xref: /openbmc/qemu/hw/pci/pci.c (revision 1da79ecc)
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 "hw/qdev-properties-system.h"
35 #include "migration/qemu-file-types.h"
36 #include "migration/vmstate.h"
37 #include "monitor/monitor.h"
38 #include "net/net.h"
39 #include "sysemu/numa.h"
40 #include "sysemu/sysemu.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "exec/address-spaces.h"
48 #include "hw/hotplug.h"
49 #include "hw/boards.h"
50 #include "qapi/error.h"
51 #include "qapi/qapi-commands-pci.h"
52 #include "qemu/cutils.h"
53 
54 //#define DEBUG_PCI
55 #ifdef DEBUG_PCI
56 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
57 #else
58 # define PCI_DPRINTF(format, ...)       do { } while (0)
59 #endif
60 
61 bool pci_available = true;
62 
63 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
64 static char *pcibus_get_dev_path(DeviceState *dev);
65 static char *pcibus_get_fw_dev_path(DeviceState *dev);
66 static void pcibus_reset(BusState *qbus);
67 
68 static Property pci_props[] = {
69     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
70     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
71     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
72     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
73                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
74     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
75                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
76     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
77                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
78     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
79                        failover_pair_id),
80     DEFINE_PROP_END_OF_LIST()
81 };
82 
83 static const VMStateDescription vmstate_pcibus = {
84     .name = "PCIBUS",
85     .version_id = 1,
86     .minimum_version_id = 1,
87     .fields = (VMStateField[]) {
88         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
89         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
90                              nirq, 0, vmstate_info_int32,
91                              int32_t),
92         VMSTATE_END_OF_LIST()
93     }
94 };
95 
96 static void pci_init_bus_master(PCIDevice *pci_dev)
97 {
98     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
99 
100     memory_region_init_alias(&pci_dev->bus_master_enable_region,
101                              OBJECT(pci_dev), "bus master",
102                              dma_as->root, 0, memory_region_size(dma_as->root));
103     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
104     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
105                                 &pci_dev->bus_master_enable_region);
106 }
107 
108 static void pcibus_machine_done(Notifier *notifier, void *data)
109 {
110     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
111     int i;
112 
113     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
114         if (bus->devices[i]) {
115             pci_init_bus_master(bus->devices[i]);
116         }
117     }
118 }
119 
120 static void pci_bus_realize(BusState *qbus, Error **errp)
121 {
122     PCIBus *bus = PCI_BUS(qbus);
123 
124     bus->machine_done.notify = pcibus_machine_done;
125     qemu_add_machine_init_done_notifier(&bus->machine_done);
126 
127     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
128 }
129 
130 static void pcie_bus_realize(BusState *qbus, Error **errp)
131 {
132     PCIBus *bus = PCI_BUS(qbus);
133 
134     pci_bus_realize(qbus, errp);
135 
136     /*
137      * A PCI-E bus can support extended config space if it's the root
138      * bus, or if the bus/bridge above it does as well
139      */
140     if (pci_bus_is_root(bus)) {
141         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
142     } else {
143         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
144 
145         if (pci_bus_allows_extended_config_space(parent_bus)) {
146             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
147         }
148     }
149 }
150 
151 static void pci_bus_unrealize(BusState *qbus)
152 {
153     PCIBus *bus = PCI_BUS(qbus);
154 
155     qemu_remove_machine_init_done_notifier(&bus->machine_done);
156 
157     vmstate_unregister(NULL, &vmstate_pcibus, bus);
158 }
159 
160 static int pcibus_num(PCIBus *bus)
161 {
162     if (pci_bus_is_root(bus)) {
163         return 0; /* pci host bridge */
164     }
165     return bus->parent_dev->config[PCI_SECONDARY_BUS];
166 }
167 
168 static uint16_t pcibus_numa_node(PCIBus *bus)
169 {
170     return NUMA_NODE_UNASSIGNED;
171 }
172 
173 static void pci_bus_class_init(ObjectClass *klass, void *data)
174 {
175     BusClass *k = BUS_CLASS(klass);
176     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
177 
178     k->print_dev = pcibus_dev_print;
179     k->get_dev_path = pcibus_get_dev_path;
180     k->get_fw_dev_path = pcibus_get_fw_dev_path;
181     k->realize = pci_bus_realize;
182     k->unrealize = pci_bus_unrealize;
183     k->reset = pcibus_reset;
184 
185     pbc->bus_num = pcibus_num;
186     pbc->numa_node = pcibus_numa_node;
187 }
188 
189 static const TypeInfo pci_bus_info = {
190     .name = TYPE_PCI_BUS,
191     .parent = TYPE_BUS,
192     .instance_size = sizeof(PCIBus),
193     .class_size = sizeof(PCIBusClass),
194     .class_init = pci_bus_class_init,
195 };
196 
197 static const TypeInfo pcie_interface_info = {
198     .name          = INTERFACE_PCIE_DEVICE,
199     .parent        = TYPE_INTERFACE,
200 };
201 
202 static const TypeInfo conventional_pci_interface_info = {
203     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
204     .parent        = TYPE_INTERFACE,
205 };
206 
207 static void pcie_bus_class_init(ObjectClass *klass, void *data)
208 {
209     BusClass *k = BUS_CLASS(klass);
210 
211     k->realize = pcie_bus_realize;
212 }
213 
214 static const TypeInfo pcie_bus_info = {
215     .name = TYPE_PCIE_BUS,
216     .parent = TYPE_PCI_BUS,
217     .class_init = pcie_bus_class_init,
218 };
219 
220 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
221 static void pci_update_mappings(PCIDevice *d);
222 static void pci_irq_handler(void *opaque, int irq_num, int level);
223 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
224 static void pci_del_option_rom(PCIDevice *pdev);
225 
226 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
227 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
228 
229 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
230 
231 int pci_bar(PCIDevice *d, int reg)
232 {
233     uint8_t type;
234 
235     if (reg != PCI_ROM_SLOT)
236         return PCI_BASE_ADDRESS_0 + reg * 4;
237 
238     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
239     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
240 }
241 
242 static inline int pci_irq_state(PCIDevice *d, int irq_num)
243 {
244         return (d->irq_state >> irq_num) & 0x1;
245 }
246 
247 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
248 {
249         d->irq_state &= ~(0x1 << irq_num);
250         d->irq_state |= level << irq_num;
251 }
252 
253 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
254 {
255     assert(irq_num >= 0);
256     assert(irq_num < bus->nirq);
257     bus->irq_count[irq_num] += change;
258     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
259 }
260 
261 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
262 {
263     PCIBus *bus;
264     for (;;) {
265         bus = pci_get_bus(pci_dev);
266         irq_num = bus->map_irq(pci_dev, irq_num);
267         if (bus->set_irq)
268             break;
269         pci_dev = bus->parent_dev;
270     }
271     pci_bus_change_irq_level(bus, irq_num, change);
272 }
273 
274 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
275 {
276     assert(irq_num >= 0);
277     assert(irq_num < bus->nirq);
278     return !!bus->irq_count[irq_num];
279 }
280 
281 /* Update interrupt status bit in config space on interrupt
282  * state change. */
283 static void pci_update_irq_status(PCIDevice *dev)
284 {
285     if (dev->irq_state) {
286         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
287     } else {
288         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
289     }
290 }
291 
292 void pci_device_deassert_intx(PCIDevice *dev)
293 {
294     int i;
295     for (i = 0; i < PCI_NUM_PINS; ++i) {
296         pci_irq_handler(dev, i, 0);
297     }
298 }
299 
300 static void pci_do_device_reset(PCIDevice *dev)
301 {
302     int r;
303 
304     pci_device_deassert_intx(dev);
305     assert(dev->irq_state == 0);
306 
307     /* Clear all writable bits */
308     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
309                                  pci_get_word(dev->wmask + PCI_COMMAND) |
310                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
311     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
312                                  pci_get_word(dev->wmask + PCI_STATUS) |
313                                  pci_get_word(dev->w1cmask + PCI_STATUS));
314     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
315     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
316                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
317                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
318     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
319     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
320         PCIIORegion *region = &dev->io_regions[r];
321         if (!region->size) {
322             continue;
323         }
324 
325         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
326             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
327             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
328         } else {
329             pci_set_long(dev->config + pci_bar(dev, r), region->type);
330         }
331     }
332     pci_update_mappings(dev);
333 
334     msi_reset(dev);
335     msix_reset(dev);
336 }
337 
338 /*
339  * This function is called on #RST and FLR.
340  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
341  */
342 void pci_device_reset(PCIDevice *dev)
343 {
344     qdev_reset_all(&dev->qdev);
345     pci_do_device_reset(dev);
346 }
347 
348 /*
349  * Trigger pci bus reset under a given bus.
350  * Called via qbus_reset_all on RST# assert, after the devices
351  * have been reset qdev_reset_all-ed already.
