xref: /openbmc/qemu/hw/pci/pci.c (revision 4011a686)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu-common.h"
27 #include "qemu/datadir.h"
28 #include "qemu/units.h"
29 #include "hw/irq.h"
30 #include "hw/pci/pci.h"
31 #include "hw/pci/pci_bridge.h"
32 #include "hw/pci/pci_bus.h"
33 #include "hw/pci/pci_host.h"
34 #include "hw/qdev-properties.h"
35 #include "hw/qdev-properties-system.h"
36 #include "migration/qemu-file-types.h"
37 #include "migration/vmstate.h"
38 #include "monitor/monitor.h"
39 #include "net/net.h"
40 #include "sysemu/numa.h"
41 #include "sysemu/sysemu.h"
42 #include "hw/loader.h"
43 #include "qemu/error-report.h"
44 #include "qemu/range.h"
45 #include "trace.h"
46 #include "hw/pci/msi.h"
47 #include "hw/pci/msix.h"
48 #include "exec/address-spaces.h"
49 #include "hw/hotplug.h"
50 #include "hw/boards.h"
51 #include "qapi/error.h"
52 #include "qapi/qapi-commands-pci.h"
53 #include "qemu/cutils.h"
54 
55 //#define DEBUG_PCI
56 #ifdef DEBUG_PCI
57 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
58 #else
59 # define PCI_DPRINTF(format, ...)       do { } while (0)
60 #endif
61 
62 bool pci_available = true;
63 
64 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset(BusState *qbus);
68 
69 static Property pci_props[] = {
70     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
71     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
72     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
73     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
74     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
75                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
76     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
77                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
78     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
79                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
80     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
81                        failover_pair_id),
82     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
83     DEFINE_PROP_END_OF_LIST()
84 };
85 
86 static const VMStateDescription vmstate_pcibus = {
87     .name = "PCIBUS",
88     .version_id = 1,
89     .minimum_version_id = 1,
90     .fields = (VMStateField[]) {
91         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
92         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
93                              nirq, 0, vmstate_info_int32,
94                              int32_t),
95         VMSTATE_END_OF_LIST()
96     }
97 };
98 
99 static void pci_init_bus_master(PCIDevice *pci_dev)
100 {
101     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
102 
103     memory_region_init_alias(&pci_dev->bus_master_enable_region,
104                              OBJECT(pci_dev), "bus master",
105                              dma_as->root, 0, memory_region_size(dma_as->root));
106     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
107     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
108                                 &pci_dev->bus_master_enable_region);
109 }
110 
111 static void pcibus_machine_done(Notifier *notifier, void *data)
112 {
113     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
114     int i;
115 
116     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
117         if (bus->devices[i]) {
118             pci_init_bus_master(bus->devices[i]);
119         }
120     }
121 }
122 
123 static void pci_bus_realize(BusState *qbus, Error **errp)
124 {
125     PCIBus *bus = PCI_BUS(qbus);
126 
127     bus->machine_done.notify = pcibus_machine_done;
128     qemu_add_machine_init_done_notifier(&bus->machine_done);
129 
130     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
131 }
132 
133 static void pcie_bus_realize(BusState *qbus, Error **errp)
134 {
135     PCIBus *bus = PCI_BUS(qbus);
136     Error *local_err = NULL;
137 
138     pci_bus_realize(qbus, &local_err);
139     if (local_err) {
140         error_propagate(errp, local_err);
141         return;
142     }
143 
144     /*
145      * A PCI-E bus can support extended config space if it's the root
146      * bus, or if the bus/bridge above it does as well
147      */
148     if (pci_bus_is_root(bus)) {
149         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
150     } else {
151         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
152 
153         if (pci_bus_allows_extended_config_space(parent_bus)) {
154             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
155         }
156     }
157 }
158 
159 static void pci_bus_unrealize(BusState *qbus)
160 {
161     PCIBus *bus = PCI_BUS(qbus);
162 
163     qemu_remove_machine_init_done_notifier(&bus->machine_done);
164 
165     vmstate_unregister(NULL, &vmstate_pcibus, bus);
166 }
167 
168 static int pcibus_num(PCIBus *bus)
169 {
170     if (pci_bus_is_root(bus)) {
171         return 0; /* pci host bridge */
172     }
173     return bus->parent_dev->config[PCI_SECONDARY_BUS];
174 }
175 
176 static uint16_t pcibus_numa_node(PCIBus *bus)
177 {
178     return NUMA_NODE_UNASSIGNED;
179 }
180 
181 static void pci_bus_class_init(ObjectClass *klass, void *data)
182 {
183     BusClass *k = BUS_CLASS(klass);
184     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
185 
186     k->print_dev = pcibus_dev_print;
187     k->get_dev_path = pcibus_get_dev_path;
188     k->get_fw_dev_path = pcibus_get_fw_dev_path;
189     k->realize = pci_bus_realize;
190     k->unrealize = pci_bus_unrealize;
191     k->reset = pcibus_reset;
192 
193     pbc->bus_num = pcibus_num;
194     pbc->numa_node = pcibus_numa_node;
195 }
196 
197 static const TypeInfo pci_bus_info = {
198     .name = TYPE_PCI_BUS,
199     .parent = TYPE_BUS,
200     .instance_size = sizeof(PCIBus),
201     .class_size = sizeof(PCIBusClass),
202     .class_init = pci_bus_class_init,
203 };
204 
205 static const TypeInfo pcie_interface_info = {
206     .name          = INTERFACE_PCIE_DEVICE,
207     .parent        = TYPE_INTERFACE,
208 };
209 
210 static const TypeInfo conventional_pci_interface_info = {
211     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
212     .parent        = TYPE_INTERFACE,
213 };
214 
215 static void pcie_bus_class_init(ObjectClass *klass, void *data)
216 {
217     BusClass *k = BUS_CLASS(klass);
218 
219     k->realize = pcie_bus_realize;
220 }
221 
222 static const TypeInfo pcie_bus_info = {
223     .name = TYPE_PCIE_BUS,
224     .parent = TYPE_PCI_BUS,
225     .class_init = pcie_bus_class_init,
226 };
227 
228 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
229 static void pci_update_mappings(PCIDevice *d);
230 static void pci_irq_handler(void *opaque, int irq_num, int level);
231 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
232 static void pci_del_option_rom(PCIDevice *pdev);
233 
234 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
235 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
236 
237 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
238 
239 int pci_bar(PCIDevice *d, int reg)
240 {
241     uint8_t type;
242 
243     if (reg != PCI_ROM_SLOT)
244         return PCI_BASE_ADDRESS_0 + reg * 4;
245 
246     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
247     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
248 }
249 
250 static inline int pci_irq_state(PCIDevice *d, int irq_num)
251 {
252         return (d->irq_state >> irq_num) & 0x1;
253 }
254 
255 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
256 {
257         d->irq_state &= ~(0x1 << irq_num);
258         d->irq_state |= level << irq_num;
259 }
260 
261 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
262 {
263     assert(irq_num >= 0);
264     assert(irq_num < bus->nirq);
265     bus->irq_count[irq_num] += change;
266     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
267 }
268 
269 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
270 {
271     PCIBus *bus;
272     for (;;) {
273         bus = pci_get_bus(pci_dev);
274         irq_num = bus->map_irq(pci_dev, irq_num);
275         if (bus->set_irq)
276             break;
277         pci_dev = bus->parent_dev;
278     }
279     pci_bus_change_irq_level(bus, irq_num, change);
280 }
281 
282 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
283 {
284     assert(irq_num >= 0);
285     assert(irq_num < bus->nirq);
286     return !!bus->irq_count[irq_num];
287 }
288 
289 /* Update interrupt status bit in config space on interrupt
290  * state change. */
291 static void pci_update_irq_status(PCIDevice *dev)
292 {
293     if (dev->irq_state) {
294         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
295     } else {
296         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
297     }
298 }
299 
300 void pci_device_deassert_intx(PCIDevice *dev)
301 {
302     int i;
303     for (i = 0; i < PCI_NUM_PINS; ++i) {
304         pci_irq_handler(dev, i, 0);
305     }
306 }
307 
308 static void pci_do_device_reset(PCIDevice *dev)
309 {
310     int r;
311 
312     pci_device_deassert_intx(dev);
313     assert(dev->irq_state == 0);
314 
315     /* Clear all writable bits */
316     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
317                                  pci_get_word(dev->wmask + PCI_COMMAND) |
318                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
319     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
320                                  pci_get_word(dev->wmask + PCI_STATUS) |
321                                  pci_get_word(dev->w1cmask + PCI_STATUS));
322     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
323     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
324                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
325                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
326     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
327     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
328         PCIIORegion *region = &dev->io_regions[r];
329         if (!region->size) {
330             continue;
331         }
332 
333         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
334             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
335             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
336         } else {
337             pci_set_long(dev->config + pci_bar(dev, r), region->type);
338         }
339     }
340     pci_update_mappings(dev);
341 
342     msi_reset(dev);
343     msix_reset(dev);
344 }
345 
346 /*
347  * This function is called on #RST and FLR.
