev_reset_all-ed already.
312  */
313 static void pcibus_reset(BusState *qbus)
314 {
315     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
316     int i;
317 
318     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
319         if (bus->devices[i]) {
320             pci_do_device_reset(bus->devices[i]);
321         }
322     }
323 
324     for (i = 0; i < bus->nirq; i++) {
325         assert(bus->irq_count[i] == 0);
326     }
327 }
328 
329 static void pci_host_bus_register(DeviceState *host)
330 {
331     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
332 
333     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
334 }
335 
336 static void pci_host_bus_unregister(DeviceState *host)
337 {
338     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
339 
340     QLIST_REMOVE(host_bridge, next);
341 }
342 
343 PCIBus *pci_device_root_bus(const PCIDevice *d)
344 {
345     PCIBus *bus = pci_get_bus(d);
346 
347     while (!pci_bus_is_root(bus)) {
348         d = bus->parent_dev;
349         assert(d != NULL);
350 
351         bus = pci_get_bus(d);
352     }
353 
354     return bus;
355 }
356 
357 const char *pci_root_bus_path(PCIDevice *dev)
358 {
359     PCIBus *rootbus = pci_device_root_bus(dev);
360     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
361     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
362 
363     assert(host_bridge->bus == rootbus);
364 
365     if (hc->root_bus_path) {
366         return (*hc->root_bus_path)(host_bridge, rootbus);
367     }
368 
369     return rootbus->qbus.name;
370 }
371 
372 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
373                               MemoryRegion *address_space_mem,
374                               MemoryRegion *address_space_io,
375                               uint8_t devfn_min)
376 {
377     assert(PCI_FUNC(devfn_min) == 0);
378     bus->devfn_min = devfn_min;
379     bus->slot_reserved_mask = 0x0;
380     bus->address_space_mem = address_space_mem;
381     bus->address_space_io = address_space_io;
382 
383     /* host bridge */
384     QLIST_INIT(&bus->child);
385 
386     pci_host_bus_register(parent);
387 }
388 
389 static void pci_bus_uninit(PCIBus *bus)
390 {
391     pci_host_bus_unregister(BUS(bus)->parent);
392 }
393 
394 bool pci_bus_is_express(PCIBus *bus)
395 {
396     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
397 }
398 
399 bool pci_bus_is_root(PCIBus *bus)
400 {
401     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
402 }
403 
404 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
405                               const char *name,
406                               MemoryRegion *address_space_mem,
407                               MemoryRegion *address_space_io,
408                               uint8_t devfn_min, const char *typename)
409 {
410     qbus_create_inplace(bus, bus_size, typename, parent, name);
411     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
412                       devfn_min);
413 }
414 
415 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
416                          MemoryRegion *address_space_mem,
417                          MemoryRegion *address_space_io,
418                          uint8_t devfn_min, const char *typename)
419 {
420     PCIBus *bus;
421 
422     bus = PCI_BUS(qbus_create(typename, parent, name));
423     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
424                       devfn_min);
425     return bus;
426 }
427 
428 void pci_root_bus_cleanup(PCIBus *bus)
429 {
430     pci_bus_uninit(bus);
431     /* the caller of the unplug hotplug handler will delete this device */
432     object_property_set_bool(OBJECT(bus), false, "realized", NULL);
433 }
434 
435 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
436                   void *irq_opaque, int nirq)
437 {
438     bus->set_irq = set_irq;
439     bus->map_irq = map_irq;
440     bus->irq_opaque = irq_opaque;
441     bus->nirq = nirq;
442     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
443 }
444 
445 void pci_bus_irqs_cleanup(PCIBus *bus)
446 {
447     bus->set_irq = NULL;
448     bus->map_irq = NULL;
449     bus->irq_opaque = NULL;
450     bus->nirq = 0;
451     g_free(bus->irq_count);
452 }
453 
454 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
455                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
456                               void *irq_opaque,
457                               MemoryRegion *address_space_mem,
458                               MemoryRegion *address_space_io,
459                               uint8_t devfn_min, int nirq,
460                               const char *typename)
461 {
462     PCIBus *bus;
463 
464     bus = pci_root_bus_new(parent, name, address_space_mem,
465                            address_space_io, devfn_min, typename);
466     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
467     return bus;
468 }
469 
470 void pci_unregister_root_bus(PCIBus *bus)
471 {
472     pci_bus_irqs_cleanup(bus);
473     pci_root_bus_cleanup(bus);
474 }
475 
476 int pci_bus_num(PCIBus *s)
477 {
478     return PCI_BUS_GET_CLASS(s)->bus_num(s);
479 }
480 
481 int pci_bus_numa_node(PCIBus *bus)
482 {
483     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
484 }
485 
486 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
487                                  const VMStateField *field)
488 {
489     PCIDevice *s = container_of(pv, PCIDevice, config);
490     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
491     uint8_t *config;
492     int i;
493 
494     assert(size == pci_config_size(s));
495     config = g_malloc(size);
496 
497     qemu_get_buffer(f, config, size);
498     for (i = 0; i < size; ++i) {
499         if ((config[i] ^ s->config[i]) &
500             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
501             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
502                          "cmask: %x wmask: %x w1cmask:%x", __func__,
503                          i, config[i], s->config[i],
504                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
505             g_free(config);
506             return -EINVAL;
507         }
508     }
509     memcpy(s->config, config, size);
510 
511     pci_update_mappings(s);
512     if (pc->is_bridge) {
513         PCIBridge *b = PCI_BRIDGE(s);
514         pci_bridge_update_mappings(b);
515     }
516 
517     memory_region_set_enabled(&s->bus_master_enable_region,
518                               pci_get_word(s->config + PCI_COMMAND)
519                               & PCI_COMMAND_MASTER);
520 
521     g_free(config);
522     return 0;
523 }
524 
525 /* just put buffer */
526 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
527                                  const VMStateField *field, QJSON *vmdesc)
528 {
529     const uint8_t **v = pv;
530     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
531     qemu_put_buffer(f, *v, size);
532 
533     return 0;
534 }
535 
536 static VMStateInfo vmstate_info_pci_config = {
537     .name = "pci config",
538     .get  = get_pci_config_device,
539     .put  = put_pci_config_device,
540 };
541 
542 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
543                              const VMStateField *field)
544 {
545     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
546     uint32_t irq_state[PCI_NUM_PINS];
547     int i;
548     for (i = 0; i < PCI_NUM_PINS; ++i) {
549         irq_state[i] = qemu_get_be32(f);
550         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
551             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
552                     irq_state[i]);
553             return -EINVAL;
554         }
555     }
556 
557     for (i = 0; i < PCI_NUM_PINS; ++i) {
558         pci_set_irq_state(s, i, irq_state[i]);
559     }
560 
561     return 0;
562 }
563 
564 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
565                              const VMStateField *field, QJSON *vmdesc)
566 {
567     int i;
568     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
569 
570     for (i = 0; i < PCI_NUM_PINS; ++i) {
571         qemu_put_be32(f, pci_irq_state(s, i));
572     }
573 
574     return 0;
575 }
576 
577 static VMStateInfo vmstate_info_pci_irq_state = {
578     .name = "pci irq state",
579     .get  = get_pci_irq_state,
580     .put  = put_pci_irq_state,
581 };
582 
583 static bool migrate_is_pcie(void *opaque, int version_id)
584 {
585     return pci_is_express((PCIDevice *)opaque);
586 }
587 
588 static bool migrate_is_not_pcie(void *opaque, int version_id)
589 {
590     return !pci_is_express((PCIDevice *)opaque);
591 }
592 
593 const VMStateDescription vmstate_pci_device = {
594     .name = "PCIDevice",
595     .version_id = 2,
596     .minimum_version_id = 1,
597     .fields = (VMStateField[]) {
598         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
599         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
600                                    migrate_is_not_pcie,
601                                    0, vmstate_info_pci_config,
602                                    PCI_CONFIG_SPACE_SIZE),
603         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
604                                    migrate_is_pcie,
605                                    0, vmstate_info_pci_config,
606                                    PCIE_CONFIG_SPACE_SIZE),
607         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
608                                    vmstate_info_pci_irq_state,
609                                    PCI_NUM_PINS * sizeof(int32_t)),
610         VMSTATE_END_OF_LIST()
611     }
612 };
613 
614 
615 void pci_device_save(PCIDevice *s, QEMUFile *f)
616 {
617     /* Clear interrupt status bit: it is implicit
618      * in irq_state which we are saving.