352  */
353 static void pcibus_reset(BusState *qbus)
354 {
355     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
356     int i;
357 
358     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
359         if (bus->devices[i]) {
360             pci_do_device_reset(bus->devices[i]);
361         }
362     }
363 
364     for (i = 0; i < bus->nirq; i++) {
365         assert(bus->irq_count[i] == 0);
366     }
367 }
368 
369 static void pci_host_bus_register(DeviceState *host)
370 {
371     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
372 
373     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
374 }
375 
376 static void pci_host_bus_unregister(DeviceState *host)
377 {
378     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
379 
380     QLIST_REMOVE(host_bridge, next);
381 }
382 
383 PCIBus *pci_device_root_bus(const PCIDevice *d)
384 {
385     PCIBus *bus = pci_get_bus(d);
386 
387     while (!pci_bus_is_root(bus)) {
388         d = bus->parent_dev;
389         assert(d != NULL);
390 
391         bus = pci_get_bus(d);
392     }
393 
394     return bus;
395 }
396 
397 const char *pci_root_bus_path(PCIDevice *dev)
398 {
399     PCIBus *rootbus = pci_device_root_bus(dev);
400     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
401     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
402 
403     assert(host_bridge->bus == rootbus);
404 
405     if (hc->root_bus_path) {
406         return (*hc->root_bus_path)(host_bridge, rootbus);
407     }
408 
409     return rootbus->qbus.name;
410 }
411 
412 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
413                               MemoryRegion *address_space_mem,
414                               MemoryRegion *address_space_io,
415                               uint8_t devfn_min)
416 {
417     assert(PCI_FUNC(devfn_min) == 0);
418     bus->devfn_min = devfn_min;
419     bus->slot_reserved_mask = 0x0;
420     bus->address_space_mem = address_space_mem;
421     bus->address_space_io = address_space_io;
422     bus->flags |= PCI_BUS_IS_ROOT;
423 
424     /* host bridge */
425     QLIST_INIT(&bus->child);
426 
427     pci_host_bus_register(parent);
428 }
429 
430 static void pci_bus_uninit(PCIBus *bus)
431 {
432     pci_host_bus_unregister(BUS(bus)->parent);
433 }
434 
435 bool pci_bus_is_express(PCIBus *bus)
436 {
437     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
438 }
439 
440 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
441                               const char *name,
442                               MemoryRegion *address_space_mem,
443                               MemoryRegion *address_space_io,
444                               uint8_t devfn_min, const char *typename)
445 {
446     qbus_create_inplace(bus, bus_size, typename, parent, name);
447     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
448                       devfn_min);
449 }
450 
451 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
452                          MemoryRegion *address_space_mem,
453                          MemoryRegion *address_space_io,
454                          uint8_t devfn_min, const char *typename)
455 {
456     PCIBus *bus;
457 
458     bus = PCI_BUS(qbus_create(typename, parent, name));
459     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
460                       devfn_min);
461     return bus;
462 }
463 
464 void pci_root_bus_cleanup(PCIBus *bus)
465 {
466     pci_bus_uninit(bus);
467     /* the caller of the unplug hotplug handler will delete this device */
468     qbus_unrealize(BUS(bus));
469 }
470 
471 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
472                   void *irq_opaque, int nirq)
473 {
474     bus->set_irq = set_irq;
475     bus->map_irq = map_irq;
476     bus->irq_opaque = irq_opaque;
477     bus->nirq = nirq;
478     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
479 }
480 
481 void pci_bus_irqs_cleanup(PCIBus *bus)
482 {
483     bus->set_irq = NULL;
484     bus->map_irq = NULL;
485     bus->irq_opaque = NULL;
486     bus->nirq = 0;
487     g_free(bus->irq_count);
488 }
489 
490 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
491                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
492                               void *irq_opaque,
493                               MemoryRegion *address_space_mem,
494                               MemoryRegion *address_space_io,
495                               uint8_t devfn_min, int nirq,
496                               const char *typename)
497 {
498     PCIBus *bus;
499 
500     bus = pci_root_bus_new(parent, name, address_space_mem,
501                            address_space_io, devfn_min, typename);
502     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
503     return bus;
504 }
505 
506 void pci_unregister_root_bus(PCIBus *bus)
507 {
508     pci_bus_irqs_cleanup(bus);
509     pci_root_bus_cleanup(bus);
510 }
511 
512 int pci_bus_num(PCIBus *s)
513 {
514     return PCI_BUS_GET_CLASS(s)->bus_num(s);
515 }
516 
517 int pci_bus_numa_node(PCIBus *bus)
518 {
519     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
520 }
521 
522 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
523                                  const VMStateField *field)
524 {
525     PCIDevice *s = container_of(pv, PCIDevice, config);
526     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
527     uint8_t *config;
528     int i;
529 
530     assert(size == pci_config_size(s));
531     config = g_malloc(size);
532 
533     qemu_get_buffer(f, config, size);
534     for (i = 0; i < size; ++i) {
535         if ((config[i] ^ s->config[i]) &
536             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
537             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
538                          "cmask: %x wmask: %x w1cmask:%x", __func__,
539                          i, config[i], s->config[i],
540                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
541             g_free(config);
542             return -EINVAL;
543         }
544     }
545     memcpy(s->config, config, size);
546 
547     pci_update_mappings(s);
548     if (pc->is_bridge) {
549         PCIBridge *b = PCI_BRIDGE(s);
550         pci_bridge_update_mappings(b);
551     }
552 
553     memory_region_set_enabled(&s->bus_master_enable_region,
554                               pci_get_word(s->config + PCI_COMMAND)
555                               & PCI_COMMAND_MASTER);
556 
557     g_free(config);
558     return 0;
559 }
560 
561 /* just put buffer */
562 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
563                                  const VMStateField *field, JSONWriter *vmdesc)
564 {
565     const uint8_t **v = pv;
566     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
567     qemu_put_buffer(f, *v, size);
568 
569     return 0;
570 }
571 
572 static VMStateInfo vmstate_info_pci_config = {
573     .name = "pci config",
574     .get  = get_pci_config_device,
575     .put  = put_pci_config_device,
576 };
577 
578 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
579                              const VMStateField *field)
580 {
581     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
582     uint32_t irq_state[PCI_NUM_PINS];
583     int i;
584     for (i = 0; i < PCI_NUM_PINS; ++i) {
585         irq_state[i] = qemu_get_be32(f);
586         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
587             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
588                     irq_state[i]);
589             return -EINVAL;
590         }
591     }
592 
593     for (i = 0; i < PCI_NUM_PINS; ++i) {
594         pci_set_irq_state(s, i, irq_state[i]);
595     }
596 
597     return 0;
598 }
599 
600 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
601                              const VMStateField *field, JSONWriter *vmdesc)
602 {
603     int i;
604     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
605 
606     for (i = 0; i < PCI_NUM_PINS; ++i) {
607         qemu_put_be32(f, pci_irq_state(s, i));
608     }
609 
610     return 0;
611 }
612 
613 static VMStateInfo vmstate_info_pci_irq_state = {
614     .name = "pci irq state",
615     .get  = get_pci_irq_state,
616     .put  = put_pci_irq_state,
617 };
618 
619 static bool migrate_is_pcie(void *opaque, int version_id)
620 {
621     return pci_is_express((PCIDevice *)opaque);
622 }
623 
624 static bool migrate_is_not_pcie(void *opaque, int version_id)
625 {
626     return !pci_is_express((PCIDevice *)opaque);
627 }
628 
629 const VMStateDescription vmstate_pci_device = {
630     .name = "PCIDevice",
631     .version_id = 2,
632     .minimum_version_id = 1,
633     .fields = (VMStateField[]) {
634         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
635         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
636                                    migrate_is_not_pcie,
637                                    0, vmstate_info_pci_config,
638                                    PCI_CONFIG_SPACE_SIZE),
639         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
640                                    migrate_is_pcie,
641                                    0, vmstate_info_pci_config,
642                                    PCIE_CONFIG_SPACE_SIZE),
643         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
644                                    vmstate_info_pci_irq_state,
645                                    PCI_NUM_PINS * sizeof(int32_t)),
646         VMSTATE_END_OF_LIST()
647     }
648 };
649 
650 
651 void pci_device_save(PCIDevice *s, QEMUFile *f)
652 {
653     /* Clear interrupt status bit: it is implicit
654      * in irq_state which we are saving.
655      * This makes us compatible with old devices
656      * which never set or clear this bit. */
657     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
658     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
659     /* Restore the interrupt status bit. */
660     pci_update_irq_status(s);
661 }
662 
663 int pci_device_load(PCIDevice *s, QEMUFile *f)
664 {
665     int ret;
666     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
667     /* Restore the interrupt status bit. */
668     pci_update_irq_status(s);
669     return ret;
670 }
671 
672 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
673 {
674     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
675                  pci_default_sub_vendor_id);
676     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
677                  pci_default_sub_device_id);
678 }
679 
680 /*
681  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
682  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
683  */
684 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
685                              unsigned int *slotp, unsigned int *funcp)
686 {
687     const char *p;
688     char *e;
689     unsigned long val;
690     unsigned long dom = 0, bus = 0;
691     unsigned int slot = 0;
692     unsigned int func = 0;
693 
694     p = addr;
695     val = strtoul(p, &e, 16);
696     if (e == p)
697         return -1;
698     if (*e == ':') {
699         bus = val;
700         p = e + 1;
701         val = strtoul(p, &e, 16);
702         if (e == p)
703             return -1;
704         if (*e == ':') {
705             dom = bus;
706             bus = val;
707             p = e + 1;
708             val = strtoul(p, &e, 16);
709             if (e == p)
710                 return -1;
711         }
712     }
713 
714     slot = val;
715 
716     if (funcp != NULL) {
717         if (*e != '.')
718             return -1;
719 
720         p = e + 1;
721         val = strtoul(p, &e, 16);
722         if (e == p)
723             return -1;
724 
725         func = val;
726     }
727 
728     /* if funcp == NULL func is 0 */
729     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
730         return -1;
731 
732     if (*e)
733         return -1;
734 
735     *domp = dom;
736     *busp = bus;
737     *slotp = slot;
738     if (funcp != NULL)
739         *funcp = func;
740     return 0;
741 }
742 
743 static void pci_init_cmask(PCIDevice *dev)
744 {
745     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
746     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
747     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
748     dev->cmask[PCI_REVISION_ID] = 0xff;
749     dev->cmask[PCI_CLASS_PROG] = 0xff;
750     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
751     dev->cmask[PCI_HEADER_TYPE] = 0xff;
752     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
753 }
754 
755 static void pci_init_wmask(PCIDevice *dev)
756 {
757     int config_size = pci_config_size(dev);
758 
759     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
760     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
761     pci_set_word(dev->wmask + PCI_COMMAND,
762                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
763                  PCI_COMMAND_INTX_DISABLE);
764     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
765 
766     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
767            config_size - PCI_CONFIG_HEADER_SIZE);
768 }
769 
770 static void pci_init_w1cmask(PCIDevice *dev)
771 {
772     /*
773      * Note: It's okay to set w1cmask even for readonly bits as
774      * long as their value is hardwired to 0.