348  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
349  */
350 void pci_device_reset(PCIDevice *dev)
351 {
352     qdev_reset_all(&dev->qdev);
353     pci_do_device_reset(dev);
354 }
355 
356 /*
357  * Trigger pci bus reset under a given bus.
358  * Called via qbus_reset_all on RST# assert, after the devices
359  * have been reset qdev_reset_all-ed already.
360  */
361 static void pcibus_reset(BusState *qbus)
362 {
363     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
364     int i;
365 
366     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
367         if (bus->devices[i]) {
368             pci_do_device_reset(bus->devices[i]);
369         }
370     }
371 
372     for (i = 0; i < bus->nirq; i++) {
373         assert(bus->irq_count[i] == 0);
374     }
375 }
376 
377 static void pci_host_bus_register(DeviceState *host)
378 {
379     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
380 
381     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
382 }
383 
384 static void pci_host_bus_unregister(DeviceState *host)
385 {
386     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
387 
388     QLIST_REMOVE(host_bridge, next);
389 }
390 
391 PCIBus *pci_device_root_bus(const PCIDevice *d)
392 {
393     PCIBus *bus = pci_get_bus(d);
394 
395     while (!pci_bus_is_root(bus)) {
396         d = bus->parent_dev;
397         assert(d != NULL);
398 
399         bus = pci_get_bus(d);
400     }
401 
402     return bus;
403 }
404 
405 const char *pci_root_bus_path(PCIDevice *dev)
406 {
407     PCIBus *rootbus = pci_device_root_bus(dev);
408     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
409     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
410 
411     assert(host_bridge->bus == rootbus);
412 
413     if (hc->root_bus_path) {
414         return (*hc->root_bus_path)(host_bridge, rootbus);
415     }
416 
417     return rootbus->qbus.name;
418 }
419 
420 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
421                               MemoryRegion *address_space_mem,
422                               MemoryRegion *address_space_io,
423                               uint8_t devfn_min)
424 {
425     assert(PCI_FUNC(devfn_min) == 0);
426     bus->devfn_min = devfn_min;
427     bus->slot_reserved_mask = 0x0;
428     bus->address_space_mem = address_space_mem;
429     bus->address_space_io = address_space_io;
430     bus->flags |= PCI_BUS_IS_ROOT;
431 
432     /* host bridge */
433     QLIST_INIT(&bus->child);
434 
435     pci_host_bus_register(parent);
436 }
437 
438 static void pci_bus_uninit(PCIBus *bus)
439 {
440     pci_host_bus_unregister(BUS(bus)->parent);
441 }
442 
443 bool pci_bus_is_express(PCIBus *bus)
444 {
445     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
446 }
447 
448 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
449                               const char *name,
450                               MemoryRegion *address_space_mem,
451                               MemoryRegion *address_space_io,
452                               uint8_t devfn_min, const char *typename)
453 {
454     qbus_create_inplace(bus, bus_size, typename, parent, name);
455     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
456                       devfn_min);
457 }
458 
459 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
460                          MemoryRegion *address_space_mem,
461                          MemoryRegion *address_space_io,
462                          uint8_t devfn_min, const char *typename)
463 {
464     PCIBus *bus;
465 
466     bus = PCI_BUS(qbus_create(typename, parent, name));
467     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
468                       devfn_min);
469     return bus;
470 }
471 
472 void pci_root_bus_cleanup(PCIBus *bus)
473 {
474     pci_bus_uninit(bus);
475     /* the caller of the unplug hotplug handler will delete this device */
476     qbus_unrealize(BUS(bus));
477 }
478 
479 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
480                   void *irq_opaque, int nirq)
481 {
482     bus->set_irq = set_irq;
483     bus->map_irq = map_irq;
484     bus->irq_opaque = irq_opaque;
485     bus->nirq = nirq;
486     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
487 }
488 
489 void pci_bus_irqs_cleanup(PCIBus *bus)
490 {
491     bus->set_irq = NULL;
492     bus->map_irq = NULL;
493     bus->irq_opaque = NULL;
494     bus->nirq = 0;
495     g_free(bus->irq_count);
496 }
497 
498 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
499                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
500                               void *irq_opaque,
501                               MemoryRegion *address_space_mem,
502                               MemoryRegion *address_space_io,
503                               uint8_t devfn_min, int nirq,
504                               const char *typename)
505 {
506     PCIBus *bus;
507 
508     bus = pci_root_bus_new(parent, name, address_space_mem,
509                            address_space_io, devfn_min, typename);
510     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
511     return bus;
512 }
513 
514 void pci_unregister_root_bus(PCIBus *bus)
515 {
516     pci_bus_irqs_cleanup(bus);
517     pci_root_bus_cleanup(bus);
518 }
519 
520 int pci_bus_num(PCIBus *s)
521 {
522     return PCI_BUS_GET_CLASS(s)->bus_num(s);
523 }
524 
525 int pci_bus_numa_node(PCIBus *bus)
526 {
527     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
528 }
529 
530 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
531                                  const VMStateField *field)
532 {
533     PCIDevice *s = container_of(pv, PCIDevice, config);
534     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
535     uint8_t *config;
536     int i;
537 
538     assert(size == pci_config_size(s));
539     config = g_malloc(size);
540 
541     qemu_get_buffer(f, config, size);
542     for (i = 0; i < size; ++i) {
543         if ((config[i] ^ s->config[i]) &
544             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
545             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
546                          "cmask: %x wmask: %x w1cmask:%x", __func__,
547                          i, config[i], s->config[i],
548                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
549             g_free(config);
550             return -EINVAL;
551         }
552     }
553     memcpy(s->config, config, size);
554 
555     pci_update_mappings(s);
556     if (pc->is_bridge) {
557         PCIBridge *b = PCI_BRIDGE(s);
558         pci_bridge_update_mappings(b);
559     }
560 
561     memory_region_set_enabled(&s->bus_master_enable_region,
562                               pci_get_word(s->config + PCI_COMMAND)
563                               & PCI_COMMAND_MASTER);
564 
565     g_free(config);
566     return 0;
567 }
568 
569 /* just put buffer */
570 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
571                                  const VMStateField *field, JSONWriter *vmdesc)
572 {
573     const uint8_t **v = pv;
574     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
575     qemu_put_buffer(f, *v, size);
576 
577     return 0;
578 }
579 
580 static VMStateInfo vmstate_info_pci_config = {
581     .name = "pci config",
582     .get  = get_pci_config_device,
583     .put  = put_pci_config_device,
584 };
585 
586 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
587                              const VMStateField *field)
588 {
589     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
590     uint32_t irq_state[PCI_NUM_PINS];
591     int i;
592     for (i = 0; i < PCI_NUM_PINS; ++i) {
593         irq_state[i] = qemu_get_be32(f);
594         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
595             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
596                     irq_state[i]);
597             return -EINVAL;
598         }
599     }
600 
601     for (i = 0; i < PCI_NUM_PINS; ++i) {
602         pci_set_irq_state(s, i, irq_state[i]);
603     }
604 
605     return 0;
606 }
607 
608 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
609                              const VMStateField *field, JSONWriter *vmdesc)
610 {
611     int i;
612     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
613 
614     for (i = 0; i < PCI_NUM_PINS; ++i) {
615         qemu_put_be32(f, pci_irq_state(s, i));
616     }
617 
618     return 0;
619 }
620 
621 static VMStateInfo vmstate_info_pci_irq_state = {
622     .name = "pci irq state",
623     .get  = get_pci_irq_state,
624     .put  = put_pci_irq_state,
625 };
626 
627 static bool migrate_is_pcie(void *opaque, int version_id)
628 {
629     return pci_is_express((PCIDevice *)opaque);
630 }
631 
632 static bool migrate_is_not_pcie(void *opaque, int version_id)
633 {
634     return !pci_is_express((PCIDevice *)opaque);
635 }
636 
637 const VMStateDescription vmstate_pci_device = {
638     .name = "PCIDevice",
639     .version_id = 2,
640     .minimum_version_id = 1,
641     .fields = (VMStateField[]) {
642         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
643         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
644                                    migrate_is_not_pcie,
645                                    0, vmstate_info_pci_config,
646                                    PCI_CONFIG_SPACE_SIZE),
647         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
648                                    migrate_is_pcie,
649                                    0, vmstate_info_pci_config,
650                                    PCIE_CONFIG_SPACE_SIZE),
651         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
652                                    vmstate_info_pci_irq_state,
653                                    PCI_NUM_PINS * sizeof(int32_t)),
654         VMSTATE_END_OF_LIST()
655     }
656 };
657 
658 
659 void pci_device_save(PCIDevice *s, QEMUFile *f)
660 {
661     /* Clear interrupt status bit: it is implicit
662      * in irq_state which we are saving.