619      * This makes us compatible with old devices
620      * which never set or clear this bit. */
621     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
622     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
623     /* Restore the interrupt status bit. */
624     pci_update_irq_status(s);
625 }
626 
627 int pci_device_load(PCIDevice *s, QEMUFile *f)
628 {
629     int ret;
630     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
631     /* Restore the interrupt status bit. */
632     pci_update_irq_status(s);
633     return ret;
634 }
635 
636 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
637 {
638     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
639                  pci_default_sub_vendor_id);
640     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
641                  pci_default_sub_device_id);
642 }
643 
644 /*
645  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
646  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
647  */
648 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
649                              unsigned int *slotp, unsigned int *funcp)
650 {
651     const char *p;
652     char *e;
653     unsigned long val;
654     unsigned long dom = 0, bus = 0;
655     unsigned int slot = 0;
656     unsigned int func = 0;
657 
658     p = addr;
659     val = strtoul(p, &e, 16);
660     if (e == p)
661         return -1;
662     if (*e == ':') {
663         bus = val;
664         p = e + 1;
665         val = strtoul(p, &e, 16);
666         if (e == p)
667             return -1;
668         if (*e == ':') {
669             dom = bus;
670             bus = val;
671             p = e + 1;
672             val = strtoul(p, &e, 16);
673             if (e == p)
674                 return -1;
675         }
676     }
677 
678     slot = val;
679 
680     if (funcp != NULL) {
681         if (*e != '.')
682             return -1;
683 
684         p = e + 1;
685         val = strtoul(p, &e, 16);
686         if (e == p)
687             return -1;
688 
689         func = val;
690     }
691 
692     /* if funcp == NULL func is 0 */
693     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
694         return -1;
695 
696     if (*e)
697         return -1;
698 
699     *domp = dom;
700     *busp = bus;
701     *slotp = slot;
702     if (funcp != NULL)
703         *funcp = func;
704     return 0;
705 }
706 
707 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
708                                  const char *devaddr)
709 {
710     int dom, bus;
711     unsigned slot;
712 
713     if (!root) {
714         fprintf(stderr, "No primary PCI bus\n");
715         return NULL;
716     }
717 
718     assert(!root->parent_dev);
719 
720     if (!devaddr) {
721         *devfnp = -1;
722         return pci_find_bus_nr(root, 0);
723     }
724 
725     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
726         return NULL;
727     }
728 
729     if (dom != 0) {
730         fprintf(stderr, "No support for non-zero PCI domains\n");
731         return NULL;
732     }
733 
734     *devfnp = PCI_DEVFN(slot, 0);
735     return pci_find_bus_nr(root, bus);
736 }
737 
738 static void pci_init_cmask(PCIDevice *dev)
739 {
740     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
741     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
742     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
743     dev->cmask[PCI_REVISION_ID] = 0xff;
744     dev->cmask[PCI_CLASS_PROG] = 0xff;
745     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
746     dev->cmask[PCI_HEADER_TYPE] = 0xff;
747     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
748 }
749 
750 static void pci_init_wmask(PCIDevice *dev)
751 {
752     int config_size = pci_config_size(dev);
753 
754     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
755     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
756     pci_set_word(dev->wmask + PCI_COMMAND,
757                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
758                  PCI_COMMAND_INTX_DISABLE);
759     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
760         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
761     }
762 
763     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
764            config_size - PCI_CONFIG_HEADER_SIZE);
765 }
766 
767 static void pci_init_w1cmask(PCIDevice *dev)
768 {
769     /*
770      * Note: It's okay to set w1cmask even for readonly bits as
771      * long as their value is hardwired to 0.
772      */
773     pci_set_word(dev->w1cmask + PCI_STATUS,
774                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
775                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
776                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
777 }
778 
779 static void pci_init_mask_bridge(PCIDevice *d)
780 {
781     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
782        PCI_SEC_LETENCY_TIMER */
783     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
784 
785     /* base and limit */
786     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
787     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
788     pci_set_word(d->wmask + PCI_MEMORY_BASE,
789                  PCI_MEMORY_RANGE_MASK & 0xffff);
790     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
791                  PCI_MEMORY_RANGE_MASK & 0xffff);
792     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
793                  PCI_PREF_RANGE_MASK & 0xffff);
794     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
795                  PCI_PREF_RANGE_MASK & 0xffff);
796 
797     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
798     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
799 
800     /* Supported memory and i/o types */
801     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
802     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
803     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
804                                PCI_PREF_RANGE_TYPE_64);
805     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
806                                PCI_PREF_RANGE_TYPE_64);
807 
808     /*
809      * TODO: Bridges default to 10-bit VGA decoding but we currently only
810      * implement 16-bit decoding (no alias support).
811      */
812     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
813                  PCI_BRIDGE_CTL_PARITY |
814                  PCI_BRIDGE_CTL_SERR |
815                  PCI_BRIDGE_CTL_ISA |
816                  PCI_BRIDGE_CTL_VGA |
817                  PCI_BRIDGE_CTL_VGA_16BIT |
818                  PCI_BRIDGE_CTL_MASTER_ABORT |
819                  PCI_BRIDGE_CTL_BUS_RESET |
820                  PCI_BRIDGE_CTL_FAST_BACK |
821                  PCI_BRIDGE_CTL_DISCARD |
822                  PCI_BRIDGE_CTL_SEC_DISCARD |
823                  PCI_BRIDGE_CTL_DISCARD_SERR);
824     /* Below does not do anything as we never set this bit, put here for
825      * completeness. */
826     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
827                  PCI_BRIDGE_CTL_DISCARD_STATUS);
828     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
829     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
830     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
831                                PCI_PREF_RANGE_TYPE_MASK);
832     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
833                                PCI_PREF_RANGE_TYPE_MASK);
834 }
835 
836 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
837 {
838     uint8_t slot = PCI_SLOT(dev->devfn);
839     uint8_t func;
840 
841     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
842         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
843     }
844 
845     /*
846      * multifunction bit is interpreted in two ways as follows.
847      *   - all functions must set the bit to 1.
848      *     Example: Intel X53
849      *   - function 0 must set the bit, but the rest function (> 0)
850      *     is allowed to leave the bit to 0.
851      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
852      *
853      * So OS (at least Linux) checks the bit of only function 0,
854      * and doesn't see the bit of function > 0.
855      *
856      * The below check allows both interpretation.
857      */
858     if (PCI_FUNC(dev->devfn)) {
859         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
860         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
861             /* function 0 should set multifunction bit */
862             error_setg(errp, "PCI: single function device can't be populated "
863                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
864             return;
865         }
866         return;
867     }
868 
869     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
870         return;
871     }
872     /* function 0 indicates single function, so function > 0 must be NULL */
873     for (func = 1; func < PCI_FUNC_MAX; ++func) {
874         if (bus->devices[PCI_DEVFN(slot, func)]) {
875             error_setg(errp, "PCI: %x.0 indicates single function, "
876                        "but %x.%x is already populated.",
877                        slot, slot, func);
878             return;
879         }
880     }
881 }
882 
883 static void pci_config_alloc(PCIDevice *pci_dev)
884 {
885     int config_size = pci_config_size(pci_dev);
886 
887     pci_dev->config = g_malloc0(config_size);
888     pci_dev->cmask = g_malloc0(config_size);
889     pci_dev->wmask = g_malloc0(config_size);
890     pci_dev->w1cmask = g_malloc0(config_size);
891     pci_dev->used = g_malloc0(config_size);
892 }
893 
894 static void pci_config_free(PCIDevice *pci_dev)
895 {
896     g_free(pci_dev->config);
897     g_free(pci_dev->cmask);
898     g_free(pci_dev->wmask);
899     g_free(pci_dev->w1cmask);
900     g_free(pci_dev->used);
901 }
902 
903 static void do_pci_unregister_device(PCIDevice *pci_dev)
904 {
905     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
906     pci_config_free(pci_dev);
907 
908     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
909         memory_region_del_subregion(&pci_dev->bus_master_container_region,
910                                     &pci_dev->bus_master_enable_region);
911     }
912     address_space_destroy(&pci_dev->bus_master_as);
913 }
914 
915 /* Extract PCIReqIDCache into BDF format */
916 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
917 {
918     uint8_t bus_n;
919     uint16_t result;
920 
921     switch (cache->type) {
922     case PCI_REQ_ID_BDF:
923         result = pci_get_bdf(cache->dev);
924         break;
925     case PCI_REQ_ID_SECONDARY_BUS:
926         bus_n = pci_dev_bus_num(cache->dev);
927         result = PCI_BUILD_BDF(bus_n, 0);
928         break;
929     default:
930         error_printf("Invalid PCI requester ID cache type: %d\n",
931                      cache->type);
932         exit(1);
933         break;
934     }
935 
936     return result;
937 }
938 
939 /* Parse bridges up to the root complex and return requester ID
940  * cache for specific device.  For full PCIe topology, the cache
941  * result would be exactly the same as getting BDF of the device.