775      */
776     pci_set_word(dev->w1cmask + PCI_STATUS,
777                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
778                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
779                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
780 }
781 
782 static void pci_init_mask_bridge(PCIDevice *d)
783 {
784     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
785        PCI_SEC_LETENCY_TIMER */
786     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
787 
788     /* base and limit */
789     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
790     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
791     pci_set_word(d->wmask + PCI_MEMORY_BASE,
792                  PCI_MEMORY_RANGE_MASK & 0xffff);
793     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
794                  PCI_MEMORY_RANGE_MASK & 0xffff);
795     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
796                  PCI_PREF_RANGE_MASK & 0xffff);
797     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
798                  PCI_PREF_RANGE_MASK & 0xffff);
799 
800     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
801     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
802 
803     /* Supported memory and i/o types */
804     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
805     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
806     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
807                                PCI_PREF_RANGE_TYPE_64);
808     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
809                                PCI_PREF_RANGE_TYPE_64);
810 
811     /*
812      * TODO: Bridges default to 10-bit VGA decoding but we currently only
813      * implement 16-bit decoding (no alias support).
814      */
815     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
816                  PCI_BRIDGE_CTL_PARITY |
817                  PCI_BRIDGE_CTL_SERR |
818                  PCI_BRIDGE_CTL_ISA |
819                  PCI_BRIDGE_CTL_VGA |
820                  PCI_BRIDGE_CTL_VGA_16BIT |
821                  PCI_BRIDGE_CTL_MASTER_ABORT |
822                  PCI_BRIDGE_CTL_BUS_RESET |
823                  PCI_BRIDGE_CTL_FAST_BACK |
824                  PCI_BRIDGE_CTL_DISCARD |
825                  PCI_BRIDGE_CTL_SEC_DISCARD |
826                  PCI_BRIDGE_CTL_DISCARD_SERR);
827     /* Below does not do anything as we never set this bit, put here for
828      * completeness. */
829     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
830                  PCI_BRIDGE_CTL_DISCARD_STATUS);
831     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
832     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
833     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
834                                PCI_PREF_RANGE_TYPE_MASK);
835     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
836                                PCI_PREF_RANGE_TYPE_MASK);
837 }
838 
839 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
840 {
841     uint8_t slot = PCI_SLOT(dev->devfn);
842     uint8_t func;
843 
844     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
845         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
846     }
847 
848     /*
849      * multifunction bit is interpreted in two ways as follows.
850      *   - all functions must set the bit to 1.
851      *     Example: Intel X53
852      *   - function 0 must set the bit, but the rest function (> 0)
853      *     is allowed to leave the bit to 0.
854      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
855      *
856      * So OS (at least Linux) checks the bit of only function 0,
857      * and doesn't see the bit of function > 0.
858      *
859      * The below check allows both interpretation.
860      */
861     if (PCI_FUNC(dev->devfn)) {
862         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
863         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
864             /* function 0 should set multifunction bit */
865             error_setg(errp, "PCI: single function device can't be populated "
866                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
867             return;
868         }
869         return;
870     }
871 
872     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
873         return;
874     }
875     /* function 0 indicates single function, so function > 0 must be NULL */
876     for (func = 1; func < PCI_FUNC_MAX; ++func) {
877         if (bus->devices[PCI_DEVFN(slot, func)]) {
878             error_setg(errp, "PCI: %x.0 indicates single function, "
879                        "but %x.%x is already populated.",
880                        slot, slot, func);
881             return;
882         }
883     }
884 }
885 
886 static void pci_config_alloc(PCIDevice *pci_dev)
887 {
888     int config_size = pci_config_size(pci_dev);
889 
890     pci_dev->config = g_malloc0(config_size);
891     pci_dev->cmask = g_malloc0(config_size);
892     pci_dev->wmask = g_malloc0(config_size);
893     pci_dev->w1cmask = g_malloc0(config_size);
894     pci_dev->used = g_malloc0(config_size);
895 }
896 
897 static void pci_config_free(PCIDevice *pci_dev)
898 {
899     g_free(pci_dev->config);
900     g_free(pci_dev->cmask);
901     g_free(pci_dev->wmask);
902     g_free(pci_dev->w1cmask);
903     g_free(pci_dev->used);
904 }
905 
906 static void do_pci_unregister_device(PCIDevice *pci_dev)
907 {
908     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
909     pci_config_free(pci_dev);
910 
911     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
912         memory_region_del_subregion(&pci_dev->bus_master_container_region,
913                                     &pci_dev->bus_master_enable_region);
914     }
915     address_space_destroy(&pci_dev->bus_master_as);
916 }
917 
918 /* Extract PCIReqIDCache into BDF format */
919 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
920 {
921     uint8_t bus_n;
922     uint16_t result;
923 
924     switch (cache->type) {
925     case PCI_REQ_ID_BDF:
926         result = pci_get_bdf(cache->dev);
927         break;
928     case PCI_REQ_ID_SECONDARY_BUS:
929         bus_n = pci_dev_bus_num(cache->dev);
930         result = PCI_BUILD_BDF(bus_n, 0);
931         break;
932     default:
933         error_report("Invalid PCI requester ID cache type: %d",
934                      cache->type);
935         exit(1);
936         break;
937     }
938 
939     return result;
940 }
941 
942 /* Parse bridges up to the root complex and return requester ID
943  * cache for specific device.  For full PCIe topology, the cache
944  * result would be exactly the same as getting BDF of the device.
945  * However, several tricks are required when system mixed up with
946  * legacy PCI devices and PCIe-to-PCI bridges.
947  *
948  * Here we cache the proxy device (and type) not requester ID since
949  * bus number might change from time to time.
950  */
951 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
952 {
953     PCIDevice *parent;
954     PCIReqIDCache cache = {
955         .dev = dev,
956         .type = PCI_REQ_ID_BDF,
957     };
958 
959     while (!pci_bus_is_root(pci_get_bus(dev))) {
960         /* We are under PCI/PCIe bridges */
961         parent = pci_get_bus(dev)->parent_dev;
962         if (pci_is_express(parent)) {
963             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
964                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
965                  * override the requester ID using secondary bus
966                  * number of parent bridge with zeroed devfn
967                  * (pcie-to-pci bridge spec chap 2.3). */
968                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
969                 cache.dev = dev;
970             }
971         } else {
972             /* Legacy PCI, override requester ID with the bridge's
973              * BDF upstream.  When the root complex connects to
974              * legacy PCI devices (including buses), it can only
975              * obtain requester ID info from directly attached
976              * devices.  If devices are attached under bridges, only
977              * the requester ID of the bridge that is directly
978              * attached to the root complex can be recognized. */
979             cache.type = PCI_REQ_ID_BDF;
980             cache.dev = parent;
981         }
982         dev = parent;
983     }
984 
985     return cache;
986 }
987 
988 uint16_t pci_requester_id(PCIDevice *dev)
989 {
990     return pci_req_id_cache_extract(&dev->requester_id_cache);
991 }
992 
993 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
994 {
995     return !(bus->devices[devfn]);
996 }
997 
998 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
999 {
1000     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1001 }
1002 
1003 /* -1 for devfn means auto assign */
1004 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1005                                          const char *name, int devfn,
1006                                          Error **errp)
1007 {
1008     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1009     PCIConfigReadFunc *config_read = pc->config_read;
1010     PCIConfigWriteFunc *config_write = pc->config_write;
1011     Error *local_err = NULL;
1012     DeviceState *dev = DEVICE(pci_dev);
1013     PCIBus *bus = pci_get_bus(pci_dev);
1014 
1015     /* Only pci bridges can be attached to extra PCI root buses */
1016     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1017         error_setg(errp,
1018                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1019                     bus->parent_dev->name);
1020         return NULL;
1021     }
1022 
1023     if (devfn < 0) {
1024         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1025             devfn += PCI_FUNC_MAX) {
1026             if (pci_bus_devfn_available(bus, devfn) &&
1027                    !pci_bus_devfn_reserved(bus, devfn)) {
1028                 goto found;
1029             }
1030         }
1031         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1032                    "or reserved", name);
1033         return NULL;
1034     found: ;
1035     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1036         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1037                    " reserved",
1038                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1039         return NULL;
1040     } else if (!pci_bus_devfn_available(bus, devfn)) {
1041         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1042                    " in use by %s",
1043                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1044                    bus->devices[devfn]->name);
1045         return NULL;
1046     } else if (dev->hotplugged &&
1047                pci_get_function_0(pci_dev)) {
1048         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1049                    " new func %s cannot be exposed to guest.",
1050                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1051                    pci_get_function_0(pci_dev)->name,
1052                    name);
1053 
1054        return NULL;
1055     }
1056 
1057     pci_dev->devfn = devfn;
1058     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1059     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1060 
1061     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1062                        "bus master container", UINT64_MAX);
1063     address_space_init(&pci_dev->bus_master_as,