663      * This makes us compatible with old devices
664      * which never set or clear this bit. */
665     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
666     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
667     /* Restore the interrupt status bit. */
668     pci_update_irq_status(s);
669 }
670 
671 int pci_device_load(PCIDevice *s, QEMUFile *f)
672 {
673     int ret;
674     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
675     /* Restore the interrupt status bit. */
676     pci_update_irq_status(s);
677     return ret;
678 }
679 
680 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
681 {
682     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
683                  pci_default_sub_vendor_id);
684     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
685                  pci_default_sub_device_id);
686 }
687 
688 /*
689  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
690  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
691  */
692 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
693                              unsigned int *slotp, unsigned int *funcp)
694 {
695     const char *p;
696     char *e;
697     unsigned long val;
698     unsigned long dom = 0, bus = 0;
699     unsigned int slot = 0;
700     unsigned int func = 0;
701 
702     p = addr;
703     val = strtoul(p, &e, 16);
704     if (e == p)
705         return -1;
706     if (*e == ':') {
707         bus = val;
708         p = e + 1;
709         val = strtoul(p, &e, 16);
710         if (e == p)
711             return -1;
712         if (*e == ':') {
713             dom = bus;
714             bus = val;
715             p = e + 1;
716             val = strtoul(p, &e, 16);
717             if (e == p)
718                 return -1;
719         }
720     }
721 
722     slot = val;
723 
724     if (funcp != NULL) {
725         if (*e != '.')
726             return -1;
727 
728         p = e + 1;
729         val = strtoul(p, &e, 16);
730         if (e == p)
731             return -1;
732 
733         func = val;
734     }
735 
736     /* if funcp == NULL func is 0 */
737     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
738         return -1;
739 
740     if (*e)
741         return -1;
742 
743     *domp = dom;
744     *busp = bus;
745     *slotp = slot;
746     if (funcp != NULL)
747         *funcp = func;
748     return 0;
749 }
750 
751 static void pci_init_cmask(PCIDevice *dev)
752 {
753     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
754     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
755     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
756     dev->cmask[PCI_REVISION_ID] = 0xff;
757     dev->cmask[PCI_CLASS_PROG] = 0xff;
758     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
759     dev->cmask[PCI_HEADER_TYPE] = 0xff;
760     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
761 }
762 
763 static void pci_init_wmask(PCIDevice *dev)
764 {
765     int config_size = pci_config_size(dev);
766 
767     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
768     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
769     pci_set_word(dev->wmask + PCI_COMMAND,
770                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
771                  PCI_COMMAND_INTX_DISABLE);
772     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
773 
774     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
775            config_size - PCI_CONFIG_HEADER_SIZE);
776 }
777 
778 static void pci_init_w1cmask(PCIDevice *dev)
779 {
780     /*
781      * Note: It's okay to set w1cmask even for readonly bits as
782      * long as their value is hardwired to 0.
783      */
784     pci_set_word(dev->w1cmask + PCI_STATUS,
785                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
786                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
787                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
788 }
789 
790 static void pci_init_mask_bridge(PCIDevice *d)
791 {
792     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
793        PCI_SEC_LETENCY_TIMER */
794     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
795 
796     /* base and limit */
797     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
798     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
799     pci_set_word(d->wmask + PCI_MEMORY_BASE,
800                  PCI_MEMORY_RANGE_MASK & 0xffff);
801     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
802                  PCI_MEMORY_RANGE_MASK & 0xffff);
803     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
804                  PCI_PREF_RANGE_MASK & 0xffff);
805     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
806                  PCI_PREF_RANGE_MASK & 0xffff);
807 
808     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
809     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
810 
811     /* Supported memory and i/o types */
812     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
813     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
814     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
815                                PCI_PREF_RANGE_TYPE_64);
816     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
817                                PCI_PREF_RANGE_TYPE_64);
818 
819     /*
820      * TODO: Bridges default to 10-bit VGA decoding but we currently only
821      * implement 16-bit decoding (no alias support).
822      */
823     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
824                  PCI_BRIDGE_CTL_PARITY |
825                  PCI_BRIDGE_CTL_SERR |
826                  PCI_BRIDGE_CTL_ISA |
827                  PCI_BRIDGE_CTL_VGA |
828                  PCI_BRIDGE_CTL_VGA_16BIT |
829                  PCI_BRIDGE_CTL_MASTER_ABORT |
830                  PCI_BRIDGE_CTL_BUS_RESET |
831                  PCI_BRIDGE_CTL_FAST_BACK |
832                  PCI_BRIDGE_CTL_DISCARD |
833                  PCI_BRIDGE_CTL_SEC_DISCARD |
834                  PCI_BRIDGE_CTL_DISCARD_SERR);
835     /* Below does not do anything as we never set this bit, put here for
836      * completeness. */
837     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
838                  PCI_BRIDGE_CTL_DISCARD_STATUS);
839     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
840     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
841     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
842                                PCI_PREF_RANGE_TYPE_MASK);
843     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
844                                PCI_PREF_RANGE_TYPE_MASK);
845 }
846 
847 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
848 {
849     uint8_t slot = PCI_SLOT(dev->devfn);
850     uint8_t func;
851 
852     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
853         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
854     }
855 
856     /*
857      * multifunction bit is interpreted in two ways as follows.
858      *   - all functions must set the bit to 1.
859      *     Example: Intel X53
860      *   - function 0 must set the bit, but the rest function (> 0)
861      *     is allowed to leave the bit to 0.
862      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
863      *
864      * So OS (at least Linux) checks the bit of only function 0,
865      * and doesn't see the bit of function > 0.
866      *
867      * The below check allows both interpretation.