942  * However, several tricks are required when system mixed up with
943  * legacy PCI devices and PCIe-to-PCI bridges.
944  *
945  * Here we cache the proxy device (and type) not requester ID since
946  * bus number might change from time to time.
947  */
948 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
949 {
950     PCIDevice *parent;
951     PCIReqIDCache cache = {
952         .dev = dev,
953         .type = PCI_REQ_ID_BDF,
954     };
955 
956     while (!pci_bus_is_root(pci_get_bus(dev))) {
957         /* We are under PCI/PCIe bridges */
958         parent = pci_get_bus(dev)->parent_dev;
959         if (pci_is_express(parent)) {
960             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
961                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
962                  * override the requester ID using secondary bus
963                  * number of parent bridge with zeroed devfn
964                  * (pcie-to-pci bridge spec chap 2.3). */
965                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
966                 cache.dev = dev;
967             }
968         } else {
969             /* Legacy PCI, override requester ID with the bridge's
970              * BDF upstream.  When the root complex connects to
971              * legacy PCI devices (including buses), it can only
972              * obtain requester ID info from directly attached
973              * devices.  If devices are attached under bridges, only
974              * the requester ID of the bridge that is directly
975              * attached to the root complex can be recognized. */
976             cache.type = PCI_REQ_ID_BDF;
977             cache.dev = parent;
978         }
979         dev = parent;
980     }
981 
982     return cache;
983 }
984 
985 uint16_t pci_requester_id(PCIDevice *dev)
986 {
987     return pci_req_id_cache_extract(&dev->requester_id_cache);
988 }
989 
990 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
991 {
992     return !(bus->devices[devfn]);
993 }
994 
995 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
996 {
997     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
998 }
999 
1000 /* -1 for devfn means auto assign */
1001 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1002                                          const char *name, int devfn,
1003                                          Error **errp)
1004 {
1005     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1006     PCIConfigReadFunc *config_read = pc->config_read;
1007     PCIConfigWriteFunc *config_write = pc->config_write;
1008     Error *local_err = NULL;
1009     DeviceState *dev = DEVICE(pci_dev);
1010     PCIBus *bus = pci_get_bus(pci_dev);
1011 
1012     /* Only pci bridges can be attached to extra PCI root buses */
1013     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1014         error_setg(errp,
1015                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1016                     bus->parent_dev->name);
1017         return NULL;
1018     }
1019 
1020     if (devfn < 0) {
1021         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1022             devfn += PCI_FUNC_MAX) {
1023             if (pci_bus_devfn_available(bus, devfn) &&
1024                    !pci_bus_devfn_reserved(bus, devfn)) {
1025                 goto found;
1026             }
1027         }
1028         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1029                    "or reserved", name);
1030         return NULL;
1031     found: ;
1032     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1033         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1034                    " reserved",
1035                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1036         return NULL;
1037     } else if (!pci_bus_devfn_available(bus, devfn)) {
1038         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1039                    " in use by %s",
1040                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1041                    bus->devices[devfn]->name);
1042         return NULL;
1043     } else if (dev->hotplugged &&
1044                pci_get_function_0(pci_dev)) {
1045         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1046                    " new func %s cannot be exposed to guest.",
1047                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1048                    pci_get_function_0(pci_dev)->name,
1049                    name);
1050 
1051        return NULL;
1052     }
1053 
1054     pci_dev->devfn = devfn;
1055     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1056     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1057 
1058     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1059                        "bus master container", UINT64_MAX);
1060     address_space_init(&pci_dev->bus_master_as,
1061                        &pci_dev->bus_master_container_region, pci_dev->name);
1062 
1063     if (qdev_hotplug) {
1064         pci_init_bus_master(pci_dev);
1065     }
1066     pci_dev->irq_state = 0;
1067     pci_config_alloc(pci_dev);
1068 
1069     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1070     pci_config_set_device_id(pci_dev->config, pc->device_id);
1071     pci_config_set_revision(pci_dev->config, pc->revision);
1072     pci_config_set_class(pci_dev->config, pc->class_id);
1073 
1074     if (!pc->is_bridge) {
1075         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1076             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1077                          pc->subsystem_vendor_id);
1078             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1079                          pc->subsystem_id);
1080         } else {
1081             pci_set_default_subsystem_id(pci_dev);
1082         }
1083     } else {
1084         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1085         assert(!pc->subsystem_vendor_id);
1086         assert(!pc->subsystem_id);
1087     }
1088     pci_init_cmask(pci_dev);
1089     pci_init_wmask(pci_dev);
1090     pci_init_w1cmask(pci_dev);
1091     if (pc->is_bridge) {
1092         pci_init_mask_bridge(pci_dev);
1093     }
1094     pci_init_multifunction(bus, pci_dev, &local_err);
1095     if (local_err) {
1096         error_propagate(errp, local_err);
1097         do_pci_unregister_device(pci_dev);
1098         return NULL;
1099     }
1100 
1101     if (!config_read)
1102         config_read = pci_default_read_config;
1103     if (!config_write)
1104         config_write = pci_default_write_config;
1105     pci_dev->config_read = config_read;
1106     pci_dev->config_write = config_write;
1107     bus->devices[devfn] = pci_dev;
1108     pci_dev->version_id = 2; /* Current pci device vmstate version */
1109     return pci_dev;
1110 }
1111 
1112 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1113 {
1114     PCIIORegion *r;
1115     int i;
1116 
1117     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1118         r = &pci_dev->io_regions[i];
1119         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1120             continue;
1121         memory_region_del_subregion(r->address_space, r->memory);
1122     }
1123 
1124     pci_unregister_vga(pci_dev);
1125 }
1126 
1127 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1128 {
1129     PCIDevice *pci_dev = PCI_DEVICE(dev);
1130     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1131 
1132     pci_unregister_io_regions(pci_dev);
1133     pci_del_option_rom(pci_dev);
1134 
1135     if (pc->exit) {
1136         pc->exit(pci_dev);
1137     }
1138 
1139     pci_device_deassert_intx(pci_dev);
1140     do_pci_unregister_device(pci_dev);
1141 }
1142 
1143 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1144                       uint8_t type, MemoryRegion *memory)
1145 {
1146     PCIIORegion *r;
1147     uint32_t addr; /* offset in pci config space */
1148     uint64_t wmask;
1149     pcibus_t size = memory_region_size(memory);
1150 
1151     assert(region_num >= 0);
1152     assert(region_num < PCI_NUM_REGIONS);
1153     if (size & (size-1)) {
1154         error_report("ERROR: PCI region size must be pow2 "
1155                     "type=0x%x, size=0x%"FMT_PCIBUS"", type, size);
1156         exit(1);
1157     }
1158 
1159     r = &pci_dev->io_regions[region_num];
1160     r->addr = PCI_BAR_UNMAPPED;
1161     r->size = size;
1162     r->type = type;
1163     r->memory = memory;
1164     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1165                         ? pci_get_bus(pci_dev)->address_space_io
1166                         : pci_get_bus(pci_dev)->address_space_mem;
1167 
1168     wmask = ~(size - 1);
1169     if (region_num == PCI_ROM_SLOT) {
1170         /* ROM enable bit is writable */
1171         wmask |= PCI_ROM_ADDRESS_ENABLE;
1172     }
1173 
1174     addr = pci_bar(pci_dev, region_num);
1175     pci_set_long(pci_dev->config + addr, type);
1176 
1177     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1178         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1179         pci_set_quad(pci_dev->wmask + addr, wmask);
1180         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1181     } else {
1182         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1183         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1184     }
1185 }
1186 
1187 static void pci_update_vga(PCIDevice *pci_dev)
1188 {
1189     uint16_t cmd;
1190 
1191     if (!pci_dev->has_vga) {
1192         return;
1193     }
1194 
1195     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1196 
1197     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1198                               cmd & PCI_COMMAND_MEMORY);
1199     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1200                               cmd & PCI_COMMAND_IO);
1201     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1202                               cmd & PCI_COMMAND_IO);
1203 }
1204 
1205 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1206                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1207 {
1208     PCIBus *bus = pci_get_bus(pci_dev);
1209 
1210     assert(!pci_dev->has_vga);
1211 
1212     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1213     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1214     memory_region_add_subregion_overlap(bus->address_space_mem,
1215                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1216 
1217     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1218     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1219     memory_region_add_subregion_overlap(bus->address_space_io,
1220                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1221 
1222     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1223     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1224     memory_region_add_subregion_overlap(bus->address_space_io,
1225                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1226     pci_dev->has_vga = true;
1227 
1228     pci_update_vga(pci_dev);
1229 }
1230 
1231 void pci_unregister_vga(PCIDevice *pci_dev)
1232 {
1233     PCIBus *bus = pci_get_bus(pci_dev);
1234 
1235     if (!pci_dev->has_vga) {
1236         return;
1237     }
1238 
1239     memory_region_del_subregion(bus->address_space_mem,
1240                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1241     memory_region_del_subregion(bus->address_space_io,
1242                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1243     memory_region_del_subregion(bus->address_space_io,
1244                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1245     pci_dev->has_vga = false;
1246 }
1247 
1248 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1249 {
1250     return pci_dev->io_regions[region_num].addr;
1251 }
1252 
1253 static pcibus_t pci_bar_address(PCIDevice *d,
1254                                 int reg, uint8_t type, pcibus_t size)
1255 {
1256     pcibus_t new_addr, last_addr;
1257     int bar = pci_bar(d, reg);
1258     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1259     Object *machine = qdev_get_machine();
1260     ObjectClass *oc = object_get_class(machine);
1261     MachineClass *mc = MACHINE_CLASS(oc);
1262     bool allow_0_address = mc->pci_allow_0_address;
1263 
1264     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1265         if (!(cmd & PCI_COMMAND_IO)) {
1266             return PCI_BAR_UNMAPPED;
1267         }
1268         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1269         last_addr = new_addr + size - 1;
1270         /* Check if 32 bit BAR wraps around explicitly.
