xref: /openbmc/qemu/hw/pci/pci.c (revision 335d52f4)
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 PCIBus *pci_device_root_bus(const PCIDevice *d)
337 {
338     PCIBus *bus = pci_get_bus(d);
339 
340     while (!pci_bus_is_root(bus)) {
341         d = bus->parent_dev;
342         assert(d != NULL);
343 
344         bus = pci_get_bus(d);
345     }
346 
347     return bus;
348 }
349 
350 const char *pci_root_bus_path(PCIDevice *dev)
351 {
352     PCIBus *rootbus = pci_device_root_bus(dev);
353     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
354     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
355 
356     assert(host_bridge->bus == rootbus);
357 
358     if (hc->root_bus_path) {
359         return (*hc->root_bus_path)(host_bridge, rootbus);
360     }
361 
362     return rootbus->qbus.name;
363 }
364 
365 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
366                               MemoryRegion *address_space_mem,
367                               MemoryRegion *address_space_io,
368                               uint8_t devfn_min)
369 {
370     assert(PCI_FUNC(devfn_min) == 0);
371     bus->devfn_min = devfn_min;
372     bus->slot_reserved_mask = 0x0;
373     bus->address_space_mem = address_space_mem;
374     bus->address_space_io = address_space_io;
375 
376     /* host bridge */
377     QLIST_INIT(&bus->child);
378 
379     pci_host_bus_register(parent);
380 }
381 
382 bool pci_bus_is_express(PCIBus *bus)
383 {
384     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
385 }
386 
387 bool pci_bus_is_root(PCIBus *bus)
388 {
389     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
390 }
391 
392 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
393                               const char *name,
394                               MemoryRegion *address_space_mem,
395                               MemoryRegion *address_space_io,
396                               uint8_t devfn_min, const char *typename)
397 {
398     qbus_create_inplace(bus, bus_size, typename, parent, name);
399     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
400                       devfn_min);
401 }
402 
403 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
404                          MemoryRegion *address_space_mem,
405                          MemoryRegion *address_space_io,
406                          uint8_t devfn_min, const char *typename)
407 {
408     PCIBus *bus;
409 
410     bus = PCI_BUS(qbus_create(typename, parent, name));
411     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
412                       devfn_min);
413     return bus;
414 }
415 
416 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
417                   void *irq_opaque, int nirq)
418 {
419     bus->set_irq = set_irq;
420     bus->map_irq = map_irq;
421     bus->irq_opaque = irq_opaque;
422     bus->nirq = nirq;
423     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
424 }
425 
426 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
427                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
428                               void *irq_opaque,
429                               MemoryRegion *address_space_mem,
430                               MemoryRegion *address_space_io,
431                               uint8_t devfn_min, int nirq,
432                               const char *typename)
433 {
434     PCIBus *bus;
435 
436     bus = pci_root_bus_new(parent, name, address_space_mem,
437                            address_space_io, devfn_min, typename);
438     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
439     return bus;
440 }
441 
442 int pci_bus_num(PCIBus *s)
443 {
444     return PCI_BUS_GET_CLASS(s)->bus_num(s);
445 }
446 
447 int pci_bus_numa_node(PCIBus *bus)
448 {
449     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
450 }
451 
452 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
453                                  const VMStateField *field)
454 {
455     PCIDevice *s = container_of(pv, PCIDevice, config);
456     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
457     uint8_t *config;
458     int i;
459 
460     assert(size == pci_config_size(s));
461     config = g_malloc(size);
462 
463     qemu_get_buffer(f, config, size);
464     for (i = 0; i < size; ++i) {
465         if ((config[i] ^ s->config[i]) &
466             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
467             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
468                          "cmask: %x wmask: %x w1cmask:%x", __func__,
469                          i, config[i], s->config[i],
470                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
471             g_free(config);
472             return -EINVAL;
473         }
474     }
475     memcpy(s->config, config, size);
476 
477     pci_update_mappings(s);
478     if (pc->is_bridge) {
479         PCIBridge *b = PCI_BRIDGE(s);
480         pci_bridge_update_mappings(b);
481     }
482 
483     memory_region_set_enabled(&s->bus_master_enable_region,
484                               pci_get_word(s->config + PCI_COMMAND)
485                               & PCI_COMMAND_MASTER);
486 
487     g_free(config);
488     return 0;
489 }
490 
491 /* just put buffer */
492 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
493                                  const VMStateField *field, QJSON *vmdesc)
494 {
495     const uint8_t **v = pv;
496     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
497     qemu_put_buffer(f, *v, size);
498 
499     return 0;
500 }
501 
502 static VMStateInfo vmstate_info_pci_config = {
503     .name = "pci config",
504     .get  = get_pci_config_device,
505     .put  = put_pci_config_device,
506 };
507 
508 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
509                              const VMStateField *field)
510 {
511     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
512     uint32_t irq_state[PCI_NUM_PINS];
513     int i;
514     for (i = 0; i < PCI_NUM_PINS; ++i) {
515         irq_state[i] = qemu_get_be32(f);
516         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
517             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
518                     irq_state[i]);
519             return -EINVAL;
520         }
521     }
522 
523     for (i = 0; i < PCI_NUM_PINS; ++i) {
524         pci_set_irq_state(s, i, irq_state[i]);
525     }
526 
527     return 0;
528 }
529 
530 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
531                              const VMStateField *field, QJSON *vmdesc)
532 {
533     int i;
534     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
535 
536     for (i = 0; i < PCI_NUM_PINS; ++i) {
537         qemu_put_be32(f, pci_irq_state(s, i));
538     }
539 
540     return 0;
541 }
542 
543 static VMStateInfo vmstate_info_pci_irq_state = {
544     .name = "pci irq state",
545     .get  = get_pci_irq_state,
546     .put  = put_pci_irq_state,
547 };
548 
549 static bool migrate_is_pcie(void *opaque, int version_id)
550 {
551     return pci_is_express((PCIDevice *)opaque);
552 }
553 
554 static bool migrate_is_not_pcie(void *opaque, int version_id)
555 {
556     return !pci_is_express((PCIDevice *)opaque);
557 }
558 
559 const VMStateDescription vmstate_pci_device = {
560     .name = "PCIDevice",
561     .version_id = 2,
562     .minimum_version_id = 1,
563     .fields = (VMStateField[]) {
564         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
565         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
566                                    migrate_is_not_pcie,
567                                    0, vmstate_info_pci_config,
568                                    PCI_CONFIG_SPACE_SIZE),
569         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
570                                    migrate_is_pcie,
571                                    0, vmstate_info_pci_config,
572                                    PCIE_CONFIG_SPACE_SIZE),
573         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
574 				   vmstate_info_pci_irq_state,
575 				   PCI_NUM_PINS * sizeof(int32_t)),
576         VMSTATE_END_OF_LIST()
577     }
578 };
579 
580 
581 void pci_device_save(PCIDevice *s, QEMUFile *f)
582 {
583     /* Clear interrupt status bit: it is implicit
584      * in irq_state which we are saving.
585      * This makes us compatible with old devices
586      * which never set or clear this bit. */
587     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
588     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
589     /* Restore the interrupt status bit. */
590     pci_update_irq_status(s);
591 }
592 
593 int pci_device_load(PCIDevice *s, QEMUFile *f)
594 {
595     int ret;
596     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
597     /* Restore the interrupt status bit. */
598     pci_update_irq_status(s);
599     return ret;
600 }
601 
602 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
603 {
604     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
605                  pci_default_sub_vendor_id);
606     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
607                  pci_default_sub_device_id);
608 }
609 
610 /*
611  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
612  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
613  */
614 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
615                              unsigned int *slotp, unsigned int *funcp)
616 {
617     const char *p;
618     char *e;
619     unsigned long val;
620     unsigned long dom = 0, bus = 0;
621     unsigned int slot = 0;
622     unsigned int func = 0;
623 
624     p = addr;
625     val = strtoul(p, &e, 16);
626     if (e == p)
627 	return -1;
628     if (*e == ':') {
629 	bus = val;
630 	p = e + 1;
631 	val = strtoul(p, &e, 16);
632 	if (e == p)
633 	    return -1;
634 	if (*e == ':') {
635 	    dom = bus;
636 	    bus = val;
637 	    p = e + 1;
638 	    val = strtoul(p, &e, 16);
639 	    if (e == p)
640 		return -1;
641 	}
642     }
643 
644     slot = val;
645 
646     if (funcp != NULL) {
647         if (*e != '.')
