xref: /openbmc/qemu/hw/pci/pci.c (revision 6f03770d)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu-common.h"
27 #include "qemu/datadir.h"
28 #include "qemu/units.h"
29 #include "hw/irq.h"
30 #include "hw/pci/pci.h"
31 #include "hw/pci/pci_bridge.h"
32 #include "hw/pci/pci_bus.h"
33 #include "hw/pci/pci_host.h"
34 #include "hw/qdev-properties.h"
35 #include "hw/qdev-properties-system.h"
36 #include "migration/qemu-file-types.h"
37 #include "migration/vmstate.h"
38 #include "monitor/monitor.h"
39 #include "net/net.h"
40 #include "sysemu/numa.h"
41 #include "sysemu/sysemu.h"
42 #include "hw/loader.h"
43 #include "qemu/error-report.h"
44 #include "qemu/range.h"
45 #include "trace.h"
46 #include "hw/pci/msi.h"
47 #include "hw/pci/msix.h"
48 #include "exec/address-spaces.h"
49 #include "hw/hotplug.h"
50 #include "hw/boards.h"
51 #include "qapi/error.h"
52 #include "qapi/qapi-commands-pci.h"
53 #include "qemu/cutils.h"
54 
55 //#define DEBUG_PCI
56 #ifdef DEBUG_PCI
57 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
58 #else
59 # define PCI_DPRINTF(format, ...)       do { } while (0)
60 #endif
61 
62 bool pci_available = true;
63 
64 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset(BusState *qbus);
68 
69 static Property pci_props[] = {
70     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
71     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
72     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
73     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
74     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
75                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
76     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
77                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
78     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
79                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
80     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
81                        failover_pair_id),
82     DEFINE_PROP_END_OF_LIST()
83 };
84 
85 static const VMStateDescription vmstate_pcibus = {
86     .name = "PCIBUS",
87     .version_id = 1,
88     .minimum_version_id = 1,
89     .fields = (VMStateField[]) {
90         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
91         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
92                              nirq, 0, vmstate_info_int32,
93                              int32_t),
94         VMSTATE_END_OF_LIST()
95     }
96 };
97 
98 static void pci_init_bus_master(PCIDevice *pci_dev)
99 {
100     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
101 
102     memory_region_init_alias(&pci_dev->bus_master_enable_region,
103                              OBJECT(pci_dev), "bus master",
104                              dma_as->root, 0, memory_region_size(dma_as->root));
105     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
106     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
107                                 &pci_dev->bus_master_enable_region);
108 }
109 
110 static void pcibus_machine_done(Notifier *notifier, void *data)
111 {
112     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
113     int i;
114 
115     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
116         if (bus->devices[i]) {
117             pci_init_bus_master(bus->devices[i]);
118         }
119     }
120 }
121 
122 static void pci_bus_realize(BusState *qbus, Error **errp)
123 {
124     PCIBus *bus = PCI_BUS(qbus);
125 
126     bus->machine_done.notify = pcibus_machine_done;
127     qemu_add_machine_init_done_notifier(&bus->machine_done);
128 
129     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
130 }
131 
132 static void pcie_bus_realize(BusState *qbus, Error **errp)
133 {
134     PCIBus *bus = PCI_BUS(qbus);
135     Error *local_err = NULL;
136 
137     pci_bus_realize(qbus, &local_err);
138     if (local_err) {
139         error_propagate(errp, local_err);
140         return;
141     }
142 
143     /*
144      * A PCI-E bus can support extended config space if it's the root
145      * bus, or if the bus/bridge above it does as well
146      */
147     if (pci_bus_is_root(bus)) {
148         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
149     } else {
150         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
151 
152         if (pci_bus_allows_extended_config_space(parent_bus)) {
153             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
154         }
155     }
156 }
157 
158 static void pci_bus_unrealize(BusState *qbus)
159 {
160     PCIBus *bus = PCI_BUS(qbus);
161 
162     qemu_remove_machine_init_done_notifier(&bus->machine_done);
163 
164     vmstate_unregister(NULL, &vmstate_pcibus, bus);
165 }
166 
167 static int pcibus_num(PCIBus *bus)
168 {
169     if (pci_bus_is_root(bus)) {
170         return 0; /* pci host bridge */
171     }
172     return bus->parent_dev->config[PCI_SECONDARY_BUS];
173 }
174 
175 static uint16_t pcibus_numa_node(PCIBus *bus)
176 {
177     return NUMA_NODE_UNASSIGNED;
178 }
179 
180 static void pci_bus_class_init(ObjectClass *klass, void *data)
181 {
182     BusClass *k = BUS_CLASS(klass);
183     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
184 
185     k->print_dev = pcibus_dev_print;
186     k->get_dev_path = pcibus_get_dev_path;
187     k->get_fw_dev_path = pcibus_get_fw_dev_path;
188     k->realize = pci_bus_realize;
189     k->unrealize = pci_bus_unrealize;
190     k->reset = pcibus_reset;
191 
192     pbc->bus_num = pcibus_num;
193     pbc->numa_node = pcibus_numa_node;
194 }
195 
196 static const TypeInfo pci_bus_info = {
197     .name = TYPE_PCI_BUS,
198     .parent = TYPE_BUS,
199     .instance_size = sizeof(PCIBus),
200     .class_size = sizeof(PCIBusClass),
201     .class_init = pci_bus_class_init,
202 };
203 
204 static const TypeInfo pcie_interface_info = {
205     .name          = INTERFACE_PCIE_DEVICE,
206     .parent        = TYPE_INTERFACE,
207 };
208 
209 static const TypeInfo conventional_pci_interface_info = {
210     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
211     .parent        = TYPE_INTERFACE,
212 };
213 
214 static void pcie_bus_class_init(ObjectClass *klass, void *data)
215 {
216     BusClass *k = BUS_CLASS(klass);
217 
218     k->realize = pcie_bus_realize;
219 }
220 
221 static const TypeInfo pcie_bus_info = {
222     .name = TYPE_PCIE_BUS,
223     .parent = TYPE_PCI_BUS,
224     .class_init = pcie_bus_class_init,
225 };
226 
227 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
228 static void pci_update_mappings(PCIDevice *d);
229 static void pci_irq_handler(void *opaque, int irq_num, int level);
230 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
231 static void pci_del_option_rom(PCIDevice *pdev);
232 
233 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
234 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
235 
236 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
237 
238 int pci_bar(PCIDevice *d, int reg)
239 {
240     uint8_t type;
241 
242     if (reg != PCI_ROM_SLOT)
243         return PCI_BASE_ADDRESS_0 + reg * 4;
244 
245     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
246     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
247 }
248 
249 static inline int pci_irq_state(PCIDevice *d, int irq_num)
250 {
251         return (d->irq_state >> irq_num) & 0x1;
252 }
253 
254 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
255 {
256         d->irq_state &= ~(0x1 << irq_num);
257         d->irq_state |= level << irq_num;
258 }
259 
260 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
261 {
262     assert(irq_num >= 0);
263     assert(irq_num < bus->nirq);
264     bus->irq_count[irq_num] += change;
265     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
266 }
267 
268 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
269 {
270     PCIBus *bus;
271     for (;;) {
272         bus = pci_get_bus(pci_dev);
273         irq_num = bus->map_irq(pci_dev, irq_num);
274         if (bus->set_irq)
275             break;
276         pci_dev = bus->parent_dev;
277     }
278     pci_bus_change_irq_level(bus, irq_num, change);
279 }
280 
281 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
282 {
283     assert(irq_num >= 0);
284     assert(irq_num < bus->nirq);
285     return !!bus->irq_count[irq_num];
286 }
287 
288 /* Update interrupt status bit in config space on interrupt
289  * state change. */
290 static void pci_update_irq_status(PCIDevice *dev)
291 {
292     if (dev->irq_state) {
293         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
294     } else {
295         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
296     }
297 }
298 
299 void pci_device_deassert_intx(PCIDevice *dev)
300 {
301     int i;
302     for (i = 0; i < PCI_NUM_PINS; ++i) {
303         pci_irq_handler(dev, i, 0);
304     }
305 }
306 
307 static void pci_do_device_reset(PCIDevice *dev)
308 {
309     int r;
310 
311     pci_device_deassert_intx(dev);
312     assert(dev->irq_state == 0);
313 
314     /* Clear all writable bits */
315     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
316                                  pci_get_word(dev->wmask + PCI_COMMAND) |
317                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
318     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
319                                  pci_get_word(dev->wmask + PCI_STATUS) |
320                                  pci_get_word(dev->w1cmask + PCI_STATUS));
321     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
322     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
323                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
324                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
325     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
326     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
327         PCIIORegion *region = &dev->io_regions[r];
328         if (!region->size) {
329             continue;
330         }
331 
332         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
333             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
334             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
335         } else {
336             pci_set_long(dev->config + pci_bar(dev, r), region->type);
337         }
338     }
339     pci_update_mappings(dev);
340 
341     msi_reset(dev);
342     msix_reset(dev);
343 }
344 
345 /*
346  * This function is called on #RST and FLR.
347  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
348  */
349 void pci_device_reset(PCIDevice *dev)
350 {
351     qdev_reset_all(&dev->qdev);
352     pci_do_device_reset(dev);
353 }
354 
355 /*
356  * Trigger pci bus reset under a given bus.
357  * Called via qbus_reset_all on RST# assert, after the devices
358  * have been reset qdev_reset_all-ed already.