1064                        &pci_dev->bus_master_container_region, pci_dev->name);
1065 
1066     if (phase_check(PHASE_MACHINE_READY)) {
1067         pci_init_bus_master(pci_dev);
1068     }
1069     pci_dev->irq_state = 0;
1070     pci_config_alloc(pci_dev);
1071 
1072     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1073     pci_config_set_device_id(pci_dev->config, pc->device_id);
1074     pci_config_set_revision(pci_dev->config, pc->revision);
1075     pci_config_set_class(pci_dev->config, pc->class_id);
1076 
1077     if (!pc->is_bridge) {
1078         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1079             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1080                          pc->subsystem_vendor_id);
1081             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1082                          pc->subsystem_id);
1083         } else {
1084             pci_set_default_subsystem_id(pci_dev);
1085         }
1086     } else {
1087         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1088         assert(!pc->subsystem_vendor_id);
1089         assert(!pc->subsystem_id);
1090     }
1091     pci_init_cmask(pci_dev);
1092     pci_init_wmask(pci_dev);
1093     pci_init_w1cmask(pci_dev);
1094     if (pc->is_bridge) {
1095         pci_init_mask_bridge(pci_dev);
1096     }
1097     pci_init_multifunction(bus, pci_dev, &local_err);
1098     if (local_err) {
1099         error_propagate(errp, local_err);
1100         do_pci_unregister_device(pci_dev);
1101         return NULL;
1102     }
1103 
1104     if (!config_read)
1105         config_read = pci_default_read_config;
1106     if (!config_write)
1107         config_write = pci_default_write_config;
1108     pci_dev->config_read = config_read;
1109     pci_dev->config_write = config_write;
1110     bus->devices[devfn] = pci_dev;
1111     pci_dev->version_id = 2; /* Current pci device vmstate version */
1112     return pci_dev;
1113 }
1114 
1115 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1116 {
1117     PCIIORegion *r;
1118     int i;
1119 
1120     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1121         r = &pci_dev->io_regions[i];
1122         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1123             continue;
1124         memory_region_del_subregion(r->address_space, r->memory);
1125     }
1126 
1127     pci_unregister_vga(pci_dev);
1128 }
1129 
1130 static void pci_qdev_unrealize(DeviceState *dev)
1131 {
1132     PCIDevice *pci_dev = PCI_DEVICE(dev);
1133     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1134 
1135     pci_unregister_io_regions(pci_dev);
1136     pci_del_option_rom(pci_dev);
1137 
1138     if (pc->exit) {
1139         pc->exit(pci_dev);
1140     }
1141 
1142     pci_device_deassert_intx(pci_dev);
1143     do_pci_unregister_device(pci_dev);
1144 }
1145 
1146 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1147                       uint8_t type, MemoryRegion *memory)
1148 {
1149     PCIIORegion *r;
1150     uint32_t addr; /* offset in pci config space */
1151     uint64_t wmask;
1152     pcibus_t size = memory_region_size(memory);
1153     uint8_t hdr_type;
1154 
1155     assert(region_num >= 0);
1156     assert(region_num < PCI_NUM_REGIONS);
1157     assert(is_power_of_2(size));
1158 
1159     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1160     hdr_type =
1161         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1162     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1163 
1164     r = &pci_dev->io_regions[region_num];
1165     r->addr = PCI_BAR_UNMAPPED;
1166     r->size = size;
1167     r->type = type;
1168     r->memory = memory;
1169     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1170                         ? pci_get_bus(pci_dev)->address_space_io
1171                         : pci_get_bus(pci_dev)->address_space_mem;
1172 
1173     wmask = ~(size - 1);
1174     if (region_num == PCI_ROM_SLOT) {
1175         /* ROM enable bit is writable */
1176         wmask |= PCI_ROM_ADDRESS_ENABLE;
1177     }
1178 
1179     addr = pci_bar(pci_dev, region_num);
1180     pci_set_long(pci_dev->config + addr, type);
1181 
1182     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1183         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1184         pci_set_quad(pci_dev->wmask + addr, wmask);
1185         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1186     } else {
1187         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1188         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1189     }
1190 }
1191 
1192 static void pci_update_vga(PCIDevice *pci_dev)
1193 {
1194     uint16_t cmd;
1195 
1196     if (!pci_dev->has_vga) {
1197         return;
1198     }
1199 
1200     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1201 
1202     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1203                               cmd & PCI_COMMAND_MEMORY);
1204     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1205                               cmd & PCI_COMMAND_IO);
1206     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1207                               cmd & PCI_COMMAND_IO);
1208 }
1209 
1210 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1211                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1212 {
1213     PCIBus *bus = pci_get_bus(pci_dev);
1214 
1215     assert(!pci_dev->has_vga);
1216 
1217     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1218     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1219     memory_region_add_subregion_overlap(bus->address_space_mem,
1220                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1221 
1222     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1223     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1224     memory_region_add_subregion_overlap(bus->address_space_io,
1225                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1226 
1227     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1228     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1229     memory_region_add_subregion_overlap(bus->address_space_io,
1230                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1231     pci_dev->has_vga = true;
1232 
1233     pci_update_vga(pci_dev);
1234 }
1235 
1236 void pci_unregister_vga(PCIDevice *pci_dev)
1237 {
1238     PCIBus *bus = pci_get_bus(pci_dev);
1239 
1240     if (!pci_dev->has_vga) {
1241         return;
1242     }
1243 
1244     memory_region_del_subregion(bus->address_space_mem,
1245                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1246     memory_region_del_subregion(bus->address_space_io,
1247                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1248     memory_region_del_subregion(bus->address_space_io,
1249                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1250     pci_dev->has_vga = false;
1251 }
1252 
1253 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1254 {
1255     return pci_dev->io_regions[region_num].addr;
1256 }
1257 
1258 static pcibus_t pci_bar_address(PCIDevice *d,
1259                                 int reg, uint8_t type, pcibus_t size)
1260 {
1261     pcibus_t new_addr, last_addr;
1262     int bar = pci_bar(d, reg);
1263     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1264     Object *machine = qdev_get_machine();
1265     ObjectClass *oc = object_get_class(machine);
1266     MachineClass *mc = MACHINE_CLASS(oc);
1267     bool allow_0_address = mc->pci_allow_0_address;
1268 
1269     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1270         if (!(cmd & PCI_COMMAND_IO)) {
1271             return PCI_BAR_UNMAPPED;
1272         }
1273         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1274         last_addr = new_addr + size - 1;
1275         /* Check if 32 bit BAR wraps around explicitly.
1276          * TODO: make priorities correct and remove this work around.
1277          */
1278         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1279             (!allow_0_address && new_addr == 0)) {
1280             return PCI_BAR_UNMAPPED;
1281         }
1282         return new_addr;
1283     }
1284 
1285     if (!(cmd & PCI_COMMAND_MEMORY)) {
1286         return PCI_BAR_UNMAPPED;
1287     }
1288     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1289         new_addr = pci_get_quad(d->config + bar);
1290     } else {
1291         new_addr = pci_get_long(d->config + bar);
1292     }
1293     /* the ROM slot has a specific enable bit */
1294     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1295         return PCI_BAR_UNMAPPED;
1296     }
1297     new_addr &= ~(size - 1);
1298     last_addr = new_addr + size - 1;
1299     /* NOTE: we do not support wrapping */
1300     /* XXX: as we cannot support really dynamic
1301        mappings, we handle specific values as invalid
1302        mappings. */
1303     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1304         (!allow_0_address && new_addr == 0)) {
1305         return PCI_BAR_UNMAPPED;
1306     }
1307 
1308     /* Now pcibus_t is 64bit.
1309      * Check if 32 bit BAR wraps around explicitly.
1310      * Without this, PC ide doesn't work well.
1311      * TODO: remove this work around.
1312      */
1313     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1314         return PCI_BAR_UNMAPPED;
1315     }
1316 
1317     /*
1318      * OS is allowed to set BAR beyond its addressable
1319      * bits. For example, 32 bit OS can set 64bit bar
1320      * to >4G. Check it. TODO: we might need to support
1321      * it in the future for e.g. PAE.
1322      */
1323     if (last_addr >= HWADDR_MAX) {
1324         return PCI_BAR_UNMAPPED;
1325     }
1326 
1327     return new_addr;
1328 }
1329 
1330 static void pci_update_mappings(PCIDevice *d)
1331 {
1332     PCIIORegion *r;
1333     int i;
1334     pcibus_t new_addr;
1335 
1336     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1337         r = &d->io_regions[i];
1338 
1339         /* this region isn't registered */
1340         if (!r->size)
1341             continue;
1342 
1343         new_addr = pci_bar_address(d, i, r->type, r->size);
1344 
1345         /* This bar isn't changed */
1346         if (new_addr == r->addr)
1347             continue;
1348 
1349         /* now do the real mapping */
1350         if (r->addr != PCI_BAR_UNMAPPED) {
1351             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1352                                           PCI_SLOT(d->devfn),
1353                                           PCI_FUNC(d->devfn),
1354                                           i, r->addr, r->size);
1355             memory_region_del_subregion(r->address_space, r->memory);
1356         }
1357         r->addr = new_addr;
1358         if (r->addr != PCI_BAR_UNMAPPED) {
1359             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1360                                           PCI_SLOT(d->devfn),
1361                                           PCI_FUNC(d->devfn),
1362                                           i, r->addr, r->size);
1363             memory_region_add_subregion_overlap(r->address_space,
1364                                                 r->addr, r->memory, 1);
1365         }
1366     }
1367 
1368     pci_update_vga(d);
1369 }
1370 
1371 static inline int pci_irq_disabled(PCIDevice *d)
1372 {
1373     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1374 }
1375 
1376 /* Called after interrupt disabled field update in config space,
1377  * assert/deassert interrupts if necessary.