648             return -1;
649 
650         p = e + 1;
651         val = strtoul(p, &e, 16);
652         if (e == p)
653             return -1;
654 
655         func = val;
656     }
657 
658     /* if funcp == NULL func is 0 */
659     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
660 	return -1;
661 
662     if (*e)
663 	return -1;
664 
665     *domp = dom;
666     *busp = bus;
667     *slotp = slot;
668     if (funcp != NULL)
669         *funcp = func;
670     return 0;
671 }
672 
673 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
674                                  const char *devaddr)
675 {
676     int dom, bus;
677     unsigned slot;
678 
679     if (!root) {
680         fprintf(stderr, "No primary PCI bus\n");
681         return NULL;
682     }
683 
684     assert(!root->parent_dev);
685 
686     if (!devaddr) {
687         *devfnp = -1;
688         return pci_find_bus_nr(root, 0);
689     }
690 
691     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
692         return NULL;
693     }
694 
695     if (dom != 0) {
696         fprintf(stderr, "No support for non-zero PCI domains\n");
697         return NULL;
698     }
699 
700     *devfnp = PCI_DEVFN(slot, 0);
701     return pci_find_bus_nr(root, bus);
702 }
703 
704 static void pci_init_cmask(PCIDevice *dev)
705 {
706     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
707     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
708     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
709     dev->cmask[PCI_REVISION_ID] = 0xff;
710     dev->cmask[PCI_CLASS_PROG] = 0xff;
711     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
712     dev->cmask[PCI_HEADER_TYPE] = 0xff;
713     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
714 }
715 
716 static void pci_init_wmask(PCIDevice *dev)
717 {
718     int config_size = pci_config_size(dev);
719 
720     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
721     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
722     pci_set_word(dev->wmask + PCI_COMMAND,
723                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
724                  PCI_COMMAND_INTX_DISABLE);
725     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
726         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
727     }
728 
729     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
730            config_size - PCI_CONFIG_HEADER_SIZE);
731 }
732 
733 static void pci_init_w1cmask(PCIDevice *dev)
734 {
735     /*
736      * Note: It's okay to set w1cmask even for readonly bits as
737      * long as their value is hardwired to 0.
738      */
739     pci_set_word(dev->w1cmask + PCI_STATUS,
740                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
741                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
742                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
743 }
744 
745 static void pci_init_mask_bridge(PCIDevice *d)
746 {
747     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
748        PCI_SEC_LETENCY_TIMER */
749     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
750 
751     /* base and limit */
752     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
753     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
754     pci_set_word(d->wmask + PCI_MEMORY_BASE,
755                  PCI_MEMORY_RANGE_MASK & 0xffff);
756     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
757                  PCI_MEMORY_RANGE_MASK & 0xffff);
758     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
759                  PCI_PREF_RANGE_MASK & 0xffff);
760     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
761                  PCI_PREF_RANGE_MASK & 0xffff);
762 
763     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
764     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
765 
766     /* Supported memory and i/o types */
767     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
768     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
769     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
770                                PCI_PREF_RANGE_TYPE_64);
771     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
772                                PCI_PREF_RANGE_TYPE_64);
773 
774     /*
775      * TODO: Bridges default to 10-bit VGA decoding but we currently only
776      * implement 16-bit decoding (no alias support).
777      */
778     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
779                  PCI_BRIDGE_CTL_PARITY |
780                  PCI_BRIDGE_CTL_SERR |
781                  PCI_BRIDGE_CTL_ISA |
782                  PCI_BRIDGE_CTL_VGA |
783                  PCI_BRIDGE_CTL_VGA_16BIT |
784                  PCI_BRIDGE_CTL_MASTER_ABORT |
785                  PCI_BRIDGE_CTL_BUS_RESET |
786                  PCI_BRIDGE_CTL_FAST_BACK |
787                  PCI_BRIDGE_CTL_DISCARD |
788                  PCI_BRIDGE_CTL_SEC_DISCARD |
789                  PCI_BRIDGE_CTL_DISCARD_SERR);
790     /* Below does not do anything as we never set this bit, put here for
791      * completeness. */
792     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
793                  PCI_BRIDGE_CTL_DISCARD_STATUS);
794     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
795     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
796     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
797                                PCI_PREF_RANGE_TYPE_MASK);
798     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
799                                PCI_PREF_RANGE_TYPE_MASK);
800 }
801 
802 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
803 {
804     uint8_t slot = PCI_SLOT(dev->devfn);
805     uint8_t func;
806 
807     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
808         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
809     }
810 
811     /*
812      * multifunction bit is interpreted in two ways as follows.
813      *   - all functions must set the bit to 1.
814      *     Example: Intel X53
815      *   - function 0 must set the bit, but the rest function (> 0)
816      *     is allowed to leave the bit to 0.
817      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
818      *
819      * So OS (at least Linux) checks the bit of only function 0,
820      * and doesn't see the bit of function > 0.
821      *
822      * The below check allows both interpretation.
823      */
824     if (PCI_FUNC(dev->devfn)) {
825         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
826         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
827             /* function 0 should set multifunction bit */
828             error_setg(errp, "PCI: single function device can't be populated "
829                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
830             return;
831         }
832         return;
833     }
834 
835     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
836         return;
837     }
838     /* function 0 indicates single function, so function > 0 must be NULL */
839     for (func = 1; func < PCI_FUNC_MAX; ++func) {
840         if (bus->devices[PCI_DEVFN(slot, func)]) {
841             error_setg(errp, "PCI: %x.0 indicates single function, "
842                        "but %x.%x is already populated.",
843                        slot, slot, func);
844             return;
845         }
846     }
847 }
848 
849 static void pci_config_alloc(PCIDevice *pci_dev)
850 {
851     int config_size = pci_config_size(pci_dev);
852 
853     pci_dev->config = g_malloc0(config_size);
854     pci_dev->cmask = g_malloc0(config_size);
855     pci_dev->wmask = g_malloc0(config_size);
856     pci_dev->w1cmask = g_malloc0(config_size);
857     pci_dev->used = g_malloc0(config_size);
858 }
859 
860 static void pci_config_free(PCIDevice *pci_dev)
861 {
862     g_free(pci_dev->config);
863     g_free(pci_dev->cmask);
864     g_free(pci_dev->wmask);
865     g_free(pci_dev->w1cmask);
866     g_free(pci_dev->used);
867 }
868 
869 static void do_pci_unregister_device(PCIDevice *pci_dev)
870 {
871     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
872     pci_config_free(pci_dev);
873 
874     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
875         memory_region_del_subregion(&pci_dev->bus_master_container_region,
876                                     &pci_dev->bus_master_enable_region);
877     }
878     address_space_destroy(&pci_dev->bus_master_as);
879 }
880 
881 /* Extract PCIReqIDCache into BDF format */
882 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
883 {
884     uint8_t bus_n;
885     uint16_t result;
886 
887     switch (cache->type) {
888     case PCI_REQ_ID_BDF:
889         result = pci_get_bdf(cache->dev);
890         break;
891     case PCI_REQ_ID_SECONDARY_BUS:
892         bus_n = pci_dev_bus_num(cache->dev);
893         result = PCI_BUILD_BDF(bus_n, 0);
894         break;
895     default:
896         error_printf("Invalid PCI requester ID cache type: %d\n",
897                      cache->type);
898         exit(1);
899         break;
900     }
901 
902     return result;
903 }
904 
905 /* Parse bridges up to the root complex and return requester ID
906  * cache for specific device.  For full PCIe topology, the cache
907  * result would be exactly the same as getting BDF of the device.
908  * However, several tricks are required when system mixed up with
909  * legacy PCI devices and PCIe-to-PCI bridges.
910  *
911  * Here we cache the proxy device (and type) not requester ID since
912  * bus number might change from time to time.