359  */
360 static void pcibus_reset(BusState *qbus)
361 {
362     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
363     int i;
364 
365     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
366         if (bus->devices[i]) {
367             pci_do_device_reset(bus->devices[i]);
368         }
369     }
370 
371     for (i = 0; i < bus->nirq; i++) {
372         assert(bus->irq_count[i] == 0);
373     }
374 }
375 
376 static void pci_host_bus_register(DeviceState *host)
377 {
378     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
379 
380     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
381 }
382 
383 static void pci_host_bus_unregister(DeviceState *host)
384 {
385     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
386 
387     QLIST_REMOVE(host_bridge, next);
388 }
389 
390 PCIBus *pci_device_root_bus(const PCIDevice *d)
391 {
392     PCIBus *bus = pci_get_bus(d);
393 
394     while (!pci_bus_is_root(bus)) {
395         d = bus->parent_dev;
396         assert(d != NULL);
397 
398         bus = pci_get_bus(d);
399     }
400 
401     return bus;
402 }
403 
404 const char *pci_root_bus_path(PCIDevice *dev)
405 {
406     PCIBus *rootbus = pci_device_root_bus(dev);
407     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
408     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
409 
410     assert(host_bridge->bus == rootbus);
411 
412     if (hc->root_bus_path) {
413         return (*hc->root_bus_path)(host_bridge, rootbus);
414     }
415 
416     return rootbus->qbus.name;
417 }
418 
419 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
420                               MemoryRegion *address_space_mem,
421                               MemoryRegion *address_space_io,
422                               uint8_t devfn_min)
423 {
424     assert(PCI_FUNC(devfn_min) == 0);
425     bus->devfn_min = devfn_min;
426     bus->slot_reserved_mask = 0x0;
427     bus->address_space_mem = address_space_mem;
428     bus->address_space_io = address_space_io;
429     bus->flags |= PCI_BUS_IS_ROOT;
430 
431     /* host bridge */
432     QLIST_INIT(&bus->child);
433 
434     pci_host_bus_register(parent);
435 }
436 
437 static void pci_bus_uninit(PCIBus *bus)
438 {
439     pci_host_bus_unregister(BUS(bus)->parent);
440 }
441 
442 bool pci_bus_is_express(PCIBus *bus)
443 {
444     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
445 }
446 
447 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
448                               const char *name,
449                               MemoryRegion *address_space_mem,
450                               MemoryRegion *address_space_io,
451                               uint8_t devfn_min, const char *typename)
452 {
453     qbus_create_inplace(bus, bus_size, typename, parent, name);
454     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
455                       devfn_min);
456 }
457 
458 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
459                          MemoryRegion *address_space_mem,
460                          MemoryRegion *address_space_io,
461                          uint8_t devfn_min, const char *typename)
462 {
463     PCIBus *bus;
464 
465     bus = PCI_BUS(qbus_create(typename, parent, name));
466     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
467                       devfn_min);
468     return bus;
469 }
470 
471 void pci_root_bus_cleanup(PCIBus *bus)
472 {
473     pci_bus_uninit(bus);
474     /* the caller of the unplug hotplug handler will delete this device */
475     qbus_unrealize(BUS(bus));
476 }
477 
478 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
479                   void *irq_opaque, int nirq)
480 {
481     bus->set_irq = set_irq;
482     bus->map_irq = map_irq;
483     bus->irq_opaque = irq_opaque;
484     bus->nirq = nirq;
485     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
486 }
487 
488 void pci_bus_irqs_cleanup(PCIBus *bus)
489 {
490     bus->set_irq = NULL;
491     bus->map_irq = NULL;
492     bus->irq_opaque = NULL;
493     bus->nirq = 0;
494     g_free(bus->irq_count);
495 }
496 
497 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
498                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
499                               void *irq_opaque,
500                               MemoryRegion *address_space_mem,
501                               MemoryRegion *address_space_io,
502                               uint8_t devfn_min, int nirq,
503                               const char *typename)
504 {
505     PCIBus *bus;
506 
507     bus = pci_root_bus_new(parent, name, address_space_mem,
508                            address_space_io, devfn_min, typename);
509     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
510     return bus;
511 }
512 
513 void pci_unregister_root_bus(PCIBus *bus)
514 {
515     pci_bus_irqs_cleanup(bus);
516     pci_root_bus_cleanup(bus);
517 }
518 
519 int pci_bus_num(PCIBus *s)
520 {
521     return PCI_BUS_GET_CLASS(s)->bus_num(s);
522 }
523 
524 int pci_bus_numa_node(PCIBus *bus)
525 {
526     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
527 }
528 
529 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
530                                  const VMStateField *field)
531 {
532     PCIDevice *s = container_of(pv, PCIDevice, config);
533     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
534     uint8_t *config;
535     int i;
536 
537     assert(size == pci_config_size(s));
538     config = g_malloc(size);
539 
540     qemu_get_buffer(f, config, size);
541     for (i = 0; i < size; ++i) {
542         if ((config[i] ^ s->config[i]) &
543             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
544             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
545                          "cmask: %x wmask: %x w1cmask:%x", __func__,
546                          i, config[i], s->config[i],
547                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
548             g_free(config);
549             return -EINVAL;
550         }
551     }
552     memcpy(s->config, config, size);
553 
554     pci_update_mappings(s);
555     if (pc->is_bridge) {
556         PCIBridge *b = PCI_BRIDGE(s);
557         pci_bridge_update_mappings(b);
558     }
559 
560     memory_region_set_enabled(&s->bus_master_enable_region,
561                               pci_get_word(s->config + PCI_COMMAND)
562                               & PCI_COMMAND_MASTER);
563 
564     g_free(config);
565     return 0;
566 }
567 
568 /* just put buffer */
569 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
570                                  const VMStateField *field, JSONWriter *vmdesc)
571 {
572     const uint8_t **v = pv;
573     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
574     qemu_put_buffer(f, *v, size);
575 
576     return 0;
577 }
578 
579 static VMStateInfo vmstate_info_pci_config = {
580     .name = "pci config",
581     .get  = get_pci_config_device,
582     .put  = put_pci_config_device,
583 };
584 
585 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
586                              const VMStateField *field)
587 {
588     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
589     uint32_t irq_state[PCI_NUM_PINS];
590     int i;
591     for (i = 0; i < PCI_NUM_PINS; ++i) {
592         irq_state[i] = qemu_get_be32(f);
593         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
594             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
595                     irq_state[i]);
596             return -EINVAL;
597         }
598     }
599 
600     for (i = 0; i < PCI_NUM_PINS; ++i) {
601         pci_set_irq_state(s, i, irq_state[i]);
602     }
603 
604     return 0;
605 }
606 
607 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
608                              const VMStateField *field, JSONWriter *vmdesc)
609 {
610     int i;
611     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
612 
613     for (i = 0; i < PCI_NUM_PINS; ++i) {
614         qemu_put_be32(f, pci_irq_state(s, i));
615     }
616 
617     return 0;
618 }
619 
620 static VMStateInfo vmstate_info_pci_irq_state = {
621     .name = "pci irq state",
622     .get  = get_pci_irq_state,
623     .put  = put_pci_irq_state,
624 };
625 
626 static bool migrate_is_pcie(void *opaque, int version_id)
627 {
628     return pci_is_express((PCIDevice *)opaque);
629 }
630 
631 static bool migrate_is_not_pcie(void *opaque, int version_id)
632 {
633     return !pci_is_express((PCIDevice *)opaque);
634 }
635 
636 const VMStateDescription vmstate_pci_device = {
637     .name = "PCIDevice",
638     .version_id = 2,
639     .minimum_version_id = 1,
640     .fields = (VMStateField[]) {
641         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
642         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
643                                    migrate_is_not_pcie,
644                                    0, vmstate_info_pci_config,
645                                    PCI_CONFIG_SPACE_SIZE),
646         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
647                                    migrate_is_pcie,
648                                    0, vmstate_info_pci_config,
649                                    PCIE_CONFIG_SPACE_SIZE),
650         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
651                                    vmstate_info_pci_irq_state,
652                                    PCI_NUM_PINS * sizeof(int32_t)),
653         VMSTATE_END_OF_LIST()
654     }
655 };
656 
657 
658 void pci_device_save(PCIDevice *s, QEMUFile *f)
659 {
660     /* Clear interrupt status bit: it is implicit
661      * in irq_state which we are saving.
662      * This makes us compatible with old devices
663      * which never set or clear this bit. */
664     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
665     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
666     /* Restore the interrupt status bit. */
667     pci_update_irq_status(s);
668 }
669 
670 int pci_device_load(PCIDevice *s, QEMUFile *f)
671 {
672     int ret;
673     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
674     /* Restore the interrupt status bit. */
675     pci_update_irq_status(s);
676     return ret;
677 }
678 
679 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
680 {
681     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
682                  pci_default_sub_vendor_id);
683     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
684                  pci_default_sub_device_id);
685 }
686 
687 /*
688  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
689  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
690  */
691 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
692                              unsigned int *slotp, unsigned int *funcp)
693 {
694     const char *p;
695     char *e;
696     unsigned long val;
697     unsigned long dom = 0, bus = 0;
698     unsigned int slot = 0;
699     unsigned int func = 0;
700 
701     p = addr;
702     val = strtoul(p, &e, 16);
703     if (e == p)
704         return -1;
705     if (*e == ':') {
706         bus = val;
707         p = e + 1;
708         val = strtoul(p, &e, 16);
709         if (e == p)
710             return -1;
711         if (*e == ':') {
712             dom = bus;
713             bus = val;
714             p = e + 1;
715             val = strtoul(p, &e, 16);
716             if (e == p)
717                 return -1;
718         }
719     }
720 
721     slot = val;
722 
723     if (funcp != NULL) {
724         if (*e != '.')
725             return -1;
726 
727         p = e + 1;
728         val = strtoul(p, &e, 16);
729         if (e == p)
730             return -1;
731 
732         func = val;
733     }
734 
735     /* if funcp == NULL func is 0 */
736     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
737         return -1;
738 
739     if (*e)
740         return -1;
741 
742     *domp = dom;
743     *busp = bus;
744     *slotp = slot;
745     if (funcp != NULL)
746         *funcp = func;
747     return 0;
748 }
749 
750 static void pci_init_cmask(PCIDevice *dev)
751 {
752     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
753     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
754     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
755     dev->cmask[PCI_REVISION_ID] = 0xff;
756     dev->cmask[PCI_CLASS_PROG] = 0xff;
757     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
758     dev->cmask[PCI_HEADER_TYPE] = 0xff;
759     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
760 }
761 
762 static void pci_init_wmask(PCIDevice *dev)
763 {
764     int config_size = pci_config_size(dev);
765 
766     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
767     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
768     pci_set_word(dev->wmask + PCI_COMMAND,
769                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
770                  PCI_COMMAND_INTX_DISABLE);
771     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
772 
773     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
774            config_size - PCI_CONFIG_HEADER_SIZE);
775 }
776 
777 static void pci_init_w1cmask(PCIDevice *dev)
778 {
779     /*
780      * Note: It's okay to set w1cmask even for readonly bits as
781      * long as their value is hardwired to 0.