1271          * TODO: make priorities correct and remove this work around.
1272          */
1273         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1274             (!allow_0_address && new_addr == 0)) {
1275             return PCI_BAR_UNMAPPED;
1276         }
1277         return new_addr;
1278     }
1279 
1280     if (!(cmd & PCI_COMMAND_MEMORY)) {
1281         return PCI_BAR_UNMAPPED;
1282     }
1283     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1284         new_addr = pci_get_quad(d->config + bar);
1285     } else {
1286         new_addr = pci_get_long(d->config + bar);
1287     }
1288     /* the ROM slot has a specific enable bit */
1289     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1290         return PCI_BAR_UNMAPPED;
1291     }
1292     new_addr &= ~(size - 1);
1293     last_addr = new_addr + size - 1;
1294     /* NOTE: we do not support wrapping */
1295     /* XXX: as we cannot support really dynamic
1296        mappings, we handle specific values as invalid
1297        mappings. */
1298     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1299         (!allow_0_address && new_addr == 0)) {
1300         return PCI_BAR_UNMAPPED;
1301     }
1302 
1303     /* Now pcibus_t is 64bit.
1304      * Check if 32 bit BAR wraps around explicitly.
1305      * Without this, PC ide doesn't work well.
1306      * TODO: remove this work around.
1307      */
1308     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1309         return PCI_BAR_UNMAPPED;
1310     }
1311 
1312     /*
1313      * OS is allowed to set BAR beyond its addressable
1314      * bits. For example, 32 bit OS can set 64bit bar
1315      * to >4G. Check it. TODO: we might need to support
1316      * it in the future for e.g. PAE.
1317      */
1318     if (last_addr >= HWADDR_MAX) {
1319         return PCI_BAR_UNMAPPED;
1320     }
1321 
1322     return new_addr;
1323 }
1324 
1325 static void pci_update_mappings(PCIDevice *d)
1326 {
1327     PCIIORegion *r;
1328     int i;
1329     pcibus_t new_addr;
1330 
1331     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1332         r = &d->io_regions[i];
1333 
1334         /* this region isn't registered */
1335         if (!r->size)
1336             continue;
1337 
1338         new_addr = pci_bar_address(d, i, r->type, r->size);
1339 
1340         /* This bar isn't changed */
1341         if (new_addr == r->addr)
1342             continue;
1343 
1344         /* now do the real mapping */
1345         if (r->addr != PCI_BAR_UNMAPPED) {
1346             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1347                                           PCI_SLOT(d->devfn),
1348                                           PCI_FUNC(d->devfn),
1349                                           i, r->addr, r->size);
1350             memory_region_del_subregion(r->address_space, r->memory);
1351         }
1352         r->addr = new_addr;
1353         if (r->addr != PCI_BAR_UNMAPPED) {
1354             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1355                                           PCI_SLOT(d->devfn),
1356                                           PCI_FUNC(d->devfn),
1357                                           i, r->addr, r->size);
1358             memory_region_add_subregion_overlap(r->address_space,
1359                                                 r->addr, r->memory, 1);
1360         }
1361     }
1362 
1363     pci_update_vga(d);
1364 }
1365 
1366 static inline int pci_irq_disabled(PCIDevice *d)
1367 {
1368     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1369 }
1370 
1371 /* Called after interrupt disabled field update in config space,
1372  * assert/deassert interrupts if necessary.
1373  * Gets original interrupt disable bit value (before update). */
1374 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1375 {
1376     int i, disabled = pci_irq_disabled(d);
1377     if (disabled == was_irq_disabled)
1378         return;
1379     for (i = 0; i < PCI_NUM_PINS; ++i) {
1380         int state = pci_irq_state(d, i);
1381         pci_change_irq_level(d, i, disabled ? -state : state);
1382     }
1383 }
1384 
1385 uint32_t pci_default_read_config(PCIDevice *d,
1386                                  uint32_t address, int len)
1387 {
1388     uint32_t val = 0;
1389 
1390     if (pci_is_express_downstream_port(d) &&
1391         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1392         pcie_sync_bridge_lnk(d);
1393     }
1394     memcpy(&val, d->config + address, len);
1395     return le32_to_cpu(val);
1396 }
1397 
1398 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1399 {
1400     int i, was_irq_disabled = pci_irq_disabled(d);
1401     uint32_t val = val_in;
1402 
1403     for (i = 0; i < l; val >>= 8, ++i) {
1404         uint8_t wmask = d->wmask[addr + i];
1405         uint8_t w1cmask = d->w1cmask[addr + i];
1406         assert(!(wmask & w1cmask));
1407         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1408         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1409     }
1410     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1411         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1412         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1413         range_covers_byte(addr, l, PCI_COMMAND))
1414         pci_update_mappings(d);
1415 
1416     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1417         pci_update_irq_disabled(d, was_irq_disabled);
1418         memory_region_set_enabled(&d->bus_master_enable_region,
1419                                   pci_get_word(d->config + PCI_COMMAND)
1420                                     & PCI_COMMAND_MASTER);
1421     }
1422 
1423     msi_write_config(d, addr, val_in, l);
1424     msix_write_config(d, addr, val_in, l);
1425 }
1426 
1427 /***********************************************************/
1428 /* generic PCI irq support */
1429 
1430 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1431 static void pci_irq_handler(void *opaque, int irq_num, int level)
1432 {
1433     PCIDevice *pci_dev = opaque;
1434     int change;
1435 
1436     change = level - pci_irq_state(pci_dev, irq_num);
1437     if (!change)
1438         return;
1439 
1440     pci_set_irq_state(pci_dev, irq_num, level);
1441     pci_update_irq_status(pci_dev);
1442     if (pci_irq_disabled(pci_dev))
1443         return;
1444     pci_change_irq_level(pci_dev, irq_num, change);
1445 }
1446 
1447 static inline int pci_intx(PCIDevice *pci_dev)
1448 {
1449     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1450 }
1451 
1452 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1453 {
1454     int intx = pci_intx(pci_dev);
1455 
1456     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1457 }
1458 
1459 void pci_set_irq(PCIDevice *pci_dev, int level)
1460 {
1461     int intx = pci_intx(pci_dev);
1462     pci_irq_handler(pci_dev, intx, level);
1463 }
1464 
1465 /* Special hooks used by device assignment */
1466 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1467 {
1468     assert(pci_bus_is_root(bus));
1469     bus->route_intx_to_irq = route_intx_to_irq;
1470 }
1471 
1472 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1473 {
1474     PCIBus *bus;
1475 
1476     do {
1477         bus = pci_get_bus(dev);
1478         pin = bus->map_irq(dev, pin);
1479         dev = bus->parent_dev;
1480     } while (dev);
1481 
1482     if (!bus->route_intx_to_irq) {
1483         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1484                      object_get_typename(OBJECT(bus->qbus.parent)));
1485         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1486     }
1487 
1488     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1489 }
1490 
1491 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1492 {
1493     return old->mode != new->mode || old->irq != new->irq;
1494 }
1495 
1496 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1497 {
1498     PCIDevice *dev;
1499     PCIBus *sec;
1500     int i;
1501 
1502     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1503         dev = bus->devices[i];
1504         if (dev && dev->intx_routing_notifier) {
1505             dev->intx_routing_notifier(dev);
1506         }
1507     }
1508 
1509     QLIST_FOREACH(sec, &bus->child, sibling) {
1510         pci_bus_fire_intx_routing_notifier(sec);
1511     }
1512 }
1513 
1514 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1515                                           PCIINTxRoutingNotifier notifier)
1516 {
1517     dev->intx_routing_notifier = notifier;
1518 }
1519 
1520 /*
1521  * PCI-to-PCI bridge specification
1522  * 9.1: Interrupt routing. Table 9-1
1523  *
1524  * the PCI Express Base Specification, Revision 2.1
1525  * 2.2.8.1: INTx interrutp signaling - Rules
1526  *          the Implementation Note
1527  *          Table 2-20
1528  */
1529 /*
1530  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1531  * 0-origin unlike PCI interrupt pin register.