913  */
914 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
915 {
916     PCIDevice *parent;
917     PCIReqIDCache cache = {
918         .dev = dev,
919         .type = PCI_REQ_ID_BDF,
920     };
921 
922     while (!pci_bus_is_root(pci_get_bus(dev))) {
923         /* We are under PCI/PCIe bridges */
924         parent = pci_get_bus(dev)->parent_dev;
925         if (pci_is_express(parent)) {
926             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
927                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
928                  * override the requester ID using secondary bus
929                  * number of parent bridge with zeroed devfn
930                  * (pcie-to-pci bridge spec chap 2.3). */
931                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
932                 cache.dev = dev;
933             }
934         } else {
935             /* Legacy PCI, override requester ID with the bridge's
936              * BDF upstream.  When the root complex connects to
937              * legacy PCI devices (including buses), it can only
938              * obtain requester ID info from directly attached
939              * devices.  If devices are attached under bridges, only
940              * the requester ID of the bridge that is directly
941              * attached to the root complex can be recognized. */
942             cache.type = PCI_REQ_ID_BDF;
943             cache.dev = parent;
944         }
945         dev = parent;
946     }
947 
948     return cache;
949 }
950 
951 uint16_t pci_requester_id(PCIDevice *dev)
952 {
953     return pci_req_id_cache_extract(&dev->requester_id_cache);
954 }
955 
956 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
957 {
958     return !(bus->devices[devfn]);
959 }
960 
961 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
962 {
963     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
964 }
965 
966 /* -1 for devfn means auto assign */
967 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
968                                          const char *name, int devfn,
969                                          Error **errp)
970 {
971     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
972     PCIConfigReadFunc *config_read = pc->config_read;
973     PCIConfigWriteFunc *config_write = pc->config_write;
974     Error *local_err = NULL;
975     DeviceState *dev = DEVICE(pci_dev);
976     PCIBus *bus = pci_get_bus(pci_dev);
977 
978     /* Only pci bridges can be attached to extra PCI root buses */
979     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
980         error_setg(errp,
981                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
982                     bus->parent_dev->name);
983         return NULL;
984     }
985 
986     if (devfn < 0) {
987         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
988             devfn += PCI_FUNC_MAX) {
989             if (pci_bus_devfn_available(bus, devfn) &&
990                    !pci_bus_devfn_reserved(bus, devfn)) {
991                 goto found;
992             }
993         }
994         error_setg(errp, "PCI: no slot/function available for %s, all in use "
995                    "or reserved", name);
996         return NULL;
997     found: ;
998     } else if (pci_bus_devfn_reserved(bus, devfn)) {
999         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1000                    " reserved",
1001                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1002         return NULL;
1003     } else if (!pci_bus_devfn_available(bus, devfn)) {
1004         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1005                    " in use by %s",
1006                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1007                    bus->devices[devfn]->name);
1008         return NULL;
1009     } else if (dev->hotplugged &&
1010                pci_get_function_0(pci_dev)) {
1011         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1012                    " new func %s cannot be exposed to guest.",
1013                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1014                    pci_get_function_0(pci_dev)->name,
1015                    name);
1016 
1017        return NULL;
1018     }
1019 
1020     pci_dev->devfn = devfn;
1021     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1022     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1023 
1024     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1025                        "bus master container", UINT64_MAX);
1026     address_space_init(&pci_dev->bus_master_as,
1027                        &pci_dev->bus_master_container_region, pci_dev->name);
1028 
1029     if (qdev_hotplug) {
1030         pci_init_bus_master(pci_dev);
1031     }
1032     pci_dev->irq_state = 0;
1033     pci_config_alloc(pci_dev);
1034 
1035     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1036     pci_config_set_device_id(pci_dev->config, pc->device_id);
1037     pci_config_set_revision(pci_dev->config, pc->revision);
1038     pci_config_set_class(pci_dev->config, pc->class_id);
1039 
1040     if (!pc->is_bridge) {
1041         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1042             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1043                          pc->subsystem_vendor_id);
1044             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1045                          pc->subsystem_id);
1046         } else {
1047             pci_set_default_subsystem_id(pci_dev);
1048         }
1049     } else {
1050         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1051         assert(!pc->subsystem_vendor_id);
1052         assert(!pc->subsystem_id);
1053     }
1054     pci_init_cmask(pci_dev);
1055     pci_init_wmask(pci_dev);
1056     pci_init_w1cmask(pci_dev);
1057     if (pc->is_bridge) {
1058         pci_init_mask_bridge(pci_dev);
1059     }
1060     pci_init_multifunction(bus, pci_dev, &local_err);
1061     if (local_err) {
1062         error_propagate(errp, local_err);
1063         do_pci_unregister_device(pci_dev);
1064         return NULL;
1065     }
1066 
1067     if (!config_read)
1068         config_read = pci_default_read_config;
1069     if (!config_write)
1070         config_write = pci_default_write_config;
1071     pci_dev->config_read = config_read;
1072     pci_dev->config_write = config_write;
1073     bus->devices[devfn] = pci_dev;
1074     pci_dev->version_id = 2; /* Current pci device vmstate version */
1075     return pci_dev;
1076 }
1077 
1078 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1079 {
1080     PCIIORegion *r;
1081     int i;
1082 
1083     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1084         r = &pci_dev->io_regions[i];
1085         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1086             continue;
1087         memory_region_del_subregion(r->address_space, r->memory);
1088     }
1089 
1090     pci_unregister_vga(pci_dev);
1091 }
1092 
1093 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1094 {
1095     PCIDevice *pci_dev = PCI_DEVICE(dev);
1096     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1097 
1098     pci_unregister_io_regions(pci_dev);
1099     pci_del_option_rom(pci_dev);
1100 
1101     if (pc->exit) {
1102         pc->exit(pci_dev);
1103     }
1104 
1105     pci_device_deassert_intx(pci_dev);
1106     do_pci_unregister_device(pci_dev);
1107 }
1108 
1109 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1110                       uint8_t type, MemoryRegion *memory)
1111 {
1112     PCIIORegion *r;
1113     uint32_t addr; /* offset in pci config space */
1114     uint64_t wmask;
1115     pcibus_t size = memory_region_size(memory);
1116 
1117     assert(region_num >= 0);
1118     assert(region_num < PCI_NUM_REGIONS);
1119     if (size & (size-1)) {
1120         error_report("ERROR: PCI region size must be pow2 "
1121                     "type=0x%x, size=0x%"FMT_PCIBUS"", type, size);
1122         exit(1);
1123     }
1124 
1125     r = &pci_dev->io_regions[region_num];
1126     r->addr = PCI_BAR_UNMAPPED;
1127     r->size = size;
1128     r->type = type;
1129     r->memory = memory;
1130     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1131                         ? pci_get_bus(pci_dev)->address_space_io
1132                         : pci_get_bus(pci_dev)->address_space_mem;
1133 
1134     wmask = ~(size - 1);
1135     if (region_num == PCI_ROM_SLOT) {
1136         /* ROM enable bit is writable */
1137         wmask |= PCI_ROM_ADDRESS_ENABLE;
1138     }
1139 
1140     addr = pci_bar(pci_dev, region_num);
1141     pci_set_long(pci_dev->config + addr, type);
1142 
1143     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1144         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1145         pci_set_quad(pci_dev->wmask + addr, wmask);
1146         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1147     } else {
1148         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1149         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1150     }
1151 }
1152 
1153 static void pci_update_vga(PCIDevice *pci_dev)
1154 {
1155     uint16_t cmd;
1156 
1157     if (!pci_dev->has_vga) {
1158         return;
1159     }
1160 
1161     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1162 
1163     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1164                               cmd & PCI_COMMAND_MEMORY);
1165     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1166                               cmd & PCI_COMMAND_IO);
1167     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1168                               cmd & PCI_COMMAND_IO);
1169 }
1170 
1171 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1172                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1173 {
1174     PCIBus *bus = pci_get_bus(pci_dev);
1175 
1176     assert(!pci_dev->has_vga);
1177 
1178     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1179     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1180     memory_region_add_subregion_overlap(bus->address_space_mem,
1181                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1182 
1183     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1184     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1185     memory_region_add_subregion_overlap(bus->address_space_io,
1186                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1187 
1188     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1189     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1190     memory_region_add_subregion_overlap(bus->address_space_io,
1191                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1192     pci_dev->has_vga = true;
1193 
1194     pci_update_vga(pci_dev);
1195 }
1196 
1197 void pci_unregister_vga(PCIDevice *pci_dev)
1198 {
1199     PCIBus *bus = pci_get_bus(pci_dev);
1200 
1201     if (!pci_dev->has_vga) {
1202         return;
1203     }
1204 
1205     memory_region_del_subregion(bus->address_space_mem,
1206                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1207     memory_region_del_subregion(bus->address_space_io,
1208                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1209     memory_region_del_subregion(bus->address_space_io,
1210                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1211     pci_dev->has_vga = false;
1212 }
1213 
1214 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1215 {
1216     return pci_dev->io_regions[region_num].addr;
1217 }
1218 
1219 static pcibus_t pci_bar_address(PCIDevice *d,
1220 				int reg, uint8_t type, pcibus_t size)
1221 {
1222     pcibus_t new_addr, last_addr;
1223     int bar = pci_bar(d, reg);
1224     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1225     Object *machine = qdev_get_machine();
1226     ObjectClass *oc = object_get_class(machine);
1227     MachineClass *mc = MACHINE_CLASS(oc);
1228     bool allow_0_address = mc->pci_allow_0_address;
1229 
1230     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1231         if (!(cmd & PCI_COMMAND_IO)) {
1232             return PCI_BAR_UNMAPPED;
1233         }
1234         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1235         last_addr = new_addr + size - 1;
1236         /* Check if 32 bit BAR wraps around explicitly.