782      */
783     pci_set_word(dev->w1cmask + PCI_STATUS,
784                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
785                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
786                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
787 }
788 
789 static void pci_init_mask_bridge(PCIDevice *d)
790 {
791     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
792        PCI_SEC_LETENCY_TIMER */
793     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
794 
795     /* base and limit */
796     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
797     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
798     pci_set_word(d->wmask + PCI_MEMORY_BASE,
799                  PCI_MEMORY_RANGE_MASK & 0xffff);
800     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
801                  PCI_MEMORY_RANGE_MASK & 0xffff);
802     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
803                  PCI_PREF_RANGE_MASK & 0xffff);
804     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
805                  PCI_PREF_RANGE_MASK & 0xffff);
806 
807     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
808     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
809 
810     /* Supported memory and i/o types */
811     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
812     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
813     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
814                                PCI_PREF_RANGE_TYPE_64);
815     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
816                                PCI_PREF_RANGE_TYPE_64);
817 
818     /*
819      * TODO: Bridges default to 10-bit VGA decoding but we currently only
820      * implement 16-bit decoding (no alias support).
821      */
822     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
823                  PCI_BRIDGE_CTL_PARITY |
824                  PCI_BRIDGE_CTL_SERR |
825                  PCI_BRIDGE_CTL_ISA |
826                  PCI_BRIDGE_CTL_VGA |
827                  PCI_BRIDGE_CTL_VGA_16BIT |
828                  PCI_BRIDGE_CTL_MASTER_ABORT |
829                  PCI_BRIDGE_CTL_BUS_RESET |
830                  PCI_BRIDGE_CTL_FAST_BACK |
831                  PCI_BRIDGE_CTL_DISCARD |
832                  PCI_BRIDGE_CTL_SEC_DISCARD |
833                  PCI_BRIDGE_CTL_DISCARD_SERR);
834     /* Below does not do anything as we never set this bit, put here for
835      * completeness. */
836     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
837                  PCI_BRIDGE_CTL_DISCARD_STATUS);
838     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
839     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
840     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
841                                PCI_PREF_RANGE_TYPE_MASK);
842     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
843                                PCI_PREF_RANGE_TYPE_MASK);
844 }
845 
846 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
847 {
848     uint8_t slot = PCI_SLOT(dev->devfn);
849     uint8_t func;
850 
851     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
852         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
853     }
854 
855     /*
856      * multifunction bit is interpreted in two ways as follows.
857      *   - all functions must set the bit to 1.
858      *     Example: Intel X53
859      *   - function 0 must set the bit, but the rest function (> 0)
860      *     is allowed to leave the bit to 0.
861      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
862      *
863      * So OS (at least Linux) checks the bit of only function 0,
864      * and doesn't see the bit of function > 0.
865      *
866      * The below check allows both interpretation.
867      */
868     if (PCI_FUNC(dev->devfn)) {
869         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
870         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
871             /* function 0 should set multifunction bit */
872             error_setg(errp, "PCI: single function device can't be populated "
873                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
874             return;
875         }
876         return;
877     }
878 
879     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
880         return;
881     }
882     /* function 0 indicates single function, so function > 0 must be NULL */
883     for (func = 1; func < PCI_FUNC_MAX; ++func) {
884         if (bus->devices[PCI_DEVFN(slot, func)]) {
885             error_setg(errp, "PCI: %x.0 indicates single function, "
886                        "but %x.%x is already populated.",
887                        slot, slot, func);
888             return;
889         }
890     }
891 }
892 
893 static void pci_config_alloc(PCIDevice *pci_dev)
894 {
895     int config_size = pci_config_size(pci_dev);
896 
897     pci_dev->config = g_malloc0(config_size);
898     pci_dev->cmask = g_malloc0(config_size);
899     pci_dev->wmask = g_malloc0(config_size);
900     pci_dev->w1cmask = g_malloc0(config_size);
901     pci_dev->used = g_malloc0(config_size);
902 }
903 
904 static void pci_config_free(PCIDevice *pci_dev)
905 {
906     g_free(pci_dev->config);
907     g_free(pci_dev->cmask);
908     g_free(pci_dev->wmask);
909     g_free(pci_dev->w1cmask);
910     g_free(pci_dev->used);
911 }
912 
913 static void do_pci_unregister_device(PCIDevice *pci_dev)
914 {
915     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
916     pci_config_free(pci_dev);
917 
918     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
919         memory_region_del_subregion(&pci_dev->bus_master_container_region,
920                                     &pci_dev->bus_master_enable_region);
921     }
922     address_space_destroy(&pci_dev->bus_master_as);
923 }
924 
925 /* Extract PCIReqIDCache into BDF format */
926 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
927 {
928     uint8_t bus_n;
929     uint16_t result;
930 
931     switch (cache->type) {
932     case PCI_REQ_ID_BDF:
933         result = pci_get_bdf(cache->dev);
934         break;
935     case PCI_REQ_ID_SECONDARY_BUS:
936         bus_n = pci_dev_bus_num(cache->dev);
937         result = PCI_BUILD_BDF(bus_n, 0);
938         break;
939     default:
940         error_report("Invalid PCI requester ID cache type: %d",
941                      cache->type);
942         exit(1);
943         break;
944     }
945 
946     return result;
947 }
948 
949 /* Parse bridges up to the root complex and return requester ID
950  * cache for specific device.  For full PCIe topology, the cache
951  * result would be exactly the same as getting BDF of the device.
952  * However, several tricks are required when system mixed up with
953  * legacy PCI devices and PCIe-to-PCI bridges.
954  *
955  * Here we cache the proxy device (and type) not requester ID since
956  * bus number might change from time to time.
957  */
958 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
959 {
960     PCIDevice *parent;
961     PCIReqIDCache cache = {
962         .dev = dev,
963         .type = PCI_REQ_ID_BDF,
964     };
965 
966     while (!pci_bus_is_root(pci_get_bus(dev))) {
967         /* We are under PCI/PCIe bridges */
968         parent = pci_get_bus(dev)->parent_dev;
969         if (pci_is_express(parent)) {
970             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
971                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
972                  * override the requester ID using secondary bus
973                  * number of parent bridge with zeroed devfn
974                  * (pcie-to-pci bridge spec chap 2.3). */
975                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
976                 cache.dev = dev;
977             }
978         } else {
979             /* Legacy PCI, override requester ID with the bridge's
980              * BDF upstream.  When the root complex connects to
981              * legacy PCI devices (including buses), it can only
982              * obtain requester ID info from directly attached
983              * devices.  If devices are attached under bridges, only
984              * the requester ID of the bridge that is directly
985              * attached to the root complex can be recognized. */
986             cache.type = PCI_REQ_ID_BDF;
987             cache.dev = parent;
988         }
989         dev = parent;
990     }
991 
992     return cache;
993 }
994 
995 uint16_t pci_requester_id(PCIDevice *dev)
996 {
997     return pci_req_id_cache_extract(&dev->requester_id_cache);
998 }
999 
1000 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1001 {
1002     return !(bus->devices[devfn]);
1003 }
1004 
1005 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1006 {
1007     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1008 }
1009 
1010 /* -1 for devfn means auto assign */
1011 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1012                                          const char *name, int devfn,
1013                                          Error **errp)
1014 {
1015     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1016     PCIConfigReadFunc *config_read = pc->config_read;
1017     PCIConfigWriteFunc *config_write = pc->config_write;
1018     Error *local_err = NULL;
1019     DeviceState *dev = DEVICE(pci_dev);
1020     PCIBus *bus = pci_get_bus(pci_dev);
1021 
1022     /* Only pci bridges can be attached to extra PCI root buses */
1023     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1024         error_setg(errp,
1025                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1026                     bus->parent_dev->name);
1027         return NULL;
1028     }
1029 
1030     if (devfn < 0) {
1031         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1032             devfn += PCI_FUNC_MAX) {
1033             if (pci_bus_devfn_available(bus, devfn) &&
1034                    !pci_bus_devfn_reserved(bus, devfn)) {
1035                 goto found;
1036             }
1037         }
1038         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1039                    "or reserved", name);
1040         return NULL;
1041     found: ;
1042     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1043         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1044                    " reserved",
1045                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1046         return NULL;
1047     } else if (!pci_bus_devfn_available(bus, devfn)) {
1048         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1049                    " in use by %s",
1050                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1051                    bus->devices[devfn]->name);
1052         return NULL;
1053     } else if (dev->hotplugged &&
1054                pci_get_function_0(pci_dev)) {
1055         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1056                    " new func %s cannot be exposed to guest.",
1057                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1058                    pci_get_function_0(pci_dev)->name,
1059                    name);
1060 
1061        return NULL;
1062     }
1063 
1064     pci_dev->devfn = devfn;
1065     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1066     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1067 
1068     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1069                        "bus master container", UINT64_MAX);
1070     address_space_init(&pci_dev->bus_master_as,
