xref: /openbmc/qemu/hw/pci/pci.c (revision 7cc25f6c)
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 "hw/irq.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bridge.h"
31 #include "hw/pci/pci_bus.h"
32 #include "hw/pci/pci_host.h"
33 #include "hw/qdev-properties.h"
34 #include "migration/qemu-file-types.h"
35 #include "migration/vmstate.h"
36 #include "monitor/monitor.h"
37 #include "net/net.h"
38 #include "sysemu/numa.h"
39 #include "sysemu/sysemu.h"
40 #include "hw/loader.h"
41 #include "qemu/error-report.h"
42 #include "qemu/range.h"
43 #include "trace.h"
44 #include "hw/pci/msi.h"
45 #include "hw/pci/msix.h"
46 #include "exec/address-spaces.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "qapi/error.h"
50 #include "qapi/qapi-commands-pci.h"
51 #include "qemu/cutils.h"
52 
53 //#define DEBUG_PCI
54 #ifdef DEBUG_PCI
55 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
56 #else
57 # define PCI_DPRINTF(format, ...)       do { } while (0)
58 #endif
59 
60 bool pci_available = true;
61 
62 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
63 static char *pcibus_get_dev_path(DeviceState *dev);
64 static char *pcibus_get_fw_dev_path(DeviceState *dev);
65 static void pcibus_reset(BusState *qbus);
66 
67 static Property pci_props[] = {
68     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
69     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
70     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
71     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
72                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
73     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
74                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
75     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
76                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
77     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
78                        failover_pair_id),
79     DEFINE_PROP_END_OF_LIST()
80 };
81 
82 static const VMStateDescription vmstate_pcibus = {
83     .name = "PCIBUS",
84     .version_id = 1,
85     .minimum_version_id = 1,
86     .fields = (VMStateField[]) {
87         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
88         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
89                              nirq, 0, vmstate_info_int32,
90                              int32_t),
91         VMSTATE_END_OF_LIST()
92     }
93 };
94 
95 static void pci_init_bus_master(PCIDevice *pci_dev)
96 {
97     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
98 
99     memory_region_init_alias(&pci_dev->bus_master_enable_region,
100                              OBJECT(pci_dev), "bus master",
101                              dma_as->root, 0, memory_region_size(dma_as->root));
102     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
103     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
104                                 &pci_dev->bus_master_enable_region);
105 }
106 
107 static void pcibus_machine_done(Notifier *notifier, void *data)
108 {
109     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
110     int i;
111 
112     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
113         if (bus->devices[i]) {
114             pci_init_bus_master(bus->devices[i]);
115         }
116     }
117 }
118 
119 static void pci_bus_realize(BusState *qbus, Error **errp)
120 {
121     PCIBus *bus = PCI_BUS(qbus);
122 
123     bus->machine_done.notify = pcibus_machine_done;
124     qemu_add_machine_init_done_notifier(&bus->machine_done);
125 
126     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
127 }
128 
129 static void pcie_bus_realize(BusState *qbus, Error **errp)
130 {
131     PCIBus *bus = PCI_BUS(qbus);
132 
133     pci_bus_realize(qbus, errp);
134 
135     /*
136      * A PCI-E bus can support extended config space if it's the root
137      * bus, or if the bus/bridge above it does as well
138      */
139     if (pci_bus_is_root(bus)) {
140         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
141     } else {
142         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
143 
144         if (pci_bus_allows_extended_config_space(parent_bus)) {
145             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
146         }
147     }
148 }
149 
150 static void pci_bus_unrealize(BusState *qbus)
151 {
152     PCIBus *bus = PCI_BUS(qbus);
153 
154     qemu_remove_machine_init_done_notifier(&bus->machine_done);
155 
156     vmstate_unregister(NULL, &vmstate_pcibus, bus);
157 }
158 
159 static int pcibus_num(PCIBus *bus)
160 {
161     if (pci_bus_is_root(bus)) {
162         return 0; /* pci host bridge */
163     }
164     return bus->parent_dev->config[PCI_SECONDARY_BUS];
165 }
166 
167 static uint16_t pcibus_numa_node(PCIBus *bus)
168 {
169     return NUMA_NODE_UNASSIGNED;
170 }
171 
172 static void pci_bus_class_init(ObjectClass *klass, void *data)
173 {
174     BusClass *k = BUS_CLASS(klass);
175     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
176 
177     k->print_dev = pcibus_dev_print;
178     k->get_dev_path = pcibus_get_dev_path;
179     k->get_fw_dev_path = pcibus_get_fw_dev_path;
180     k->realize = pci_bus_realize;
181     k->unrealize = pci_bus_unrealize;
182     k->reset = pcibus_reset;
183 
184     pbc->bus_num = pcibus_num;
185     pbc->numa_node = pcibus_numa_node;
186 }
187 
188 static const TypeInfo pci_bus_info = {
189     .name = TYPE_PCI_BUS,
190     .parent = TYPE_BUS,
191     .instance_size = sizeof(PCIBus),
192     .class_size = sizeof(PCIBusClass),
193     .class_init = pci_bus_class_init,
194 };
195 
196 static const TypeInfo pcie_interface_info = {
197     .name          = INTERFACE_PCIE_DEVICE,
198     .parent        = TYPE_INTERFACE,
199 };
200 
201 static const TypeInfo conventional_pci_interface_info = {
202     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
203     .parent        = TYPE_INTERFACE,
204 };
205 
206 static void pcie_bus_class_init(ObjectClass *klass, void *data)
207 {
208     BusClass *k = BUS_CLASS(klass);
209 
210     k->realize = pcie_bus_realize;
211 }
212 
213 static const TypeInfo pcie_bus_info = {
214     .name = TYPE_PCIE_BUS,
215     .parent = TYPE_PCI_BUS,
216     .class_init = pcie_bus_class_init,
217 };
218 
219 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
220 static void pci_update_mappings(PCIDevice *d);
221 static void pci_irq_handler(void *opaque, int irq_num, int level);
222 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
223 static void pci_del_option_rom(PCIDevice *pdev);
224 
225 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
226 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
227 
228 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
229 
230 int pci_bar(PCIDevice *d, int reg)
231 {
232     uint8_t type;
233 
234     if (reg != PCI_ROM_SLOT)
235         return PCI_BASE_ADDRESS_0 + reg * 4;
236 
237     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
238     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
239 }
240 
241 static inline int pci_irq_state(PCIDevice *d, int irq_num)
242 {
243         return (d->irq_state >> irq_num) & 0x1;
244 }
245 
246 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
247 {
248         d->irq_state &= ~(0x1 << irq_num);
249         d->irq_state |= level << irq_num;
250 }
251 
252 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
253 {
254     assert(irq_num >= 0);
255     assert(irq_num < bus->nirq);
256     bus->irq_count[irq_num] += change;
257     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
258 }
259 
260 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
261 {
262     PCIBus *bus;
263     for (;;) {
264         bus = pci_get_bus(pci_dev);
265         irq_num = bus->map_irq(pci_dev, irq_num);
266         if (bus->set_irq)
267             break;
268         pci_dev = bus->parent_dev;
269     }
270     pci_bus_change_irq_level(bus, irq_num, change);
271 }
272 
273 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
274 {
275     assert(irq_num >= 0);
276     assert(irq_num < bus->nirq);
277     return !!bus->irq_count[irq_num];
278 }
279 
280 /* Update interrupt status bit in config space on interrupt
281  * state change. */
282 static void pci_update_irq_status(PCIDevice *dev)
283 {
284     if (dev->irq_state) {
285         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
286     } else {
287         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
288     }
289 }
290 
291 void pci_device_deassert_intx(PCIDevice *dev)
292 {
293     int i;
294     for (i = 0; i < PCI_NUM_PINS; ++i) {
295         pci_irq_handler(dev, i, 0);
296     }
297 }
298 
299 static void pci_do_device_reset(PCIDevice *dev)
300 {
301     int r;
302 
303     pci_device_deassert_intx(dev);
304     assert(dev->irq_state == 0);
305 
306     /* Clear all writable bits */
307     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
308                                  pci_get_word(dev->wmask + PCI_COMMAND) |
309                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
310     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
311                                  pci_get_word(dev->wmask + PCI_STATUS) |
312                                  pci_get_word(dev->w1cmask + PCI_STATUS));
313     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
314     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
315                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
316                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
317     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
318     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
319         PCIIORegion *region = &dev->io_regions[r];
320         if (!region->size) {
321             continue;
322         }
323 
324         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
325             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
326             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
327         } else {
328             pci_set_long(dev->config + pci_bar(dev, r), region->type);
329         }
330     }
331     pci_update_mappings(dev);
332 
333     msi_reset(dev);
334     msix_reset(dev);
335 }
336 
337 /*
338  * This function is called on #RST and FLR.
339  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
340  */
341 void pci_device_reset(PCIDevice *dev)
342 {
343     qdev_reset_all(&dev->qdev);
344     pci_do_device_reset(dev);
345 }
346 
347 /*
348  * Trigger pci bus reset under a given bus.
349  * Called via qbus_reset_all on RST# assert, after the devices
350  * have been reset qdev_reset_all-ed already.