1378  * Gets original interrupt disable bit value (before update). */
1379 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1380 {
1381     int i, disabled = pci_irq_disabled(d);
1382     if (disabled == was_irq_disabled)
1383         return;
1384     for (i = 0; i < PCI_NUM_PINS; ++i) {
1385         int state = pci_irq_state(d, i);
1386         pci_change_irq_level(d, i, disabled ? -state : state);
1387     }
1388 }
1389 
1390 uint32_t pci_default_read_config(PCIDevice *d,
1391                                  uint32_t address, int len)
1392 {
1393     uint32_t val = 0;
1394 
1395     assert(address + len <= pci_config_size(d));
1396 
1397     if (pci_is_express_downstream_port(d) &&
1398         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1399         pcie_sync_bridge_lnk(d);
1400     }
1401     memcpy(&val, d->config + address, len);
1402     return le32_to_cpu(val);
1403 }
1404 
1405 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1406 {
1407     int i, was_irq_disabled = pci_irq_disabled(d);
1408     uint32_t val = val_in;
1409 
1410     assert(addr + l <= pci_config_size(d));
1411 
1412     for (i = 0; i < l; val >>= 8, ++i) {
1413         uint8_t wmask = d->wmask[addr + i];
1414         uint8_t w1cmask = d->w1cmask[addr + i];
1415         assert(!(wmask & w1cmask));
1416         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1417         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1418     }
1419     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1420         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1421         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1422         range_covers_byte(addr, l, PCI_COMMAND))
1423         pci_update_mappings(d);
1424 
1425     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1426         pci_update_irq_disabled(d, was_irq_disabled);
1427         memory_region_set_enabled(&d->bus_master_enable_region,
1428                                   pci_get_word(d->config + PCI_COMMAND)
1429                                     & PCI_COMMAND_MASTER);
1430     }
1431 
1432     msi_write_config(d, addr, val_in, l);
1433     msix_write_config(d, addr, val_in, l);
1434 }
1435 
1436 /***********************************************************/
1437 /* generic PCI irq support */
1438 
1439 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1440 static void pci_irq_handler(void *opaque, int irq_num, int level)
1441 {
1442     PCIDevice *pci_dev = opaque;
1443     int change;
1444 
1445     change = level - pci_irq_state(pci_dev, irq_num);
1446     if (!change)
1447         return;
1448 
1449     pci_set_irq_state(pci_dev, irq_num, level);
1450     pci_update_irq_status(pci_dev);
1451     if (pci_irq_disabled(pci_dev))
1452         return;
1453     pci_change_irq_level(pci_dev, irq_num, change);
1454 }
1455 
1456 static inline int pci_intx(PCIDevice *pci_dev)
1457 {
1458     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1459 }
1460 
1461 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1462 {
1463     int intx = pci_intx(pci_dev);
1464 
1465     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1466 }
1467 
1468 void pci_set_irq(PCIDevice *pci_dev, int level)
1469 {
1470     int intx = pci_intx(pci_dev);
1471     pci_irq_handler(pci_dev, intx, level);
1472 }
1473 
1474 /* Special hooks used by device assignment */
1475 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1476 {
1477     assert(pci_bus_is_root(bus));
1478     bus->route_intx_to_irq = route_intx_to_irq;
1479 }
1480 
1481 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1482 {
1483     PCIBus *bus;
1484 
1485     do {
1486         bus = pci_get_bus(dev);
1487         pin = bus->map_irq(dev, pin);
1488         dev = bus->parent_dev;
1489     } while (dev);
1490 
1491     if (!bus->route_intx_to_irq) {
1492         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1493                      object_get_typename(OBJECT(bus->qbus.parent)));
1494         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1495     }
1496 
1497     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1498 }
1499 
1500 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1501 {
1502     return old->mode != new->mode || old->irq != new->irq;
1503 }
1504 
1505 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1506 {
1507     PCIDevice *dev;
1508     PCIBus *sec;
1509     int i;
1510 
1511     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1512         dev = bus->devices[i];
1513         if (dev && dev->intx_routing_notifier) {
1514             dev->intx_routing_notifier(dev);
1515         }
1516     }
1517 
1518     QLIST_FOREACH(sec, &bus->child, sibling) {
1519         pci_bus_fire_intx_routing_notifier(sec);
1520     }
1521 }
1522 
1523 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1524                                           PCIINTxRoutingNotifier notifier)
1525 {
1526     dev->intx_routing_notifier = notifier;
1527 }
1528 
1529 /*
1530  * PCI-to-PCI bridge specification
1531  * 9.1: Interrupt routing. Table 9-1
1532  *
1533  * the PCI Express Base Specification, Revision 2.1
1534  * 2.2.8.1: INTx interrutp signaling - Rules
1535  *          the Implementation Note
1536  *          Table 2-20
1537  */
1538 /*
1539  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1540  * 0-origin unlike PCI interrupt pin register.
1541  */
1542 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1543 {
1544     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1545 }
1546 
1547 /***********************************************************/
1548 /* monitor info on PCI */
1549 
1550 typedef struct {
1551     uint16_t class;
1552     const char *desc;
1553     const char *fw_name;
1554     uint16_t fw_ign_bits;
1555 } pci_class_desc;
1556 
1557 static const pci_class_desc pci_class_descriptions[] =
1558 {
1559     { 0x0001, "VGA controller", "display"},
1560     { 0x0100, "SCSI controller", "scsi"},
1561     { 0x0101, "IDE controller", "ide"},
1562     { 0x0102, "Floppy controller", "fdc"},
1563     { 0x0103, "IPI controller", "ipi"},
1564     { 0x0104, "RAID controller", "raid"},
1565     { 0x0106, "SATA controller"},
1566     { 0x0107, "SAS controller"},
1567     { 0x0180, "Storage controller"},
1568     { 0x0200, "Ethernet controller", "ethernet"},
1569     { 0x0201, "Token Ring controller", "token-ring"},
1570     { 0x0202, "FDDI controller", "fddi"},
1571     { 0x0203, "ATM controller", "atm"},
1572     { 0x0280, "Network controller"},
1573     { 0x0300, "VGA controller", "display", 0x00ff},
1574     { 0x0301, "XGA controller"},
1575     { 0x0302, "3D controller"},
1576     { 0x0380, "Display controller"},
1577     { 0x0400, "Video controller", "video"},
1578     { 0x0401, "Audio controller", "sound"},
1579     { 0x0402, "Phone"},
1580     { 0x0403, "Audio controller", "sound"},
1581     { 0x0480, "Multimedia controller"},
1582     { 0x0500, "RAM controller", "memory"},
1583     { 0x0501, "Flash controller", "flash"},
1584     { 0x0580, "Memory controller"},
1585     { 0x0600, "Host bridge", "host"},
1586     { 0x0601, "ISA bridge", "isa"},
1587     { 0x0602, "EISA bridge", "eisa"},
1588     { 0x0603, "MC bridge", "mca"},
1589     { 0x0604, "PCI bridge", "pci-bridge"},
1590     { 0x0605, "PCMCIA bridge", "pcmcia"},
1591     { 0x0606, "NUBUS bridge", "nubus"},
1592     { 0x0607, "CARDBUS bridge", "cardbus"},
1593     { 0x0608, "RACEWAY bridge"},
1594     { 0x0680, "Bridge"},
1595     { 0x0700, "Serial port", "serial"},
1596     { 0x0701, "Parallel port", "parallel"},
1597     { 0x0800, "Interrupt controller", "interrupt-controller"},
1598     { 0x0801, "DMA controller", "dma-controller"},
1599     { 0x0802, "Timer", "timer"},
1600     { 0x0803, "RTC", "rtc"},
1601     { 0x0900, "Keyboard", "keyboard"},
1602     { 0x0901, "Pen", "pen"},
1603     { 0x0902, "Mouse", "mouse"},
1604     { 0x0A00, "Dock station", "dock", 0x00ff},
1605     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1606     { 0x0c00, "Fireware contorller", "fireware"},
1607     { 0x0c01, "Access bus controller", "access-bus"},
1608     { 0x0c02, "SSA controller", "ssa"},
1609     { 0x0c03, "USB controller", "usb"},
1610     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1611     { 0x0c05, "SMBus"},
1612     { 0, NULL}
1613 };
1614 
1615 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1616                                                   void (*fn)(PCIBus *b,
1617                                                              PCIDevice *d,
1618                                                              void *opaque),
1619                                                   void *opaque)
1620 {
1621     PCIDevice *d;
1622     int devfn;
1623 
1624     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1625         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1626         if (d) {
1627             fn(bus, d, opaque);
1628         }
1629     }
1630 }
1631 
1632 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1633                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1634                          void *opaque)
1635 {
1636     bus = pci_find_bus_nr(bus, bus_num);
1637 
1638     if (bus) {
1639         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1640     }
1641 }
1642 
1643 static void pci_for_each_device_under_bus(PCIBus *bus,
1644                                           void (*fn)(PCIBus *b, PCIDevice *d,
1645                                                      void *opaque),
1646                                           void *opaque)
1647 {
1648     PCIDevice *d;
1649     int devfn;
1650 
1651     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1652         d = bus->devices[devfn];
1653         if (d) {
1654             fn(bus, d, opaque);
1655         }
1656     }
1657 }
1658 
1659 void pci_for_each_device(PCIBus *bus, int bus_num,
1660                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1661                          void *opaque)
1662 {
1663     bus = pci_find_bus_nr(bus, bus_num);
1664 
1665     if (bus) {
1666         pci_for_each_device_under_bus(bus, fn, opaque);
1667     }
1668 }
1669 
1670 static const pci_class_desc *get_class_desc(int class)
1671 {
1672     const pci_class_desc *desc;
1673 
1674     desc = pci_class_descriptions;
1675     while (desc->desc && class != desc->class) {
1676         desc++;
1677     }
1678 
1679     return desc;
1680 }
1681 
1682 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1683 
1684 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1685 {
1686     PciMemoryRegionList *head = NULL, **tail = &head;
1687     int i;
1688 
1689     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1690         const PCIIORegion *r = &dev->io_regions[i];
1691         PciMemoryRegion *region;
1692 
1693         if (!r->size) {
1694             continue;
1695         }
1696 
1697         region = g_malloc0(sizeof(*region));
1698 
1699         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1700             region->type = g_strdup("io");
1701         } else {
1702             region->type = g_strdup("memory");
1703             region->has_prefetch = true;
1704             region->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1705             region->has_mem_type_64 = true;
1706             region->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1707         }
1708 
1709         region->bar = i;
1710         region->address = r->addr;
1711         region->size = r->size;
1712 
1713         QAPI_LIST_APPEND(tail, region);
1714     }
1715 
1716     return head;
1717 }
1718 