1532  */
1533 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1534 {
1535     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1536 }
1537 
1538 /***********************************************************/
1539 /* monitor info on PCI */
1540 
1541 typedef struct {
1542     uint16_t class;
1543     const char *desc;
1544     const char *fw_name;
1545     uint16_t fw_ign_bits;
1546 } pci_class_desc;
1547 
1548 static const pci_class_desc pci_class_descriptions[] =
1549 {
1550     { 0x0001, "VGA controller", "display"},
1551     { 0x0100, "SCSI controller", "scsi"},
1552     { 0x0101, "IDE controller", "ide"},
1553     { 0x0102, "Floppy controller", "fdc"},
1554     { 0x0103, "IPI controller", "ipi"},
1555     { 0x0104, "RAID controller", "raid"},
1556     { 0x0106, "SATA controller"},
1557     { 0x0107, "SAS controller"},
1558     { 0x0180, "Storage controller"},
1559     { 0x0200, "Ethernet controller", "ethernet"},
1560     { 0x0201, "Token Ring controller", "token-ring"},
1561     { 0x0202, "FDDI controller", "fddi"},
1562     { 0x0203, "ATM controller", "atm"},
1563     { 0x0280, "Network controller"},
1564     { 0x0300, "VGA controller", "display", 0x00ff},
1565     { 0x0301, "XGA controller"},
1566     { 0x0302, "3D controller"},
1567     { 0x0380, "Display controller"},
1568     { 0x0400, "Video controller", "video"},
1569     { 0x0401, "Audio controller", "sound"},
1570     { 0x0402, "Phone"},
1571     { 0x0403, "Audio controller", "sound"},
1572     { 0x0480, "Multimedia controller"},
1573     { 0x0500, "RAM controller", "memory"},
1574     { 0x0501, "Flash controller", "flash"},
1575     { 0x0580, "Memory controller"},
1576     { 0x0600, "Host bridge", "host"},
1577     { 0x0601, "ISA bridge", "isa"},
1578     { 0x0602, "EISA bridge", "eisa"},
1579     { 0x0603, "MC bridge", "mca"},
1580     { 0x0604, "PCI bridge", "pci-bridge"},
1581     { 0x0605, "PCMCIA bridge", "pcmcia"},
1582     { 0x0606, "NUBUS bridge", "nubus"},
1583     { 0x0607, "CARDBUS bridge", "cardbus"},
1584     { 0x0608, "RACEWAY bridge"},
1585     { 0x0680, "Bridge"},
1586     { 0x0700, "Serial port", "serial"},
1587     { 0x0701, "Parallel port", "parallel"},
1588     { 0x0800, "Interrupt controller", "interrupt-controller"},
1589     { 0x0801, "DMA controller", "dma-controller"},
1590     { 0x0802, "Timer", "timer"},
1591     { 0x0803, "RTC", "rtc"},
1592     { 0x0900, "Keyboard", "keyboard"},
1593     { 0x0901, "Pen", "pen"},
1594     { 0x0902, "Mouse", "mouse"},
1595     { 0x0A00, "Dock station", "dock", 0x00ff},
1596     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1597     { 0x0c00, "Fireware contorller", "fireware"},
1598     { 0x0c01, "Access bus controller", "access-bus"},
1599     { 0x0c02, "SSA controller", "ssa"},
1600     { 0x0c03, "USB controller", "usb"},
1601     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1602     { 0x0c05, "SMBus"},
1603     { 0, NULL}
1604 };
1605 
1606 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1607                                                   void (*fn)(PCIBus *b,
1608                                                              PCIDevice *d,
1609                                                              void *opaque),
1610                                                   void *opaque)
1611 {
1612     PCIDevice *d;
1613     int devfn;
1614 
1615     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1616         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1617         if (d) {
1618             fn(bus, d, opaque);
1619         }
1620     }
1621 }
1622 
1623 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1624                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1625                          void *opaque)
1626 {
1627     bus = pci_find_bus_nr(bus, bus_num);
1628 
1629     if (bus) {
1630         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1631     }
1632 }
1633 
1634 static void pci_for_each_device_under_bus(PCIBus *bus,
1635                                           void (*fn)(PCIBus *b, PCIDevice *d,
1636                                                      void *opaque),
1637                                           void *opaque)
1638 {
1639     PCIDevice *d;
1640     int devfn;
1641 
1642     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1643         d = bus->devices[devfn];
1644         if (d) {
1645             fn(bus, d, opaque);
1646         }
1647     }
1648 }
1649 
1650 void pci_for_each_device(PCIBus *bus, int bus_num,
1651                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1652                          void *opaque)
1653 {
1654     bus = pci_find_bus_nr(bus, bus_num);
1655 
1656     if (bus) {
1657         pci_for_each_device_under_bus(bus, fn, opaque);
1658     }
1659 }
1660 
1661 static const pci_class_desc *get_class_desc(int class)
1662 {
1663     const pci_class_desc *desc;
1664 
1665     desc = pci_class_descriptions;
1666     while (desc->desc && class != desc->class) {
1667         desc++;
1668     }
1669 
1670     return desc;
1671 }
1672 
1673 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1674 
1675 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1676 {
1677     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1678     int i;
1679 
1680     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1681         const PCIIORegion *r = &dev->io_regions[i];
1682         PciMemoryRegionList *region;
1683 
1684         if (!r->size) {
1685             continue;
1686         }
1687 
1688         region = g_malloc0(sizeof(*region));
1689         region->value = g_malloc0(sizeof(*region->value));
1690 
1691         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1692             region->value->type = g_strdup("io");
1693         } else {
1694             region->value->type = g_strdup("memory");
1695             region->value->has_prefetch = true;
1696             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1697             region->value->has_mem_type_64 = true;
1698             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1699         }
1700 
1701         region->value->bar = i;
1702         region->value->address = r->addr;
1703         region->value->size = r->size;
1704 
1705         /* XXX: waiting for the qapi to support GSList */
1706         if (!cur_item) {
1707             head = cur_item = region;
1708         } else {
1709             cur_item->next = region;
1710             cur_item = region;
1711         }
1712     }
1713 
1714     return head;
1715 }
1716 
1717 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1718                                            int bus_num)
1719 {
1720     PciBridgeInfo *info;
1721     PciMemoryRange *range;
1722 
1723     info = g_new0(PciBridgeInfo, 1);
1724 
1725     info->bus = g_new0(PciBusInfo, 1);
1726     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1727     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1728     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1729 
1730     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1731     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1732     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1733 
1734     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1735     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1736     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1737 
1738     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1739     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1740     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1741 
1742     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1743         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1744         if (child_bus) {
1745             info->has_devices = true;
1746             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1747         }
1748     }
1749 
1750     return info;
1751 }
1752 
1753 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1754                                            int bus_num)
1755 {
1756     const pci_class_desc *desc;
1757     PciDeviceInfo *info;
1758     uint8_t type;
1759     int class;
1760 
1761     info = g_new0(PciDeviceInfo, 1);
1762     info->bus = bus_num;
1763     info->slot = PCI_SLOT(dev->devfn);
1764     info->function = PCI_FUNC(dev->devfn);
1765 
1766     info->class_info = g_new0(PciDeviceClass, 1);
1767     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1768     info->class_info->q_class = class;
1769     desc = get_class_desc(class);
1770     if (desc->desc) {
1771         info->class_info->has_desc = true;
1772         info->class_info->desc = g_strdup(desc->desc);
1773     }
1774 
1775     info->id = g_new0(PciDeviceId, 1);
1776     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1777     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1778     info->regions = qmp_query_pci_regions(dev);
1779     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1780 
1781     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1782         info->has_irq = true;
1783         info->irq = dev->config[PCI_INTERRUPT_LINE];
1784     }
1785 
1786     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1787     if (type == PCI_HEADER_TYPE_BRIDGE) {