351  */
352 static void pcibus_reset(BusState *qbus)
353 {
354     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
355     int i;
356 
357     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
358         if (bus->devices[i]) {
359             pci_do_device_reset(bus->devices[i]);
360         }
361     }
362 
363     for (i = 0; i < bus->nirq; i++) {
364         assert(bus->irq_count[i] == 0);
365     }
366 }
367 
368 static void pci_host_bus_register(DeviceState *host)
369 {
370     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
371 
372     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
373 }
374 
375 static void pci_host_bus_unregister(DeviceState *host)
376 {
377     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
378 
379     QLIST_REMOVE(host_bridge, next);
380 }
381 
382 PCIBus *pci_device_root_bus(const PCIDevice *d)
383 {
384     PCIBus *bus = pci_get_bus(d);
385 
386     while (!pci_bus_is_root(bus)) {
387         d = bus->parent_dev;
388         assert(d != NULL);
389 
390         bus = pci_get_bus(d);
391     }
392 
393     return bus;
394 }
395 
396 const char *pci_root_bus_path(PCIDevice *dev)
397 {
398     PCIBus *rootbus = pci_device_root_bus(dev);
399     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
400     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
401 
402     assert(host_bridge->bus == rootbus);
403 
404     if (hc->root_bus_path) {
405         return (*hc->root_bus_path)(host_bridge, rootbus);
406     }
407 
408     return rootbus->qbus.name;
409 }
410 
411 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
412                               MemoryRegion *address_space_mem,
413                               MemoryRegion *address_space_io,
414                               uint8_t devfn_min)
415 {
416     assert(PCI_FUNC(devfn_min) == 0);
417     bus->devfn_min = devfn_min;
418     bus->slot_reserved_mask = 0x0;
419     bus->address_space_mem = address_space_mem;
420     bus->address_space_io = address_space_io;
421     bus->flags |= PCI_BUS_IS_ROOT;
422 
423     /* host bridge */
424     QLIST_INIT(&bus->child);
425 
426     pci_host_bus_register(parent);
427 }
428 
429 static void pci_bus_uninit(PCIBus *bus)
430 {
431     pci_host_bus_unregister(BUS(bus)->parent);
432 }
433 
434 bool pci_bus_is_express(PCIBus *bus)
435 {
436     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
437 }
438 
439 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
440                               const char *name,
441                               MemoryRegion *address_space_mem,
442                               MemoryRegion *address_space_io,
443                               uint8_t devfn_min, const char *typename)
444 {
445     qbus_create_inplace(bus, bus_size, typename, parent, name);
446     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
447                       devfn_min);
448 }
449 
450 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
451                          MemoryRegion *address_space_mem,
452                          MemoryRegion *address_space_io,
453                          uint8_t devfn_min, const char *typename)
454 {
455     PCIBus *bus;
456 
457     bus = PCI_BUS(qbus_create(typename, parent, name));
458     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
459                       devfn_min);
460     return bus;
461 }
462 
463 void pci_root_bus_cleanup(PCIBus *bus)
464 {
465     pci_bus_uninit(bus);
466     /* the caller of the unplug hotplug handler will delete this device */
467     qbus_unrealize(BUS(bus));
468 }
469 
470 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
471                   void *irq_opaque, int nirq)
472 {
473     bus->set_irq = set_irq;
474     bus->map_irq = map_irq;
475     bus->irq_opaque = irq_opaque;
476     bus->nirq = nirq;
477     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
478 }
479 
480 void pci_bus_irqs_cleanup(PCIBus *bus)
481 {
482     bus->set_irq = NULL;
483     bus->map_irq = NULL;
484     bus->irq_opaque = NULL;
485     bus->nirq = 0;
486     g_free(bus->irq_count);
487 }
488 
489 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
490                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
491                               void *irq_opaque,
492                               MemoryRegion *address_space_mem,
493                               MemoryRegion *address_space_io,
494                               uint8_t devfn_min, int nirq,
495                               const char *typename)
496 {
497     PCIBus *bus;
498 
499     bus = pci_root_bus_new(parent, name, address_space_mem,
500                            address_space_io, devfn_min, typename);
501     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
502     return bus;
503 }
504 
505 void pci_unregister_root_bus(PCIBus *bus)
506 {
507     pci_bus_irqs_cleanup(bus);
508     pci_root_bus_cleanup(bus);
509 }
510 
511 int pci_bus_num(PCIBus *s)
512 {
513     return PCI_BUS_GET_CLASS(s)->bus_num(s);
514 }
515 
516 int pci_bus_numa_node(PCIBus *bus)
517 {
518     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
519 }
520 
521 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
522                                  const VMStateField *field)
523 {
524     PCIDevice *s = container_of(pv, PCIDevice, config);
525     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
526     uint8_t *config;
527     int i;
528 
529     assert(size == pci_config_size(s));
530     config = g_malloc(size);
531 
532     qemu_get_buffer(f, config, size);
533     for (i = 0; i < size; ++i) {
534         if ((config[i] ^ s->config[i]) &
535             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
536             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
537                          "cmask: %x wmask: %x w1cmask:%x", __func__,
538                          i, config[i], s->config[i],
539                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
540             g_free(config);
541             return -EINVAL;
542         }
543     }
544     memcpy(s->config, config, size);
545 
546     pci_update_mappings(s);
547     if (pc->is_bridge) {
548         PCIBridge *b = PCI_BRIDGE(s);
549         pci_bridge_update_mappings(b);
550     }
551 
552     memory_region_set_enabled(&s->bus_master_enable_region,
553                               pci_get_word(s->config + PCI_COMMAND)
554                               & PCI_COMMAND_MASTER);
555 
556     g_free(config);
557     return 0;
558 }
559 
560 /* just put buffer */
561 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
562                                  const VMStateField *field, QJSON *vmdesc)
563 {
564     const uint8_t **v = pv;
565     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
566     qemu_put_buffer(f, *v, size);
567 
568     return 0;
569 }
570 
571 static VMStateInfo vmstate_info_pci_config = {
572     .name = "pci config",
573     .get  = get_pci_config_device,
574     .put  = put_pci_config_device,
575 };
576 
577 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
578                              const VMStateField *field)
579 {
580     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
581     uint32_t irq_state[PCI_NUM_PINS];
582     int i;
583     for (i = 0; i < PCI_NUM_PINS; ++i) {
584         irq_state[i] = qemu_get_be32(f);
585         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
586             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
587                     irq_state[i]);
588             return -EINVAL;
589         }
590     }
591 
592     for (i = 0; i < PCI_NUM_PINS; ++i) {
593         pci_set_irq_state(s, i, irq_state[i]);
594     }
595 
596     return 0;
597 }
598 
599 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
600                              const VMStateField *field, QJSON *vmdesc)
601 {
602     int i;
603     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
604 
605     for (i = 0; i < PCI_NUM_PINS; ++i) {
606         qemu_put_be32(f, pci_irq_state(s, i));
607     }
608 
609     return 0;
610 }
611 
612 static VMStateInfo vmstate_info_pci_irq_state = {
613     .name = "pci irq state",
614     .get  = get_pci_irq_state,
615     .put  = put_pci_irq_state,
616 };
617 
618 static bool migrate_is_pcie(void *opaque, int version_id)
619 {
620     return pci_is_express((PCIDevice *)opaque);
621 }
622 
623 static bool migrate_is_not_pcie(void *opaque, int version_id)
624 {
625     return !pci_is_express((PCIDevice *)opaque);
626 }
627 
628 const VMStateDescription vmstate_pci_device = {
629     .name = "PCIDevice",
630     .version_id = 2,
631     .minimum_version_id = 1,
632     .fields = (VMStateField[]) {
633         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
634         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
635                                    migrate_is_not_pcie,
636                                    0, vmstate_info_pci_config,
637                                    PCI_CONFIG_SPACE_SIZE),
638         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
639                                    migrate_is_pcie,
640                                    0, vmstate_info_pci_config,
641                                    PCIE_CONFIG_SPACE_SIZE),
642         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
643                                    vmstate_info_pci_irq_state,
644                                    PCI_NUM_PINS * sizeof(int32_t)),
645         VMSTATE_END_OF_LIST()
646     }
647 };
648 
649 
650 void pci_device_save(PCIDevice *s, QEMUFile *f)
651 {
652     /* Clear interrupt status bit: it is implicit
653      * in irq_state which we are saving.
654      * This makes us compatible with old devices
655      * which never set or clear this bit. */
656     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
657     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
658     /* Restore the interrupt status bit. */
659     pci_update_irq_status(s);
660 }
661 
662 int pci_device_load(PCIDevice *s, QEMUFile *f)
663 {
664     int ret;
665     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
666     /* Restore the interrupt status bit. */
667     pci_update_irq_status(s);
668     return ret;
669 }
670 
671 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
672 {
673     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
674                  pci_default_sub_vendor_id);
675     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
676                  pci_default_sub_device_id);
677 }
678 
679 /*
680  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
681  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
682  */
683 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
684                              unsigned int *slotp, unsigned int *funcp)
685 {
686     const char *p;
687     char *e;
688     unsigned long val;
689     unsigned long dom = 0, bus = 0;
690     unsigned int slot = 0;
691     unsigned int func = 0;
692 
693     p = addr;
694     val = strtoul(p, &e, 16);
695     if (e == p)
696         return -1;
697     if (*e == ':') {
698         bus = val;
699         p = e + 1;
700         val = strtoul(p, &e, 16);
701         if (e == p)
702             return -1;
703         if (*e == ':') {
704             dom = bus;
705             bus = val;
706             p = e + 1;
707             val = strtoul(p, &e, 16);
708             if (e == p)
709                 return -1;
710         }
711     }
712 
713     slot = val;
714 
715     if (funcp != NULL) {
716         if (*e != '.')
717             return -1;
718 
719         p = e + 1;
720         val = strtoul(p, &e, 16);
721         if (e == p)
722             return -1;
723 
724         func = val;
725     }
726 
727     /* if funcp == NULL func is 0 */
728     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
729         return -1;
730 
731     if (*e)
732         return -1;
733 
734     *domp = dom;
735     *busp = bus;
736     *slotp = slot;
737     if (funcp != NULL)
738         *funcp = func;
739     return 0;
740 }
741 
742 static void pci_init_cmask(PCIDevice *dev)
743 {
744     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
745     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
746     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
747     dev->cmask[PCI_REVISION_ID] = 0xff;
748     dev->cmask[PCI_CLASS_PROG] = 0xff;
749     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
750     dev->cmask[PCI_HEADER_TYPE] = 0xff;
751     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
752 }
753 
754 static void pci_init_wmask(PCIDevice *dev)
755 {
756     int config_size = pci_config_size(dev);
757 
758     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
759     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
760     pci_set_word(dev->wmask + PCI_COMMAND,
761                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
762                  PCI_COMMAND_INTX_DISABLE);
763     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
764 
765     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
766            config_size - PCI_CONFIG_HEADER_SIZE);
767 }
768 
769 static void pci_init_w1cmask(PCIDevice *dev)
770 {
771     /*
772      * Note: It's okay to set w1cmask even for readonly bits as
773      * long as their value is hardwired to 0.