1071                        &pci_dev->bus_master_container_region, pci_dev->name);
1072 
1073     if (phase_check(PHASE_MACHINE_READY)) {
1074         pci_init_bus_master(pci_dev);
1075     }
1076     pci_dev->irq_state = 0;
1077     pci_config_alloc(pci_dev);
1078 
1079     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1080     pci_config_set_device_id(pci_dev->config, pc->device_id);
1081     pci_config_set_revision(pci_dev->config, pc->revision);
1082     pci_config_set_class(pci_dev->config, pc->class_id);
1083 
1084     if (!pc->is_bridge) {
1085         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1086             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1087                          pc->subsystem_vendor_id);
1088             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1089                          pc->subsystem_id);
1090         } else {
1091             pci_set_default_subsystem_id(pci_dev);
1092         }
1093     } else {
1094         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1095         assert(!pc->subsystem_vendor_id);
1096         assert(!pc->subsystem_id);
1097     }
1098     pci_init_cmask(pci_dev);
1099     pci_init_wmask(pci_dev);
1100     pci_init_w1cmask(pci_dev);
1101     if (pc->is_bridge) {
1102         pci_init_mask_bridge(pci_dev);
1103     }
1104     pci_init_multifunction(bus, pci_dev, &local_err);
1105     if (local_err) {
1106         error_propagate(errp, local_err);
1107         do_pci_unregister_device(pci_dev);
1108         return NULL;
1109     }
1110 
1111     if (!config_read)
1112         config_read = pci_default_read_config;
1113     if (!config_write)
1114         config_write = pci_default_write_config;
1115     pci_dev->config_read = config_read;
1116     pci_dev->config_write = config_write;
1117     bus->devices[devfn] = pci_dev;
1118     pci_dev->version_id = 2; /* Current pci device vmstate version */
1119     return pci_dev;
1120 }
1121 
1122 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1123 {
1124     PCIIORegion *r;
1125     int i;
1126 
1127     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1128         r = &pci_dev->io_regions[i];
1129         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1130             continue;
1131         memory_region_del_subregion(r->address_space, r->memory);
1132     }
1133 
1134     pci_unregister_vga(pci_dev);
1135 }
1136 
1137 static void pci_qdev_unrealize(DeviceState *dev)
1138 {
1139     PCIDevice *pci_dev = PCI_DEVICE(dev);
1140     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1141 
1142     pci_unregister_io_regions(pci_dev);
1143     pci_del_option_rom(pci_dev);
1144 
1145     if (pc->exit) {
1146         pc->exit(pci_dev);
1147     }
1148 
1149     pci_device_deassert_intx(pci_dev);
1150     do_pci_unregister_device(pci_dev);
1151 }
1152 
1153 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1154                       uint8_t type, MemoryRegion *memory)
1155 {
1156     PCIIORegion *r;
1157     uint32_t addr; /* offset in pci config space */
1158     uint64_t wmask;
1159     pcibus_t size = memory_region_size(memory);
1160     uint8_t hdr_type;
1161 
1162     assert(region_num >= 0);
1163     assert(region_num < PCI_NUM_REGIONS);
1164     assert(is_power_of_2(size));
1165 
1166     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1167     hdr_type =
1168         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1169     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1170 
1171     r = &pci_dev->io_regions[region_num];
1172     r->addr = PCI_BAR_UNMAPPED;
1173     r->size = size;
1174     r->type = type;
1175     r->memory = memory;
1176     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1177                         ? pci_get_bus(pci_dev)->address_space_io
1178                         : pci_get_bus(pci_dev)->address_space_mem;
1179 
1180     wmask = ~(size - 1);
1181     if (region_num == PCI_ROM_SLOT) {
1182         /* ROM enable bit is writable */
1183         wmask |= PCI_ROM_ADDRESS_ENABLE;
1184     }
1185 
1186     addr = pci_bar(pci_dev, region_num);
1187     pci_set_long(pci_dev->config + addr, type);
1188 
1189     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1190         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1191         pci_set_quad(pci_dev->wmask + addr, wmask);
1192         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1193     } else {
1194         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1195         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1196     }
1197 }
1198 
1199 static void pci_update_vga(PCIDevice *pci_dev)
1200 {
1201     uint16_t cmd;
1202 
1203     if (!pci_dev->has_vga) {
1204         return;
1205     }
1206 
1207     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1208 
1209     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1210                               cmd & PCI_COMMAND_MEMORY);
1211     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1212                               cmd & PCI_COMMAND_IO);
1213     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1214                               cmd & PCI_COMMAND_IO);
1215 }
1216 
1217 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1218                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1219 {
1220     PCIBus *bus = pci_get_bus(pci_dev);
1221 
1222     assert(!pci_dev->has_vga);
1223 
1224     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1225     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1226     memory_region_add_subregion_overlap(bus->address_space_mem,
1227                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1228 
1229     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1230     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1231     memory_region_add_subregion_overlap(bus->address_space_io,
1232                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1233 
1234     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1235     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1236     memory_region_add_subregion_overlap(bus->address_space_io,
1237                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1238     pci_dev->has_vga = true;
1239 
1240     pci_update_vga(pci_dev);
1241 }
1242 
1243 void pci_unregister_vga(PCIDevice *pci_dev)
1244 {
1245     PCIBus *bus = pci_get_bus(pci_dev);
1246 
1247     if (!pci_dev->has_vga) {
1248         return;
1249     }
1250 
1251     memory_region_del_subregion(bus->address_space_mem,
1252                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1253     memory_region_del_subregion(bus->address_space_io,
1254                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1255     memory_region_del_subregion(bus->address_space_io,
1256                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1257     pci_dev->has_vga = false;
1258 }
1259 
1260 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1261 {
1262     return pci_dev->io_regions[region_num].addr;
1263 }
1264 
1265 static pcibus_t pci_bar_address(PCIDevice *d,
1266                                 int reg, uint8_t type, pcibus_t size)
1267 {
1268     pcibus_t new_addr, last_addr;
1269     int bar = pci_bar(d, reg);
1270     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1271     Object *machine = qdev_get_machine();
1272     ObjectClass *oc = object_get_class(machine);
1273     MachineClass *mc = MACHINE_CLASS(oc);
1274     bool allow_0_address = mc->pci_allow_0_address;
1275 
1276     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1277         if (!(cmd & PCI_COMMAND_IO)) {
1278             return PCI_BAR_UNMAPPED;
1279         }
1280         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1281         last_addr = new_addr + size - 1;
1282         /* Check if 32 bit BAR wraps around explicitly.
1283          * TODO: make priorities correct and remove this work around.
1284          */
1285         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1286             (!allow_0_address && new_addr == 0)) {
1287             return PCI_BAR_UNMAPPED;
1288         }
1289         return new_addr;
1290     }
1291 
1292     if (!(cmd & PCI_COMMAND_MEMORY)) {
1293         return PCI_BAR_UNMAPPED;
1294     }
1295     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1296         new_addr = pci_get_quad(d->config + bar);
1297     } else {
1298         new_addr = pci_get_long(d->config + bar);
1299     }
1300     /* the ROM slot has a specific enable bit */
1301     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1302         return PCI_BAR_UNMAPPED;
1303     }
1304     new_addr &= ~(size - 1);
1305     last_addr = new_addr + size - 1;
1306     /* NOTE: we do not support wrapping */
1307     /* XXX: as we cannot support really dynamic
1308        mappings, we handle specific values as invalid
1309        mappings. */
1310     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1311         (!allow_0_address && new_addr == 0)) {
1312         return PCI_BAR_UNMAPPED;
1313     }
1314 
1315     /* Now pcibus_t is 64bit.
1316      * Check if 32 bit BAR wraps around explicitly.
1317      * Without this, PC ide doesn't work well.
1318      * TODO: remove this work around.
1319      */
1320     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1321         return PCI_BAR_UNMAPPED;
1322     }
1323 
1324     /*
1325      * OS is allowed to set BAR beyond its addressable
1326      * bits. For example, 32 bit OS can set 64bit bar
1327      * to >4G. Check it. TODO: we might need to support
1328      * it in the future for e.g. PAE.
1329      */
1330     if (last_addr >= HWADDR_MAX) {
1331         return PCI_BAR_UNMAPPED;
1332     }
1333 
1334     return new_addr;
1335 }
1336 
1337 static void pci_update_mappings(PCIDevice *d)
1338 {
1339     PCIIORegion *r;
1340     int i;
1341     pcibus_t new_addr;
1342 
1343     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1344         r = &d->io_regions[i];
1345 
1346         /* this region isn't registered */
1347         if (!r->size)
1348             continue;
1349 
1350         new_addr = pci_bar_address(d, i, r->type, r->size);
1351 
1352         /* This bar isn't changed */
1353         if (new_addr == r->addr)
1354             continue;
1355 
1356         /* now do the real mapping */
1357         if (r->addr != PCI_BAR_UNMAPPED) {
1358             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1359                                           PCI_SLOT(d->devfn),
1360                                           PCI_FUNC(d->devfn),
1361                                           i, r->addr, r->size);
1362             memory_region_del_subregion(r->address_space, r->memory);
1363         }
1364         r->addr = new_addr;
1365         if (r->addr != PCI_BAR_UNMAPPED) {
1366             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1367                                           PCI_SLOT(d->devfn),
1368                                           PCI_FUNC(d->devfn),
1369                                           i, r->addr, r->size);
1370             memory_region_add_subregion_overlap(r->address_space,
1371                                                 r->addr, r->memory, 1);
1372         }
1373     }
1374 
1375     pci_update_vga(d);
1376 }
1377 
1378 static inline int pci_irq_disabled(PCIDevice *d)
1379 {
1380     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1381 }
1382 
1383 /* Called after interrupt disabled field update in config space,
1384  * assert/deassert interrupts if necessary.