868      */
869     if (PCI_FUNC(dev->devfn)) {
870         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
871         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
872             /* function 0 should set multifunction bit */
873             error_setg(errp, "PCI: single function device can't be populated "
874                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
875             return;
876         }
877         return;
878     }
879 
880     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
881         return;
882     }
883     /* function 0 indicates single function, so function > 0 must be NULL */
884     for (func = 1; func < PCI_FUNC_MAX; ++func) {
885         if (bus->devices[PCI_DEVFN(slot, func)]) {
886             error_setg(errp, "PCI: %x.0 indicates single function, "
887                        "but %x.%x is already populated.",
888                        slot, slot, func);
889             return;
890         }
891     }
892 }
893 
894 static void pci_config_alloc(PCIDevice *pci_dev)
895 {
896     int config_size = pci_config_size(pci_dev);
897 
898     pci_dev->config = g_malloc0(config_size);
899     pci_dev->cmask = g_malloc0(config_size);
900     pci_dev->wmask = g_malloc0(config_size);
901     pci_dev->w1cmask = g_malloc0(config_size);
902     pci_dev->used = g_malloc0(config_size);
903 }
904 
905 static void pci_config_free(PCIDevice *pci_dev)
906 {
907     g_free(pci_dev->config);
908     g_free(pci_dev->cmask);
909     g_free(pci_dev->wmask);
910     g_free(pci_dev->w1cmask);
911     g_free(pci_dev->used);
912 }
913 
914 static void do_pci_unregister_device(PCIDevice *pci_dev)
915 {
916     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
917     pci_config_free(pci_dev);
918 
919     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
920         memory_region_del_subregion(&pci_dev->bus_master_container_region,
921                                     &pci_dev->bus_master_enable_region);
922     }
923     address_space_destroy(&pci_dev->bus_master_as);
924 }
925 
926 /* Extract PCIReqIDCache into BDF format */
927 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
928 {
929     uint8_t bus_n;
930     uint16_t result;
931 
932     switch (cache->type) {
933     case PCI_REQ_ID_BDF:
934         result = pci_get_bdf(cache->dev);
935         break;
936     case PCI_REQ_ID_SECONDARY_BUS:
937         bus_n = pci_dev_bus_num(cache->dev);
938         result = PCI_BUILD_BDF(bus_n, 0);
939         break;
940     default:
941         error_report("Invalid PCI requester ID cache type: %d",
942                      cache->type);
943         exit(1);
944         break;
945     }
946 
947     return result;
948 }
949 
950 /* Parse bridges up to the root complex and return requester ID
951  * cache for specific device.  For full PCIe topology, the cache
952  * result would be exactly the same as getting BDF of the device.
953  * However, several tricks are required when system mixed up with
954  * legacy PCI devices and PCIe-to-PCI bridges.
955  *
956  * Here we cache the proxy device (and type) not requester ID since
957  * bus number might change from time to time.
958  */
959 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
960 {
961     PCIDevice *parent;
962     PCIReqIDCache cache = {
963         .dev = dev,
964         .type = PCI_REQ_ID_BDF,
965     };
966 
967     while (!pci_bus_is_root(pci_get_bus(dev))) {
968         /* We are under PCI/PCIe bridges */
969         parent = pci_get_bus(dev)->parent_dev;
970         if (pci_is_express(parent)) {
971             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
972                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
973                  * override the requester ID using secondary bus
974                  * number of parent bridge with zeroed devfn
975                  * (pcie-to-pci bridge spec chap 2.3). */
976                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
977                 cache.dev = dev;
978             }
979         } else {
980             /* Legacy PCI, override requester ID with the bridge's
981              * BDF upstream.  When the root complex connects to
982              * legacy PCI devices (including buses), it can only
983              * obtain requester ID info from directly attached
984              * devices.  If devices are attached under bridges, only
985              * the requester ID of the bridge that is directly
986              * attached to the root complex can be recognized. */
987             cache.type = PCI_REQ_ID_BDF;
988             cache.dev = parent;
989         }
990         dev = parent;
991     }
992 
993     return cache;
994 }
995 
996 uint16_t pci_requester_id(PCIDevice *dev)
997 {
998     return pci_req_id_cache_extract(&dev->requester_id_cache);
999 }
1000 
1001 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1002 {
1003     return !(bus->devices[devfn]);
1004 }
1005 
1006 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1007 {
1008     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1009 }
1010 
1011 /* -1 for devfn means auto assign */
1012 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1013                                          const char *name, int devfn,
1014                                          Error **errp)
1015 {
1016     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1017     PCIConfigReadFunc *config_read = pc->config_read;
1018     PCIConfigWriteFunc *config_write = pc->config_write;
1019     Error *local_err = NULL;
1020     DeviceState *dev = DEVICE(pci_dev);
1021     PCIBus *bus = pci_get_bus(pci_dev);
1022 
1023     /* Only pci bridges can be attached to extra PCI root buses */
1024     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1025         error_setg(errp,
1026                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1027                     bus->parent_dev->name);
1028         return NULL;
1029     }
1030 
1031     if (devfn < 0) {
1032         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1033             devfn += PCI_FUNC_MAX) {
1034             if (pci_bus_devfn_available(bus, devfn) &&
1035                    !pci_bus_devfn_reserved(bus, devfn)) {
1036                 goto found;
1037             }
1038         }
1039         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1040                    "or reserved", name);
1041         return NULL;
1042     found: ;
1043     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1044         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1045                    " reserved",
1046                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1047         return NULL;
1048     } else if (!pci_bus_devfn_available(bus, devfn)) {
1049         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1050                    " in use by %s",
1051                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1052                    bus->devices[devfn]->name);
1053         return NULL;
1054     } else if (dev->hotplugged &&
1055                pci_get_function_0(pci_dev)) {
1056         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1057                    " new func %s cannot be exposed to guest.",
1058                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1059                    pci_get_function_0(pci_dev)->name,
1060                    name);
1061 
1062        return NULL;
1063     }
1064 
1065     pci_dev->devfn = devfn;
1066     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1067     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1068 
1069     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1070                        "bus master container", UINT64_MAX);
1071     address_space_init(&pci_dev->bus_master_as,
1072                        &pci_dev->bus_master_container_region, pci_dev->name);
1073 
1074     if (phase_check(PHASE_MACHINE_READY)) {
1075         pci_init_bus_master(pci_dev);
1076     }
1077     pci_dev->irq_state = 0;
1078     pci_config_alloc(pci_dev);
1079 
1080     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1081     pci_config_set_device_id(pci_dev->config, pc->device_id);
1082     pci_config_set_revision(pci_dev->config, pc->revision);
1083     pci_config_set_class(pci_dev->config, pc->class_id);
1084 
1085     if (!pc->is_bridge) {
1086         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1087             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1088                          pc->subsystem_vendor_id);
1089             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1090                          pc->subsystem_id);
1091         } else {
1092             pci_set_default_subsystem_id(pci_dev);
1093         }
1094     } else {
1095         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1096         assert(!pc->subsystem_vendor_id);
1097         assert(!pc->subsystem_id);
1098     }
1099     pci_init_cmask(pci_dev);
1100     pci_init_wmask(pci_dev);
1101     pci_init_w1cmask(pci_dev);
1102     if (pc->is_bridge) {
1103         pci_init_mask_bridge(pci_dev);
1104     }
1105     pci_init_multifunction(bus, pci_dev, &local_err);
1106     if (local_err) {
1107         error_propagate(errp, local_err);
1108         do_pci_unregister_device(pci_dev);
1109         return NULL;
1110     }
1111 
1112     if (!config_read)
1113         config_read = pci_default_read_config;
1114     if (!config_write)
1115         config_write = pci_default_write_config;
1116     pci_dev->config_read = config_read;
1117     pci_dev->config_write = config_write;