774      */
775     pci_set_word(dev->w1cmask + PCI_STATUS,
776                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
777                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
778                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
779 }
780 
781 static void pci_init_mask_bridge(PCIDevice *d)
782 {
783     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
784        PCI_SEC_LETENCY_TIMER */
785     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
786 
787     /* base and limit */
788     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
789     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
790     pci_set_word(d->wmask + PCI_MEMORY_BASE,
791                  PCI_MEMORY_RANGE_MASK & 0xffff);
792     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
793                  PCI_MEMORY_RANGE_MASK & 0xffff);
794     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
795                  PCI_PREF_RANGE_MASK & 0xffff);
796     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
797                  PCI_PREF_RANGE_MASK & 0xffff);
798 
799     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
800     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
801 
802     /* Supported memory and i/o types */
803     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
804     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
805     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
806                                PCI_PREF_RANGE_TYPE_64);
807     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
808                                PCI_PREF_RANGE_TYPE_64);
809 
810     /*
811      * TODO: Bridges default to 10-bit VGA decoding but we currently only
812      * implement 16-bit decoding (no alias support).
813      */
814     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
815                  PCI_BRIDGE_CTL_PARITY |
816                  PCI_BRIDGE_CTL_SERR |
817                  PCI_BRIDGE_CTL_ISA |
818                  PCI_BRIDGE_CTL_VGA |
819                  PCI_BRIDGE_CTL_VGA_16BIT |
820                  PCI_BRIDGE_CTL_MASTER_ABORT |
821                  PCI_BRIDGE_CTL_BUS_RESET |
822                  PCI_BRIDGE_CTL_FAST_BACK |
823                  PCI_BRIDGE_CTL_DISCARD |
824                  PCI_BRIDGE_CTL_SEC_DISCARD |
825                  PCI_BRIDGE_CTL_DISCARD_SERR);
826     /* Below does not do anything as we never set this bit, put here for
827      * completeness. */
828     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
829                  PCI_BRIDGE_CTL_DISCARD_STATUS);
830     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
831     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
832     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
833                                PCI_PREF_RANGE_TYPE_MASK);
834     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
835                                PCI_PREF_RANGE_TYPE_MASK);
836 }
837 
838 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
839 {
840     uint8_t slot = PCI_SLOT(dev->devfn);
841     uint8_t func;
842 
843     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
844         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
845     }
846 
847     /*
848      * multifunction bit is interpreted in two ways as follows.
849      *   - all functions must set the bit to 1.
850      *     Example: Intel X53
851      *   - function 0 must set the bit, but the rest function (> 0)
852      *     is allowed to leave the bit to 0.
853      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
854      *
855      * So OS (at least Linux) checks the bit of only function 0,
856      * and doesn't see the bit of function > 0.
857      *
858      * The below check allows both interpretation.
859      */
860     if (PCI_FUNC(dev->devfn)) {
861         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
862         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
863             /* function 0 should set multifunction bit */
864             error_setg(errp, "PCI: single function device can't be populated "
865                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
866             return;
867         }
868         return;
869     }
870 
871     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
872         return;
873     }
874     /* function 0 indicates single function, so function > 0 must be NULL */
875     for (func = 1; func < PCI_FUNC_MAX; ++func) {
876         if (bus->devices[PCI_DEVFN(slot, func)]) {
877             error_setg(errp, "PCI: %x.0 indicates single function, "
878                        "but %x.%x is already populated.",
879                        slot, slot, func);
880             return;
881         }
882     }
883 }
884 
885 static void pci_config_alloc(PCIDevice *pci_dev)
886 {
887     int config_size = pci_config_size(pci_dev);
888 
889     pci_dev->config = g_malloc0(config_size);
890     pci_dev->cmask = g_malloc0(config_size);
891     pci_dev->wmask = g_malloc0(config_size);
892     pci_dev->w1cmask = g_malloc0(config_size);
893     pci_dev->used = g_malloc0(config_size);
894 }
895 
896 static void pci_config_free(PCIDevice *pci_dev)
897 {
898     g_free(pci_dev->config);
899     g_free(pci_dev->cmask);
900     g_free(pci_dev->wmask);
901     g_free(pci_dev->w1cmask);
902     g_free(pci_dev->used);
903 }
904 
905 static void do_pci_unregister_device(PCIDevice *pci_dev)
906 {
907     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
908     pci_config_free(pci_dev);
909 
910     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
911         memory_region_del_subregion(&pci_dev->bus_master_container_region,
912                                     &pci_dev->bus_master_enable_region);
913     }
914     address_space_destroy(&pci_dev->bus_master_as);
915 }
916 
917 /* Extract PCIReqIDCache into BDF format */
918 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
919 {
920     uint8_t bus_n;
921     uint16_t result;
922 
923     switch (cache->type) {
924     case PCI_REQ_ID_BDF:
925         result = pci_get_bdf(cache->dev);
926         break;
927     case PCI_REQ_ID_SECONDARY_BUS:
928         bus_n = pci_dev_bus_num(cache->dev);
929         result = PCI_BUILD_BDF(bus_n, 0);
930         break;
931     default:
932         error_report("Invalid PCI requester ID cache type: %d",
933                      cache->type);
934         exit(1);
935         break;
936     }
937 
938     return result;
939 }
940 
941 /* Parse bridges up to the root complex and return requester ID
942  * cache for specific device.  For full PCIe topology, the cache
943  * result would be exactly the same as getting BDF of the device.
944  * However, several tricks are required when system mixed up with
945  * legacy PCI devices and PCIe-to-PCI bridges.
946  *
947  * Here we cache the proxy device (and type) not requester ID since
948  * bus number might change from time to time.
949  */
950 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
951 {
952     PCIDevice *parent;
953     PCIReqIDCache cache = {
954         .dev = dev,
955         .type = PCI_REQ_ID_BDF,
956     };
957 
958     while (!pci_bus_is_root(pci_get_bus(dev))) {
959         /* We are under PCI/PCIe bridges */
960         parent = pci_get_bus(dev)->parent_dev;
961         if (pci_is_express(parent)) {
962             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
963                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
964                  * override the requester ID using secondary bus
965                  * number of parent bridge with zeroed devfn
966                  * (pcie-to-pci bridge spec chap 2.3). */
967                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
968                 cache.dev = dev;
969             }
970         } else {
971             /* Legacy PCI, override requester ID with the bridge's
972              * BDF upstream.  When the root complex connects to
973              * legacy PCI devices (including buses), it can only
974              * obtain requester ID info from directly attached
975              * devices.  If devices are attached under bridges, only
976              * the requester ID of the bridge that is directly
977              * attached to the root complex can be recognized. */
978             cache.type = PCI_REQ_ID_BDF;
979             cache.dev = parent;
980         }
981         dev = parent;
982     }
983 
984     return cache;
985 }
986 
987 uint16_t pci_requester_id(PCIDevice *dev)
988 {
989     return pci_req_id_cache_extract(&dev->requester_id_cache);
990 }
991 
992 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
993 {
994     return !(bus->devices[devfn]);
995 }
996 
997 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
998 {
999     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1000 }
1001 
1002 /* -1 for devfn means auto assign */
1003 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1004                                          const char *name, int devfn,
1005                                          Error **errp)
1006 {
1007     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1008     PCIConfigReadFunc *config_read = pc->config_read;
1009     PCIConfigWriteFunc *config_write = pc->config_write;
1010     Error *local_err = NULL;
1011     DeviceState *dev = DEVICE(pci_dev);
1012     PCIBus *bus = pci_get_bus(pci_dev);
1013 
1014     /* Only pci bridges can be attached to extra PCI root buses */
1015     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1016         error_setg(errp,
1017                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1018                     bus->parent_dev->name);
1019         return NULL;
1020     }
1021 
1022     if (devfn < 0) {
1023         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1024             devfn += PCI_FUNC_MAX) {
1025             if (pci_bus_devfn_available(bus, devfn) &&
1026                    !pci_bus_devfn_reserved(bus, devfn)) {
1027                 goto found;
1028             }
1029         }
1030         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1031                    "or reserved", name);
1032         return NULL;
1033     found: ;
1034     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1035         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1036                    " reserved",
1037                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1038         return NULL;
1039     } else if (!pci_bus_devfn_available(bus, devfn)) {
1040         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1041                    " in use by %s",
1042                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1043                    bus->devices[devfn]->name);
1044         return NULL;
1045     } else if (dev->hotplugged &&
1046                pci_get_function_0(pci_dev)) {
1047         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1048                    " new func %s cannot be exposed to guest.",
1049                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1050                    pci_get_function_0(pci_dev)->name,
1051                    name);
1052 
1053        return NULL;
1054     }
1055 
1056     pci_dev->devfn = devfn;
1057     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1058     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1059 