1385  * Gets original interrupt disable bit value (before update). */
1386 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1387 {
1388     int i, disabled = pci_irq_disabled(d);
1389     if (disabled == was_irq_disabled)
1390         return;
1391     for (i = 0; i < PCI_NUM_PINS; ++i) {
1392         int state = pci_irq_state(d, i);
1393         pci_change_irq_level(d, i, disabled ? -state : state);
1394     }
1395 }
1396 
1397 uint32_t pci_default_read_config(PCIDevice *d,
1398                                  uint32_t address, int len)
1399 {
1400     uint32_t val = 0;
1401 
1402     assert(address + len <= pci_config_size(d));
1403 
1404     if (pci_is_express_downstream_port(d) &&
1405         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1406         pcie_sync_bridge_lnk(d);
1407     }
1408     memcpy(&val, d->config + address, len);
1409     return le32_to_cpu(val);
1410 }
1411 
1412 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1413 {
1414     int i, was_irq_disabled = pci_irq_disabled(d);
1415     uint32_t val = val_in;
1416 
1417     assert(addr + l <= pci_config_size(d));
1418 
1419     for (i = 0; i < l; val >>= 8, ++i) {
1420         uint8_t wmask = d->wmask[addr + i];
1421         uint8_t w1cmask = d->w1cmask[addr + i];
1422         assert(!(wmask & w1cmask));
1423         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1424         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1425     }
1426     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1427         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1428         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1429         range_covers_byte(addr, l, PCI_COMMAND))
1430         pci_update_mappings(d);
1431 
1432     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1433         pci_update_irq_disabled(d, was_irq_disabled);
1434         memory_region_set_enabled(&d->bus_master_enable_region,
1435                                   pci_get_word(d->config + PCI_COMMAND)
1436                                     & PCI_COMMAND_MASTER);
1437     }
1438 
1439     msi_write_config(d, addr, val_in, l);
1440     msix_write_config(d, addr, val_in, l);
1441 }
1442 
1443 /***********************************************************/
1444 /* generic PCI irq support */
1445 
1446 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1447 static void pci_irq_handler(void *opaque, int irq_num, int level)
1448 {
1449     PCIDevice *pci_dev = opaque;
1450     int change;
1451 
1452     change = level - pci_irq_state(pci_dev, irq_num);
1453     if (!change)
1454         return;
1455 
1456     pci_set_irq_state(pci_dev, irq_num, level);
1457     pci_update_irq_status(pci_dev);
1458     if (pci_irq_disabled(pci_dev))
1459         return;
1460     pci_change_irq_level(pci_dev, irq_num, change);
1461 }
1462 
1463 static inline int pci_intx(PCIDevice *pci_dev)
1464 {
1465     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1466 }
1467 
1468 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1469 {
1470     int intx = pci_intx(pci_dev);
1471 
1472     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1473 }
1474 
1475 void pci_set_irq(PCIDevice *pci_dev, int level)
1476 {
1477     int intx = pci_intx(pci_dev);
1478     pci_irq_handler(pci_dev, intx, level);
1479 }
1480 
1481 /* Special hooks used by device assignment */
1482 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1483 {
1484     assert(pci_bus_is_root(bus));
1485     bus->route_intx_to_irq = route_intx_to_irq;
1486 }
1487 
1488 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1489 {
1490     PCIBus *bus;
1491 
1492     do {
1493         bus = pci_get_bus(dev);
1494         pin = bus->map_irq(dev, pin);
1495         dev = bus->parent_dev;
1496     } while (dev);
1497 
1498     if (!bus->route_intx_to_irq) {
1499         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1500                      object_get_typename(OBJECT(bus->qbus.parent)));
1501         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1502     }
1503 
1504     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1505 }
1506 
1507 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1508 {
1509     return old->mode != new->mode || old->irq != new->irq;
1510 }
1511 
1512 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1513 {
1514     PCIDevice *dev;
1515     PCIBus *sec;
1516     int i;
1517 
1518     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1519         dev = bus->devices[i];
1520         if (dev && dev->intx_routing_notifier) {
1521             dev->intx_routing_notifier(dev);
1522         }
1523     }
1524 
1525     QLIST_FOREACH(sec, &bus->child, sibling) {
1526         pci_bus_fire_intx_routing_notifier(sec);
1527     }
1528 }
1529 
1530 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1531                                           PCIINTxRoutingNotifier notifier)
1532 {
1533     dev->intx_routing_notifier = notifier;
1534 }
1535 
1536 /*
1537  * PCI-to-PCI bridge specification
1538  * 9.1: Interrupt routing. Table 9-1
1539  *
1540  * the PCI Express Base Specification, Revision 2.1
1541  * 2.2.8.1: INTx interrutp signaling - Rules
1542  *          the Implementation Note
1543  *          Table 2-20
1544  */
1545 /*
1546  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1547  * 0-origin unlike PCI interrupt pin register.
1548  */
1549 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1550 {
1551     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1552 }
1553 
1554 /***********************************************************/
1555 /* monitor info on PCI */
1556 
1557 typedef struct {
1558     uint16_t class;
1559     const char *desc;
1560     const char *fw_name;
1561     uint16_t fw_ign_bits;
1562 } pci_class_desc;
1563 
1564 static const pci_class_desc pci_class_descriptions[] =
1565 {
1566     { 0x0001, "VGA controller", "display"},
1567     { 0x0100, "SCSI controller", "scsi"},
1568     { 0x0101, "IDE controller", "ide"},
1569     { 0x0102, "Floppy controller", "fdc"},
1570     { 0x0103, "IPI controller", "ipi"},
1571     { 0x0104, "RAID controller", "raid"},
1572     { 0x0106, "SATA controller"},
1573     { 0x0107, "SAS controller"},
1574     { 0x0180, "Storage controller"},
1575     { 0x0200, "Ethernet controller", "ethernet"},
1576     { 0x0201, "Token Ring controller", "token-ring"},
1577     { 0x0202, "FDDI controller", "fddi"},
1578     { 0x0203, "ATM controller", "atm"},
1579     { 0x0280, "Network controller"},
1580     { 0x0300, "VGA controller", "display", 0x00ff},
1581     { 0x0301, "XGA controller"},
1582     { 0x0302, "3D controller"},
1583     { 0x0380, "Display controller"},
1584     { 0x0400, "Video controller", "video"},
1585     { 0x0401, "Audio controller", "sound"},
1586     { 0x0402, "Phone"},
1587     { 0x0403, "Audio controller", "sound"},
1588     { 0x0480, "Multimedia controller"},
1589     { 0x0500, "RAM controller", "memory"},
1590     { 0x0501, "Flash controller", "flash"},
1591     { 0x0580, "Memory controller"},
1592     { 0x0600, "Host bridge", "host"},
1593     { 0x0601, "ISA bridge", "isa"},
1594     { 0x0602, "EISA bridge", "eisa"},
1595     { 0x0603, "MC bridge", "mca"},
1596     { 0x0604, "PCI bridge", "pci-bridge"},
1597     { 0x0605, "PCMCIA bridge", "pcmcia"},
1598     { 0x0606, "NUBUS bridge", "nubus"},
1599     { 0x0607, "CARDBUS bridge", "cardbus"},
1600     { 0x0608, "RACEWAY bridge"},
1601     { 0x0680, "Bridge"},
1602     { 0x0700, "Serial port", "serial"},
1603     { 0x0701, "Parallel port", "parallel"},
1604     { 0x0800, "Interrupt controller", "interrupt-controller"},
1605     { 0x0801, "DMA controller", "dma-controller"},
1606     { 0x0802, "Timer", "timer"},
1607     { 0x0803, "RTC", "rtc"},
1608     { 0x0900, "Keyboard", "keyboard"},
1609     { 0x0901, "Pen", "pen"},
1610     { 0x0902, "Mouse", "mouse"},
1611     { 0x0A00, "Dock station", "dock", 0x00ff},
1612     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1613     { 0x0c00, "Fireware contorller", "fireware"},
1614     { 0x0c01, "Access bus controller", "access-bus"},
1615     { 0x0c02, "SSA controller", "ssa"},
1616     { 0x0c03, "USB controller", "usb"},
1617     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1618     { 0x0c05, "SMBus"},
1619     { 0, NULL}
1620 };
1621 
1622 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1623                                                   void (*fn)(PCIBus *b,
1624                                                              PCIDevice *d,
1625                                                              void *opaque),
1626                                                   void *opaque)
1627 {
1628     PCIDevice *d;
1629     int devfn;
1630 
1631     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1632         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1633         if (d) {
1634             fn(bus, d, opaque);
1635         }
1636     }
1637 }
1638 
1639 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1640                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1641                          void *opaque)
1642 {
1643     bus = pci_find_bus_nr(bus, bus_num);
1644 
1645     if (bus) {
1646         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1647     }
1648 }
1649 
1650 static void pci_for_each_device_under_bus(PCIBus *bus,
1651                                           void (*fn)(PCIBus *b, PCIDevice *d,
1652                                                      void *opaque),
1653                                           void *opaque)
1654 {
1655     PCIDevice *d;
1656     int devfn;
1657 
1658     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1659         d = bus->devices[devfn];
1660         if (d) {
1661             fn(bus, d, opaque);
1662         }
1663     }
1664 }
1665 
1666 void pci_for_each_device(PCIBus *bus, int bus_num,
1667                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1668                          void *opaque)
1669 {
1670     bus = pci_find_bus_nr(bus, bus_num);
1671 
1672     if (bus) {
1673         pci_for_each_device_under_bus(bus, fn, opaque);
1674     }
1675 }
1676 
1677 static const pci_class_desc *get_class_desc(int class)
1678 {
1679     const pci_class_desc *desc;
1680 
1681     desc = pci_class_descriptions;
1682     while (desc->desc && class != desc->class) {
1683         desc++;
1684     }
1685 
1686     return desc;
1687 }
1688 
1689 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1690 
1691 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1692 {
1693     PciMemoryRegionList *head = NULL, **tail = &head;
1694     int i;
1695 
1696     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1697         const PCIIORegion *r = &dev->io_regions[i];
1698         PciMemoryRegion *region;
1699 
1700         if (!r->size) {
1701             continue;
1702         }
1703 
1704         region = g_malloc0(sizeof(*region));
1705 
1706         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1707             region->type = g_strdup("io");
1708         } else {
1709             region->type = g_strdup("memory");
1710             region->has_prefetch = true;
1711             region->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1712             region->has_mem_type_64 = true;
1713             region->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1714         }
1715 
1716         region->bar = i;
1717         region->address = r->addr;
1718         region->size = r->size;
1719 
1720         QAPI_LIST_APPEND(tail, region);
1721     }
1722 
1723     return head;
1724 }
1725 
1726 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1727                                            int bus_num)
1728 {
1729     PciBridgeInfo *info;
1730     PciMemoryRange *range;
1731 
1732     info = g_new0(PciBridgeInfo, 1);
1733 
1734     info->bus = g_new0(PciBusInfo, 1);