1118     bus->devices[devfn] = pci_dev;
1119     pci_dev->version_id = 2; /* Current pci device vmstate version */
1120     return pci_dev;
1121 }
1122 
1123 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1124 {
1125     PCIIORegion *r;
1126     int i;
1127 
1128     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1129         r = &pci_dev->io_regions[i];
1130         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1131             continue;
1132         memory_region_del_subregion(r->address_space, r->memory);
1133     }
1134 
1135     pci_unregister_vga(pci_dev);
1136 }
1137 
1138 static void pci_qdev_unrealize(DeviceState *dev)
1139 {
1140     PCIDevice *pci_dev = PCI_DEVICE(dev);
1141     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1142 
1143     pci_unregister_io_regions(pci_dev);
1144     pci_del_option_rom(pci_dev);
1145 
1146     if (pc->exit) {
1147         pc->exit(pci_dev);
1148     }
1149 
1150     pci_device_deassert_intx(pci_dev);
1151     do_pci_unregister_device(pci_dev);
1152 }
1153 
1154 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1155                       uint8_t type, MemoryRegion *memory)
1156 {
1157     PCIIORegion *r;
1158     uint32_t addr; /* offset in pci config space */
1159     uint64_t wmask;
1160     pcibus_t size = memory_region_size(memory);
1161     uint8_t hdr_type;
1162 
1163     assert(region_num >= 0);
1164     assert(region_num < PCI_NUM_REGIONS);
1165     assert(is_power_of_2(size));
1166 
1167     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1168     hdr_type =
1169         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1170     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1171 
1172     r = &pci_dev->io_regions[region_num];
1173     r->addr = PCI_BAR_UNMAPPED;
1174     r->size = size;
1175     r->type = type;
1176     r->memory = memory;
1177     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1178                         ? pci_get_bus(pci_dev)->address_space_io
1179                         : pci_get_bus(pci_dev)->address_space_mem;
1180 
1181     wmask = ~(size - 1);
1182     if (region_num == PCI_ROM_SLOT) {
1183         /* ROM enable bit is writable */
1184         wmask |= PCI_ROM_ADDRESS_ENABLE;
1185     }
1186 
1187     addr = pci_bar(pci_dev, region_num);
1188     pci_set_long(pci_dev->config + addr, type);
1189 
1190     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1191         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1192         pci_set_quad(pci_dev->wmask + addr, wmask);
1193         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1194     } else {
1195         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1196         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1197     }
1198 }
1199 
1200 static void pci_update_vga(PCIDevice *pci_dev)
1201 {
1202     uint16_t cmd;
1203 
1204     if (!pci_dev->has_vga) {
1205         return;
1206     }
1207 
1208     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1209 
1210     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1211                               cmd & PCI_COMMAND_MEMORY);
1212     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1213                               cmd & PCI_COMMAND_IO);
1214     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1215                               cmd & PCI_COMMAND_IO);
1216 }
1217 
1218 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1219                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1220 {
1221     PCIBus *bus = pci_get_bus(pci_dev);
1222 
1223     assert(!pci_dev->has_vga);
1224 
1225     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1226     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1227     memory_region_add_subregion_overlap(bus->address_space_mem,
1228                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1229 
1230     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1231     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1232     memory_region_add_subregion_overlap(bus->address_space_io,
1233                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1234 
1235     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1236     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1237     memory_region_add_subregion_overlap(bus->address_space_io,
1238                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1239     pci_dev->has_vga = true;
1240 
1241     pci_update_vga(pci_dev);
1242 }
1243 
1244 void pci_unregister_vga(PCIDevice *pci_dev)
1245 {
1246     PCIBus *bus = pci_get_bus(pci_dev);
1247 
1248     if (!pci_dev->has_vga) {
1249         return;
1250     }
1251 
1252     memory_region_del_subregion(bus->address_space_mem,
1253                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1254     memory_region_del_subregion(bus->address_space_io,
1255                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1256     memory_region_del_subregion(bus->address_space_io,
1257                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1258     pci_dev->has_vga = false;
1259 }
1260 
1261 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1262 {
1263     return pci_dev->io_regions[region_num].addr;
1264 }
1265 
1266 static pcibus_t pci_bar_address(PCIDevice *d,
1267                                 int reg, uint8_t type, pcibus_t size)
1268 {
1269     pcibus_t new_addr, last_addr;
1270     int bar = pci_bar(d, reg);
1271     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1272     Object *machine = qdev_get_machine();
1273     ObjectClass *oc = object_get_class(machine);
1274     MachineClass *mc = MACHINE_CLASS(oc);
1275     bool allow_0_address = mc->pci_allow_0_address;
1276 
1277     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1278         if (!(cmd & PCI_COMMAND_IO)) {
1279             return PCI_BAR_UNMAPPED;
1280         }
1281         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1282         last_addr = new_addr + size - 1;
1283         /* Check if 32 bit BAR wraps around explicitly.
1284          * TODO: make priorities correct and remove this work around.
1285          */
1286         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1287             (!allow_0_address && new_addr == 0)) {
1288             return PCI_BAR_UNMAPPED;
1289         }
1290         return new_addr;
1291     }
1292 
1293     if (!(cmd & PCI_COMMAND_MEMORY)) {
1294         return PCI_BAR_UNMAPPED;
1295     }
1296     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1297         new_addr = pci_get_quad(d->config + bar);
1298     } else {
1299         new_addr = pci_get_long(d->config + bar);
1300     }
1301     /* the ROM slot has a specific enable bit */
1302     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1303         return PCI_BAR_UNMAPPED;
1304     }
1305     new_addr &= ~(size - 1);
1306     last_addr = new_addr + size - 1;
1307     /* NOTE: we do not support wrapping */
1308     /* XXX: as we cannot support really dynamic
1309        mappings, we handle specific values as invalid
1310        mappings. */
1311     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1312         (!allow_0_address && new_addr == 0)) {
1313         return PCI_BAR_UNMAPPED;
1314     }
1315 
1316     /* Now pcibus_t is 64bit.
1317      * Check if 32 bit BAR wraps around explicitly.
1318      * Without this, PC ide doesn't work well.
1319      * TODO: remove this work around.
1320      */
1321     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1322         return PCI_BAR_UNMAPPED;
1323     }
1324 
1325     /*
1326      * OS is allowed to set BAR beyond its addressable
1327      * bits. For example, 32 bit OS can set 64bit bar
1328      * to >4G. Check it. TODO: we might need to support
1329      * it in the future for e.g. PAE.
1330      */
1331     if (last_addr >= HWADDR_MAX) {
1332         return PCI_BAR_UNMAPPED;
1333     }
1334 
1335     return new_addr;
1336 }
1337 
1338 static void pci_update_mappings(PCIDevice *d)
1339 {
1340     PCIIORegion *r;
1341     int i;
1342     pcibus_t new_addr;
1343 
1344     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1345         r = &d->io_regions[i];
1346 
1347         /* this region isn't registered */
1348         if (!r->size)
1349             continue;
1350 
1351         new_addr = pci_bar_address(d, i, r->type, r->size);
1352 
1353         /* This bar isn't changed */
1354         if (new_addr == r->addr)
1355             continue;
1356 
1357         /* now do the real mapping */
1358         if (r->addr != PCI_BAR_UNMAPPED) {
1359             trace_pci_update_mappings_del(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_del_subregion(r->address_space, r->memory);
1364         }
1365         r->addr = new_addr;
1366         if (r->addr != PCI_BAR_UNMAPPED) {
1367             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1368                                           PCI_SLOT(d->devfn),
1369                                           PCI_FUNC(d->devfn),
1370                                           i, r->addr, r->size);
1371             memory_region_add_subregion_overlap(r->address_space,
1372                                                 r->addr, r->memory, 1);
1373         }
1374     }
1375 
1376     pci_update_vga(d);
1377 }
1378 
1379 static inline int pci_irq_disabled(PCIDevice *d)
1380 {
1381     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1382 }
1383 
1384 /* Called after interrupt disabled field update in config space,
1385  * assert/deassert interrupts if necessary.