1060     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1061                        "bus master container", UINT64_MAX);
1062     address_space_init(&pci_dev->bus_master_as,
1063                        &pci_dev->bus_master_container_region, pci_dev->name);
1064 
1065     if (qdev_hotplug) {
1066         pci_init_bus_master(pci_dev);
1067     }
1068     pci_dev->irq_state = 0;
1069     pci_config_alloc(pci_dev);
1070 
1071     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1072     pci_config_set_device_id(pci_dev->config, pc->device_id);
1073     pci_config_set_revision(pci_dev->config, pc->revision);
1074     pci_config_set_class(pci_dev->config, pc->class_id);
1075 
1076     if (!pc->is_bridge) {
1077         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1078             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1079                          pc->subsystem_vendor_id);
1080             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1081                          pc->subsystem_id);
1082         } else {
1083             pci_set_default_subsystem_id(pci_dev);
1084         }
1085     } else {
1086         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1087         assert(!pc->subsystem_vendor_id);
1088         assert(!pc->subsystem_id);
1089     }
1090     pci_init_cmask(pci_dev);
1091     pci_init_wmask(pci_dev);
1092     pci_init_w1cmask(pci_dev);
1093     if (pc->is_bridge) {
1094         pci_init_mask_bridge(pci_dev);
1095     }
1096     pci_init_multifunction(bus, pci_dev, &local_err);
1097     if (local_err) {
1098         error_propagate(errp, local_err);
1099         do_pci_unregister_device(pci_dev);
1100         return NULL;
1101     }
1102 
1103     if (!config_read)
1104         config_read = pci_default_read_config;
1105     if (!config_write)
1106         config_write = pci_default_write_config;
1107     pci_dev->config_read = config_read;
1108     pci_dev->config_write = config_write;
1109     bus->devices[devfn] = pci_dev;
1110     pci_dev->version_id = 2; /* Current pci device vmstate version */
1111     return pci_dev;
1112 }
1113 
1114 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1115 {
1116     PCIIORegion *r;
1117     int i;
1118 
1119     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1120         r = &pci_dev->io_regions[i];
1121         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1122             continue;
1123         memory_region_del_subregion(r->address_space, r->memory);
1124     }
1125 
1126     pci_unregister_vga(pci_dev);
1127 }
1128 
1129 static void pci_qdev_unrealize(DeviceState *dev)
1130 {
1131     PCIDevice *pci_dev = PCI_DEVICE(dev);
1132     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1133 
1134     pci_unregister_io_regions(pci_dev);
1135     pci_del_option_rom(pci_dev);
1136 
1137     if (pc->exit) {
1138         pc->exit(pci_dev);
1139     }
1140 
1141     pci_device_deassert_intx(pci_dev);
1142     do_pci_unregister_device(pci_dev);
1143 }
1144 
1145 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1146                       uint8_t type, MemoryRegion *memory)
1147 {
1148     PCIIORegion *r;
1149     uint32_t addr; /* offset in pci config space */
1150     uint64_t wmask;
1151     pcibus_t size = memory_region_size(memory);
1152     uint8_t hdr_type;
1153 
1154     assert(region_num >= 0);
1155     assert(region_num < PCI_NUM_REGIONS);
1156     assert(is_power_of_2(size));
1157 
1158     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1159     hdr_type =
1160         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1161     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1162 
1163     r = &pci_dev->io_regions[region_num];
1164     r->addr = PCI_BAR_UNMAPPED;
1165     r->size = size;
1166     r->type = type;
1167     r->memory = memory;
1168     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1169                         ? pci_get_bus(pci_dev)->address_space_io
1170                         : pci_get_bus(pci_dev)->address_space_mem;
1171 
1172     wmask = ~(size - 1);
1173     if (region_num == PCI_ROM_SLOT) {
1174         /* ROM enable bit is writable */
1175         wmask |= PCI_ROM_ADDRESS_ENABLE;
1176     }
1177 
1178     addr = pci_bar(pci_dev, region_num);
1179     pci_set_long(pci_dev->config + addr, type);
1180 
1181     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1182         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1183         pci_set_quad(pci_dev->wmask + addr, wmask);
1184         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1185     } else {
1186         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1187         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1188     }
1189 }
1190 
1191 static void pci_update_vga(PCIDevice *pci_dev)
1192 {
1193     uint16_t cmd;
1194 
1195     if (!pci_dev->has_vga) {
1196         return;
1197     }
1198 
1199     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1200 
1201     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1202                               cmd & PCI_COMMAND_MEMORY);
1203     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1204                               cmd & PCI_COMMAND_IO);
1205     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1206                               cmd & PCI_COMMAND_IO);
1207 }
1208 
1209 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1210                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1211 {
1212     PCIBus *bus = pci_get_bus(pci_dev);
1213 
1214     assert(!pci_dev->has_vga);
1215 
1216     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1217     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1218     memory_region_add_subregion_overlap(bus->address_space_mem,
1219                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1220 
1221     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1222     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1223     memory_region_add_subregion_overlap(bus->address_space_io,
1224                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1225 
1226     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1227     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1228     memory_region_add_subregion_overlap(bus->address_space_io,
1229                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1230     pci_dev->has_vga = true;
1231 
1232     pci_update_vga(pci_dev);
1233 }
1234 
1235 void pci_unregister_vga(PCIDevice *pci_dev)
1236 {
1237     PCIBus *bus = pci_get_bus(pci_dev);
1238 
1239     if (!pci_dev->has_vga) {
1240         return;
1241     }
1242 
1243     memory_region_del_subregion(bus->address_space_mem,
1244                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1245     memory_region_del_subregion(bus->address_space_io,
1246                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1247     memory_region_del_subregion(bus->address_space_io,
1248                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1249     pci_dev->has_vga = false;
1250 }
1251 
1252 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1253 {
1254     return pci_dev->io_regions[region_num].addr;
1255 }
1256 
1257 static pcibus_t pci_bar_address(PCIDevice *d,
1258                                 int reg, uint8_t type, pcibus_t size)
1259 {
1260     pcibus_t new_addr, last_addr;
1261     int bar = pci_bar(d, reg);
1262     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1263     Object *machine = qdev_get_machine();
1264     ObjectClass *oc = object_get_class(machine);
1265     MachineClass *mc = MACHINE_CLASS(oc);
1266     bool allow_0_address = mc->pci_allow_0_address;
1267 
1268     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1269         if (!(cmd & PCI_COMMAND_IO)) {
1270             return PCI_BAR_UNMAPPED;
1271         }
1272         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1273         last_addr = new_addr + size - 1;
1274         /* Check if 32 bit BAR wraps around explicitly.
1275          * TODO: make priorities correct and remove this work around.
1276          */
1277         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1278             (!allow_0_address && new_addr == 0)) {
1279             return PCI_BAR_UNMAPPED;
1280         }
1281         return new_addr;
1282     }
1283 
1284     if (!(cmd & PCI_COMMAND_MEMORY)) {
1285         return PCI_BAR_UNMAPPED;
1286     }
1287     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1288         new_addr = pci_get_quad(d->config + bar);
1289     } else {
1290         new_addr = pci_get_long(d->config + bar);
1291     }
1292     /* the ROM slot has a specific enable bit */
1293     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1294         return PCI_BAR_UNMAPPED;
1295     }
1296     new_addr &= ~(size - 1);
1297     last_addr = new_addr + size - 1;
1298     /* NOTE: we do not support wrapping */
1299     /* XXX: as we cannot support really dynamic
1300        mappings, we handle specific values as invalid
1301        mappings. */
1302     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1303         (!allow_0_address && new_addr == 0)) {
1304         return PCI_BAR_UNMAPPED;
1305     }
1306 
1307     /* Now pcibus_t is 64bit.
1308      * Check if 32 bit BAR wraps around explicitly.
1309      * Without this, PC ide doesn't work well.
1310      * TODO: remove this work around.
1311      */
1312     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1313         return PCI_BAR_UNMAPPED;
1314     }
1315 
1316     /*
1317      * OS is allowed to set BAR beyond its addressable
1318      * bits. For example, 32 bit OS can set 64bit bar
1319      * to >4G. Check it. TODO: we might need to support
1320      * it in the future for e.g. PAE.
1321      */
1322     if (last_addr >= HWADDR_MAX) {
1323         return PCI_BAR_UNMAPPED;
1324     }
1325 
1326     return new_addr;
1327 }
1328 
1329 static void pci_update_mappings(PCIDevice *d)
1330 {
1331     PCIIORegion *r;
1332     int i;
1333     pcibus_t new_addr;
1334 
1335     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1336         r = &d->io_regions[i];
1337 
1338         /* this region isn't registered */
1339         if (!r->size)
1340             continue;
1341 
1342         new_addr = pci_bar_address(d, i, r->type, r->size);
1343 
1344         /* This bar isn't changed */
1345         if (new_addr == r->addr)
1346             continue;
1347 
1348         /* now do the real mapping */
1349         if (r->addr != PCI_BAR_UNMAPPED) {
1350             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1351                                           PCI_SLOT(d->devfn),
1352                                           PCI_FUNC(d->devfn),
1353                                           i, r->addr, r->size);
1354             memory_region_del_subregion(r->address_space, r->memory);
1355         }
1356         r->addr = new_addr;
1357         if (r->addr != PCI_BAR_UNMAPPED) {
1358             trace_pci_update_mappings_add(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_add_subregion_overlap(r->address_space,
1363                                                 r->addr, r->memory, 1);
1364         }
1365     }
1366 
1367     pci_update_vga(d);
1368 }
1369 
1370 static inline int pci_irq_disabled(PCIDevice *d)
1371 {
1372     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1373 }
1374 
1375 /* Called after interrupt disabled field update in config space,
1376  * assert/deassert interrupts if necessary.