1735     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1736     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1737     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1738 
1739     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1740     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1741     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1742 
1743     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1744     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1745     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1746 
1747     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1748     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1749     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1750 
1751     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1752         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1753         if (child_bus) {
1754             info->has_devices = true;
1755             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1756         }
1757     }
1758 
1759     return info;
1760 }
1761 
1762 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1763                                            int bus_num)
1764 {
1765     const pci_class_desc *desc;
1766     PciDeviceInfo *info;
1767     uint8_t type;
1768     int class;
1769 
1770     info = g_new0(PciDeviceInfo, 1);
1771     info->bus = bus_num;
1772     info->slot = PCI_SLOT(dev->devfn);
1773     info->function = PCI_FUNC(dev->devfn);
1774 
1775     info->class_info = g_new0(PciDeviceClass, 1);
1776     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1777     info->class_info->q_class = class;
1778     desc = get_class_desc(class);
1779     if (desc->desc) {
1780         info->class_info->has_desc = true;
1781         info->class_info->desc = g_strdup(desc->desc);
1782     }
1783 
1784     info->id = g_new0(PciDeviceId, 1);
1785     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1786     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1787     info->regions = qmp_query_pci_regions(dev);
1788     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1789 
1790     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1791     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1792         info->has_irq = true;
1793         info->irq = dev->config[PCI_INTERRUPT_LINE];
1794     }
1795 
1796     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1797     if (type == PCI_HEADER_TYPE_BRIDGE) {
1798         info->has_pci_bridge = true;
1799         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1800     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1801         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1802         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1803         info->id->subsystem_vendor =
1804             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1805     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1806         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1807         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1808         info->id->subsystem_vendor =
1809             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1810     }
1811 
1812     return info;
1813 }
1814 
1815 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1816 {
1817     PciDeviceInfoList *head = NULL, **tail = &head;
1818     PCIDevice *dev;
1819     int devfn;
1820 
1821     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1822         dev = bus->devices[devfn];
1823         if (dev) {
1824             QAPI_LIST_APPEND(tail, qmp_query_pci_device(dev, bus, bus_num));
1825         }
1826     }
1827 
1828     return head;
1829 }
1830 
1831 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1832 {
1833     PciInfo *info = NULL;
1834 
1835     bus = pci_find_bus_nr(bus, bus_num);
1836     if (bus) {
1837         info = g_malloc0(sizeof(*info));
1838         info->bus = bus_num;
1839         info->devices = qmp_query_pci_devices(bus, bus_num);
1840     }
1841 
1842     return info;
1843 }
1844 
1845 PciInfoList *qmp_query_pci(Error **errp)
1846 {
1847     PciInfoList *head = NULL, **tail = &head;
1848     PCIHostState *host_bridge;
1849 
1850     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1851         QAPI_LIST_APPEND(tail,
1852                          qmp_query_pci_bus(host_bridge->bus,
1853                                            pci_bus_num(host_bridge->bus)));
1854     }
1855 
1856     return head;
1857 }
1858 
1859 /* Initialize a PCI NIC.  */
1860 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1861                                const char *default_model,
1862                                const char *default_devaddr)
1863 {
1864     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1865     GSList *list;
1866     GPtrArray *pci_nic_models;
1867     PCIBus *bus;
1868     PCIDevice *pci_dev;
1869     DeviceState *dev;
1870     int devfn;
1871     int i;
1872     int dom, busnr;
1873     unsigned slot;
1874 
1875     if (nd->model && !strcmp(nd->model, "virtio")) {
1876         g_free(nd->model);
1877         nd->model = g_strdup("virtio-net-pci");
1878     }
1879 
1880     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1881     pci_nic_models = g_ptr_array_new();
1882     while (list) {
1883         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1884                                              TYPE_DEVICE);
1885         GSList *next;
1886         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1887             dc->user_creatable) {
1888             const char *name = object_class_get_name(list->data);
1889             /*
1890              * A network device might also be something else than a NIC, see
1891              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1892              * property, too. Unfortunately, some devices like virtio-net only
1893              * create this property during instance_init, so we have to create
1894              * a temporary instance here to be able to check it.
1895              */
1896             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1897             if (object_property_find(obj, "netdev")) {
1898                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1899             }
1900             object_unref(obj);
1901         }
1902         next = list->next;
1903         g_slist_free_1(list);
1904         list = next;
1905     }
1906     g_ptr_array_add(pci_nic_models, NULL);
1907 
1908     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1909         exit(0);
1910     }
1911 
1912     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1913                             default_model);
1914     if (i < 0) {
1915         exit(1);
1916     }
1917 
1918     if (!rootbus) {
1919         error_report("No primary PCI bus");
1920         exit(1);
1921     }
1922 
1923     assert(!rootbus->parent_dev);
1924 
1925     if (!devaddr) {
1926         devfn = -1;
1927         busnr = 0;
1928     } else {
1929         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1930             error_report("Invalid PCI device address %s for device %s",
1931                          devaddr, nd->model);
1932             exit(1);
1933         }
1934 
1935         if (dom != 0) {
1936             error_report("No support for non-zero PCI domains");
1937             exit(1);
1938         }
1939 
1940         devfn = PCI_DEVFN(slot, 0);
1941     }
1942 
1943     bus = pci_find_bus_nr(rootbus, busnr);
1944     if (!bus) {
1945         error_report("Invalid PCI device address %s for device %s",
1946                      devaddr, nd->model);
1947         exit(1);
1948     }
1949 
1950     pci_dev = pci_new(devfn, nd->model);
1951     dev = &pci_dev->qdev;
1952     qdev_set_nic_properties(dev, nd);
1953     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1954     g_ptr_array_free(pci_nic_models, true);
1955     return pci_dev;
1956 }
1957 
1958 PCIDevice *pci_vga_init(PCIBus *bus)
1959 {
1960     switch (vga_interface_type) {
1961     case VGA_CIRRUS:
1962         return pci_create_simple(bus, -1, "cirrus-vga");
1963     case VGA_QXL:
1964         return pci_create_simple(bus, -1, "qxl-vga");
1965     case VGA_STD:
1966         return pci_create_simple(bus, -1, "VGA");
1967     case VGA_VMWARE:
1968         return pci_create_simple(bus, -1, "vmware-svga");
1969     case VGA_VIRTIO:
1970         return pci_create_simple(bus, -1, "virtio-vga");
1971     case VGA_NONE:
1972     default: /* Other non-PCI types. Checking for unsupported types is already
1973                 done in vl.c. */
1974         return NULL;
1975     }
1976 }
1977 
1978 /* Whether a given bus number is in range of the secondary
1979  * bus of the given bridge device. */
1980 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1981 {
1982     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1983              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1984         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1985         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1986 }
1987 
1988 /* Whether a given bus number is in a range of a root bus */
1989 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1990 {
1991     int i;
1992 
1993     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1994         PCIDevice *dev = bus->devices[i];
1995 
1996         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1997             if (pci_secondary_bus_in_range(dev, bus_num)) {
1998                 return true;
1999             }
2000         }
2001     }
2002 
2003     return false;
2004 }
2005 
2006 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2007 {
2008     PCIBus *sec;
2009 
2010     if (!bus) {
2011         return NULL;
2012     }
2013 
2014     if (pci_bus_num(bus) == bus_num) {
2015         return bus;
2016     }
2017 
2018     /* Consider all bus numbers in range for the host pci bridge. */
2019     if (!pci_bus_is_root(bus) &&
2020         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2021         return NULL;
2022     }
2023 
2024     /* try child bus */
2025     for (; bus; bus = sec) {
2026         QLIST_FOREACH(sec, &bus->child, sibling) {
2027             if (pci_bus_num(sec) == bus_num) {
2028                 return sec;
2029             }
2030             /* PXB buses assumed to be children of bus 0 */
2031             if (pci_bus_is_root(sec)) {
2032                 if (pci_root_bus_in_range(sec, bus_num)) {
2033                     break;
2034                 }
2035             } else {
2036                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2037                     break;
2038                 }
2039             }
2040         }
2041     }
2042 
2043     return NULL;
2044 }
2045 
2046 void pci_for_each_bus_depth_first(PCIBus *bus,
2047                                   void *(*begin)(PCIBus *bus, void *parent_state),
2048                                   void (*end)(PCIBus *bus, void *state),
2049                                   void *parent_state)
2050 {
2051     PCIBus *sec;
2052     void *state;
2053 
2054     if (!bus) {
2055         return;
2056     }
2057 
2058     if (begin) {
2059         state = begin(bus, parent_state);
2060     } else {
2061         state = parent_state;
2062     }
2063 
2064     QLIST_FOREACH(sec, &bus->child, sibling) {
2065         pci_for_each_bus_depth_first(sec, begin, end, state);
2066     }
2067 
2068     if (end) {
2069         end(bus, state);
2070     }
2071 }
2072 
2073 
2074 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2075 {
2076     bus = pci_find_bus_nr(bus, bus_num);
2077 
2078     if (!bus)
2079         return NULL;
2080 
2081     return bus->devices[devfn];
2082 }
2083 
2084 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2085 {
2086     PCIDevice *pci_dev = (PCIDevice *)qdev;