1719 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1720                                            int bus_num)
1721 {
1722     PciBridgeInfo *info;
1723     PciMemoryRange *range;
1724 
1725     info = g_new0(PciBridgeInfo, 1);
1726 
1727     info->bus = g_new0(PciBusInfo, 1);
1728     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1729     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1730     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1731 
1732     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1733     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1734     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1735 
1736     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1737     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1738     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1739 
1740     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1741     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1742     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1743 
1744     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1745         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1746         if (child_bus) {
1747             info->has_devices = true;
1748             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1749         }
1750     }
1751 
1752     return info;
1753 }
1754 
1755 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1756                                            int bus_num)
1757 {
1758     const pci_class_desc *desc;
1759     PciDeviceInfo *info;
1760     uint8_t type;
1761     int class;
1762 
1763     info = g_new0(PciDeviceInfo, 1);
1764     info->bus = bus_num;
1765     info->slot = PCI_SLOT(dev->devfn);
1766     info->function = PCI_FUNC(dev->devfn);
1767 
1768     info->class_info = g_new0(PciDeviceClass, 1);
1769     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1770     info->class_info->q_class = class;
1771     desc = get_class_desc(class);
1772     if (desc->desc) {
1773         info->class_info->has_desc = true;
1774         info->class_info->desc = g_strdup(desc->desc);
1775     }
1776 
1777     info->id = g_new0(PciDeviceId, 1);
1778     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1779     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1780     info->regions = qmp_query_pci_regions(dev);
1781     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1782 
1783     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1784     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1785         info->has_irq = true;
1786         info->irq = dev->config[PCI_INTERRUPT_LINE];
1787     }
1788 
1789     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1790     if (type == PCI_HEADER_TYPE_BRIDGE) {
1791         info->has_pci_bridge = true;
1792         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1793     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1794         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1795         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1796         info->id->subsystem_vendor =
1797             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1798     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1799         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1800         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1801         info->id->subsystem_vendor =
1802             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1803     }
1804 
1805     return info;
1806 }
1807 
1808 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1809 {
1810     PciDeviceInfoList *head = NULL, **tail = &head;
1811     PCIDevice *dev;
1812     int devfn;
1813 
1814     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1815         dev = bus->devices[devfn];
1816         if (dev) {
1817             QAPI_LIST_APPEND(tail, qmp_query_pci_device(dev, bus, bus_num));
1818         }
1819     }
1820 
1821     return head;
1822 }
1823 
1824 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1825 {
1826     PciInfo *info = NULL;
1827 
1828     bus = pci_find_bus_nr(bus, bus_num);
1829     if (bus) {
1830         info = g_malloc0(sizeof(*info));
1831         info->bus = bus_num;
1832         info->devices = qmp_query_pci_devices(bus, bus_num);
1833     }
1834 
1835     return info;
1836 }
1837 
1838 PciInfoList *qmp_query_pci(Error **errp)
1839 {
1840     PciInfoList *head = NULL, **tail = &head;
1841     PCIHostState *host_bridge;
1842 
1843     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1844         QAPI_LIST_APPEND(tail,
1845                          qmp_query_pci_bus(host_bridge->bus,
1846                                            pci_bus_num(host_bridge->bus)));
1847     }
1848 
1849     return head;
1850 }
1851 
1852 /* Initialize a PCI NIC.  */
1853 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1854                                const char *default_model,
1855                                const char *default_devaddr)
1856 {
1857     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1858     GSList *list;
1859     GPtrArray *pci_nic_models;
1860     PCIBus *bus;
1861     PCIDevice *pci_dev;
1862     DeviceState *dev;
1863     int devfn;
1864     int i;
1865     int dom, busnr;
1866     unsigned slot;
1867 
1868     if (nd->model && !strcmp(nd->model, "virtio")) {
1869         g_free(nd->model);
1870         nd->model = g_strdup("virtio-net-pci");
1871     }
1872 
1873     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1874     pci_nic_models = g_ptr_array_new();
1875     while (list) {
1876         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1877                                              TYPE_DEVICE);
1878         GSList *next;
1879         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1880             dc->user_creatable) {
1881             const char *name = object_class_get_name(list->data);
1882             /*
1883              * A network device might also be something else than a NIC, see
1884              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1885              * property, too. Unfortunately, some devices like virtio-net only
1886              * create this property during instance_init, so we have to create
1887              * a temporary instance here to be able to check it.
1888              */
1889             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1890             if (object_property_find(obj, "netdev")) {
1891                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1892             }
1893             object_unref(obj);
1894         }
1895         next = list->next;
1896         g_slist_free_1(list);
1897         list = next;
1898     }
1899     g_ptr_array_add(pci_nic_models, NULL);
1900 
1901     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1902         exit(0);
1903     }
1904 
1905     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1906                             default_model);
1907     if (i < 0) {
1908         exit(1);
1909     }
1910 
1911     if (!rootbus) {
1912         error_report("No primary PCI bus");
1913         exit(1);
1914     }
1915 
1916     assert(!rootbus->parent_dev);
1917 
1918     if (!devaddr) {
1919         devfn = -1;
1920         busnr = 0;
1921     } else {
1922         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1923             error_report("Invalid PCI device address %s for device %s",
1924                          devaddr, nd->model);
1925             exit(1);
1926         }
1927 
1928         if (dom != 0) {
1929             error_report("No support for non-zero PCI domains");
1930             exit(1);
1931         }
1932 
1933         devfn = PCI_DEVFN(slot, 0);
1934     }
1935 
1936     bus = pci_find_bus_nr(rootbus, busnr);
1937     if (!bus) {
1938         error_report("Invalid PCI device address %s for device %s",
1939                      devaddr, nd->model);
1940         exit(1);
1941     }
1942 
1943     pci_dev = pci_new(devfn, nd->model);
1944     dev = &pci_dev->qdev;
1945     qdev_set_nic_properties(dev, nd);
1946     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1947     g_ptr_array_free(pci_nic_models, true);
1948     return pci_dev;
1949 }
1950 
1951 PCIDevice *pci_vga_init(PCIBus *bus)
1952 {
1953     switch (vga_interface_type) {
1954     case VGA_CIRRUS:
1955         return pci_create_simple(bus, -1, "cirrus-vga");
1956     case VGA_QXL:
1957         return pci_create_simple(bus, -1, "qxl-vga");
1958     case VGA_STD:
1959         return pci_create_simple(bus, -1, "VGA");
1960     case VGA_VMWARE:
1961         return pci_create_simple(bus, -1, "vmware-svga");
1962     case VGA_VIRTIO:
1963         return pci_create_simple(bus, -1, "virtio-vga");
1964     case VGA_NONE:
1965     default: /* Other non-PCI types. Checking for unsupported types is already
1966                 done in vl.c. */
1967         return NULL;
1968     }
1969 }
1970 
1971 /* Whether a given bus number is in range of the secondary
1972  * bus of the given bridge device. */
1973 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1974 {
1975     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1976              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1977         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1978         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1979 }
1980 
1981 /* Whether a given bus number is in a range of a root bus */
1982 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1983 {
1984     int i;
1985 
1986     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1987         PCIDevice *dev = bus->devices[i];
1988 
1989         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1990             if (pci_secondary_bus_in_range(dev, bus_num)) {
1991                 return true;
1992             }
1993         }
1994     }
1995 
1996     return false;
1997 }
1998 
1999 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2000 {
2001     PCIBus *sec;
2002 
2003     if (!bus) {
2004         return NULL;
2005     }
2006 
2007     if (pci_bus_num(bus) == bus_num) {
2008         return bus;
2009     }
2010 
2011     /* Consider all bus numbers in range for the host pci bridge. */
2012     if (!pci_bus_is_root(bus) &&
2013         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2014         return NULL;
2015     }
2016 
2017     /* try child bus */
2018     for (; bus; bus = sec) {
2019         QLIST_FOREACH(sec, &bus->child, sibling) {
2020             if (pci_bus_num(sec) == bus_num) {
2021                 return sec;
2022             }
2023             /* PXB buses assumed to be children of bus 0 */
2024             if (pci_bus_is_root(sec)) {
2025                 if (pci_root_bus_in_range(sec, bus_num)) {
2026                     break;
2027                 }
2028             } else {
2029                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2030                     break;
2031                 }
2032             }
2033         }
2034     }
2035 
2036     return NULL;
2037 }
2038 
2039 void pci_for_each_bus_depth_first(PCIBus *bus,
2040                                   void *(*begin)(PCIBus *bus, void *parent_state),
2041                                   void (*end)(PCIBus *bus, void *state),
2042                                   void *parent_state)
2043 {
2044     PCIBus *sec;
2045     void *state;
2046 
2047     if (!bus) {
2048         return;
2049     }
2050 
2051     if (begin) {
2052         state = begin(bus, parent_state);
2053     } else {
2054         state = parent_state;
2055     }
2056 
2057     QLIST_FOREACH(sec, &bus->child, sibling) {