1788         info->has_pci_bridge = true;
1789         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1790     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1791         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1792         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1793         info->id->subsystem_vendor =
1794             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1795     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1796         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1797         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1798         info->id->subsystem_vendor =
1799             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1800     }
1801 
1802     return info;
1803 }
1804 
1805 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1806 {
1807     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1808     PCIDevice *dev;
1809     int devfn;
1810 
1811     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1812         dev = bus->devices[devfn];
1813         if (dev) {
1814             info = g_malloc0(sizeof(*info));
1815             info->value = qmp_query_pci_device(dev, bus, bus_num);
1816 
1817             /* XXX: waiting for the qapi to support GSList */
1818             if (!cur_item) {
1819                 head = cur_item = info;
1820             } else {
1821                 cur_item->next = info;
1822                 cur_item = info;
1823             }
1824         }
1825     }
1826 
1827     return head;
1828 }
1829 
1830 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1831 {
1832     PciInfo *info = NULL;
1833 
1834     bus = pci_find_bus_nr(bus, bus_num);
1835     if (bus) {
1836         info = g_malloc0(sizeof(*info));
1837         info->bus = bus_num;
1838         info->devices = qmp_query_pci_devices(bus, bus_num);
1839     }
1840 
1841     return info;
1842 }
1843 
1844 PciInfoList *qmp_query_pci(Error **errp)
1845 {
1846     PciInfoList *info, *head = NULL, *cur_item = NULL;
1847     PCIHostState *host_bridge;
1848 
1849     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1850         info = g_malloc0(sizeof(*info));
1851         info->value = qmp_query_pci_bus(host_bridge->bus,
1852                                         pci_bus_num(host_bridge->bus));
1853 
1854         /* XXX: waiting for the qapi to support GSList */
1855         if (!cur_item) {
1856             head = cur_item = info;
1857         } else {
1858             cur_item->next = info;
1859             cur_item = info;
1860         }
1861     }
1862 
1863     return head;
1864 }
1865 
1866 /* Initialize a PCI NIC.  */
1867 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1868                                const char *default_model,
1869                                const char *default_devaddr)
1870 {
1871     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1872     GSList *list;
1873     GPtrArray *pci_nic_models;
1874     PCIBus *bus;
1875     PCIDevice *pci_dev;
1876     DeviceState *dev;
1877     int devfn;
1878     int i;
1879 
1880     if (nd->model && !strcmp(nd->model, "virtio")) {
1881         g_free(nd->model);
1882         nd->model = g_strdup("virtio-net-pci");
1883     }
1884 
1885     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1886     pci_nic_models = g_ptr_array_new();
1887     while (list) {
1888         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1889                                              TYPE_DEVICE);
1890         GSList *next;
1891         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1892             dc->user_creatable) {
1893             const char *name = object_class_get_name(list->data);
1894             g_ptr_array_add(pci_nic_models, (gpointer)name);
1895         }
1896         next = list->next;
1897         g_slist_free_1(list);
1898         list = next;
1899     }
1900     g_ptr_array_add(pci_nic_models, NULL);
1901 
1902     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1903         exit(0);
1904     }
1905 
1906     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1907                             default_model);
1908     if (i < 0) {
1909         exit(1);
1910     }
1911 
1912     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1913     if (!bus) {
1914         error_report("Invalid PCI device address %s for device %s",
1915                      devaddr, nd->model);
1916         exit(1);
1917     }
1918 
1919     pci_dev = pci_create(bus, devfn, nd->model);
1920     dev = &pci_dev->qdev;
1921     qdev_set_nic_properties(dev, nd);
1922     qdev_init_nofail(dev);
1923     g_ptr_array_free(pci_nic_models, true);
1924     return pci_dev;
1925 }
1926 
1927 PCIDevice *pci_vga_init(PCIBus *bus)
1928 {
1929     switch (vga_interface_type) {
1930     case VGA_CIRRUS:
1931         return pci_create_simple(bus, -1, "cirrus-vga");
1932     case VGA_QXL:
1933         return pci_create_simple(bus, -1, "qxl-vga");
1934     case VGA_STD:
1935         return pci_create_simple(bus, -1, "VGA");
1936     case VGA_VMWARE:
1937         return pci_create_simple(bus, -1, "vmware-svga");
1938     case VGA_VIRTIO:
1939         return pci_create_simple(bus, -1, "virtio-vga");
1940     case VGA_NONE:
1941     default: /* Other non-PCI types. Checking for unsupported types is already
1942                 done in vl.c. */
1943         return NULL;
1944     }
1945 }
1946 
1947 /* Whether a given bus number is in range of the secondary
1948  * bus of the given bridge device. */
1949 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1950 {
1951     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1952              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1953         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1954         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1955 }
1956 
1957 /* Whether a given bus number is in a range of a root bus */
1958 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1959 {
1960     int i;
1961 
1962     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1963         PCIDevice *dev = bus->devices[i];
1964 
1965         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1966             if (pci_secondary_bus_in_range(dev, bus_num)) {
1967                 return true;
1968             }
1969         }
1970     }
1971 
1972     return false;
1973 }
1974 
1975 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1976 {
1977     PCIBus *sec;
1978 
1979     if (!bus) {
1980         return NULL;
1981     }
1982 
1983     if (pci_bus_num(bus) == bus_num) {
1984         return bus;
1985     }
1986 
1987     /* Consider all bus numbers in range for the host pci bridge. */
1988     if (!pci_bus_is_root(bus) &&
1989         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1990         return NULL;
1991     }
1992 
1993     /* try child bus */
1994     for (; bus; bus = sec) {
1995         QLIST_FOREACH(sec, &bus->child, sibling) {
1996             if (pci_bus_num(sec) == bus_num) {
1997                 return sec;
1998             }
1999             /* PXB buses assumed to be children of bus 0 */
2000             if (pci_bus_is_root(sec)) {
2001                 if (pci_root_bus_in_range(sec, bus_num)) {
2002                     break;
2003                 }
2004             } else {
2005                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2006                     break;
2007                 }
2008             }
2009         }
2010     }
2011 
2012     return NULL;
2013 }
2014 
2015 void pci_for_each_bus_depth_first(PCIBus *bus,
2016                                   void *(*begin)(PCIBus *bus, void *parent_state),
2017                                   void (*end)(PCIBus *bus, void *state),
2018                                   void *parent_state)
2019 {
2020     PCIBus *sec;
2021     void *state;
2022 
2023     if (!bus) {
2024         return;
2025     }
2026 
2027     if (begin) {
2028         state = begin(bus, parent_state);
2029     } else {
2030         state = parent_state;
2031     }
2032 
2033     QLIST_FOREACH(sec, &bus->child, sibling) {
2034         pci_for_each_bus_depth_first(sec, begin, end, state);
2035     }
2036 
2037     if (end) {
2038         end(bus, state);
2039     }
2040 }
2041 
2042 
2043 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2044 {
2045     bus = pci_find_bus_nr(bus, bus_num);
2046 
2047     if (!bus)
2048         return NULL;
2049 
2050     return bus->devices[devfn];
2051 }
2052 
2053 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2054 {
2055     PCIDevice *pci_dev = (PCIDevice *)qdev;
2056     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2057     ObjectClass *klass = OBJECT_CLASS(pc);
2058     Error *local_err = NULL;
2059     bool is_default_rom;
2060 
2061     /* initialize cap_present for pci_is_express() and pci_config_size(),
2062      * Note that hybrid PCIs are not set automatically and need to manage
2063      * QEMU_PCI_CAP_EXPRESS manually */
2064     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2065        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2066         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2067     }
2068 
2069     pci_dev = do_pci_register_device(pci_dev,