1386  * Gets original interrupt disable bit value (before update). */
1387 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1388 {
1389     int i, disabled = pci_irq_disabled(d);
1390     if (disabled == was_irq_disabled)
1391         return;
1392     for (i = 0; i < PCI_NUM_PINS; ++i) {
1393         int state = pci_irq_state(d, i);
1394         pci_change_irq_level(d, i, disabled ? -state : state);
1395     }
1396 }
1397 
1398 uint32_t pci_default_read_config(PCIDevice *d,
1399                                  uint32_t address, int len)
1400 {
1401     uint32_t val = 0;
1402 
1403     assert(address + len <= pci_config_size(d));
1404 
1405     if (pci_is_express_downstream_port(d) &&
1406         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1407         pcie_sync_bridge_lnk(d);
1408     }
1409     memcpy(&val, d->config + address, len);
1410     return le32_to_cpu(val);
1411 }
1412 
1413 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1414 {
1415     int i, was_irq_disabled = pci_irq_disabled(d);
1416     uint32_t val = val_in;
1417 
1418     assert(addr + l <= pci_config_size(d));
1419 
1420     for (i = 0; i < l; val >>= 8, ++i) {
1421         uint8_t wmask = d->wmask[addr + i];
1422         uint8_t w1cmask = d->w1cmask[addr + i];
1423         assert(!(wmask & w1cmask));
1424         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1425         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1426     }
1427     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1428         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1429         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1430         range_covers_byte(addr, l, PCI_COMMAND))
1431         pci_update_mappings(d);
1432 
1433     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1434         pci_update_irq_disabled(d, was_irq_disabled);
1435         memory_region_set_enabled(&d->bus_master_enable_region,
1436                                   pci_get_word(d->config + PCI_COMMAND)
1437                                     & PCI_COMMAND_MASTER);
1438     }
1439 
1440     msi_write_config(d, addr, val_in, l);
1441     msix_write_config(d, addr, val_in, l);
1442 }
1443 
1444 /***********************************************************/
1445 /* generic PCI irq support */
1446 
1447 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1448 static void pci_irq_handler(void *opaque, int irq_num, int level)
1449 {
1450     PCIDevice *pci_dev = opaque;
1451     int change;
1452 
1453     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1454     assert(level == 0 || level == 1);
1455     change = level - pci_irq_state(pci_dev, irq_num);
1456     if (!change)
1457         return;
1458 
1459     pci_set_irq_state(pci_dev, irq_num, level);
1460     pci_update_irq_status(pci_dev);
1461     if (pci_irq_disabled(pci_dev))
1462         return;
1463     pci_change_irq_level(pci_dev, irq_num, change);
1464 }
1465 
1466 static inline int pci_intx(PCIDevice *pci_dev)
1467 {
1468     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1469 }
1470 
1471 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1472 {
1473     int intx = pci_intx(pci_dev);
1474     assert(0 <= intx && intx < PCI_NUM_PINS);
1475 
1476     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1477 }
1478 
1479 void pci_set_irq(PCIDevice *pci_dev, int level)
1480 {
1481     int intx = pci_intx(pci_dev);
1482     pci_irq_handler(pci_dev, intx, level);
1483 }
1484 
1485 /* Special hooks used by device assignment */
1486 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1487 {
1488     assert(pci_bus_is_root(bus));
1489     bus->route_intx_to_irq = route_intx_to_irq;
1490 }
1491 
1492 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1493 {
1494     PCIBus *bus;
1495 
1496     do {
1497         bus = pci_get_bus(dev);
1498         pin = bus->map_irq(dev, pin);
1499         dev = bus->parent_dev;
1500     } while (dev);
1501 
1502     if (!bus->route_intx_to_irq) {
1503         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1504                      object_get_typename(OBJECT(bus->qbus.parent)));
1505         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1506     }
1507 
1508     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1509 }
1510 
1511 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1512 {
1513     return old->mode != new->mode || old->irq != new->irq;
1514 }
1515 
1516 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1517 {
1518     PCIDevice *dev;
1519     PCIBus *sec;
1520     int i;
1521 
1522     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1523         dev = bus->devices[i];
1524         if (dev && dev->intx_routing_notifier) {
1525             dev->intx_routing_notifier(dev);
1526         }
1527     }
1528 
1529     QLIST_FOREACH(sec, &bus->child, sibling) {
1530         pci_bus_fire_intx_routing_notifier(sec);
1531     }
1532 }
1533 
1534 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1535                                           PCIINTxRoutingNotifier notifier)
1536 {
1537     dev->intx_routing_notifier = notifier;
1538 }
1539 
1540 /*
1541  * PCI-to-PCI bridge specification
1542  * 9.1: Interrupt routing. Table 9-1
1543  *
1544  * the PCI Express Base Specification, Revision 2.1
1545  * 2.2.8.1: INTx interrutp signaling - Rules
1546  *          the Implementation Note
1547  *          Table 2-20
1548  */
1549 /*
1550  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1551  * 0-origin unlike PCI interrupt pin register.
1552  */
1553 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1554 {
1555     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1556 }
1557 
1558 /***********************************************************/
1559 /* monitor info on PCI */
1560 
1561 typedef struct {
1562     uint16_t class;
1563     const char *desc;
1564     const char *fw_name;
1565     uint16_t fw_ign_bits;
1566 } pci_class_desc;
1567 
1568 static const pci_class_desc pci_class_descriptions[] =
1569 {
1570     { 0x0001, "VGA controller", "display"},
1571     { 0x0100, "SCSI controller", "scsi"},
1572     { 0x0101, "IDE controller", "ide"},
1573     { 0x0102, "Floppy controller", "fdc"},
1574     { 0x0103, "IPI controller", "ipi"},
1575     { 0x0104, "RAID controller", "raid"},
1576     { 0x0106, "SATA controller"},
1577     { 0x0107, "SAS controller"},
1578     { 0x0180, "Storage controller"},
1579     { 0x0200, "Ethernet controller", "ethernet"},
1580     { 0x0201, "Token Ring controller", "token-ring"},
1581     { 0x0202, "FDDI controller", "fddi"},
1582     { 0x0203, "ATM controller", "atm"},
1583     { 0x0280, "Network controller"},
1584     { 0x0300, "VGA controller", "display", 0x00ff},
1585     { 0x0301, "XGA controller"},
1586     { 0x0302, "3D controller"},
1587     { 0x0380, "Display controller"},
1588     { 0x0400, "Video controller", "video"},
1589     { 0x0401, "Audio controller", "sound"},
1590     { 0x0402, "Phone"},
1591     { 0x0403, "Audio controller", "sound"},
1592     { 0x0480, "Multimedia controller"},
1593     { 0x0500, "RAM controller", "memory"},
1594     { 0x0501, "Flash controller", "flash"},
1595     { 0x0580, "Memory controller"},
1596     { 0x0600, "Host bridge", "host"},
1597     { 0x0601, "ISA bridge", "isa"},
1598     { 0x0602, "EISA bridge", "eisa"},
1599     { 0x0603, "MC bridge", "mca"},
1600     { 0x0604, "PCI bridge", "pci-bridge"},
1601     { 0x0605, "PCMCIA bridge", "pcmcia"},
1602     { 0x0606, "NUBUS bridge", "nubus"},
1603     { 0x0607, "CARDBUS bridge", "cardbus"},
1604     { 0x0608, "RACEWAY bridge"},
1605     { 0x0680, "Bridge"},
1606     { 0x0700, "Serial port", "serial"},
1607     { 0x0701, "Parallel port", "parallel"},
1608     { 0x0800, "Interrupt controller", "interrupt-controller"},
1609     { 0x0801, "DMA controller", "dma-controller"},
1610     { 0x0802, "Timer", "timer"},
1611     { 0x0803, "RTC", "rtc"},
1612     { 0x0900, "Keyboard", "keyboard"},
1613     { 0x0901, "Pen", "pen"},
1614     { 0x0902, "Mouse", "mouse"},
1615     { 0x0A00, "Dock station", "dock", 0x00ff},
1616     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1617     { 0x0c00, "Fireware contorller", "fireware"},
1618     { 0x0c01, "Access bus controller", "access-bus"},
1619     { 0x0c02, "SSA controller", "ssa"},
1620     { 0x0c03, "USB controller", "usb"},
1621     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1622     { 0x0c05, "SMBus"},
1623     { 0, NULL}
1624 };
1625 
1626 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1627                                                   void (*fn)(PCIBus *b,
1628                                                              PCIDevice *d,
1629                                                              void *opaque),
1630                                                   void *opaque)
1631 {
1632     PCIDevice *d;
1633     int devfn;
1634 
1635     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1636         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1637         if (d) {
1638             fn(bus, d, opaque);
1639         }
1640     }
1641 }
1642 
1643 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1644                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1645                          void *opaque)
1646 {
1647     bus = pci_find_bus_nr(bus, bus_num);
1648 
1649     if (bus) {
1650         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1651     }
1652 }
1653 
1654 static void pci_for_each_device_under_bus(PCIBus *bus,
1655                                           void (*fn)(PCIBus *b, PCIDevice *d,
1656                                                      void *opaque),
1657                                           void *opaque)
1658 {
1659     PCIDevice *d;
1660     int devfn;
1661 
1662     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1663         d = bus->devices[devfn];
1664         if (d) {
1665             fn(bus, d, opaque);
1666         }
1667     }
1668 }
1669 
1670 void pci_for_each_device(PCIBus *bus, int bus_num,
1671                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1672                          void *opaque)
1673 {
1674     bus = pci_find_bus_nr(bus, bus_num);
1675 
1676     if (bus) {
1677         pci_for_each_device_under_bus(bus, fn, opaque);
1678     }
1679 }
1680 
1681 static const pci_class_desc *get_class_desc(int class)
1682 {
1683     const pci_class_desc *desc;
1684 
1685     desc = pci_class_descriptions;
1686     while (desc->desc && class != desc->class) {
1687         desc++;
1688     }
1689 
1690     return desc;
1691 }
1692 
1693 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1694 