2058         pci_for_each_bus_depth_first(sec, begin, end, state);
2059     }
2060 
2061     if (end) {
2062         end(bus, state);
2063     }
2064 }
2065 
2066 
2067 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2068 {
2069     bus = pci_find_bus_nr(bus, bus_num);
2070 
2071     if (!bus)
2072         return NULL;
2073 
2074     return bus->devices[devfn];
2075 }
2076 
2077 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2078 {
2079     PCIDevice *pci_dev = (PCIDevice *)qdev;
2080     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2081     ObjectClass *klass = OBJECT_CLASS(pc);
2082     Error *local_err = NULL;
2083     bool is_default_rom;
2084     uint16_t class_id;
2085 
2086     /* initialize cap_present for pci_is_express() and pci_config_size(),
2087      * Note that hybrid PCIs are not set automatically and need to manage
2088      * QEMU_PCI_CAP_EXPRESS manually */
2089     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2090        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2091         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2092     }
2093 
2094     pci_dev = do_pci_register_device(pci_dev,
2095                                      object_get_typename(OBJECT(qdev)),
2096                                      pci_dev->devfn, errp);
2097     if (pci_dev == NULL)
2098         return;
2099 
2100     if (pc->realize) {
2101         pc->realize(pci_dev, &local_err);
2102         if (local_err) {
2103             error_propagate(errp, local_err);
2104             do_pci_unregister_device(pci_dev);
2105             return;
2106         }
2107     }
2108 
2109     if (pci_dev->failover_pair_id) {
2110         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2111             error_setg(errp, "failover primary device must be on "
2112                              "PCIExpress bus");
2113             pci_qdev_unrealize(DEVICE(pci_dev));
2114             return;
2115         }
2116         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2117         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2118             error_setg(errp, "failover primary device is not an "
2119                              "Ethernet device");
2120             pci_qdev_unrealize(DEVICE(pci_dev));
2121             return;
2122         }
2123         if (!(pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2124             && (PCI_FUNC(pci_dev->devfn) == 0)) {
2125             qdev->allow_unplug_during_migration = true;
2126         } else {
2127             error_setg(errp, "failover: primary device must be in its own "
2128                               "PCI slot");
2129             pci_qdev_unrealize(DEVICE(pci_dev));
2130             return;
2131         }
2132         qdev->allow_unplug_during_migration = true;
2133     }
2134 
2135     /* rom loading */
2136     is_default_rom = false;
2137     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2138         pci_dev->romfile = g_strdup(pc->romfile);
2139         is_default_rom = true;
2140     }
2141 
2142     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2143     if (local_err) {
2144         error_propagate(errp, local_err);
2145         pci_qdev_unrealize(DEVICE(pci_dev));
2146         return;
2147     }
2148 }
2149 
2150 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2151                                  const char *name)
2152 {
2153     DeviceState *dev;
2154 
2155     dev = qdev_new(name);
2156     qdev_prop_set_int32(dev, "addr", devfn);
2157     qdev_prop_set_bit(dev, "multifunction", multifunction);
2158     return PCI_DEVICE(dev);
2159 }
2160 
2161 PCIDevice *pci_new(int devfn, const char *name)
2162 {
2163     return pci_new_multifunction(devfn, false, name);
2164 }
2165 
2166 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2167 {
2168     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2169 }
2170 
2171 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2172                                            bool multifunction,
2173                                            const char *name)
2174 {
2175     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2176     pci_realize_and_unref(dev, bus, &error_fatal);
2177     return dev;
2178 }
2179 
2180 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2181 {
2182     return pci_create_simple_multifunction(bus, devfn, false, name);
2183 }
2184 
2185 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2186 {
2187     int offset = PCI_CONFIG_HEADER_SIZE;
2188     int i;
2189     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2190         if (pdev->used[i])
2191             offset = i + 1;
2192         else if (i - offset + 1 == size)
2193             return offset;
2194     }
2195     return 0;
2196 }
2197 
2198 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2199                                         uint8_t *prev_p)
2200 {
2201     uint8_t next, prev;
2202 
2203     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2204         return 0;
2205 
2206     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2207          prev = next + PCI_CAP_LIST_NEXT)
2208         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2209             break;
2210 
2211     if (prev_p)
2212         *prev_p = prev;
2213     return next;
2214 }
2215 
2216 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2217 {
2218     uint8_t next, prev, found = 0;
2219 
2220     if (!(pdev->used[offset])) {
2221         return 0;
2222     }
2223 
2224     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2225 
2226     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2227          prev = next + PCI_CAP_LIST_NEXT) {
2228         if (next <= offset && next > found) {
2229             found = next;
2230         }
2231     }
2232     return found;
2233 }
2234 
2235 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2236    This is needed for an option rom which is used for more than one device. */
2237 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2238 {
2239     uint16_t vendor_id;
2240     uint16_t device_id;
2241     uint16_t rom_vendor_id;
2242     uint16_t rom_device_id;
2243     uint16_t rom_magic;
2244     uint16_t pcir_offset;
2245     uint8_t checksum;
2246 
2247     /* Words in rom data are little endian (like in PCI configuration),
2248        so they can be read / written with pci_get_word / pci_set_word. */
2249 
2250     /* Only a valid rom will be patched. */
2251     rom_magic = pci_get_word(ptr);
2252     if (rom_magic != 0xaa55) {
2253         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2254         return;
2255     }
2256     pcir_offset = pci_get_word(ptr + 0x18);
2257     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2258         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2259         return;
2260     }
2261 
2262     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2263     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2264     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2265     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2266 
2267     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2268                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2269 
2270     checksum = ptr[6];
2271 
2272     if (vendor_id != rom_vendor_id) {
2273         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2274         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2275         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2276         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2277         ptr[6] = checksum;
2278         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2279     }
2280 
2281     if (device_id != rom_device_id) {
2282         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2283         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2284         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2285         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2286         ptr[6] = checksum;
2287         pci_set_word(ptr + pcir_offset + 6, device_id);
2288     }
2289 }
2290 
2291 /* Add an option rom for the device */
2292 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2293                                Error **errp)
2294 {
2295     int size;
2296     char *path;
2297     void *ptr;
2298     char name[32];
2299     const VMStateDescription *vmsd;
2300 
2301     if (!pdev->romfile)
2302         return;
2303     if (strlen(pdev->romfile) == 0)
2304         return;
2305 
2306     if (!pdev->rom_bar) {
2307         /*
2308          * Load rom via fw_cfg instead of creating a rom bar,
2309          * for 0.11 compatibility.
2310          */
2311         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2312 
2313         /*
2314          * Hot-plugged devices can't use the option ROM
2315          * if the rom bar is disabled.
2316          */
2317         if (DEVICE(pdev)->hotplugged) {
2318             error_setg(errp, "Hot-plugged device without ROM bar"
2319                        " can't have an option ROM");
2320             return;
2321         }
2322 
2323         if (class == 0x0300) {
2324             rom_add_vga(pdev->romfile);
2325         } else {
2326             rom_add_option(pdev->romfile, -1);
2327         }
2328         return;
2329     }
2330 
2331     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2332     if (path == NULL) {
2333         path = g_strdup(pdev->romfile);
2334     }
2335 
2336     size = get_image_size(path);
2337     if (size < 0) {
2338         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2339         g_free(path);
2340         return;
2341     } else if (size == 0) {
2342         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2343         g_free(path);
2344         return;
2345     }
2346     size = pow2ceil(size);
2347 
2348     vmsd = qdev_get_vmsd(DEVICE(pdev));
2349 
2350     if (vmsd) {
2351         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2352     } else {
2353         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2354     }
2355     pdev->has_rom = true;
2356     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2357     ptr = memory_region_get_ram_ptr(&pdev->rom);
2358     if (load_image_size(path, ptr, size) < 0) {
2359         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2360         g_free(path);
2361         return;
2362     }
2363     g_free(path);
2364 
2365     if (is_default_rom) {
2366         /* Only the default rom images will be patched (if needed). */
2367         pci_patch_ids(pdev, ptr, size);
2368     }
2369 
2370     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2371 }
2372 
2373 static void pci_del_option_rom(PCIDevice *pdev)
2374 {
2375     if (!pdev->has_rom)
2376         return;
2377 
2378     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2379     pdev->has_rom = false;
2380 }
2381 
2382 /*
2383  * On success, pci_add_capability() returns a positive value
2384  * that the offset of the pci capability.
2385  * On failure, it sets an error and returns a negative error
2386  * code.
2387  */
2388 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2389                        uint8_t offset, uint8_t size,
2390                        Error **errp)
2391 {
2392     uint8_t *config;
2393     int i, overlapping_cap;
2394 
2395     if (!offset) {
2396         offset = pci_find_space(pdev, size);
2397         /* out of PCI config space is programming error */
2398         assert(offset);
2399     } else {
2400         /* Verify that capabilities don't overlap.  Note: device assignment
2401          * depends on this check to verify that the device is not broken.