2070                                      object_get_typename(OBJECT(qdev)),
2071                                      pci_dev->devfn, errp);
2072     if (pci_dev == NULL)
2073         return;
2074 
2075     if (pc->realize) {
2076         pc->realize(pci_dev, &local_err);
2077         if (local_err) {
2078             error_propagate(errp, local_err);
2079             do_pci_unregister_device(pci_dev);
2080             return;
2081         }
2082     }
2083 
2084     /* rom loading */
2085     is_default_rom = false;
2086     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2087         pci_dev->romfile = g_strdup(pc->romfile);
2088         is_default_rom = true;
2089     }
2090 
2091     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2092     if (local_err) {
2093         error_propagate(errp, local_err);
2094         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2095         return;
2096     }
2097 }
2098 
2099 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2100                                     const char *name)
2101 {
2102     DeviceState *dev;
2103 
2104     dev = qdev_create(&bus->qbus, name);
2105     qdev_prop_set_int32(dev, "addr", devfn);
2106     qdev_prop_set_bit(dev, "multifunction", multifunction);
2107     return PCI_DEVICE(dev);
2108 }
2109 
2110 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2111                                            bool multifunction,
2112                                            const char *name)
2113 {
2114     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2115     qdev_init_nofail(&dev->qdev);
2116     return dev;
2117 }
2118 
2119 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2120 {
2121     return pci_create_multifunction(bus, devfn, false, name);
2122 }
2123 
2124 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2125 {
2126     return pci_create_simple_multifunction(bus, devfn, false, name);
2127 }
2128 
2129 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2130 {
2131     int offset = PCI_CONFIG_HEADER_SIZE;
2132     int i;
2133     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2134         if (pdev->used[i])
2135             offset = i + 1;
2136         else if (i - offset + 1 == size)
2137             return offset;
2138     }
2139     return 0;
2140 }
2141 
2142 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2143                                         uint8_t *prev_p)
2144 {
2145     uint8_t next, prev;
2146 
2147     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2148         return 0;
2149 
2150     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2151          prev = next + PCI_CAP_LIST_NEXT)
2152         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2153             break;
2154 
2155     if (prev_p)
2156         *prev_p = prev;
2157     return next;
2158 }
2159 
2160 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2161 {
2162     uint8_t next, prev, found = 0;
2163 
2164     if (!(pdev->used[offset])) {
2165         return 0;
2166     }
2167 
2168     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2169 
2170     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2171          prev = next + PCI_CAP_LIST_NEXT) {
2172         if (next <= offset && next > found) {
2173             found = next;
2174         }
2175     }
2176     return found;
2177 }
2178 
2179 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2180    This is needed for an option rom which is used for more than one device. */
2181 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2182 {
2183     uint16_t vendor_id;
2184     uint16_t device_id;
2185     uint16_t rom_vendor_id;
2186     uint16_t rom_device_id;
2187     uint16_t rom_magic;
2188     uint16_t pcir_offset;
2189     uint8_t checksum;
2190 
2191     /* Words in rom data are little endian (like in PCI configuration),
2192        so they can be read / written with pci_get_word / pci_set_word. */
2193 
2194     /* Only a valid rom will be patched. */
2195     rom_magic = pci_get_word(ptr);
2196     if (rom_magic != 0xaa55) {
2197         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2198         return;
2199     }
2200     pcir_offset = pci_get_word(ptr + 0x18);
2201     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2202         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2203         return;
2204     }
2205 
2206     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2207     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2208     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2209     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2210 
2211     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2212                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2213 
2214     checksum = ptr[6];
2215 
2216     if (vendor_id != rom_vendor_id) {
2217         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2218         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2219         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2220         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2221         ptr[6] = checksum;
2222         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2223     }
2224 
2225     if (device_id != rom_device_id) {
2226         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2227         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2228         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2229         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2230         ptr[6] = checksum;
2231         pci_set_word(ptr + pcir_offset + 6, device_id);
2232     }
2233 }
2234 
2235 /* Add an option rom for the device */
2236 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2237                                Error **errp)
2238 {
2239     int size;
2240     char *path;
2241     void *ptr;
2242     char name[32];
2243     const VMStateDescription *vmsd;
2244 
2245     if (!pdev->romfile)
2246         return;
2247     if (strlen(pdev->romfile) == 0)
2248         return;
2249 
2250     if (!pdev->rom_bar) {
2251         /*
2252          * Load rom via fw_cfg instead of creating a rom bar,
2253          * for 0.11 compatibility.
2254          */
2255         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2256 
2257         /*
2258          * Hot-plugged devices can't use the option ROM
2259          * if the rom bar is disabled.
2260          */
2261         if (DEVICE(pdev)->hotplugged) {
2262             error_setg(errp, "Hot-plugged device without ROM bar"
2263                        " can't have an option ROM");
2264             return;
2265         }
2266 
2267         if (class == 0x0300) {
2268             rom_add_vga(pdev->romfile);
2269         } else {
2270             rom_add_option(pdev->romfile, -1);
2271         }
2272         return;
2273     }
2274 
2275     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2276     if (path == NULL) {
2277         path = g_strdup(pdev->romfile);
2278     }
2279 
2280     size = get_image_size(path);
2281     if (size < 0) {
2282         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2283         g_free(path);
2284         return;
2285     } else if (size == 0) {
2286         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2287         g_free(path);
2288         return;
2289     }
2290     size = pow2ceil(size);
2291 
2292     vmsd = qdev_get_vmsd(DEVICE(pdev));
2293 
2294     if (vmsd) {
2295         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2296     } else {
2297         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2298     }
2299     pdev->has_rom = true;
2300     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2301     ptr = memory_region_get_ram_ptr(&pdev->rom);
2302     if (load_image_size(path, ptr, size) < 0) {
2303         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2304         g_free(path);
2305         return;
2306     }
2307     g_free(path);
2308 
2309     if (is_default_rom) {
2310         /* Only the default rom images will be patched (if needed). */
2311         pci_patch_ids(pdev, ptr, size);
2312     }
2313 
2314     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2315 }
2316 
2317 static void pci_del_option_rom(PCIDevice *pdev)
2318 {
2319     if (!pdev->has_rom)
2320         return;
2321 
2322     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2323     pdev->has_rom = false;
2324 }
2325 
2326 /*
2327  * On success, pci_add_capability() returns a positive value
2328  * that the offset of the pci capability.
2329  * On failure, it sets an error and returns a negative error
2330  * code.
2331  */
2332 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2333                        uint8_t offset, uint8_t size,
2334                        Error **errp)
2335 {
2336     uint8_t *config;
2337     int i, overlapping_cap;
2338 
2339     if (!offset) {
2340         offset = pci_find_space(pdev, size);
2341         /* out of PCI config space is programming error */
2342         assert(offset);
2343     } else {
2344         /* Verify that capabilities don't overlap.  Note: device assignment
2345          * depends on this check to verify that the device is not broken.