1695 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1696 {
1697     PciMemoryRegionList *head = NULL, **tail = &head;
1698     int i;
1699 
1700     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1701         const PCIIORegion *r = &dev->io_regions[i];
1702         PciMemoryRegion *region;
1703 
1704         if (!r->size) {
1705             continue;
1706         }
1707 
1708         region = g_malloc0(sizeof(*region));
1709 
1710         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1711             region->type = g_strdup("io");
1712         } else {
1713             region->type = g_strdup("memory");
1714             region->has_prefetch = true;
1715             region->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1716             region->has_mem_type_64 = true;
1717             region->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1718         }
1719 
1720         region->bar = i;
1721         region->address = r->addr;
1722         region->size = r->size;
1723 
1724         QAPI_LIST_APPEND(tail, region);
1725     }
1726 
1727     return head;
1728 }
1729 
1730 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1731                                            int bus_num)
1732 {
1733     PciBridgeInfo *info;
1734     PciMemoryRange *range;
1735 
1736     info = g_new0(PciBridgeInfo, 1);
1737 
1738     info->bus = g_new0(PciBusInfo, 1);
1739     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1740     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1741     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1742 
1743     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1744     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1745     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1746 
1747     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1748     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1749     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1750 
1751     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1752     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1753     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1754 
1755     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1756         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1757         if (child_bus) {
1758             info->has_devices = true;
1759             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1760         }
1761     }
1762 
1763     return info;
1764 }
1765 
1766 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1767                                            int bus_num)
1768 {
1769     const pci_class_desc *desc;
1770     PciDeviceInfo *info;
1771     uint8_t type;
1772     int class;
1773 
1774     info = g_new0(PciDeviceInfo, 1);
1775     info->bus = bus_num;
1776     info->slot = PCI_SLOT(dev->devfn);
1777     info->function = PCI_FUNC(dev->devfn);
1778 
1779     info->class_info = g_new0(PciDeviceClass, 1);
1780     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1781     info->class_info->q_class = class;
1782     desc = get_class_desc(class);
1783     if (desc->desc) {
1784         info->class_info->has_desc = true;
1785         info->class_info->desc = g_strdup(desc->desc);
1786     }
1787 
1788     info->id = g_new0(PciDeviceId, 1);
1789     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1790     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1791     info->regions = qmp_query_pci_regions(dev);
1792     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1793 
1794     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1795     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1796         info->has_irq = true;
1797         info->irq = dev->config[PCI_INTERRUPT_LINE];
1798     }
1799 
1800     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1801     if (type == PCI_HEADER_TYPE_BRIDGE) {
1802         info->has_pci_bridge = true;
1803         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1804     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1805         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1806         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1807         info->id->subsystem_vendor =
1808             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1809     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1810         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1811         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1812         info->id->subsystem_vendor =
1813             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1814     }
1815 
1816     return info;
1817 }
1818 
1819 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1820 {
1821     PciDeviceInfoList *head = NULL, **tail = &head;
1822     PCIDevice *dev;
1823     int devfn;
1824 
1825     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1826         dev = bus->devices[devfn];
1827         if (dev) {
1828             QAPI_LIST_APPEND(tail, qmp_query_pci_device(dev, bus, bus_num));
1829         }
1830     }
1831 
1832     return head;
1833 }
1834 
1835 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1836 {
1837     PciInfo *info = NULL;
1838 
1839     bus = pci_find_bus_nr(bus, bus_num);
1840     if (bus) {
1841         info = g_malloc0(sizeof(*info));
1842         info->bus = bus_num;
1843         info->devices = qmp_query_pci_devices(bus, bus_num);
1844     }
1845 
1846     return info;
1847 }
1848 
1849 PciInfoList *qmp_query_pci(Error **errp)
1850 {
1851     PciInfoList *head = NULL, **tail = &head;
1852     PCIHostState *host_bridge;
1853 
1854     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1855         QAPI_LIST_APPEND(tail,
1856                          qmp_query_pci_bus(host_bridge->bus,
1857                                            pci_bus_num(host_bridge->bus)));
1858     }
1859 
1860     return head;
1861 }
1862 
1863 /* Initialize a PCI NIC.  */
1864 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1865                                const char *default_model,
1866                                const char *default_devaddr)
1867 {
1868     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1869     GSList *list;
1870     GPtrArray *pci_nic_models;
1871     PCIBus *bus;
1872     PCIDevice *pci_dev;
1873     DeviceState *dev;
1874     int devfn;
1875     int i;
1876     int dom, busnr;
1877     unsigned slot;
1878 
1879     if (nd->model && !strcmp(nd->model, "virtio")) {
1880         g_free(nd->model);
1881         nd->model = g_strdup("virtio-net-pci");
1882     }
1883 
1884     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1885     pci_nic_models = g_ptr_array_new();
1886     while (list) {
1887         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1888                                              TYPE_DEVICE);
1889         GSList *next;
1890         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1891             dc->user_creatable) {
1892             const char *name = object_class_get_name(list->data);
1893             /*
1894              * A network device might also be something else than a NIC, see
1895              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1896              * property, too. Unfortunately, some devices like virtio-net only
1897              * create this property during instance_init, so we have to create
1898              * a temporary instance here to be able to check it.
1899              */
1900             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1901             if (object_property_find(obj, "netdev")) {
1902                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1903             }
1904             object_unref(obj);
1905         }
1906         next = list->next;
1907         g_slist_free_1(list);
1908         list = next;
1909     }
1910     g_ptr_array_add(pci_nic_models, NULL);
1911 
1912     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1913         exit(0);
1914     }
1915 
1916     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1917                             default_model);
1918     if (i < 0) {
1919         exit(1);
1920     }
1921 
1922     if (!rootbus) {
1923         error_report("No primary PCI bus");
1924         exit(1);
1925     }
1926 
1927     assert(!rootbus->parent_dev);
1928 
1929     if (!devaddr) {
1930         devfn = -1;
1931         busnr = 0;
1932     } else {
1933         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1934             error_report("Invalid PCI device address %s for device %s",
1935                          devaddr, nd->model);
1936             exit(1);
1937         }
1938 
1939         if (dom != 0) {
1940             error_report("No support for non-zero PCI domains");
1941             exit(1);
1942         }
1943 
1944         devfn = PCI_DEVFN(slot, 0);
1945     }
1946 
1947     bus = pci_find_bus_nr(rootbus, busnr);
1948     if (!bus) {
1949         error_report("Invalid PCI device address %s for device %s",
1950                      devaddr, nd->model);
1951         exit(1);
1952     }
1953 
1954     pci_dev = pci_new(devfn, nd->model);
1955     dev = &pci_dev->qdev;
1956     qdev_set_nic_properties(dev, nd);
1957     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1958     g_ptr_array_free(pci_nic_models, true);
1959     return pci_dev;
1960 }
1961 
1962 PCIDevice *pci_vga_init(PCIBus *bus)
1963 {
1964     switch (vga_interface_type) {
1965     case VGA_CIRRUS:
1966         return pci_create_simple(bus, -1, "cirrus-vga");
1967     case VGA_QXL:
1968         return pci_create_simple(bus, -1, "qxl-vga");
1969     case VGA_STD:
1970         return pci_create_simple(bus, -1, "VGA");
1971     case VGA_VMWARE:
1972         return pci_create_simple(bus, -1, "vmware-svga");
1973     case VGA_VIRTIO:
1974         return pci_create_simple(bus, -1, "virtio-vga");
1975     case VGA_NONE:
1976     default: /* Other non-PCI types. Checking for unsupported types is already
1977                 done in vl.c. */
1978         return NULL;
1979     }
1980 }
1981 
1982 /* Whether a given bus number is in range of the secondary
1983  * bus of the given bridge device. */
1984 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1985 {
1986     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1987              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1988         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1989         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1990 }
1991 
1992 /* Whether a given bus number is in a range of a root bus */
1993 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1994 {
1995     int i;
1996 