1237          * TODO: make priorities correct and remove this work around.
1238          */
1239         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1240             (!allow_0_address && new_addr == 0)) {
1241             return PCI_BAR_UNMAPPED;
1242         }
1243         return new_addr;
1244     }
1245 
1246     if (!(cmd & PCI_COMMAND_MEMORY)) {
1247         return PCI_BAR_UNMAPPED;
1248     }
1249     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1250         new_addr = pci_get_quad(d->config + bar);
1251     } else {
1252         new_addr = pci_get_long(d->config + bar);
1253     }
1254     /* the ROM slot has a specific enable bit */
1255     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1256         return PCI_BAR_UNMAPPED;
1257     }
1258     new_addr &= ~(size - 1);
1259     last_addr = new_addr + size - 1;
1260     /* NOTE: we do not support wrapping */
1261     /* XXX: as we cannot support really dynamic
1262        mappings, we handle specific values as invalid
1263        mappings. */
1264     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1265         (!allow_0_address && new_addr == 0)) {
1266         return PCI_BAR_UNMAPPED;
1267     }
1268 
1269     /* Now pcibus_t is 64bit.
1270      * Check if 32 bit BAR wraps around explicitly.
1271      * Without this, PC ide doesn't work well.
1272      * TODO: remove this work around.
1273      */
1274     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1275         return PCI_BAR_UNMAPPED;
1276     }
1277 
1278     /*
1279      * OS is allowed to set BAR beyond its addressable
1280      * bits. For example, 32 bit OS can set 64bit bar
1281      * to >4G. Check it. TODO: we might need to support
1282      * it in the future for e.g. PAE.
1283      */
1284     if (last_addr >= HWADDR_MAX) {
1285         return PCI_BAR_UNMAPPED;
1286     }
1287 
1288     return new_addr;
1289 }
1290 
1291 static void pci_update_mappings(PCIDevice *d)
1292 {
1293     PCIIORegion *r;
1294     int i;
1295     pcibus_t new_addr;
1296 
1297     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1298         r = &d->io_regions[i];
1299 
1300         /* this region isn't registered */
1301         if (!r->size)
1302             continue;
1303 
1304         new_addr = pci_bar_address(d, i, r->type, r->size);
1305 
1306         /* This bar isn't changed */
1307         if (new_addr == r->addr)
1308             continue;
1309 
1310         /* now do the real mapping */
1311         if (r->addr != PCI_BAR_UNMAPPED) {
1312             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1313                                           PCI_SLOT(d->devfn),
1314                                           PCI_FUNC(d->devfn),
1315                                           i, r->addr, r->size);
1316             memory_region_del_subregion(r->address_space, r->memory);
1317         }
1318         r->addr = new_addr;
1319         if (r->addr != PCI_BAR_UNMAPPED) {
1320             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1321                                           PCI_SLOT(d->devfn),
1322                                           PCI_FUNC(d->devfn),
1323                                           i, r->addr, r->size);
1324             memory_region_add_subregion_overlap(r->address_space,
1325                                                 r->addr, r->memory, 1);
1326         }
1327     }
1328 
1329     pci_update_vga(d);
1330 }
1331 
1332 static inline int pci_irq_disabled(PCIDevice *d)
1333 {
1334     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1335 }
1336 
1337 /* Called after interrupt disabled field update in config space,
1338  * assert/deassert interrupts if necessary.
1339  * Gets original interrupt disable bit value (before update). */
1340 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1341 {
1342     int i, disabled = pci_irq_disabled(d);
1343     if (disabled == was_irq_disabled)
1344         return;
1345     for (i = 0; i < PCI_NUM_PINS; ++i) {
1346         int state = pci_irq_state(d, i);
1347         pci_change_irq_level(d, i, disabled ? -state : state);
1348     }
1349 }
1350 
1351 uint32_t pci_default_read_config(PCIDevice *d,
1352                                  uint32_t address, int len)
1353 {
1354     uint32_t val = 0;
1355 
1356     if (pci_is_express_downstream_port(d) &&
1357         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1358         pcie_sync_bridge_lnk(d);
1359     }
1360     memcpy(&val, d->config + address, len);
1361     return le32_to_cpu(val);
1362 }
1363 
1364 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1365 {
1366     int i, was_irq_disabled = pci_irq_disabled(d);
1367     uint32_t val = val_in;
1368 
1369     for (i = 0; i < l; val >>= 8, ++i) {
1370         uint8_t wmask = d->wmask[addr + i];
1371         uint8_t w1cmask = d->w1cmask[addr + i];
1372         assert(!(wmask & w1cmask));
1373         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1374         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1375     }
1376     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1377         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1378         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1379         range_covers_byte(addr, l, PCI_COMMAND))
1380         pci_update_mappings(d);
1381 
1382     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1383         pci_update_irq_disabled(d, was_irq_disabled);
1384         memory_region_set_enabled(&d->bus_master_enable_region,
1385                                   pci_get_word(d->config + PCI_COMMAND)
1386                                     & PCI_COMMAND_MASTER);
1387     }
1388 
1389     msi_write_config(d, addr, val_in, l);
1390     msix_write_config(d, addr, val_in, l);
1391 }
1392 
1393 /***********************************************************/
1394 /* generic PCI irq support */
1395 
1396 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1397 static void pci_irq_handler(void *opaque, int irq_num, int level)
1398 {
1399     PCIDevice *pci_dev = opaque;
1400     int change;
1401 
1402     change = level - pci_irq_state(pci_dev, irq_num);
1403     if (!change)
1404         return;
1405 
1406     pci_set_irq_state(pci_dev, irq_num, level);
1407     pci_update_irq_status(pci_dev);
1408     if (pci_irq_disabled(pci_dev))
1409         return;
1410     pci_change_irq_level(pci_dev, irq_num, change);
1411 }
1412 
1413 static inline int pci_intx(PCIDevice *pci_dev)
1414 {
1415     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1416 }
1417 
1418 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1419 {
1420     int intx = pci_intx(pci_dev);
1421 
1422     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1423 }
1424 
1425 void pci_set_irq(PCIDevice *pci_dev, int level)
1426 {
1427     int intx = pci_intx(pci_dev);
1428     pci_irq_handler(pci_dev, intx, level);
1429 }
1430 
1431 /* Special hooks used by device assignment */
1432 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1433 {
1434     assert(pci_bus_is_root(bus));
1435     bus->route_intx_to_irq = route_intx_to_irq;
1436 }
1437 
1438 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1439 {
1440     PCIBus *bus;
1441 
1442     do {
1443         bus = pci_get_bus(dev);
1444         pin = bus->map_irq(dev, pin);
1445         dev = bus->parent_dev;
1446     } while (dev);
1447 
1448     if (!bus->route_intx_to_irq) {
1449         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1450                      object_get_typename(OBJECT(bus->qbus.parent)));
1451         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1452     }
1453 
1454     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1455 }
1456 
1457 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1458 {
1459     return old->mode != new->mode || old->irq != new->irq;
1460 }
1461 
1462 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1463 {
1464     PCIDevice *dev;
1465     PCIBus *sec;
1466     int i;
1467 
1468     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1469         dev = bus->devices[i];
1470         if (dev && dev->intx_routing_notifier) {
1471             dev->intx_routing_notifier(dev);
1472         }
1473     }
1474 
1475     QLIST_FOREACH(sec, &bus->child, sibling) {
1476         pci_bus_fire_intx_routing_notifier(sec);
1477     }
1478 }
1479 
1480 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1481                                           PCIINTxRoutingNotifier notifier)
1482 {
1483     dev->intx_routing_notifier = notifier;
1484 }
1485 
1486 /*
1487  * PCI-to-PCI bridge specification
1488  * 9.1: Interrupt routing. Table 9-1
1489  *
1490  * the PCI Express Base Specification, Revision 2.1
1491  * 2.2.8.1: INTx interrutp signaling - Rules
1492  *          the Implementation Note
1493  *          Table 2-20
1494  */
1495 /*
1496  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1497  * 0-origin unlike PCI interrupt pin register.