2087     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2088     ObjectClass *klass = OBJECT_CLASS(pc);
2089     Error *local_err = NULL;
2090     bool is_default_rom;
2091     uint16_t class_id;
2092 
2093     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2094         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2095         return;
2096     }
2097 
2098     /* initialize cap_present for pci_is_express() and pci_config_size(),
2099      * Note that hybrid PCIs are not set automatically and need to manage
2100      * QEMU_PCI_CAP_EXPRESS manually */
2101     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2102        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2103         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2104     }
2105 
2106     pci_dev = do_pci_register_device(pci_dev,
2107                                      object_get_typename(OBJECT(qdev)),
2108                                      pci_dev->devfn, errp);
2109     if (pci_dev == NULL)
2110         return;
2111 
2112     if (pc->realize) {
2113         pc->realize(pci_dev, &local_err);
2114         if (local_err) {
2115             error_propagate(errp, local_err);
2116             do_pci_unregister_device(pci_dev);
2117             return;
2118         }
2119     }
2120 
2121     if (pci_dev->failover_pair_id) {
2122         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2123             error_setg(errp, "failover primary device must be on "
2124                              "PCIExpress bus");
2125             pci_qdev_unrealize(DEVICE(pci_dev));
2126             return;
2127         }
2128         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2129         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2130             error_setg(errp, "failover primary device is not an "
2131                              "Ethernet device");
2132             pci_qdev_unrealize(DEVICE(pci_dev));
2133             return;
2134         }
2135         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2136             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2137             error_setg(errp, "failover: primary device must be in its own "
2138                               "PCI slot");
2139             pci_qdev_unrealize(DEVICE(pci_dev));
2140             return;
2141         }
2142         qdev->allow_unplug_during_migration = true;
2143     }
2144 
2145     /* rom loading */
2146     is_default_rom = false;
2147     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2148         pci_dev->romfile = g_strdup(pc->romfile);
2149         is_default_rom = true;
2150     }
2151 
2152     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2153     if (local_err) {
2154         error_propagate(errp, local_err);
2155         pci_qdev_unrealize(DEVICE(pci_dev));
2156         return;
2157     }
2158 }
2159 
2160 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2161                                  const char *name)
2162 {
2163     DeviceState *dev;
2164 
2165     dev = qdev_new(name);
2166     qdev_prop_set_int32(dev, "addr", devfn);
2167     qdev_prop_set_bit(dev, "multifunction", multifunction);
2168     return PCI_DEVICE(dev);
2169 }
2170 
2171 PCIDevice *pci_new(int devfn, const char *name)
2172 {
2173     return pci_new_multifunction(devfn, false, name);
2174 }
2175 
2176 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2177 {
2178     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2179 }
2180 
2181 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2182                                            bool multifunction,
2183                                            const char *name)
2184 {
2185     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2186     pci_realize_and_unref(dev, bus, &error_fatal);
2187     return dev;
2188 }
2189 
2190 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2191 {
2192     return pci_create_simple_multifunction(bus, devfn, false, name);
2193 }
2194 
2195 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2196 {
2197     int offset = PCI_CONFIG_HEADER_SIZE;
2198     int i;
2199     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2200         if (pdev->used[i])
2201             offset = i + 1;
2202         else if (i - offset + 1 == size)
2203             return offset;
2204     }
2205     return 0;
2206 }
2207 
2208 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2209                                         uint8_t *prev_p)
2210 {
2211     uint8_t next, prev;
2212 
2213     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2214         return 0;
2215 
2216     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2217          prev = next + PCI_CAP_LIST_NEXT)
2218         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2219             break;
2220 
2221     if (prev_p)
2222         *prev_p = prev;
2223     return next;
2224 }
2225 
2226 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2227 {
2228     uint8_t next, prev, found = 0;
2229 
2230     if (!(pdev->used[offset])) {
2231         return 0;
2232     }
2233 
2234     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2235 
2236     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2237          prev = next + PCI_CAP_LIST_NEXT) {
2238         if (next <= offset && next > found) {
2239             found = next;
2240         }
2241     }
2242     return found;
2243 }
2244 
2245 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2246    This is needed for an option rom which is used for more than one device. */
2247 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2248 {
2249     uint16_t vendor_id;
2250     uint16_t device_id;
2251     uint16_t rom_vendor_id;
2252     uint16_t rom_device_id;
2253     uint16_t rom_magic;
2254     uint16_t pcir_offset;
2255     uint8_t checksum;
2256 
2257     /* Words in rom data are little endian (like in PCI configuration),
2258        so they can be read / written with pci_get_word / pci_set_word. */
2259 
2260     /* Only a valid rom will be patched. */
2261     rom_magic = pci_get_word(ptr);
2262     if (rom_magic != 0xaa55) {
2263         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2264         return;
2265     }
2266     pcir_offset = pci_get_word(ptr + 0x18);
2267     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2268         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2269         return;
2270     }
2271 
2272     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2273     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2274     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2275     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2276 
2277     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2278                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2279 
2280     checksum = ptr[6];
2281 
2282     if (vendor_id != rom_vendor_id) {
2283         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2284         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2285         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2286         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2287         ptr[6] = checksum;
2288         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2289     }
2290 
2291     if (device_id != rom_device_id) {
2292         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2293         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2294         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2295         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2296         ptr[6] = checksum;
2297         pci_set_word(ptr + pcir_offset + 6, device_id);
2298     }
2299 }
2300 
2301 /* Add an option rom for the device */
2302 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2303                                Error **errp)
2304 {
2305     int64_t size;
2306     char *path;
2307     void *ptr;
2308     char name[32];
2309     const VMStateDescription *vmsd;
2310 
2311     if (!pdev->romfile)
2312         return;
2313     if (strlen(pdev->romfile) == 0)
2314         return;
2315 
2316     if (!pdev->rom_bar) {
2317         /*
2318          * Load rom via fw_cfg instead of creating a rom bar,
2319          * for 0.11 compatibility.
2320          */
2321         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2322 
2323         /*
2324          * Hot-plugged devices can't use the option ROM
2325          * if the rom bar is disabled.
2326          */
2327         if (DEVICE(pdev)->hotplugged) {
2328             error_setg(errp, "Hot-plugged device without ROM bar"
2329                        " can't have an option ROM");
2330             return;
2331         }
2332 
2333         if (class == 0x0300) {
2334             rom_add_vga(pdev->romfile);
2335         } else {
2336             rom_add_option(pdev->romfile, -1);
2337         }
2338         return;
2339     }
2340 
2341     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2342     if (path == NULL) {
2343         path = g_strdup(pdev->romfile);
2344     }
2345 
2346     size = get_image_size(path);
2347     if (size < 0) {
2348         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2349         g_free(path);
2350         return;
2351     } else if (size == 0) {
2352         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2353         g_free(path);
2354         return;
2355     } else if (size > 2 * GiB) {
2356         error_setg(errp, "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2357                    pdev->romfile);
2358         g_free(path);
2359         return;
2360     }
2361     if (pdev->romsize != -1) {
2362         if (size > pdev->romsize) {
2363             error_setg(errp, "romfile \"%s\" (%u bytes) is too large for ROM size %u",
2364                        pdev->romfile, (uint32_t)size, pdev->romsize);
2365             g_free(path);
2366             return;
2367         }
2368     } else {
2369         pdev->romsize = pow2ceil(size);
2370     }
2371 
2372     vmsd = qdev_get_vmsd(DEVICE(pdev));
2373 
2374     if (vmsd) {
2375         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2376     } else {
2377         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2378     }
2379     pdev->has_rom = true;
2380     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize, &error_fatal);
2381     ptr = memory_region_get_ram_ptr(&pdev->rom);
2382     if (load_image_size(path, ptr, size) < 0) {
2383         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2384         g_free(path);
2385         return;
2386     }
2387     g_free(path);
2388 
2389     if (is_default_rom) {
2390         /* Only the default rom images will be patched (if needed). */
2391         pci_patch_ids(pdev, ptr, size);
2392     }
2393 
2394     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2395 }
2396 
2397 static void pci_del_option_rom(PCIDevice *pdev)
2398 {
2399     if (!pdev->has_rom)
2400         return;
2401 
2402     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2403     pdev->has_rom = false;
2404 }
2405 
2406 /*
2407  * On success, pci_add_capability() returns a positive value
2408  * that the offset of the pci capability.
2409  * On failure, it sets an error and returns a negative error
2410  * code.
2411  */
2412 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2413                        uint8_t offset, uint8_t size,
2414                        Error **errp)
2415 {
2416     uint8_t *config;
2417     int i, overlapping_cap;
2418 
2419     if (!offset) {
2420         offset = pci_find_space(pdev, size);
2421         /* out of PCI config space is programming error */
2422         assert(offset);
2423     } else {
2424         /* Verify that capabilities don't overlap.  Note: device assignment
2425          * depends on this check to verify that the device is not broken.