1377  * Gets original interrupt disable bit value (before update). */
1378 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1379 {
1380     int i, disabled = pci_irq_disabled(d);
1381     if (disabled == was_irq_disabled)
1382         return;
1383     for (i = 0; i < PCI_NUM_PINS; ++i) {
1384         int state = pci_irq_state(d, i);
1385         pci_change_irq_level(d, i, disabled ? -state : state);
1386     }
1387 }
1388 
1389 uint32_t pci_default_read_config(PCIDevice *d,
1390                                  uint32_t address, int len)
1391 {
1392     uint32_t val = 0;
1393 
1394     assert(address + len <= pci_config_size(d));
1395 
1396     if (pci_is_express_downstream_port(d) &&
1397         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1398         pcie_sync_bridge_lnk(d);
1399     }
1400     memcpy(&val, d->config + address, len);
1401     return le32_to_cpu(val);
1402 }
1403 
1404 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1405 {
1406     int i, was_irq_disabled = pci_irq_disabled(d);
1407     uint32_t val = val_in;
1408 
1409     assert(addr + l <= pci_config_size(d));
1410 
1411     for (i = 0; i < l; val >>= 8, ++i) {
1412         uint8_t wmask = d->wmask[addr + i];
1413         uint8_t w1cmask = d->w1cmask[addr + i];
1414         assert(!(wmask & w1cmask));
1415         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1416         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1417     }
1418     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1419         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1420         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1421         range_covers_byte(addr, l, PCI_COMMAND))
1422         pci_update_mappings(d);
1423 
1424     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1425         pci_update_irq_disabled(d, was_irq_disabled);
1426         memory_region_set_enabled(&d->bus_master_enable_region,
1427                                   pci_get_word(d->config + PCI_COMMAND)
1428                                     & PCI_COMMAND_MASTER);
1429     }
1430 
1431     msi_write_config(d, addr, val_in, l);
1432     msix_write_config(d, addr, val_in, l);
1433 }
1434 
1435 /***********************************************************/
1436 /* generic PCI irq support */
1437 
1438 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1439 static void pci_irq_handler(void *opaque, int irq_num, int level)
1440 {
1441     PCIDevice *pci_dev = opaque;
1442     int change;
1443 
1444     change = level - pci_irq_state(pci_dev, irq_num);
1445     if (!change)
1446         return;
1447 
1448     pci_set_irq_state(pci_dev, irq_num, level);
1449     pci_update_irq_status(pci_dev);
1450     if (pci_irq_disabled(pci_dev))
1451         return;
1452     pci_change_irq_level(pci_dev, irq_num, change);
1453 }
1454 
1455 static inline int pci_intx(PCIDevice *pci_dev)
1456 {
1457     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1458 }
1459 
1460 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1461 {
1462     int intx = pci_intx(pci_dev);
1463 
1464     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1465 }
1466 
1467 void pci_set_irq(PCIDevice *pci_dev, int level)
1468 {
1469     int intx = pci_intx(pci_dev);
1470     pci_irq_handler(pci_dev, intx, level);
1471 }
1472 
1473 /* Special hooks used by device assignment */
1474 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1475 {
1476     assert(pci_bus_is_root(bus));
1477     bus->route_intx_to_irq = route_intx_to_irq;
1478 }
1479 
1480 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1481 {
1482     PCIBus *bus;
1483 
1484     do {
1485         bus = pci_get_bus(dev);
1486         pin = bus->map_irq(dev, pin);
1487         dev = bus->parent_dev;
1488     } while (dev);
1489 
1490     if (!bus->route_intx_to_irq) {
1491         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1492                      object_get_typename(OBJECT(bus->qbus.parent)));
1493         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1494     }
1495 
1496     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1497 }
1498 
1499 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1500 {
1501     return old->mode != new->mode || old->irq != new->irq;
1502 }
1503 
1504 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1505 {
1506     PCIDevice *dev;
1507     PCIBus *sec;
1508     int i;
1509 
1510     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1511         dev = bus->devices[i];
1512         if (dev && dev->intx_routing_notifier) {
1513             dev->intx_routing_notifier(dev);
1514         }
1515     }
1516 
1517     QLIST_FOREACH(sec, &bus->child, sibling) {
1518         pci_bus_fire_intx_routing_notifier(sec);
1519     }
1520 }
1521 
1522 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1523                                           PCIINTxRoutingNotifier notifier)
1524 {
1525     dev->intx_routing_notifier = notifier;
1526 }
1527 
1528 /*
1529  * PCI-to-PCI bridge specification
1530  * 9.1: Interrupt routing. Table 9-1
1531  *
1532  * the PCI Express Base Specification, Revision 2.1
1533  * 2.2.8.1: INTx interrutp signaling - Rules
1534  *          the Implementation Note
1535  *          Table 2-20
1536  */
1537 /*
1538  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1539  * 0-origin unlike PCI interrupt pin register.
1540  */
1541 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1542 {
1543     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1544 }
1545 
1546 /***********************************************************/
1547 /* monitor info on PCI */
1548 
1549 typedef struct {
1550     uint16_t class;
1551     const char *desc;
1552     const char *fw_name;
1553     uint16_t fw_ign_bits;
1554 } pci_class_desc;
1555 
1556 static const pci_class_desc pci_class_descriptions[] =
1557 {
1558     { 0x0001, "VGA controller", "display"},
1559     { 0x0100, "SCSI controller", "scsi"},
1560     { 0x0101, "IDE controller", "ide"},
1561     { 0x0102, "Floppy controller", "fdc"},
1562     { 0x0103, "IPI controller", "ipi"},
1563     { 0x0104, "RAID controller", "raid"},
1564     { 0x0106, "SATA controller"},
1565     { 0x0107, "SAS controller"},
1566     { 0x0180, "Storage controller"},
1567     { 0x0200, "Ethernet controller", "ethernet"},
1568     { 0x0201, "Token Ring controller", "token-ring"},
1569     { 0x0202, "FDDI controller", "fddi"},
1570     { 0x0203, "ATM controller", "atm"},
1571     { 0x0280, "Network controller"},
1572     { 0x0300, "VGA controller", "display", 0x00ff},
1573     { 0x0301, "XGA controller"},
1574     { 0x0302, "3D controller"},
1575     { 0x0380, "Display controller"},
1576     { 0x0400, "Video controller", "video"},
1577     { 0x0401, "Audio controller", "sound"},
1578     { 0x0402, "Phone"},
1579     { 0x0403, "Audio controller", "sound"},
1580     { 0x0480, "Multimedia controller"},
1581     { 0x0500, "RAM controller", "memory"},
1582     { 0x0501, "Flash controller", "flash"},
1583     { 0x0580, "Memory controller"},
1584     { 0x0600, "Host bridge", "host"},
1585     { 0x0601, "ISA bridge", "isa"},
1586     { 0x0602, "EISA bridge", "eisa"},
1587     { 0x0603, "MC bridge", "mca"},
1588     { 0x0604, "PCI bridge", "pci-bridge"},
1589     { 0x0605, "PCMCIA bridge", "pcmcia"},
1590     { 0x0606, "NUBUS bridge", "nubus"},
1591     { 0x0607, "CARDBUS bridge", "cardbus"},
1592     { 0x0608, "RACEWAY bridge"},
1593     { 0x0680, "Bridge"},
1594     { 0x0700, "Serial port", "serial"},
1595     { 0x0701, "Parallel port", "parallel"},
1596     { 0x0800, "Interrupt controller", "interrupt-controller"},
1597     { 0x0801, "DMA controller", "dma-controller"},
1598     { 0x0802, "Timer", "timer"},
1599     { 0x0803, "RTC", "rtc"},
1600     { 0x0900, "Keyboard", "keyboard"},
1601     { 0x0901, "Pen", "pen"},
1602     { 0x0902, "Mouse", "mouse"},
1603     { 0x0A00, "Dock station", "dock", 0x00ff},
1604     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1605     { 0x0c00, "Fireware contorller", "fireware"},
1606     { 0x0c01, "Access bus controller", "access-bus"},
1607     { 0x0c02, "SSA controller", "ssa"},
1608     { 0x0c03, "USB controller", "usb"},
1609     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1610     { 0x0c05, "SMBus"},
1611     { 0, NULL}
1612 };
1613 
1614 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1615                                                   void (*fn)(PCIBus *b,
1616                                                              PCIDevice *d,
1617                                                              void *opaque),
1618                                                   void *opaque)
1619 {
1620     PCIDevice *d;
1621     int devfn;
1622 
1623     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1624         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1625         if (d) {
1626             fn(bus, d, opaque);
1627         }
1628     }
1629 }
1630 
1631 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1632                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1633                          void *opaque)
1634 {
1635     bus = pci_find_bus_nr(bus, bus_num);
1636 
1637     if (bus) {
1638         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1639     }
1640 }
1641 
1642 static void pci_for_each_device_under_bus(PCIBus *bus,
1643                                           void (*fn)(PCIBus *b, PCIDevice *d,
1644                                                      void *opaque),
1645                                           void *opaque)
1646 {
1647     PCIDevice *d;
1648     int devfn;
1649 
1650     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1651         d = bus->devices[devfn];
1652         if (d) {
1653             fn(bus, d, opaque);
1654         }
1655     }
1656 }
1657 
1658 void pci_for_each_device(PCIBus *bus, int bus_num,
1659                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1660                          void *opaque)
1661 {
1662     bus = pci_find_bus_nr(bus, bus_num);
1663 
1664     if (bus) {
1665         pci_for_each_device_under_bus(bus, fn, opaque);
1666     }
1667 }
1668 
1669 static const pci_class_desc *get_class_desc(int class)
1670 {
1671     const pci_class_desc *desc;
1672 
1673     desc = pci_class_descriptions;
1674     while (desc->desc && class != desc->class) {
1675         desc++;
1676     }
1677 
1678     return desc;
1679 }
1680 
1681 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1682 
1683 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1684 {
1685     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1686     int i;
1687 
1688     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1689         const PCIIORegion *r = &dev->io_regions[i];
1690         PciMemoryRegionList *region;
1691 
1692         if (!r->size) {
1693             continue;
1694         }
1695 
1696         region = g_malloc0(sizeof(*region));
1697         region->value = g_malloc0(sizeof(*region->value));
1698 
1699         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1700             region->value->type = g_strdup("io");
1701         } else {
1702             region->value->type = g_strdup("memory");
1703             region->value->has_prefetch = true;
1704             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1705             region->value->has_mem_type_64 = true;
1706             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1707         }
1708 
1709         region->value->bar = i;
1710         region->value->address = r->addr;
1711         region->value->size = r->size;
1712 
1713         /* XXX: waiting for the qapi to support GSList */
1714         if (!cur_item) {
1715             head = cur_item = region;
1716         } else {
1717             cur_item->next = region;
1718             cur_item = region;
1719         }
1720     }
1721 
1722     return head;
1723 }
1724 