1498  */
1499 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1500 {
1501     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1502 }
1503 
1504 /***********************************************************/
1505 /* monitor info on PCI */
1506 
1507 typedef struct {
1508     uint16_t class;
1509     const char *desc;
1510     const char *fw_name;
1511     uint16_t fw_ign_bits;
1512 } pci_class_desc;
1513 
1514 static const pci_class_desc pci_class_descriptions[] =
1515 {
1516     { 0x0001, "VGA controller", "display"},
1517     { 0x0100, "SCSI controller", "scsi"},
1518     { 0x0101, "IDE controller", "ide"},
1519     { 0x0102, "Floppy controller", "fdc"},
1520     { 0x0103, "IPI controller", "ipi"},
1521     { 0x0104, "RAID controller", "raid"},
1522     { 0x0106, "SATA controller"},
1523     { 0x0107, "SAS controller"},
1524     { 0x0180, "Storage controller"},
1525     { 0x0200, "Ethernet controller", "ethernet"},
1526     { 0x0201, "Token Ring controller", "token-ring"},
1527     { 0x0202, "FDDI controller", "fddi"},
1528     { 0x0203, "ATM controller", "atm"},
1529     { 0x0280, "Network controller"},
1530     { 0x0300, "VGA controller", "display", 0x00ff},
1531     { 0x0301, "XGA controller"},
1532     { 0x0302, "3D controller"},
1533     { 0x0380, "Display controller"},
1534     { 0x0400, "Video controller", "video"},
1535     { 0x0401, "Audio controller", "sound"},
1536     { 0x0402, "Phone"},
1537     { 0x0403, "Audio controller", "sound"},
1538     { 0x0480, "Multimedia controller"},
1539     { 0x0500, "RAM controller", "memory"},
1540     { 0x0501, "Flash controller", "flash"},
1541     { 0x0580, "Memory controller"},
1542     { 0x0600, "Host bridge", "host"},
1543     { 0x0601, "ISA bridge", "isa"},
1544     { 0x0602, "EISA bridge", "eisa"},
1545     { 0x0603, "MC bridge", "mca"},
1546     { 0x0604, "PCI bridge", "pci-bridge"},
1547     { 0x0605, "PCMCIA bridge", "pcmcia"},
1548     { 0x0606, "NUBUS bridge", "nubus"},
1549     { 0x0607, "CARDBUS bridge", "cardbus"},
1550     { 0x0608, "RACEWAY bridge"},
1551     { 0x0680, "Bridge"},
1552     { 0x0700, "Serial port", "serial"},
1553     { 0x0701, "Parallel port", "parallel"},
1554     { 0x0800, "Interrupt controller", "interrupt-controller"},
1555     { 0x0801, "DMA controller", "dma-controller"},
1556     { 0x0802, "Timer", "timer"},
1557     { 0x0803, "RTC", "rtc"},
1558     { 0x0900, "Keyboard", "keyboard"},
1559     { 0x0901, "Pen", "pen"},
1560     { 0x0902, "Mouse", "mouse"},
1561     { 0x0A00, "Dock station", "dock", 0x00ff},
1562     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1563     { 0x0c00, "Fireware contorller", "fireware"},
1564     { 0x0c01, "Access bus controller", "access-bus"},
1565     { 0x0c02, "SSA controller", "ssa"},
1566     { 0x0c03, "USB controller", "usb"},
1567     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1568     { 0x0c05, "SMBus"},
1569     { 0, NULL}
1570 };
1571 
1572 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1573                                                   void (*fn)(PCIBus *b,
1574                                                              PCIDevice *d,
1575                                                              void *opaque),
1576                                                   void *opaque)
1577 {
1578     PCIDevice *d;
1579     int devfn;
1580 
1581     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1582         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1583         if (d) {
1584             fn(bus, d, opaque);
1585         }
1586     }
1587 }
1588 
1589 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1590                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1591                          void *opaque)
1592 {
1593     bus = pci_find_bus_nr(bus, bus_num);
1594 
1595     if (bus) {
1596         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1597     }
1598 }
1599 
1600 static void pci_for_each_device_under_bus(PCIBus *bus,
1601                                           void (*fn)(PCIBus *b, PCIDevice *d,
1602                                                      void *opaque),
1603                                           void *opaque)
1604 {
1605     PCIDevice *d;
1606     int devfn;
1607 
1608     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1609         d = bus->devices[devfn];
1610         if (d) {
1611             fn(bus, d, opaque);
1612         }
1613     }
1614 }
1615 
1616 void pci_for_each_device(PCIBus *bus, int bus_num,
1617                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1618                          void *opaque)
1619 {
1620     bus = pci_find_bus_nr(bus, bus_num);
1621 
1622     if (bus) {
1623         pci_for_each_device_under_bus(bus, fn, opaque);
1624     }
1625 }
1626 
1627 static const pci_class_desc *get_class_desc(int class)
1628 {
1629     const pci_class_desc *desc;
1630 
1631     desc = pci_class_descriptions;
1632     while (desc->desc && class != desc->class) {
1633         desc++;
1634     }
1635 
1636     return desc;
1637 }
1638 
1639 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1640 
1641 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1642 {
1643     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1644     int i;
1645 
1646     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1647         const PCIIORegion *r = &dev->io_regions[i];
1648         PciMemoryRegionList *region;
1649 
1650         if (!r->size) {
1651             continue;
1652         }
1653 
1654         region = g_malloc0(sizeof(*region));
1655         region->value = g_malloc0(sizeof(*region->value));
1656 
1657         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1658             region->value->type = g_strdup("io");
1659         } else {
1660             region->value->type = g_strdup("memory");
1661             region->value->has_prefetch = true;
1662             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1663             region->value->has_mem_type_64 = true;
1664             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1665         }
1666 
1667         region->value->bar = i;
1668         region->value->address = r->addr;
1669         region->value->size = r->size;
1670 
1671         /* XXX: waiting for the qapi to support GSList */
1672         if (!cur_item) {
1673             head = cur_item = region;
1674         } else {
1675             cur_item->next = region;
1676             cur_item = region;
1677         }
1678     }
1679 
1680     return head;
1681 }
1682 
1683 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1684                                            int bus_num)
1685 {
1686     PciBridgeInfo *info;
1687     PciMemoryRange *range;
1688 
1689     info = g_new0(PciBridgeInfo, 1);
1690 
1691     info->bus = g_new0(PciBusInfo, 1);
1692     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1693     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1694     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1695 
1696     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1697     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1698     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1699 
1700     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1701     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1702     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1703 
1704     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1705     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1706     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1707 
1708     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1709         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1710         if (child_bus) {
1711             info->has_devices = true;
1712             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1713         }
1714     }
1715 
1716     return info;
1717 }
1718 
1719 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1720                                            int bus_num)
1721 {
1722     const pci_class_desc *desc;
1723     PciDeviceInfo *info;
1724     uint8_t type;
1725     int class;
1726 
1727     info = g_new0(PciDeviceInfo, 1);
1728     info->bus = bus_num;
1729     info->slot = PCI_SLOT(dev->devfn);
1730     info->function = PCI_FUNC(dev->devfn);
1731 
1732     info->class_info = g_new0(PciDeviceClass, 1);
1733     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1734     info->class_info->q_class = class;
1735     desc = get_class_desc(class);
1736     if (desc->desc) {
1737         info->class_info->has_desc = true;
1738         info->class_info->desc = g_strdup(desc->desc);
1739     }
1740 
1741     info->id = g_new0(PciDeviceId, 1);
1742     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1743     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1744     info->regions = qmp_query_pci_regions(dev);
1745     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1746 
1747     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1748         info->has_irq = true;
1749         info->irq = dev->config[PCI_INTERRUPT_LINE];
1750     }
1751 
1752     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1753     if (type == PCI_HEADER_TYPE_BRIDGE) {
1754         info->has_pci_bridge = true;
1755         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1756     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1757         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1758         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1759         info->id->subsystem_vendor =
1760             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1761     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1762         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1763         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1764         info->id->subsystem_vendor =
1765             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1766     }
1767 
1768     return info;
1769 }
1770 
1771 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1772 {
1773     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1774     PCIDevice *dev;
1775     int devfn;
1776 
1777     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1778         dev = bus->devices[devfn];