2346          * Should never trigger for emulated devices, but it's helpful
2347          * for debugging these. */
2348         for (i = offset; i < offset + size; i++) {
2349             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2350             if (overlapping_cap) {
2351                 error_setg(errp, "%s:%02x:%02x.%x "
2352                            "Attempt to add PCI capability %x at offset "
2353                            "%x overlaps existing capability %x at offset %x",
2354                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2355                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2356                            cap_id, offset, overlapping_cap, i);
2357                 return -EINVAL;
2358             }
2359         }
2360     }
2361 
2362     config = pdev->config + offset;
2363     config[PCI_CAP_LIST_ID] = cap_id;
2364     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2365     pdev->config[PCI_CAPABILITY_LIST] = offset;
2366     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2367     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2368     /* Make capability read-only by default */
2369     memset(pdev->wmask + offset, 0, size);
2370     /* Check capability by default */
2371     memset(pdev->cmask + offset, 0xFF, size);
2372     return offset;
2373 }
2374 
2375 /* Unlink capability from the pci config space. */
2376 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2377 {
2378     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2379     if (!offset)
2380         return;
2381     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2382     /* Make capability writable again */
2383     memset(pdev->wmask + offset, 0xff, size);
2384     memset(pdev->w1cmask + offset, 0, size);
2385     /* Clear cmask as device-specific registers can't be checked */
2386     memset(pdev->cmask + offset, 0, size);
2387     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2388 
2389     if (!pdev->config[PCI_CAPABILITY_LIST])
2390         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2391 }
2392 
2393 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2394 {
2395     return pci_find_capability_list(pdev, cap_id, NULL);
2396 }
2397 
2398 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2399 {
2400     PCIDevice *d = (PCIDevice *)dev;
2401     const pci_class_desc *desc;
2402     char ctxt[64];
2403     PCIIORegion *r;
2404     int i, class;
2405 
2406     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2407     desc = pci_class_descriptions;
2408     while (desc->desc && class != desc->class)
2409         desc++;
2410     if (desc->desc) {
2411         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2412     } else {
2413         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2414     }
2415 
2416     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2417                    "pci id %04x:%04x (sub %04x:%04x)\n",
2418                    indent, "", ctxt, pci_dev_bus_num(d),
2419                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2420                    pci_get_word(d->config + PCI_VENDOR_ID),
2421                    pci_get_word(d->config + PCI_DEVICE_ID),
2422                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2423                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2424     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2425         r = &d->io_regions[i];
2426         if (!r->size)
2427             continue;
2428         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2429                        " [0x%"FMT_PCIBUS"]\n",
2430                        indent, "",
2431                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2432                        r->addr, r->addr + r->size - 1);
2433     }
2434 }
2435 
2436 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2437 {
2438     PCIDevice *d = (PCIDevice *)dev;
2439     const char *name = NULL;
2440     const pci_class_desc *desc =  pci_class_descriptions;
2441     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2442 
2443     while (desc->desc &&
2444           (class & ~desc->fw_ign_bits) !=
2445           (desc->class & ~desc->fw_ign_bits)) {
2446         desc++;
2447     }
2448 
2449     if (desc->desc) {
2450         name = desc->fw_name;
2451     }
2452 
2453     if (name) {
2454         pstrcpy(buf, len, name);
2455     } else {
2456         snprintf(buf, len, "pci%04x,%04x",
2457                  pci_get_word(d->config + PCI_VENDOR_ID),
2458                  pci_get_word(d->config + PCI_DEVICE_ID));
2459     }
2460 
2461     return buf;
2462 }
2463 
2464 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2465 {
2466     PCIDevice *d = (PCIDevice *)dev;
2467     char path[50], name[33];
2468     int off;
2469 
2470     off = snprintf(path, sizeof(path), "%s@%x",
2471                    pci_dev_fw_name(dev, name, sizeof name),
2472                    PCI_SLOT(d->devfn));
2473     if (PCI_FUNC(d->devfn))
2474         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2475     return g_strdup(path);
2476 }
2477 
2478 static char *pcibus_get_dev_path(DeviceState *dev)
2479 {
2480     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2481     PCIDevice *t;
2482     int slot_depth;
2483     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2484      * 00 is added here to make this format compatible with
2485      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2486      * Slot.Function list specifies the slot and function numbers for all
2487      * devices on the path from root to the specific device. */
2488     const char *root_bus_path;
2489     int root_bus_len;
2490     char slot[] = ":SS.F";
2491     int slot_len = sizeof slot - 1 /* For '\0' */;
2492     int path_len;
2493     char *path, *p;
2494     int s;
2495 
2496     root_bus_path = pci_root_bus_path(d);
2497     root_bus_len = strlen(root_bus_path);
2498 
2499     /* Calculate # of slots on path between device and root. */;
2500     slot_depth = 0;
2501     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2502         ++slot_depth;
2503     }
2504 
2505     path_len = root_bus_len + slot_len * slot_depth;
2506 
2507     /* Allocate memory, fill in the terminating null byte. */
2508     path = g_malloc(path_len + 1 /* For '\0' */);
2509     path[path_len] = '\0';
2510 
2511     memcpy(path, root_bus_path, root_bus_len);
2512 
2513     /* Fill in slot numbers. We walk up from device to root, so need to print
2514      * them in the reverse order, last to first. */
2515     p = path + path_len;
2516     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2517         p -= slot_len;
2518         s = snprintf(slot, sizeof slot, ":%02x.%x",
2519                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2520         assert(s == slot_len);
2521         memcpy(p, slot, slot_len);
2522     }
2523 
2524     return path;
2525 }
2526 
2527 static int pci_qdev_find_recursive(PCIBus *bus,
2528                                    const char *id, PCIDevice **pdev)
2529 {
2530     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2531     if (!qdev) {
2532         return -ENODEV;
2533     }
2534 
2535     /* roughly check if given qdev is pci device */
2536     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2537         *pdev = PCI_DEVICE(qdev);
2538         return 0;
2539     }
2540     return -EINVAL;
2541 }
2542 
2543 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2544 {
2545     PCIHostState *host_bridge;
2546     int rc = -ENODEV;
2547 
2548     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2549         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2550         if (!tmp) {
2551             rc = 0;
2552             break;
2553         }
2554         if (tmp != -ENODEV) {
2555             rc = tmp;
2556         }
2557     }
2558 
2559     return rc;
2560 }
2561 
2562 MemoryRegion *pci_address_space(PCIDevice *dev)
2563 {
2564     return pci_get_bus(dev)->address_space_mem;
2565 }
2566 
2567 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2568 {
2569     return pci_get_bus(dev)->address_space_io;
2570 }
2571 
2572 static void pci_device_class_init(ObjectClass *klass, void *data)
2573 {
2574     DeviceClass *k = DEVICE_CLASS(klass);
2575 
2576     k->realize = pci_qdev_realize;
2577     k->unrealize = pci_qdev_unrealize;
2578     k->bus_type = TYPE_PCI_BUS;
2579     k->props = pci_props;
2580 }
2581 
2582 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2583 {
2584     if (!object_class_is_abstract(klass)) {
2585         ObjectClass *conventional =
2586             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2587         ObjectClass *pcie =
2588             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2589         assert(conventional || pcie);
2590     }
2591 }
2592 
2593 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2594 {
2595     PCIBus *bus = pci_get_bus(dev);
2596     PCIBus *iommu_bus = bus;
2597 
2598     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2599         iommu_bus = pci_get_bus(iommu_bus->parent_dev);
2600     }
2601     if (iommu_bus && iommu_bus->iommu_fn) {
2602         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2603     }
2604     return &address_space_memory;
2605 }
2606 
2607 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2608 {
2609     bus->iommu_fn = fn;
2610     bus->iommu_opaque = opaque;
2611 }
2612 
2613 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2614 {
2615     Range *range = opaque;
2616     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2617     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2618     int i;
2619 
2620     if (!(cmd & PCI_COMMAND_MEMORY)) {
2621         return;
2622     }
2623 
2624     if (pc->is_bridge) {
2625         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2626         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2627 
2628         base = MAX(base, 0x1ULL << 32);
2629 
2630         if (limit >= base) {
2631             Range pref_range;
2632             range_set_bounds(&pref_range, base, limit);
2633             range_extend(range, &pref_range);
2634         }
2635     }
2636     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2637         PCIIORegion *r = &dev->io_regions[i];
2638         pcibus_t lob, upb;
2639         Range region_range;
2640 
2641         if (!r->size ||
2642             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2643             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2644             continue;
2645         }
2646 
2647         lob = pci_bar_address(dev, i, r->type, r->size);
2648         upb = lob + r->size - 1;
2649         if (lob == PCI_BAR_UNMAPPED) {
2650             continue;
2651         }
2652 
2653         lob = MAX(lob, 0x1ULL << 32);
2654 
2655         if (upb >= lob) {
2656             range_set_bounds(&region_range, lob, upb);
2657             range_extend(range, &region_range);
2658         }
2659     }
2660 }
2661 
2662 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2663 {
2664     range_make_empty(range);
2665     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2666 }
2667 
2668 static bool pcie_has_upstream_port(PCIDevice *dev)
2669 {
2670     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2671 
2672     /* Device associated with an upstream port.
2673      * As there are several types of these, it's easier to check the
2674      * parent device: upstream ports are always connected to
2675      * root or downstream ports.
2676      */
2677     return parent_dev &&
2678         pci_is_express(parent_dev) &&
2679         parent_dev->exp.exp_cap &&
2680         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2681          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2682 }
2683 
2684 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2685 {
2686     PCIBus *bus = pci_get_bus(pci_dev);
2687 
2688     if(pcie_has_upstream_port(pci_dev)) {
2689         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2690         return bus->devices[0];
2691     } else {
2692         /* Other bus types might support multiple devices at slots 0-31 */
2693         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2694     }
2695 }
2696 
2697 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2698 {
2699     MSIMessage msg;
2700     if (msix_enabled(dev)) {
2701         msg = msix_get_message(dev, vector);
2702     } else if (msi_enabled(dev)) {
2703         msg = msi_get_message(dev, vector);
2704     } else {
2705         /* Should never happen */
2706         error_report("%s: unknown interrupt type", __func__);
2707         abort();
2708     }
2709     return msg;
2710 }
2711 
2712 static const TypeInfo pci_device_type_info = {
2713     .name = TYPE_PCI_DEVICE,
2714     .parent = TYPE_DEVICE,
2715     .instance_size = sizeof(PCIDevice),
2716     .abstract = true,
2717     .class_size = sizeof(PCIDeviceClass),
2718     .class_init = pci_device_class_init,
2719     .class_base_init = pci_device_class_base_init,
2720 };
2721 
2722 static void pci_register_types(void)
2723 {
2724     type_register_static(&pci_bus_info);
2725     type_register_static(&pcie_bus_info);
2726     type_register_static(&conventional_pci_interface_info);
2727     type_register_static(&pcie_interface_info);
2728     type_register_static(&pci_device_type_info);
2729 }
2730 
2731 type_init(pci_register_types)
2732