1725 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1726                                            int bus_num)
1727 {
1728     PciBridgeInfo *info;
1729     PciMemoryRange *range;
1730 
1731     info = g_new0(PciBridgeInfo, 1);
1732 
1733     info->bus = g_new0(PciBusInfo, 1);
1734     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1735     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1736     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1737 
1738     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1739     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1740     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1741 
1742     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1743     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1744     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1745 
1746     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1747     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1748     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1749 
1750     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1751         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1752         if (child_bus) {
1753             info->has_devices = true;
1754             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1755         }
1756     }
1757 
1758     return info;
1759 }
1760 
1761 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1762                                            int bus_num)
1763 {
1764     const pci_class_desc *desc;
1765     PciDeviceInfo *info;
1766     uint8_t type;
1767     int class;
1768 
1769     info = g_new0(PciDeviceInfo, 1);
1770     info->bus = bus_num;
1771     info->slot = PCI_SLOT(dev->devfn);
1772     info->function = PCI_FUNC(dev->devfn);
1773 
1774     info->class_info = g_new0(PciDeviceClass, 1);
1775     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1776     info->class_info->q_class = class;
1777     desc = get_class_desc(class);
1778     if (desc->desc) {
1779         info->class_info->has_desc = true;
1780         info->class_info->desc = g_strdup(desc->desc);
1781     }
1782 
1783     info->id = g_new0(PciDeviceId, 1);
1784     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1785     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1786     info->regions = qmp_query_pci_regions(dev);
1787     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1788 
1789     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1790     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1791         info->has_irq = true;
1792         info->irq = dev->config[PCI_INTERRUPT_LINE];
1793     }
1794 
1795     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1796     if (type == PCI_HEADER_TYPE_BRIDGE) {
1797         info->has_pci_bridge = true;
1798         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1799     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1800         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1801         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1802         info->id->subsystem_vendor =
1803             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1804     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1805         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1806         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1807         info->id->subsystem_vendor =
1808             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1809     }
1810 
1811     return info;
1812 }
1813 
1814 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1815 {
1816     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1817     PCIDevice *dev;
1818     int devfn;
1819 
1820     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1821         dev = bus->devices[devfn];
1822         if (dev) {
1823             info = g_malloc0(sizeof(*info));
1824             info->value = qmp_query_pci_device(dev, bus, bus_num);
1825 
1826             /* XXX: waiting for the qapi to support GSList */
1827             if (!cur_item) {
1828                 head = cur_item = info;
1829             } else {
1830                 cur_item->next = info;
1831                 cur_item = info;
1832             }
1833         }
1834     }
1835 
1836     return head;
1837 }
1838 
1839 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1840 {
1841     PciInfo *info = NULL;
1842 
1843     bus = pci_find_bus_nr(bus, bus_num);
1844     if (bus) {
1845         info = g_malloc0(sizeof(*info));
1846         info->bus = bus_num;
1847         info->devices = qmp_query_pci_devices(bus, bus_num);
1848     }
1849 
1850     return info;
1851 }
1852 
1853 PciInfoList *qmp_query_pci(Error **errp)
1854 {
1855     PciInfoList *info, *head = NULL, *cur_item = NULL;
1856     PCIHostState *host_bridge;
1857 
1858     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1859         info = g_malloc0(sizeof(*info));
1860         info->value = qmp_query_pci_bus(host_bridge->bus,
1861                                         pci_bus_num(host_bridge->bus));
1862 
1863         /* XXX: waiting for the qapi to support GSList */
1864         if (!cur_item) {
1865             head = cur_item = info;
1866         } else {
1867             cur_item->next = info;
1868             cur_item = info;
1869         }
1870     }
1871 
1872     return head;
1873 }
1874 
1875 /* Initialize a PCI NIC.  */
1876 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1877                                const char *default_model,
1878                                const char *default_devaddr)
1879 {
1880     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1881     GSList *list;
1882     GPtrArray *pci_nic_models;
1883     PCIBus *bus;
1884     PCIDevice *pci_dev;
1885     DeviceState *dev;
1886     int devfn;
1887     int i;
1888     int dom, busnr;
1889     unsigned slot;
1890 
1891     if (nd->model && !strcmp(nd->model, "virtio")) {
1892         g_free(nd->model);
1893         nd->model = g_strdup("virtio-net-pci");
1894     }
1895 
1896     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1897     pci_nic_models = g_ptr_array_new();
1898     while (list) {
1899         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1900                                              TYPE_DEVICE);
1901         GSList *next;
1902         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1903             dc->user_creatable) {
1904             const char *name = object_class_get_name(list->data);
1905             /*
1906              * A network device might also be something else than a NIC, see
1907              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1908              * property, too. Unfortunately, some devices like virtio-net only
1909              * create this property during instance_init, so we have to create
1910              * a temporary instance here to be able to check it.
1911              */
1912             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1913             if (object_property_find(obj, "netdev")) {
1914                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1915             }
1916             object_unref(obj);
1917         }
1918         next = list->next;
1919         g_slist_free_1(list);
1920         list = next;
1921     }
1922     g_ptr_array_add(pci_nic_models, NULL);
1923 
1924     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1925         exit(0);
1926     }
1927 
1928     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1929                             default_model);
1930     if (i < 0) {
1931         exit(1);
1932     }
1933 
1934     if (!rootbus) {
1935         error_report("No primary PCI bus");
1936         exit(1);
1937     }
1938 
1939     assert(!rootbus->parent_dev);
1940 
1941     if (!devaddr) {
1942         devfn = -1;
1943         busnr = 0;
1944     } else {
1945         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1946             error_report("Invalid PCI device address %s for device %s",
1947                          devaddr, nd->model);
1948             exit(1);
1949         }
1950 
1951         if (dom != 0) {
1952             error_report("No support for non-zero PCI domains");
1953             exit(1);
1954         }
1955 
1956         devfn = PCI_DEVFN(slot, 0);
1957     }
1958 
1959     bus = pci_find_bus_nr(rootbus, busnr);
1960     if (!bus) {
1961         error_report("Invalid PCI device address %s for device %s",
1962                      devaddr, nd->model);
1963         exit(1);
1964     }
1965 
1966     pci_dev = pci_new(devfn, nd->model);
1967     dev = &pci_dev->qdev;
1968     qdev_set_nic_properties(dev, nd);
1969     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1970     g_ptr_array_free(pci_nic_models, true);
1971     return pci_dev;
1972 }
1973 
1974 PCIDevice *pci_vga_init(PCIBus *bus)
1975 {
1976     switch (vga_interface_type) {
1977     case VGA_CIRRUS:
1978         return pci_create_simple(bus, -1, "cirrus-vga");
1979     case VGA_QXL:
1980         return pci_create_simple(bus, -1, "qxl-vga");
1981     case VGA_STD:
1982         return pci_create_simple(bus, -1, "VGA");
1983     case VGA_VMWARE:
1984         return pci_create_simple(bus, -1, "vmware-svga");
1985     case VGA_VIRTIO:
1986         return pci_create_simple(bus, -1, "virtio-vga");
1987     case VGA_NONE:
1988     default: /* Other non-PCI types. Checking for unsupported types is already
1989                 done in vl.c. */
1990         return NULL;
1991     }
1992 }
1993 
1994 /* Whether a given bus number is in range of the secondary
1995  * bus of the given bridge device. */
1996 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1997 {
1998     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1999              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
2000         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
2001         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
2002 }
2003 
2004 /* Whether a given bus number is in a range of a root bus */
2005 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
2006 {
2007     int i;
2008 
2009     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
2010         PCIDevice *dev = bus->devices[i];
2011 
2012         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
2013             if (pci_secondary_bus_in_range(dev, bus_num)) {
2014                 return true;
2015             }
2016         }
2017     }
2018 
2019     return false;
2020 }
2021 
2022 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2023 {
2024     PCIBus *sec;
2025 
2026     if (!bus) {
2027         return NULL;
2028     }
2029 
2030     if (pci_bus_num(bus) == bus_num) {
2031         return bus;
2032     }
2033 
2034     /* Consider all bus numbers in range for the host pci bridge. */
2035     if (!pci_bus_is_root(bus) &&
2036         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2037         return NULL;
2038     }
2039 
2040     /* try child bus */
2041     for (; bus; bus = sec) {
2042         QLIST_FOREACH(sec, &bus->child, sibling) {
2043             if (pci_bus_num(sec) == bus_num) {
2044                 return sec;
2045             }
2046             /* PXB buses assumed to be children of bus 0 */
2047             if (pci_bus_is_root(sec)) {
2048                 if (pci_root_bus_in_range(sec, bus_num)) {
2049                     break;
2050                 }
2051             } else {
2052                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2053                     break;
2054                 }
2055             }
2056         }
2057     }
2058 
2059     return NULL;
2060 }
2061 
2062 void pci_for_each_bus_depth_first(PCIBus *bus,
2063                                   void *(*begin)(PCIBus *bus, void *parent_state),
2064                                   void (*end)(PCIBus *bus, void *state),
2065                                   void *parent_state)
2066 {
2067     PCIBus *sec;