2402          * Should never trigger for emulated devices, but it's helpful
2403          * for debugging these. */
2404         for (i = offset; i < offset + size; i++) {
2405             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2406             if (overlapping_cap) {
2407                 error_setg(errp, "%s:%02x:%02x.%x "
2408                            "Attempt to add PCI capability %x at offset "
2409                            "%x overlaps existing capability %x at offset %x",
2410                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2411                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2412                            cap_id, offset, overlapping_cap, i);
2413                 return -EINVAL;
2414             }
2415         }
2416     }
2417 
2418     config = pdev->config + offset;
2419     config[PCI_CAP_LIST_ID] = cap_id;
2420     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2421     pdev->config[PCI_CAPABILITY_LIST] = offset;
2422     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2423     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2424     /* Make capability read-only by default */
2425     memset(pdev->wmask + offset, 0, size);
2426     /* Check capability by default */
2427     memset(pdev->cmask + offset, 0xFF, size);
2428     return offset;
2429 }
2430 
2431 /* Unlink capability from the pci config space. */
2432 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2433 {
2434     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2435     if (!offset)
2436         return;
2437     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2438     /* Make capability writable again */
2439     memset(pdev->wmask + offset, 0xff, size);
2440     memset(pdev->w1cmask + offset, 0, size);
2441     /* Clear cmask as device-specific registers can't be checked */
2442     memset(pdev->cmask + offset, 0, size);
2443     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2444 
2445     if (!pdev->config[PCI_CAPABILITY_LIST])
2446         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2447 }
2448 
2449 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2450 {
2451     return pci_find_capability_list(pdev, cap_id, NULL);
2452 }
2453 
2454 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2455 {
2456     PCIDevice *d = (PCIDevice *)dev;
2457     const pci_class_desc *desc;
2458     char ctxt[64];
2459     PCIIORegion *r;
2460     int i, class;
2461 
2462     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2463     desc = pci_class_descriptions;
2464     while (desc->desc && class != desc->class)
2465         desc++;
2466     if (desc->desc) {
2467         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2468     } else {
2469         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2470     }
2471 
2472     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2473                    "pci id %04x:%04x (sub %04x:%04x)\n",
2474                    indent, "", ctxt, pci_dev_bus_num(d),
2475                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2476                    pci_get_word(d->config + PCI_VENDOR_ID),
2477                    pci_get_word(d->config + PCI_DEVICE_ID),
2478                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2479                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2480     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2481         r = &d->io_regions[i];
2482         if (!r->size)
2483             continue;
2484         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2485                        " [0x%"FMT_PCIBUS"]\n",
2486                        indent, "",
2487                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2488                        r->addr, r->addr + r->size - 1);
2489     }
2490 }
2491 
2492 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2493 {
2494     PCIDevice *d = (PCIDevice *)dev;
2495     const char *name = NULL;
2496     const pci_class_desc *desc =  pci_class_descriptions;
2497     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2498 
2499     while (desc->desc &&
2500           (class & ~desc->fw_ign_bits) !=
2501           (desc->class & ~desc->fw_ign_bits)) {
2502         desc++;
2503     }
2504 
2505     if (desc->desc) {
2506         name = desc->fw_name;
2507     }
2508 
2509     if (name) {
2510         pstrcpy(buf, len, name);
2511     } else {
2512         snprintf(buf, len, "pci%04x,%04x",
2513                  pci_get_word(d->config + PCI_VENDOR_ID),
2514                  pci_get_word(d->config + PCI_DEVICE_ID));
2515     }
2516 
2517     return buf;
2518 }
2519 
2520 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2521 {
2522     PCIDevice *d = (PCIDevice *)dev;
2523     char path[50], name[33];
2524     int off;
2525 
2526     off = snprintf(path, sizeof(path), "%s@%x",
2527                    pci_dev_fw_name(dev, name, sizeof name),
2528                    PCI_SLOT(d->devfn));
2529     if (PCI_FUNC(d->devfn))
2530         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2531     return g_strdup(path);
2532 }
2533 
2534 static char *pcibus_get_dev_path(DeviceState *dev)
2535 {
2536     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2537     PCIDevice *t;
2538     int slot_depth;
2539     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2540      * 00 is added here to make this format compatible with
2541      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2542      * Slot.Function list specifies the slot and function numbers for all
2543      * devices on the path from root to the specific device. */
2544     const char *root_bus_path;
2545     int root_bus_len;
2546     char slot[] = ":SS.F";
2547     int slot_len = sizeof slot - 1 /* For '\0' */;
2548     int path_len;
2549     char *path, *p;
2550     int s;
2551 
2552     root_bus_path = pci_root_bus_path(d);
2553     root_bus_len = strlen(root_bus_path);
2554 
2555     /* Calculate # of slots on path between device and root. */;
2556     slot_depth = 0;
2557     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2558         ++slot_depth;
2559     }
2560 
2561     path_len = root_bus_len + slot_len * slot_depth;
2562 
2563     /* Allocate memory, fill in the terminating null byte. */
2564     path = g_malloc(path_len + 1 /* For '\0' */);
2565     path[path_len] = '\0';
2566 
2567     memcpy(path, root_bus_path, root_bus_len);
2568 
2569     /* Fill in slot numbers. We walk up from device to root, so need to print
2570      * them in the reverse order, last to first. */
2571     p = path + path_len;
2572     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2573         p -= slot_len;
2574         s = snprintf(slot, sizeof slot, ":%02x.%x",
2575                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2576         assert(s == slot_len);
2577         memcpy(p, slot, slot_len);
2578     }
2579 
2580     return path;
2581 }
2582 
2583 static int pci_qdev_find_recursive(PCIBus *bus,
2584                                    const char *id, PCIDevice **pdev)
2585 {
2586     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2587     if (!qdev) {
2588         return -ENODEV;
2589     }
2590 
2591     /* roughly check if given qdev is pci device */
2592     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2593         *pdev = PCI_DEVICE(qdev);
2594         return 0;
2595     }
2596     return -EINVAL;
2597 }
2598 
2599 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2600 {
2601     PCIHostState *host_bridge;
2602     int rc = -ENODEV;
2603 
2604     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2605         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2606         if (!tmp) {
2607             rc = 0;
2608             break;
2609         }
2610         if (tmp != -ENODEV) {
2611             rc = tmp;
2612         }
2613     }
2614 
2615     return rc;
2616 }
2617 
2618 MemoryRegion *pci_address_space(PCIDevice *dev)
2619 {
2620     return pci_get_bus(dev)->address_space_mem;
2621 }
2622 
2623 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2624 {
2625     return pci_get_bus(dev)->address_space_io;
2626 }
2627 
2628 static void pci_device_class_init(ObjectClass *klass, void *data)
2629 {
2630     DeviceClass *k = DEVICE_CLASS(klass);
2631 
2632     k->realize = pci_qdev_realize;
2633     k->unrealize = pci_qdev_unrealize;
2634     k->bus_type = TYPE_PCI_BUS;
2635     device_class_set_props(k, pci_props);
2636 }
2637 
2638 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2639 {
2640     if (!object_class_is_abstract(klass)) {
2641         ObjectClass *conventional =
2642             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2643         ObjectClass *pcie =
2644             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2645         assert(conventional || pcie);
2646     }
2647 }
2648 
2649 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2650 {
2651     PCIBus *bus = pci_get_bus(dev);
2652     PCIBus *iommu_bus = bus;
2653     uint8_t devfn = dev->devfn;
2654 
2655     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2656         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2657 
2658         /*
2659          * The requester ID of the provided device may be aliased, as seen from
2660          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2661          * requester ID to provide a unique AddressSpace for devices, but
2662          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2663          * to-PCI bridge creates and accepts transactions on behalf of down-
2664          * stream devices.  When doing so, all downstream devices are masked
2665          * (aliased) behind a single requester ID.  The requester ID used
2666          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2667          * bridges, with a PCIe capability indicating such, follow the
2668          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2669          * where the bridge uses the seconary bus as the bridge portion of the
2670          * requester ID and devfn of 00.0.  For other bridges, typically those
2671          * found on the root complex such as the dmi-to-pci-bridge, we follow
2672          * the convention of typical bare-metal hardware, which uses the
2673          * requester ID of the bridge itself.  There are device specific
2674          * exceptions to these rules, but these are the defaults that the
2675          * Linux kernel uses when determining DMA aliases itself and believed
2676          * to be true for the bare metal equivalents of the devices emulated
2677          * in QEMU.
2678          */
2679         if (!pci_bus_is_express(iommu_bus)) {
2680             PCIDevice *parent = iommu_bus->parent_dev;
2681 
2682             if (pci_is_express(parent) &&
2683                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2684                 devfn = PCI_DEVFN(0, 0);
2685                 bus = iommu_bus;
2686             } else {
2687                 devfn = parent->devfn;
2688                 bus = parent_bus;
2689             }
2690         }
2691 
2692         iommu_bus = parent_bus;
2693     }
2694     if (iommu_bus && iommu_bus->iommu_fn) {
2695         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2696     }
2697     return &address_space_memory;
2698 }
2699 
2700 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2701 {
2702     bus->iommu_fn = fn;
2703     bus->iommu_opaque = opaque;
2704 }
2705 
2706 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2707 {
2708     Range *range = opaque;
2709     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2710     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2711     int i;
2712 
2713     if (!(cmd & PCI_COMMAND_MEMORY)) {
2714         return;
2715     }
2716 
2717     if (pc->is_bridge) {
2718         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2719         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2720 
2721         base = MAX(base, 0x1ULL << 32);
2722 
2723         if (limit >= base) {
2724             Range pref_range;
2725             range_set_bounds(&pref_range, base, limit);
2726             range_extend(range, &pref_range);
2727         }
2728     }
2729     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2730         PCIIORegion *r = &dev->io_regions[i];
2731         pcibus_t lob, upb;
2732         Range region_range;
2733 
2734         if (!r->size ||
2735             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2736             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2737             continue;
2738         }
2739 
2740         lob = pci_bar_address(dev, i, r->type, r->size);
2741         upb = lob + r->size - 1;
2742         if (lob == PCI_BAR_UNMAPPED) {
2743             continue;
2744         }
2745 
2746         lob = MAX(lob, 0x1ULL << 32);
2747 
2748         if (upb >= lob) {
2749             range_set_bounds(&region_range, lob, upb);
2750             range_extend(range, &region_range);
2751         }
2752     }
2753 }
2754 
2755 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2756 {
2757     range_make_empty(range);
2758     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2759 }
2760 
2761 static bool pcie_has_upstream_port(PCIDevice *dev)
2762 {
2763     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2764 
2765     /* Device associated with an upstream port.
2766      * As there are several types of these, it's easier to check the
2767      * parent device: upstream ports are always connected to
2768      * root or downstream ports.
2769      */
2770     return parent_dev &&
2771         pci_is_express(parent_dev) &&
2772         parent_dev->exp.exp_cap &&
2773         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2774          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2775 }
2776 
2777 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2778 {
2779     PCIBus *bus = pci_get_bus(pci_dev);
2780 
2781     if(pcie_has_upstream_port(pci_dev)) {
2782         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2783         return bus->devices[0];
2784     } else {
2785         /* Other bus types might support multiple devices at slots 0-31 */
2786         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2787     }
2788 }
2789 
2790 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2791 {
2792     MSIMessage msg;
2793     if (msix_enabled(dev)) {
2794         msg = msix_get_message(dev, vector);
2795     } else if (msi_enabled(dev)) {
2796         msg = msi_get_message(dev, vector);
2797     } else {
2798         /* Should never happen */
2799         error_report("%s: unknown interrupt type", __func__);
2800         abort();
2801     }
2802     return msg;
2803 }
2804 
2805 static const TypeInfo pci_device_type_info = {
2806     .name = TYPE_PCI_DEVICE,
2807     .parent = TYPE_DEVICE,
2808     .instance_size = sizeof(PCIDevice),
2809     .abstract = true,
2810     .class_size = sizeof(PCIDeviceClass),
2811     .class_init = pci_device_class_init,
2812     .class_base_init = pci_device_class_base_init,
2813 };
2814 
2815 static void pci_register_types(void)
2816 {
2817     type_register_static(&pci_bus_info);
2818     type_register_static(&pcie_bus_info);
2819     type_register_static(&conventional_pci_interface_info);
2820     type_register_static(&pcie_interface_info);
2821     type_register_static(&pci_device_type_info);
2822 }
2823 
2824 type_init(pci_register_types)
2825