1779         if (dev) {
1780             info = g_malloc0(sizeof(*info));
1781             info->value = qmp_query_pci_device(dev, bus, bus_num);
1782 
1783             /* XXX: waiting for the qapi to support GSList */
1784             if (!cur_item) {
1785                 head = cur_item = info;
1786             } else {
1787                 cur_item->next = info;
1788                 cur_item = info;
1789             }
1790         }
1791     }
1792 
1793     return head;
1794 }
1795 
1796 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1797 {
1798     PciInfo *info = NULL;
1799 
1800     bus = pci_find_bus_nr(bus, bus_num);
1801     if (bus) {
1802         info = g_malloc0(sizeof(*info));
1803         info->bus = bus_num;
1804         info->devices = qmp_query_pci_devices(bus, bus_num);
1805     }
1806 
1807     return info;
1808 }
1809 
1810 PciInfoList *qmp_query_pci(Error **errp)
1811 {
1812     PciInfoList *info, *head = NULL, *cur_item = NULL;
1813     PCIHostState *host_bridge;
1814 
1815     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1816         info = g_malloc0(sizeof(*info));
1817         info->value = qmp_query_pci_bus(host_bridge->bus,
1818                                         pci_bus_num(host_bridge->bus));
1819 
1820         /* XXX: waiting for the qapi to support GSList */
1821         if (!cur_item) {
1822             head = cur_item = info;
1823         } else {
1824             cur_item->next = info;
1825             cur_item = info;
1826         }
1827     }
1828 
1829     return head;
1830 }
1831 
1832 /* Initialize a PCI NIC.  */
1833 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1834                                const char *default_model,
1835                                const char *default_devaddr)
1836 {
1837     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1838     GSList *list;
1839     GPtrArray *pci_nic_models;
1840     PCIBus *bus;
1841     PCIDevice *pci_dev;
1842     DeviceState *dev;
1843     int devfn;
1844     int i;
1845 
1846     if (nd->model && !strcmp(nd->model, "virtio")) {
1847         g_free(nd->model);
1848         nd->model = g_strdup("virtio-net-pci");
1849     }
1850 
1851     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1852     pci_nic_models = g_ptr_array_new();
1853     while (list) {
1854         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1855                                              TYPE_DEVICE);
1856         GSList *next;
1857         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1858             dc->user_creatable) {
1859             const char *name = object_class_get_name(list->data);
1860             g_ptr_array_add(pci_nic_models, (gpointer)name);
1861         }
1862         next = list->next;
1863         g_slist_free_1(list);
1864         list = next;
1865     }
1866     g_ptr_array_add(pci_nic_models, NULL);
1867 
1868     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1869         exit(0);
1870     }
1871 
1872     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1873                             default_model);
1874     if (i < 0) {
1875         exit(1);
1876     }
1877 
1878     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1879     if (!bus) {
1880         error_report("Invalid PCI device address %s for device %s",
1881                      devaddr, nd->model);
1882         exit(1);
1883     }
1884 
1885     pci_dev = pci_create(bus, devfn, nd->model);
1886     dev = &pci_dev->qdev;
1887     qdev_set_nic_properties(dev, nd);
1888     qdev_init_nofail(dev);
1889     g_ptr_array_free(pci_nic_models, true);
1890     return pci_dev;
1891 }
1892 
1893 PCIDevice *pci_vga_init(PCIBus *bus)
1894 {
1895     switch (vga_interface_type) {
1896     case VGA_CIRRUS:
1897         return pci_create_simple(bus, -1, "cirrus-vga");
1898     case VGA_QXL:
1899         return pci_create_simple(bus, -1, "qxl-vga");
1900     case VGA_STD:
1901         return pci_create_simple(bus, -1, "VGA");
1902     case VGA_VMWARE:
1903         return pci_create_simple(bus, -1, "vmware-svga");
1904     case VGA_VIRTIO:
1905         return pci_create_simple(bus, -1, "virtio-vga");
1906     case VGA_NONE:
1907     default: /* Other non-PCI types. Checking for unsupported types is already
1908                 done in vl.c. */
1909         return NULL;
1910     }
1911 }
1912 
1913 /* Whether a given bus number is in range of the secondary
1914  * bus of the given bridge device. */
1915 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1916 {
1917     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1918              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1919         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1920         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1921 }
1922 
1923 /* Whether a given bus number is in a range of a root bus */
1924 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1925 {
1926     int i;
1927 
1928     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1929         PCIDevice *dev = bus->devices[i];
1930 
1931         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1932             if (pci_secondary_bus_in_range(dev, bus_num)) {
1933                 return true;
1934             }
1935         }
1936     }
1937 
1938     return false;
1939 }
1940 
1941 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1942 {
1943     PCIBus *sec;
1944 
1945     if (!bus) {
1946         return NULL;
1947     }
1948 
1949     if (pci_bus_num(bus) == bus_num) {
1950         return bus;
1951     }
1952 
1953     /* Consider all bus numbers in range for the host pci bridge. */
1954     if (!pci_bus_is_root(bus) &&
1955         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1956         return NULL;
1957     }
1958 
1959     /* try child bus */
1960     for (; bus; bus = sec) {
1961         QLIST_FOREACH(sec, &bus->child, sibling) {
1962             if (pci_bus_num(sec) == bus_num) {
1963                 return sec;
1964             }
1965             /* PXB buses assumed to be children of bus 0 */
1966             if (pci_bus_is_root(sec)) {
1967                 if (pci_root_bus_in_range(sec, bus_num)) {
1968                     break;
1969                 }
1970             } else {
1971                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1972                     break;
1973                 }
1974             }
1975         }
1976     }
1977 
1978     return NULL;
1979 }
1980 
1981 void pci_for_each_bus_depth_first(PCIBus *bus,
1982                                   void *(*begin)(PCIBus *bus, void *parent_state),
1983                                   void (*end)(PCIBus *bus, void *state),
1984                                   void *parent_state)
1985 {
1986     PCIBus *sec;
1987     void *state;
1988 
1989     if (!bus) {
1990         return;
1991     }
1992 
1993     if (begin) {
1994         state = begin(bus, parent_state);
1995     } else {
1996         state = parent_state;
1997     }
1998 
1999     QLIST_FOREACH(sec, &bus->child, sibling) {
2000         pci_for_each_bus_depth_first(sec, begin, end, state);
2001     }
2002 
2003     if (end) {
2004         end(bus, state);
2005     }
2006 }
2007 
2008 
2009 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2010 {
2011     bus = pci_find_bus_nr(bus, bus_num);
2012 
2013     if (!bus)
2014         return NULL;
2015 
2016     return bus->devices[devfn];
2017 }
2018 
2019 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2020 {
2021     PCIDevice *pci_dev = (PCIDevice *)qdev;
2022     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2023     ObjectClass *klass = OBJECT_CLASS(pc);
2024     Error *local_err = NULL;
2025     bool is_default_rom;
2026 
2027     /* initialize cap_present for pci_is_express() and pci_config_size(),
2028      * Note that hybrid PCIs are not set automatically and need to manage
2029      * QEMU_PCI_CAP_EXPRESS manually */
2030     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2031        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2032         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2033     }
2034 
2035     pci_dev = do_pci_register_device(pci_dev,
2036                                      object_get_typename(OBJECT(qdev)),
2037                                      pci_dev->devfn, errp);
2038     if (pci_dev == NULL)
2039         return;
2040 
2041     if (pc->realize) {
2042         pc->realize(pci_dev, &local_err);
2043         if (local_err) {
2044             error_propagate(errp, local_err);
2045             do_pci_unregister_device(pci_dev);
2046             return;
2047         }
2048     }
2049 
2050     /* rom loading */
2051     is_default_rom = false;
2052     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2053         pci_dev->romfile = g_strdup(pc->romfile);
2054         is_default_rom = true;
2055     }
2056 
2057     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2058     if (local_err) {
2059         error_propagate(errp, local_err);
2060         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2061         return;
2062     }
2063 }
2064 
2065 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2066                                     const char *name)
2067 {
2068     DeviceState *dev;
2069 
2070     dev = qdev_create(&bus->qbus, name);
2071     qdev_prop_set_int32(dev, "addr", devfn);
2072     qdev_prop_set_bit(dev, "multifunction", multifunction);
2073     return PCI_DEVICE(dev);
2074 }
2075 
2076 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2077                                            bool multifunction,
2078                                            const char *name)
2079 {
2080     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2081     qdev_init_nofail(&dev->qdev);
2082     return dev;
2083 }
2084 
2085 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2086 {
2087     return pci_create_multifunction(bus, devfn, false, name);
2088 }
2089 
2090 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2091 {
2092     return pci_create_simple_multifunction(bus, devfn, false, name);
2093 }
2094 
2095 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2096 {
2097     int offset = PCI_CONFIG_HEADER_SIZE;
2098     int i;