2068     void *state;
2069 
2070     if (!bus) {
2071         return;
2072     }
2073 
2074     if (begin) {
2075         state = begin(bus, parent_state);
2076     } else {
2077         state = parent_state;
2078     }
2079 
2080     QLIST_FOREACH(sec, &bus->child, sibling) {
2081         pci_for_each_bus_depth_first(sec, begin, end, state);
2082     }
2083 
2084     if (end) {
2085         end(bus, state);
2086     }
2087 }
2088 
2089 
2090 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2091 {
2092     bus = pci_find_bus_nr(bus, bus_num);
2093 
2094     if (!bus)
2095         return NULL;
2096 
2097     return bus->devices[devfn];
2098 }
2099 
2100 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2101 {
2102     PCIDevice *pci_dev = (PCIDevice *)qdev;
2103     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2104     ObjectClass *klass = OBJECT_CLASS(pc);
2105     Error *local_err = NULL;
2106     bool is_default_rom;
2107     uint16_t class_id;
2108 
2109     /* initialize cap_present for pci_is_express() and pci_config_size(),
2110      * Note that hybrid PCIs are not set automatically and need to manage
2111      * QEMU_PCI_CAP_EXPRESS manually */
2112     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2113        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2114         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2115     }
2116 
2117     pci_dev = do_pci_register_device(pci_dev,
2118                                      object_get_typename(OBJECT(qdev)),
2119                                      pci_dev->devfn, errp);
2120     if (pci_dev == NULL)
2121         return;
2122 
2123     if (pc->realize) {
2124         pc->realize(pci_dev, &local_err);
2125         if (local_err) {
2126             error_propagate(errp, local_err);
2127             do_pci_unregister_device(pci_dev);
2128             return;
2129         }
2130     }
2131 
2132     if (pci_dev->failover_pair_id) {
2133         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2134             error_setg(errp, "failover primary device must be on "
2135                              "PCIExpress bus");
2136             pci_qdev_unrealize(DEVICE(pci_dev));
2137             return;
2138         }
2139         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2140         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2141             error_setg(errp, "failover primary device is not an "
2142                              "Ethernet device");
2143             pci_qdev_unrealize(DEVICE(pci_dev));
2144             return;
2145         }
2146         if (!(pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2147             && (PCI_FUNC(pci_dev->devfn) == 0)) {
2148             qdev->allow_unplug_during_migration = true;
2149         } else {
2150             error_setg(errp, "failover: primary device must be in its own "
2151                               "PCI slot");
2152             pci_qdev_unrealize(DEVICE(pci_dev));
2153             return;
2154         }
2155         qdev->allow_unplug_during_migration = true;
2156     }
2157 
2158     /* rom loading */
2159     is_default_rom = false;
2160     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2161         pci_dev->romfile = g_strdup(pc->romfile);
2162         is_default_rom = true;
2163     }
2164 
2165     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2166     if (local_err) {
2167         error_propagate(errp, local_err);
2168         pci_qdev_unrealize(DEVICE(pci_dev));
2169         return;
2170     }
2171 }
2172 
2173 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2174                                  const char *name)
2175 {
2176     DeviceState *dev;
2177 
2178     dev = qdev_new(name);
2179     qdev_prop_set_int32(dev, "addr", devfn);
2180     qdev_prop_set_bit(dev, "multifunction", multifunction);
2181     return PCI_DEVICE(dev);
2182 }
2183 
2184 PCIDevice *pci_new(int devfn, const char *name)
2185 {
2186     return pci_new_multifunction(devfn, false, name);
2187 }
2188 
2189 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2190 {
2191     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2192 }
2193 
2194 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2195                                            bool multifunction,
2196                                            const char *name)
2197 {
2198     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2199     pci_realize_and_unref(dev, bus, &error_fatal);
2200     return dev;
2201 }
2202 
2203 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2204 {
2205     return pci_create_simple_multifunction(bus, devfn, false, name);
2206 }
2207 
2208 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2209 {
2210     int offset = PCI_CONFIG_HEADER_SIZE;
2211     int i;
2212     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2213         if (pdev->used[i])
2214             offset = i + 1;
2215         else if (i - offset + 1 == size)
2216             return offset;
2217     }
2218     return 0;
2219 }
2220 
2221 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2222                                         uint8_t *prev_p)
2223 {
2224     uint8_t next, prev;
2225 
2226     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2227         return 0;
2228 
2229     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2230          prev = next + PCI_CAP_LIST_NEXT)
2231         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2232             break;
2233 
2234     if (prev_p)
2235         *prev_p = prev;
2236     return next;
2237 }
2238 
2239 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2240 {
2241     uint8_t next, prev, found = 0;
2242 
2243     if (!(pdev->used[offset])) {
2244         return 0;
2245     }
2246 
2247     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2248 
2249     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2250          prev = next + PCI_CAP_LIST_NEXT) {
2251         if (next <= offset && next > found) {
2252             found = next;
2253         }
2254     }
2255     return found;
2256 }
2257 
2258 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2259    This is needed for an option rom which is used for more than one device. */
2260 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2261 {
2262     uint16_t vendor_id;
2263     uint16_t device_id;
2264     uint16_t rom_vendor_id;
2265     uint16_t rom_device_id;
2266     uint16_t rom_magic;
2267     uint16_t pcir_offset;
2268     uint8_t checksum;
2269 
2270     /* Words in rom data are little endian (like in PCI configuration),
2271        so they can be read / written with pci_get_word / pci_set_word. */
2272 
2273     /* Only a valid rom will be patched. */
2274     rom_magic = pci_get_word(ptr);
2275     if (rom_magic != 0xaa55) {
2276         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2277         return;
2278     }
2279     pcir_offset = pci_get_word(ptr + 0x18);
2280     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2281         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2282         return;
2283     }
2284 
2285     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2286     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2287     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2288     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2289 
2290     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2291                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2292 
2293     checksum = ptr[6];
2294 
2295     if (vendor_id != rom_vendor_id) {
2296         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2297         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2298         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2299         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2300         ptr[6] = checksum;
2301         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2302     }
2303 
2304     if (device_id != rom_device_id) {
2305         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2306         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2307         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2308         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2309         ptr[6] = checksum;
2310         pci_set_word(ptr + pcir_offset + 6, device_id);
2311     }
2312 }
2313 
2314 /* Add an option rom for the device */
2315 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2316                                Error **errp)
2317 {
2318     int size;
2319     char *path;
2320     void *ptr;
2321     char name[32];
2322     const VMStateDescription *vmsd;
2323 
2324     if (!pdev->romfile)
2325         return;
2326     if (strlen(pdev->romfile) == 0)
2327         return;
2328 
2329     if (!pdev->rom_bar) {
2330         /*
2331          * Load rom via fw_cfg instead of creating a rom bar,
2332          * for 0.11 compatibility.
2333          */
2334         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2335 
2336         /*
2337          * Hot-plugged devices can't use the option ROM
2338          * if the rom bar is disabled.
2339          */
2340         if (DEVICE(pdev)->hotplugged) {
2341             error_setg(errp, "Hot-plugged device without ROM bar"
2342                        " can't have an option ROM");
2343             return;
2344         }
2345 
2346         if (class == 0x0300) {
2347             rom_add_vga(pdev->romfile);
2348         } else {
2349             rom_add_option(pdev->romfile, -1);
2350         }
2351         return;
2352     }
2353 
2354     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2355     if (path == NULL) {
2356         path = g_strdup(pdev->romfile);
2357     }
2358 
2359     size = get_image_size(path);
2360     if (size < 0) {
2361         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2362         g_free(path);
2363         return;
2364     } else if (size == 0) {
2365         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2366         g_free(path);
2367         return;
2368     }
2369     size = pow2ceil(size);
2370 
2371     vmsd = qdev_get_vmsd(DEVICE(pdev));
2372 
2373     if (vmsd) {
2374         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2375     } else {
2376         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2377     }
2378     pdev->has_rom = true;
2379     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2380     ptr = memory_region_get_ram_ptr(&pdev->rom);
2381     if (load_image_size(path, ptr, size) < 0) {
2382         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2383         g_free(path);
2384         return;
2385     }
2386     g_free(path);
2387 
2388     if (is_default_rom) {
2389         /* Only the default rom images will be patched (if needed). */
2390         pci_patch_ids(pdev, ptr, size);
2391     }
2392 
2393     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2394 }
2395 
2396 static void pci_del_option_rom(PCIDevice *pdev)
2397 {
2398     if (!pdev->has_rom)
2399         return;
2400 
2401     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2402     pdev->has_rom = false;
2403 }
2404 
2405 /*
2406  * On success, pci_add_capability() returns a positive value
2407  * that the offset of the pci capability.
2408  * On failure, it sets an error and returns a negative error
2409  * code.
2410  */
2411 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2412                        uint8_t offset, uint8_t size,
2413                        Error **errp)
2414 {
2415     uint8_t *config;
2416     int i, overlapping_cap;
2417 
2418     if (!offset) {
2419         offset = pci_find_space(pdev, size);
2420         /* out of PCI config space is programming error */
2421         assert(offset);
2422     } else {
2423         /* Verify that capabilities don't overlap.  Note: device assignment
2424          * depends on this check to verify that the device is not broken.