2426          * Should never trigger for emulated devices, but it's helpful
2427          * for debugging these. */
2428         for (i = offset; i < offset + size; i++) {
2429             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2430             if (overlapping_cap) {
2431                 error_setg(errp, "%s:%02x:%02x.%x "
2432                            "Attempt to add PCI capability %x at offset "
2433                            "%x overlaps existing capability %x at offset %x",
2434                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2435                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2436                            cap_id, offset, overlapping_cap, i);
2437                 return -EINVAL;
2438             }
2439         }
2440     }
2441 
2442     config = pdev->config + offset;
2443     config[PCI_CAP_LIST_ID] = cap_id;
2444     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2445     pdev->config[PCI_CAPABILITY_LIST] = offset;
2446     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2447     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2448     /* Make capability read-only by default */
2449     memset(pdev->wmask + offset, 0, size);
2450     /* Check capability by default */
2451     memset(pdev->cmask + offset, 0xFF, size);
2452     return offset;
2453 }
2454 
2455 /* Unlink capability from the pci config space. */
2456 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2457 {
2458     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2459     if (!offset)
2460         return;
2461     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2462     /* Make capability writable again */
2463     memset(pdev->wmask + offset, 0xff, size);
2464     memset(pdev->w1cmask + offset, 0, size);
2465     /* Clear cmask as device-specific registers can't be checked */
2466     memset(pdev->cmask + offset, 0, size);
2467     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2468 
2469     if (!pdev->config[PCI_CAPABILITY_LIST])
2470         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2471 }
2472 
2473 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2474 {
2475     return pci_find_capability_list(pdev, cap_id, NULL);
2476 }
2477 
2478 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2479 {
2480     PCIDevice *d = (PCIDevice *)dev;
2481     const pci_class_desc *desc;
2482     char ctxt[64];
2483     PCIIORegion *r;
2484     int i, class;
2485 
2486     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2487     desc = pci_class_descriptions;
2488     while (desc->desc && class != desc->class)
2489         desc++;
2490     if (desc->desc) {
2491         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2492     } else {
2493         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2494     }
2495 
2496     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2497                    "pci id %04x:%04x (sub %04x:%04x)\n",
2498                    indent, "", ctxt, pci_dev_bus_num(d),
2499                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2500                    pci_get_word(d->config + PCI_VENDOR_ID),
2501                    pci_get_word(d->config + PCI_DEVICE_ID),
2502                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2503                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2504     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2505         r = &d->io_regions[i];
2506         if (!r->size)
2507             continue;
2508         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2509                        " [0x%"FMT_PCIBUS"]\n",
2510                        indent, "",
2511                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2512                        r->addr, r->addr + r->size - 1);
2513     }
2514 }
2515 
2516 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2517 {
2518     PCIDevice *d = (PCIDevice *)dev;
2519     const char *name = NULL;
2520     const pci_class_desc *desc =  pci_class_descriptions;
2521     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2522 
2523     while (desc->desc &&
2524           (class & ~desc->fw_ign_bits) !=
2525           (desc->class & ~desc->fw_ign_bits)) {
2526         desc++;
2527     }
2528 
2529     if (desc->desc) {
2530         name = desc->fw_name;
2531     }
2532 
2533     if (name) {
2534         pstrcpy(buf, len, name);
2535     } else {
2536         snprintf(buf, len, "pci%04x,%04x",
2537                  pci_get_word(d->config + PCI_VENDOR_ID),
2538                  pci_get_word(d->config + PCI_DEVICE_ID));
2539     }
2540 
2541     return buf;
2542 }
2543 
2544 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2545 {
2546     PCIDevice *d = (PCIDevice *)dev;
2547     char path[50], name[33];
2548     int off;
2549 
2550     off = snprintf(path, sizeof(path), "%s@%x",
2551                    pci_dev_fw_name(dev, name, sizeof name),
2552                    PCI_SLOT(d->devfn));
2553     if (PCI_FUNC(d->devfn))
2554         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2555     return g_strdup(path);
2556 }
2557 
2558 static char *pcibus_get_dev_path(DeviceState *dev)
2559 {
2560     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2561     PCIDevice *t;
2562     int slot_depth;
2563     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2564      * 00 is added here to make this format compatible with
2565      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2566      * Slot.Function list specifies the slot and function numbers for all
2567      * devices on the path from root to the specific device. */
2568     const char *root_bus_path;
2569     int root_bus_len;
2570     char slot[] = ":SS.F";
2571     int slot_len = sizeof slot - 1 /* For '\0' */;
2572     int path_len;
2573     char *path, *p;
2574     int s;
2575 
2576     root_bus_path = pci_root_bus_path(d);
2577     root_bus_len = strlen(root_bus_path);
2578 
2579     /* Calculate # of slots on path between device and root. */;
2580     slot_depth = 0;
2581     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2582         ++slot_depth;
2583     }
2584 
2585     path_len = root_bus_len + slot_len * slot_depth;
2586 
2587     /* Allocate memory, fill in the terminating null byte. */
2588     path = g_malloc(path_len + 1 /* For '\0' */);
2589     path[path_len] = '\0';
2590 
2591     memcpy(path, root_bus_path, root_bus_len);
2592 
2593     /* Fill in slot numbers. We walk up from device to root, so need to print
2594      * them in the reverse order, last to first. */
2595     p = path + path_len;
2596     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2597         p -= slot_len;
2598         s = snprintf(slot, sizeof slot, ":%02x.%x",
2599                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2600         assert(s == slot_len);
2601         memcpy(p, slot, slot_len);
2602     }
2603 
2604     return path;
2605 }
2606 
2607 static int pci_qdev_find_recursive(PCIBus *bus,
2608                                    const char *id, PCIDevice **pdev)
2609 {
2610     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2611     if (!qdev) {
2612         return -ENODEV;
2613     }
2614 
2615     /* roughly check if given qdev is pci device */
2616     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2617         *pdev = PCI_DEVICE(qdev);
2618         return 0;
2619     }
2620     return -EINVAL;
2621 }
2622 
2623 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2624 {
2625     PCIHostState *host_bridge;
2626     int rc = -ENODEV;
2627 
2628     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2629         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2630         if (!tmp) {
2631             rc = 0;
2632             break;
2633         }
2634         if (tmp != -ENODEV) {
2635             rc = tmp;
2636         }
2637     }
2638 
2639     return rc;
2640 }
2641 
2642 MemoryRegion *pci_address_space(PCIDevice *dev)
2643 {
2644     return pci_get_bus(dev)->address_space_mem;
2645 }
2646 
2647 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2648 {
2649     return pci_get_bus(dev)->address_space_io;
2650 }
2651 
2652 static void pci_device_class_init(ObjectClass *klass, void *data)
2653 {
2654     DeviceClass *k = DEVICE_CLASS(klass);
2655 
2656     k->realize = pci_qdev_realize;
2657     k->unrealize = pci_qdev_unrealize;
2658     k->bus_type = TYPE_PCI_BUS;
2659     device_class_set_props(k, pci_props);
2660 }
2661 
2662 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2663 {
2664     if (!object_class_is_abstract(klass)) {
2665         ObjectClass *conventional =
2666             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2667         ObjectClass *pcie =
2668             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2669         assert(conventional || pcie);
2670     }
2671 }
2672 
2673 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2674 {
2675     PCIBus *bus = pci_get_bus(dev);
2676     PCIBus *iommu_bus = bus;
2677     uint8_t devfn = dev->devfn;
2678 
2679     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2680         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2681 
2682         /*
2683          * The requester ID of the provided device may be aliased, as seen from
2684          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2685          * requester ID to provide a unique AddressSpace for devices, but
2686          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2687          * to-PCI bridge creates and accepts transactions on behalf of down-
2688          * stream devices.  When doing so, all downstream devices are masked
2689          * (aliased) behind a single requester ID.  The requester ID used
2690          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2691          * bridges, with a PCIe capability indicating such, follow the
2692          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2693          * where the bridge uses the seconary bus as the bridge portion of the
2694          * requester ID and devfn of 00.0.  For other bridges, typically those
2695          * found on the root complex such as the dmi-to-pci-bridge, we follow
2696          * the convention of typical bare-metal hardware, which uses the
2697          * requester ID of the bridge itself.  There are device specific
2698          * exceptions to these rules, but these are the defaults that the
2699          * Linux kernel uses when determining DMA aliases itself and believed
2700          * to be true for the bare metal equivalents of the devices emulated
2701          * in QEMU.
2702          */
2703         if (!pci_bus_is_express(iommu_bus)) {
2704             PCIDevice *parent = iommu_bus->parent_dev;
2705 
2706             if (pci_is_express(parent) &&
2707                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2708                 devfn = PCI_DEVFN(0, 0);
2709                 bus = iommu_bus;
2710             } else {
2711                 devfn = parent->devfn;
2712                 bus = parent_bus;
2713             }
2714         }
2715 
2716         iommu_bus = parent_bus;
2717     }
2718     if (iommu_bus && iommu_bus->iommu_fn) {
2719         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2720     }
2721     return &address_space_memory;
2722 }
2723 
2724 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2725 {
2726     bus->iommu_fn = fn;
2727     bus->iommu_opaque = opaque;
2728 }
2729 
2730 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2731 {
2732     Range *range = opaque;
2733     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2734     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2735     int i;
2736 
2737     if (!(cmd & PCI_COMMAND_MEMORY)) {
2738         return;
2739     }
2740 
2741     if (pc->is_bridge) {
2742         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2743         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2744 
2745         base = MAX(base, 0x1ULL << 32);
2746 
2747         if (limit >= base) {
2748             Range pref_range;
2749             range_set_bounds(&pref_range, base, limit);
2750             range_extend(range, &pref_range);
2751         }
2752     }
2753     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2754         PCIIORegion *r = &dev->io_regions[i];
2755         pcibus_t lob, upb;
2756         Range region_range;
2757 
2758         if (!r->size ||
2759             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2760             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2761             continue;
2762         }
2763 
2764         lob = pci_bar_address(dev, i, r->type, r->size);
2765         upb = lob + r->size - 1;
2766         if (lob == PCI_BAR_UNMAPPED) {
2767             continue;
2768         }
2769 
2770         lob = MAX(lob, 0x1ULL << 32);
2771 
2772         if (upb >= lob) {
2773             range_set_bounds(&region_range, lob, upb);
2774             range_extend(range, &region_range);
2775         }
2776     }
2777 }
2778 
2779 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2780 {
2781     range_make_empty(range);
2782     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2783 }
2784 
2785 static bool pcie_has_upstream_port(PCIDevice *dev)
2786 {
2787     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2788 
2789     /* Device associated with an upstream port.
2790      * As there are several types of these, it's easier to check the
2791      * parent device: upstream ports are always connected to
2792      * root or downstream ports.
2793      */
2794     return parent_dev &&
2795         pci_is_express(parent_dev) &&
2796         parent_dev->exp.exp_cap &&
2797         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2798          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2799 }
2800 
2801 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2802 {
2803     PCIBus *bus = pci_get_bus(pci_dev);
2804 
2805     if(pcie_has_upstream_port(pci_dev)) {
2806         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2807         return bus->devices[0];
2808     } else {
2809         /* Other bus types might support multiple devices at slots 0-31 */
2810         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2811     }
2812 }
2813 
2814 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2815 {
2816     MSIMessage msg;
2817     if (msix_enabled(dev)) {
2818         msg = msix_get_message(dev, vector);
2819     } else if (msi_enabled(dev)) {
2820         msg = msi_get_message(dev, vector);
2821     } else {
2822         /* Should never happen */
2823         error_report("%s: unknown interrupt type", __func__);
2824         abort();
2825     }
2826     return msg;
2827 }
2828 
2829 static const TypeInfo pci_device_type_info = {
2830     .name = TYPE_PCI_DEVICE,
2831     .parent = TYPE_DEVICE,
2832     .instance_size = sizeof(PCIDevice),
2833     .abstract = true,
2834     .class_size = sizeof(PCIDeviceClass),
2835     .class_init = pci_device_class_init,
2836     .class_base_init = pci_device_class_base_init,
2837 };
2838 
2839 static void pci_register_types(void)
2840 {
2841     type_register_static(&pci_bus_info);
2842     type_register_static(&pcie_bus_info);
2843     type_register_static(&conventional_pci_interface_info);
2844     type_register_static(&pcie_interface_info);
2845     type_register_static(&pci_device_type_info);
2846 }
2847 
2848 type_init(pci_register_types)
2849