2099     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2100         if (pdev->used[i])
2101             offset = i + 1;
2102         else if (i - offset + 1 == size)
2103             return offset;
2104     }
2105     return 0;
2106 }
2107 
2108 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2109                                         uint8_t *prev_p)
2110 {
2111     uint8_t next, prev;
2112 
2113     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2114         return 0;
2115 
2116     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2117          prev = next + PCI_CAP_LIST_NEXT)
2118         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2119             break;
2120 
2121     if (prev_p)
2122         *prev_p = prev;
2123     return next;
2124 }
2125 
2126 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2127 {
2128     uint8_t next, prev, found = 0;
2129 
2130     if (!(pdev->used[offset])) {
2131         return 0;
2132     }
2133 
2134     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2135 
2136     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2137          prev = next + PCI_CAP_LIST_NEXT) {
2138         if (next <= offset && next > found) {
2139             found = next;
2140         }
2141     }
2142     return found;
2143 }
2144 
2145 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2146    This is needed for an option rom which is used for more than one device. */
2147 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2148 {
2149     uint16_t vendor_id;
2150     uint16_t device_id;
2151     uint16_t rom_vendor_id;
2152     uint16_t rom_device_id;
2153     uint16_t rom_magic;
2154     uint16_t pcir_offset;
2155     uint8_t checksum;
2156 
2157     /* Words in rom data are little endian (like in PCI configuration),
2158        so they can be read / written with pci_get_word / pci_set_word. */
2159 
2160     /* Only a valid rom will be patched. */
2161     rom_magic = pci_get_word(ptr);
2162     if (rom_magic != 0xaa55) {
2163         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2164         return;
2165     }
2166     pcir_offset = pci_get_word(ptr + 0x18);
2167     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2168         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2169         return;
2170     }
2171 
2172     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2173     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2174     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2175     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2176 
2177     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2178                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2179 
2180     checksum = ptr[6];
2181 
2182     if (vendor_id != rom_vendor_id) {
2183         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2184         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2185         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2186         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2187         ptr[6] = checksum;
2188         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2189     }
2190 
2191     if (device_id != rom_device_id) {
2192         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2193         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2194         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2195         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2196         ptr[6] = checksum;
2197         pci_set_word(ptr + pcir_offset + 6, device_id);
2198     }
2199 }
2200 
2201 /* Add an option rom for the device */
2202 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2203                                Error **errp)
2204 {
2205     int size;
2206     char *path;
2207     void *ptr;
2208     char name[32];
2209     const VMStateDescription *vmsd;
2210 
2211     if (!pdev->romfile)
2212         return;
2213     if (strlen(pdev->romfile) == 0)
2214         return;
2215 
2216     if (!pdev->rom_bar) {
2217         /*
2218          * Load rom via fw_cfg instead of creating a rom bar,
2219          * for 0.11 compatibility.
2220          */
2221         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2222 
2223         /*
2224          * Hot-plugged devices can't use the option ROM
2225          * if the rom bar is disabled.
2226          */
2227         if (DEVICE(pdev)->hotplugged) {
2228             error_setg(errp, "Hot-plugged device without ROM bar"
2229                        " can't have an option ROM");
2230             return;
2231         }
2232 
2233         if (class == 0x0300) {
2234             rom_add_vga(pdev->romfile);
2235         } else {
2236             rom_add_option(pdev->romfile, -1);
2237         }
2238         return;
2239     }
2240 
2241     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2242     if (path == NULL) {
2243         path = g_strdup(pdev->romfile);
2244     }
2245 
2246     size = get_image_size(path);
2247     if (size < 0) {
2248         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2249         g_free(path);
2250         return;
2251     } else if (size == 0) {
2252         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2253         g_free(path);
2254         return;
2255     }
2256     size = pow2ceil(size);
2257 
2258     vmsd = qdev_get_vmsd(DEVICE(pdev));
2259 
2260     if (vmsd) {
2261         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2262     } else {
2263         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2264     }
2265     pdev->has_rom = true;
2266     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2267     ptr = memory_region_get_ram_ptr(&pdev->rom);
2268     if (load_image_size(path, ptr, size) < 0) {
2269         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2270         g_free(path);
2271         return;
2272     }
2273     g_free(path);
2274 
2275     if (is_default_rom) {
2276         /* Only the default rom images will be patched (if needed). */
2277         pci_patch_ids(pdev, ptr, size);
2278     }
2279 
2280     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2281 }
2282 
2283 static void pci_del_option_rom(PCIDevice *pdev)
2284 {
2285     if (!pdev->has_rom)
2286         return;
2287 
2288     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2289     pdev->has_rom = false;
2290 }
2291 
2292 /*
2293  * On success, pci_add_capability() returns a positive value
2294  * that the offset of the pci capability.
2295  * On failure, it sets an error and returns a negative error
2296  * code.
2297  */
2298 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2299                        uint8_t offset, uint8_t size,
2300                        Error **errp)
2301 {
2302     uint8_t *config;
2303     int i, overlapping_cap;
2304 
2305     if (!offset) {
2306         offset = pci_find_space(pdev, size);
2307         /* out of PCI config space is programming error */
2308         assert(offset);
2309     } else {
2310         /* Verify that capabilities don't overlap.  Note: device assignment
2311          * depends on this check to verify that the device is not broken.
2312          * Should never trigger for emulated devices, but it's helpful
2313          * for debugging these. */
2314         for (i = offset; i < offset + size; i++) {
2315             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2316             if (overlapping_cap) {
2317                 error_setg(errp, "%s:%02x:%02x.%x "
2318                            "Attempt to add PCI capability %x at offset "
2319                            "%x overlaps existing capability %x at offset %x",
2320                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2321                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2322                            cap_id, offset, overlapping_cap, i);
2323                 return -EINVAL;
2324             }
2325         }
2326     }
2327 
2328     config = pdev->config + offset;
2329     config[PCI_CAP_LIST_ID] = cap_id;
2330     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2331     pdev->config[PCI_CAPABILITY_LIST] = offset;
2332     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2333     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2334     /* Make capability read-only by default */
2335     memset(pdev->wmask + offset, 0, size);
2336     /* Check capability by default */
2337     memset(pdev->cmask + offset, 0xFF, size);
2338     return offset;
2339 }
2340 
2341 /* Unlink capability from the pci config space. */
2342 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2343 {
2344     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2345     if (!offset)
2346         return;
2347     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2348     /* Make capability writable again */
2349     memset(pdev->wmask + offset, 0xff, size);
2350     memset(pdev->w1cmask + offset, 0, size);
2351     /* Clear cmask as device-specific registers can't be checked */
2352     memset(pdev->cmask + offset, 0, size);
2353     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2354 
2355     if (!pdev->config[PCI_CAPABILITY_LIST])
2356         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2357 }
2358 
2359 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2360 {
2361     return pci_find_capability_list(pdev, cap_id, NULL);
2362 }
2363 
2364 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2365 {
2366     PCIDevice *d = (PCIDevice *)dev;
2367     const pci_class_desc *desc;
2368     char ctxt[64];
2369     PCIIORegion *r;
2370     int i, class;
2371 
2372     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2373     desc = pci_class_descriptions;
2374     while (desc->desc && class != desc->class)
2375         desc++;
2376     if (desc->desc) {
2377         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2378     } else {
2379         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2380     }
2381 
2382     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2383                    "pci id %04x:%04x (sub %04x:%04x)\n",
2384                    indent, "", ctxt, pci_dev_bus_num(d),
2385                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2386                    pci_get_word(d->config + PCI_VENDOR_ID),