2425          * Should never trigger for emulated devices, but it's helpful
2426          * for debugging these. */
2427         for (i = offset; i < offset + size; i++) {
2428             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2429             if (overlapping_cap) {
2430                 error_setg(errp, "%s:%02x:%02x.%x "
2431                            "Attempt to add PCI capability %x at offset "
2432                            "%x overlaps existing capability %x at offset %x",
2433                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2434                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2435                            cap_id, offset, overlapping_cap, i);
2436                 return -EINVAL;
2437             }
2438         }
2439     }
2440 
2441     config = pdev->config + offset;
2442     config[PCI_CAP_LIST_ID] = cap_id;
2443     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2444     pdev->config[PCI_CAPABILITY_LIST] = offset;
2445     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2446     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2447     /* Make capability read-only by default */
2448     memset(pdev->wmask + offset, 0, size);
2449     /* Check capability by default */
2450     memset(pdev->cmask + offset, 0xFF, size);
2451     return offset;
2452 }
2453 
2454 /* Unlink capability from the pci config space. */
2455 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2456 {
2457     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2458     if (!offset)
2459         return;
2460     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2461     /* Make capability writable again */
2462     memset(pdev->wmask + offset, 0xff, size);
2463     memset(pdev->w1cmask + offset, 0, size);
2464     /* Clear cmask as device-specific registers can't be checked */
2465     memset(pdev->cmask + offset, 0, size);
2466     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2467 
2468     if (!pdev->config[PCI_CAPABILITY_LIST])
2469         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2470 }
2471 
2472 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2473 {
2474     return pci_find_capability_list(pdev, cap_id, NULL);
2475 }
2476 
2477 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2478 {
2479     PCIDevice *d = (PCIDevice *)dev;
2480     const pci_class_desc *desc;
2481     char ctxt[64];
2482     PCIIORegion *r;
2483     int i, class;
2484 
2485     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2486     desc = pci_class_descriptions;
2487     while (desc->desc && class != desc->class)
2488         desc++;
2489     if (desc->desc) {
2490         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2491     } else {
2492         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2493     }
2494 
2495     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2496                    "pci id %04x:%04x (sub %04x:%04x)\n",
2497                    indent, "", ctxt, pci_dev_bus_num(d),
2498                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2499                    pci_get_word(d->config + PCI_VENDOR_ID),
2500                    pci_get_word(d->config + PCI_DEVICE_ID),
2501                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2502                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2503     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2504         r = &d->io_regions[i];
2505         if (!r->size)
2506             continue;
2507         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2508                        " [0x%"FMT_PCIBUS"]\n",
2509                        indent, "",
2510                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2511                        r->addr, r->addr + r->size - 1);
2512     }
2513 }
2514 
2515 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2516 {
2517     PCIDevice *d = (PCIDevice *)dev;
2518     const char *name = NULL;
2519     const pci_class_desc *desc =  pci_class_descriptions;
2520     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2521 
2522     while (desc->desc &&
2523           (class & ~desc->fw_ign_bits) !=
2524           (desc->class & ~desc->fw_ign_bits)) {
2525         desc++;
2526     }
2527 
2528     if (desc->desc) {
2529         name = desc->fw_name;
2530     }
2531 
2532     if (name) {
2533         pstrcpy(buf, len, name);
2534     } else {
2535         snprintf(buf, len, "pci%04x,%04x",
2536                  pci_get_word(d->config + PCI_VENDOR_ID),
2537                  pci_get_word(d->config + PCI_DEVICE_ID));
2538     }
2539 
2540     return buf;
2541 }
2542 
2543 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2544 {
2545     PCIDevice *d = (PCIDevice *)dev;
2546     char path[50], name[33];
2547     int off;
2548 
2549     off = snprintf(path, sizeof(path), "%s@%x",
2550                    pci_dev_fw_name(dev, name, sizeof name),
2551                    PCI_SLOT(d->devfn));
2552     if (PCI_FUNC(d->devfn))
2553         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2554     return g_strdup(path);
2555 }
2556 
2557 static char *pcibus_get_dev_path(DeviceState *dev)
2558 {
2559     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2560     PCIDevice *t;
2561     int slot_depth;
2562     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2563      * 00 is added here to make this format compatible with
2564      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2565      * Slot.Function list specifies the slot and function numbers for all
2566      * devices on the path from root to the specific device. */
2567     const char *root_bus_path;
2568     int root_bus_len;
2569     char slot[] = ":SS.F";
2570     int slot_len = sizeof slot - 1 /* For '\0' */;
2571     int path_len;
2572     char *path, *p;
2573     int s;
2574 
2575     root_bus_path = pci_root_bus_path(d);
2576     root_bus_len = strlen(root_bus_path);
2577 
2578     /* Calculate # of slots on path between device and root. */;
2579     slot_depth = 0;
2580     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2581         ++slot_depth;
2582     }
2583 
2584     path_len = root_bus_len + slot_len * slot_depth;
2585 
2586     /* Allocate memory, fill in the terminating null byte. */
2587     path = g_malloc(path_len + 1 /* For '\0' */);
2588     path[path_len] = '\0';
2589 
2590     memcpy(path, root_bus_path, root_bus_len);
2591 
2592     /* Fill in slot numbers. We walk up from device to root, so need to print
2593      * them in the reverse order, last to first. */
2594     p = path + path_len;
2595     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2596         p -= slot_len;
2597         s = snprintf(slot, sizeof slot, ":%02x.%x",
2598                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2599         assert(s == slot_len);
2600         memcpy(p, slot, slot_len);
2601     }
2602 
2603     return path;
2604 }
2605 
2606 static int pci_qdev_find_recursive(PCIBus *bus,
2607                                    const char *id, PCIDevice **pdev)
2608 {
2609     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2610     if (!qdev) {
2611         return -ENODEV;
2612     }
2613 
2614     /* roughly check if given qdev is pci device */
2615     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2616         *pdev = PCI_DEVICE(qdev);
2617         return 0;
2618     }
2619     return -EINVAL;
2620 }
2621 
2622 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2623 {
2624     PCIHostState *host_bridge;
2625     int rc = -ENODEV;
2626 
2627     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2628         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2629         if (!tmp) {
2630             rc = 0;
2631             break;
2632         }
2633         if (tmp != -ENODEV) {
2634             rc = tmp;
2635         }
2636     }
2637 
2638     return rc;
2639 }
2640 
2641 MemoryRegion *pci_address_space(PCIDevice *dev)
2642 {
2643     return pci_get_bus(dev)->address_space_mem;
2644 }
2645 
2646 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2647 {
2648     return pci_get_bus(dev)->address_space_io;
2649 }
2650 
2651 static void pci_device_class_init(ObjectClass *klass, void *data)
2652 {
2653     DeviceClass *k = DEVICE_CLASS(klass);
2654 
2655     k->realize = pci_qdev_realize;
2656     k->unrealize = pci_qdev_unrealize;
2657     k->bus_type = TYPE_PCI_BUS;
2658     device_class_set_props(k, pci_props);
2659 }
2660 
2661 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2662 {
2663     if (!object_class_is_abstract(klass)) {
2664         ObjectClass *conventional =
2665             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2666         ObjectClass *pcie =
2667             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2668         assert(conventional || pcie);
2669     }
2670 }
2671 
2672 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2673 {
2674     PCIBus *bus = pci_get_bus(dev);
2675     PCIBus *iommu_bus = bus;
2676     uint8_t devfn = dev->devfn;
2677 
2678     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2679         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2680 
2681         /*
2682          * The requester ID of the provided device may be aliased, as seen from
2683          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2684          * requester ID to provide a unique AddressSpace for devices, but
2685          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2686          * to-PCI bridge creates and accepts transactions on behalf of down-
2687          * stream devices.  When doing so, all downstream devices are masked
2688          * (aliased) behind a single requester ID.  The requester ID used
2689          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2690          * bridges, with a PCIe capability indicating such, follow the
2691          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2692          * where the bridge uses the seconary bus as the bridge portion of the
2693          * requester ID and devfn of 00.0.  For other bridges, typically those
2694          * found on the root complex such as the dmi-to-pci-bridge, we follow
2695          * the convention of typical bare-metal hardware, which uses the
2696          * requester ID of the bridge itself.  There are device specific
2697          * exceptions to these rules, but these are the defaults that the
2698          * Linux kernel uses when determining DMA aliases itself and believed
2699          * to be true for the bare metal equivalents of the devices emulated
2700          * in QEMU.
2701          */
2702         if (!pci_bus_is_express(iommu_bus)) {
2703             PCIDevice *parent = iommu_bus->parent_dev;
2704 
2705             if (pci_is_express(parent) &&
2706                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2707                 devfn = PCI_DEVFN(0, 0);
2708                 bus = iommu_bus;
2709             } else {
2710                 devfn = parent->devfn;
2711                 bus = parent_bus;
2712             }
2713         }
2714 
2715         iommu_bus = parent_bus;
2716     }
2717     if (iommu_bus && iommu_bus->iommu_fn) {
2718         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2719     }
2720     return &address_space_memory;
2721 }
2722 
2723 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2724 {
2725     bus->iommu_fn = fn;
2726     bus->iommu_opaque = opaque;
2727 }
2728 
2729 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2730 {
2731     Range *range = opaque;
2732     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2733     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2734     int i;
2735 
2736     if (!(cmd & PCI_COMMAND_MEMORY)) {
2737         return;
2738     }
2739 
2740     if (pc->is_bridge) {
2741         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2742         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2743 
2744         base = MAX(base, 0x1ULL << 32);
2745 
2746         if (limit >= base) {
2747             Range pref_range;
2748             range_set_bounds(&pref_range, base, limit);
2749             range_extend(range, &pref_range);
2750         }
2751     }
2752     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2753         PCIIORegion *r = &dev->io_regions[i];
2754         pcibus_t lob, upb;
2755         Range region_range;
2756 
2757         if (!r->size ||
2758             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2759             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2760             continue;
2761         }
2762 
2763         lob = pci_bar_address(dev, i, r->type, r->size);
2764         upb = lob + r->size - 1;
2765         if (lob == PCI_BAR_UNMAPPED) {
2766             continue;
2767         }
2768 
2769         lob = MAX(lob, 0x1ULL << 32);
2770 
2771         if (upb >= lob) {
2772             range_set_bounds(&region_range, lob, upb);
2773             range_extend(range, &region_range);
2774         }
2775     }
2776 }
2777 
2778 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2779 {
2780     range_make_empty(range);
2781     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2782 }
2783 
2784 static bool pcie_has_upstream_port(PCIDevice *dev)
2785 {
2786     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2787 
2788     /* Device associated with an upstream port.
2789      * As there are several types of these, it's easier to check the
2790      * parent device: upstream ports are always connected to
2791      * root or downstream ports.
2792      */
2793     return parent_dev &&
2794         pci_is_express(parent_dev) &&
2795         parent_dev->exp.exp_cap &&
2796         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2797          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2798 }
2799 
2800 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2801 {
2802     PCIBus *bus = pci_get_bus(pci_dev);
2803 
2804     if(pcie_has_upstream_port(pci_dev)) {
2805         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2806         return bus->devices[0];
2807     } else {
2808         /* Other bus types might support multiple devices at slots 0-31 */
2809         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2810     }
2811 }
2812 
2813 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2814 {
2815     MSIMessage msg;
2816     if (msix_enabled(dev)) {
2817         msg = msix_get_message(dev, vector);
2818     } else if (msi_enabled(dev)) {
2819         msg = msi_get_message(dev, vector);
2820     } else {
2821         /* Should never happen */
2822         error_report("%s: unknown interrupt type", __func__);
2823         abort();
2824     }
2825     return msg;
2826 }
2827 
2828 static const TypeInfo pci_device_type_info = {
2829     .name = TYPE_PCI_DEVICE,
2830     .parent = TYPE_DEVICE,
2831     .instance_size = sizeof(PCIDevice),
2832     .abstract = true,
2833     .class_size = sizeof(PCIDeviceClass),
2834     .class_init = pci_device_class_init,
2835     .class_base_init = pci_device_class_base_init,
2836 };
2837 
2838 static void pci_register_types(void)
2839 {
2840     type_register_static(&pci_bus_info);
2841     type_register_static(&pcie_bus_info);
2842     type_register_static(&conventional_pci_interface_info);
2843     type_register_static(&pcie_interface_info);
2844     type_register_static(&pci_device_type_info);
2845 }
2846 
2847 type_init(pci_register_types)
2848