2387                    pci_get_word(d->config + PCI_DEVICE_ID),
2388                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2389                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2390     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2391         r = &d->io_regions[i];
2392         if (!r->size)
2393             continue;
2394         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2395                        " [0x%"FMT_PCIBUS"]\n",
2396                        indent, "",
2397                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2398                        r->addr, r->addr + r->size - 1);
2399     }
2400 }
2401 
2402 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2403 {
2404     PCIDevice *d = (PCIDevice *)dev;
2405     const char *name = NULL;
2406     const pci_class_desc *desc =  pci_class_descriptions;
2407     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2408 
2409     while (desc->desc &&
2410           (class & ~desc->fw_ign_bits) !=
2411           (desc->class & ~desc->fw_ign_bits)) {
2412         desc++;
2413     }
2414 
2415     if (desc->desc) {
2416         name = desc->fw_name;
2417     }
2418 
2419     if (name) {
2420         pstrcpy(buf, len, name);
2421     } else {
2422         snprintf(buf, len, "pci%04x,%04x",
2423                  pci_get_word(d->config + PCI_VENDOR_ID),
2424                  pci_get_word(d->config + PCI_DEVICE_ID));
2425     }
2426 
2427     return buf;
2428 }
2429 
2430 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2431 {
2432     PCIDevice *d = (PCIDevice *)dev;
2433     char path[50], name[33];
2434     int off;
2435 
2436     off = snprintf(path, sizeof(path), "%s@%x",
2437                    pci_dev_fw_name(dev, name, sizeof name),
2438                    PCI_SLOT(d->devfn));
2439     if (PCI_FUNC(d->devfn))
2440         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2441     return g_strdup(path);
2442 }
2443 
2444 static char *pcibus_get_dev_path(DeviceState *dev)
2445 {
2446     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2447     PCIDevice *t;
2448     int slot_depth;
2449     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2450      * 00 is added here to make this format compatible with
2451      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2452      * Slot.Function list specifies the slot and function numbers for all
2453      * devices on the path from root to the specific device. */
2454     const char *root_bus_path;
2455     int root_bus_len;
2456     char slot[] = ":SS.F";
2457     int slot_len = sizeof slot - 1 /* For '\0' */;
2458     int path_len;
2459     char *path, *p;
2460     int s;
2461 
2462     root_bus_path = pci_root_bus_path(d);
2463     root_bus_len = strlen(root_bus_path);
2464 
2465     /* Calculate # of slots on path between device and root. */;
2466     slot_depth = 0;
2467     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2468         ++slot_depth;
2469     }
2470 
2471     path_len = root_bus_len + slot_len * slot_depth;
2472 
2473     /* Allocate memory, fill in the terminating null byte. */
2474     path = g_malloc(path_len + 1 /* For '\0' */);
2475     path[path_len] = '\0';
2476 
2477     memcpy(path, root_bus_path, root_bus_len);
2478 
2479     /* Fill in slot numbers. We walk up from device to root, so need to print
2480      * them in the reverse order, last to first. */
2481     p = path + path_len;
2482     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2483         p -= slot_len;
2484         s = snprintf(slot, sizeof slot, ":%02x.%x",
2485                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2486         assert(s == slot_len);
2487         memcpy(p, slot, slot_len);
2488     }
2489 
2490     return path;
2491 }
2492 
2493 static int pci_qdev_find_recursive(PCIBus *bus,
2494                                    const char *id, PCIDevice **pdev)
2495 {
2496     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2497     if (!qdev) {
2498         return -ENODEV;
2499     }
2500 
2501     /* roughly check if given qdev is pci device */
2502     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2503         *pdev = PCI_DEVICE(qdev);
2504         return 0;
2505     }
2506     return -EINVAL;
2507 }
2508 
2509 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2510 {
2511     PCIHostState *host_bridge;
2512     int rc = -ENODEV;
2513 
2514     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2515         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2516         if (!tmp) {
2517             rc = 0;
2518             break;
2519         }
2520         if (tmp != -ENODEV) {
2521             rc = tmp;
2522         }
2523     }
2524 
2525     return rc;
2526 }
2527 
2528 MemoryRegion *pci_address_space(PCIDevice *dev)
2529 {
2530     return pci_get_bus(dev)->address_space_mem;
2531 }
2532 
2533 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2534 {
2535     return pci_get_bus(dev)->address_space_io;
2536 }
2537 
2538 static void pci_device_class_init(ObjectClass *klass, void *data)
2539 {
2540     DeviceClass *k = DEVICE_CLASS(klass);
2541 
2542     k->realize = pci_qdev_realize;
2543     k->unrealize = pci_qdev_unrealize;
2544     k->bus_type = TYPE_PCI_BUS;
2545     k->props = pci_props;
2546 }
2547 
2548 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2549 {
2550     if (!object_class_is_abstract(klass)) {
2551         ObjectClass *conventional =
2552             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2553         ObjectClass *pcie =
2554             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2555         assert(conventional || pcie);
2556     }
2557 }
2558 
2559 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2560 {
2561     PCIBus *bus = pci_get_bus(dev);
2562     PCIBus *iommu_bus = bus;
2563 
2564     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2565         iommu_bus = pci_get_bus(iommu_bus->parent_dev);
2566     }
2567     if (iommu_bus && iommu_bus->iommu_fn) {
2568         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2569     }
2570     return &address_space_memory;
2571 }
2572 
2573 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2574 {
2575     bus->iommu_fn = fn;
2576     bus->iommu_opaque = opaque;
2577 }
2578 
2579 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2580 {
2581     Range *range = opaque;
2582     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2583     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2584     int i;
2585 
2586     if (!(cmd & PCI_COMMAND_MEMORY)) {
2587         return;
2588     }
2589 
2590     if (pc->is_bridge) {
2591         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2592         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2593 
2594         base = MAX(base, 0x1ULL << 32);
2595 
2596         if (limit >= base) {
2597             Range pref_range;
2598             range_set_bounds(&pref_range, base, limit);
2599             range_extend(range, &pref_range);
2600         }
2601     }
2602     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2603         PCIIORegion *r = &dev->io_regions[i];
2604         pcibus_t lob, upb;
2605         Range region_range;
2606 
2607         if (!r->size ||
2608             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2609             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2610             continue;
2611         }
2612 
2613         lob = pci_bar_address(dev, i, r->type, r->size);
2614         upb = lob + r->size - 1;
2615         if (lob == PCI_BAR_UNMAPPED) {
2616             continue;
2617         }
2618 
2619         lob = MAX(lob, 0x1ULL << 32);
2620 
2621         if (upb >= lob) {
2622             range_set_bounds(&region_range, lob, upb);
2623             range_extend(range, &region_range);
2624         }
2625     }
2626 }
2627 
2628 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2629 {
2630     range_make_empty(range);
2631     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2632 }
2633 
2634 static bool pcie_has_upstream_port(PCIDevice *dev)
2635 {
2636     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2637 
2638     /* Device associated with an upstream port.
2639      * As there are several types of these, it's easier to check the
2640      * parent device: upstream ports are always connected to
2641      * root or downstream ports.
2642      */
2643     return parent_dev &&
2644         pci_is_express(parent_dev) &&
2645         parent_dev->exp.exp_cap &&
2646         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2647          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2648 }
2649 
2650 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2651 {
2652     PCIBus *bus = pci_get_bus(pci_dev);
2653 
2654     if(pcie_has_upstream_port(pci_dev)) {
2655         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2656         return bus->devices[0];
2657     } else {
2658         /* Other bus types might support multiple devices at slots 0-31 */
2659         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2660     }
2661 }
2662 
2663 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2664 {
2665     MSIMessage msg;
2666     if (msix_enabled(dev)) {
2667         msg = msix_get_message(dev, vector);
2668     } else if (msi_enabled(dev)) {
2669         msg = msi_get_message(dev, vector);
2670     } else {
2671         /* Should never happen */
2672         error_report("%s: unknown interrupt type", __func__);
2673         abort();
2674     }
2675     return msg;
2676 }
2677 
2678 static const TypeInfo pci_device_type_info = {
2679     .name = TYPE_PCI_DEVICE,
2680     .parent = TYPE_DEVICE,
2681     .instance_size = sizeof(PCIDevice),
2682     .abstract = true,
2683     .class_size = sizeof(PCIDeviceClass),
2684     .class_init = pci_device_class_init,
2685     .class_base_init = pci_device_class_base_init,
2686 };
2687 
2688 static void pci_register_types(void)
2689 {
2690     type_register_static(&pci_bus_info);
2691     type_register_static(&pcie_bus_info);
2692     type_register_static(&conventional_pci_interface_info);
2693     type_register_static(&pcie_interface_info);
2694     type_register_static(&pci_device_type_info);
2695 }
2696 
2697 type_init(pci_register_types)
2698