1997     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1998         PCIDevice *dev = bus->devices[i];
1999 
2000         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
2001             if (pci_secondary_bus_in_range(dev, bus_num)) {
2002                 return true;
2003             }
2004         }
2005     }
2006 
2007     return false;
2008 }
2009 
2010 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2011 {
2012     PCIBus *sec;
2013 
2014     if (!bus) {
2015         return NULL;
2016     }
2017 
2018     if (pci_bus_num(bus) == bus_num) {
2019         return bus;
2020     }
2021 
2022     /* Consider all bus numbers in range for the host pci bridge. */
2023     if (!pci_bus_is_root(bus) &&
2024         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2025         return NULL;
2026     }
2027 
2028     /* try child bus */
2029     for (; bus; bus = sec) {
2030         QLIST_FOREACH(sec, &bus->child, sibling) {
2031             if (pci_bus_num(sec) == bus_num) {
2032                 return sec;
2033             }
2034             /* PXB buses assumed to be children of bus 0 */
2035             if (pci_bus_is_root(sec)) {
2036                 if (pci_root_bus_in_range(sec, bus_num)) {
2037                     break;
2038                 }
2039             } else {
2040                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2041                     break;
2042                 }
2043             }
2044         }
2045     }
2046 
2047     return NULL;
2048 }
2049 
2050 void pci_for_each_bus_depth_first(PCIBus *bus,
2051                                   void *(*begin)(PCIBus *bus, void *parent_state),
2052                                   void (*end)(PCIBus *bus, void *state),
2053                                   void *parent_state)
2054 {
2055     PCIBus *sec;
2056     void *state;
2057 
2058     if (!bus) {
2059         return;
2060     }
2061 
2062     if (begin) {
2063         state = begin(bus, parent_state);
2064     } else {
2065         state = parent_state;
2066     }
2067 
2068     QLIST_FOREACH(sec, &bus->child, sibling) {
2069         pci_for_each_bus_depth_first(sec, begin, end, state);
2070     }
2071 
2072     if (end) {
2073         end(bus, state);
2074     }
2075 }
2076 
2077 
2078 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2079 {
2080     bus = pci_find_bus_nr(bus, bus_num);
2081 
2082     if (!bus)
2083         return NULL;
2084 
2085     return bus->devices[devfn];
2086 }
2087 
2088 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2089 {
2090     PCIDevice *pci_dev = (PCIDevice *)qdev;
2091     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2092     ObjectClass *klass = OBJECT_CLASS(pc);
2093     Error *local_err = NULL;
2094     bool is_default_rom;
2095     uint16_t class_id;
2096 
2097     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2098         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2099         return;
2100     }
2101 
2102     /* initialize cap_present for pci_is_express() and pci_config_size(),
2103      * Note that hybrid PCIs are not set automatically and need to manage
2104      * QEMU_PCI_CAP_EXPRESS manually */
2105     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2106        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2107         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2108     }
2109 
2110     pci_dev = do_pci_register_device(pci_dev,
2111                                      object_get_typename(OBJECT(qdev)),
2112                                      pci_dev->devfn, errp);
2113     if (pci_dev == NULL)
2114         return;
2115 
2116     if (pc->realize) {
2117         pc->realize(pci_dev, &local_err);
2118         if (local_err) {
2119             error_propagate(errp, local_err);
2120             do_pci_unregister_device(pci_dev);
2121             return;
2122         }
2123     }
2124 
2125     if (pci_dev->failover_pair_id) {
2126         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2127             error_setg(errp, "failover primary device must be on "
2128                              "PCIExpress bus");
2129             pci_qdev_unrealize(DEVICE(pci_dev));
2130             return;
2131         }
2132         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2133         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2134             error_setg(errp, "failover primary device is not an "
2135                              "Ethernet device");
2136             pci_qdev_unrealize(DEVICE(pci_dev));
2137             return;
2138         }
2139         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2140             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2141             error_setg(errp, "failover: primary device must be in its own "
2142                               "PCI slot");
2143             pci_qdev_unrealize(DEVICE(pci_dev));
2144             return;
2145         }
2146         qdev->allow_unplug_during_migration = true;
2147     }
2148 
2149     /* rom loading */
2150     is_default_rom = false;
2151     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2152         pci_dev->romfile = g_strdup(pc->romfile);
2153         is_default_rom = true;
2154     }
2155 
2156     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2157     if (local_err) {
2158         error_propagate(errp, local_err);
2159         pci_qdev_unrealize(DEVICE(pci_dev));
2160         return;
2161     }
2162 }
2163 
2164 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2165                                  const char *name)
2166 {
2167     DeviceState *dev;
2168 
2169     dev = qdev_new(name);
2170     qdev_prop_set_int32(dev, "addr", devfn);
2171     qdev_prop_set_bit(dev, "multifunction", multifunction);
2172     return PCI_DEVICE(dev);
2173 }
2174 
2175 PCIDevice *pci_new(int devfn, const char *name)
2176 {
2177     return pci_new_multifunction(devfn, false, name);
2178 }
2179 
2180 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2181 {
2182     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2183 }
2184 
2185 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2186                                            bool multifunction,
2187                                            const char *name)
2188 {
2189     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2190     pci_realize_and_unref(dev, bus, &error_fatal);
2191     return dev;
2192 }
2193 
2194 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2195 {
2196     return pci_create_simple_multifunction(bus, devfn, false, name);
2197 }
2198 
2199 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2200 {
2201     int offset = PCI_CONFIG_HEADER_SIZE;
2202     int i;
2203     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2204         if (pdev->used[i])
2205             offset = i + 1;
2206         else if (i - offset + 1 == size)
2207             return offset;
2208     }
2209     return 0;
2210 }
2211 
2212 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2213                                         uint8_t *prev_p)
2214 {
2215     uint8_t next, prev;
2216 
2217     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2218         return 0;
2219 
2220     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2221          prev = next + PCI_CAP_LIST_NEXT)
2222         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2223             break;
2224 
2225     if (prev_p)
2226         *prev_p = prev;
2227     return next;
2228 }
2229 
2230 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2231 {
2232     uint8_t next, prev, found = 0;
2233 
2234     if (!(pdev->used[offset])) {
2235         return 0;
2236     }
2237 
2238     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2239 
2240     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2241          prev = next + PCI_CAP_LIST_NEXT) {
2242         if (next <= offset && next > found) {
2243             found = next;
2244         }
2245     }
2246     return found;
2247 }
2248 
2249 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2250    This is needed for an option rom which is used for more than one device. */
2251 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2252 {
2253     uint16_t vendor_id;
2254     uint16_t device_id;
2255     uint16_t rom_vendor_id;
2256     uint16_t rom_device_id;
2257     uint16_t rom_magic;
2258     uint16_t pcir_offset;
2259     uint8_t checksum;
2260 
2261     /* Words in rom data are little endian (like in PCI configuration),
2262        so they can be read / written with pci_get_word / pci_set_word. */
2263 
2264     /* Only a valid rom will be patched. */
2265     rom_magic = pci_get_word(ptr);
2266     if (rom_magic != 0xaa55) {
2267         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2268         return;
2269     }
2270     pcir_offset = pci_get_word(ptr + 0x18);
2271     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2272         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2273         return;
2274     }
2275 
2276     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2277     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2278     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2279     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2280 
2281     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2282                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2283 
2284     checksum = ptr[6];
2285 
2286     if (vendor_id != rom_vendor_id) {
2287         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2288         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2289         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2290         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2291         ptr[6] = checksum;
2292         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2293     }
2294 
2295     if (device_id != rom_device_id) {
2296         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2297         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2298         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2299         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2300         ptr[6] = checksum;
2301         pci_set_word(ptr + pcir_offset + 6, device_id);
2302     }
2303 }
2304 
2305 /* Add an option rom for the device */
2306 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2307                                Error **errp)
2308 {
2309     int64_t size;
2310     char *path;
2311     void *ptr;
2312     char name[32];
2313     const VMStateDescription *vmsd;
2314 
2315     if (!pdev->romfile)
2316         return;
2317     if (strlen(pdev->romfile) == 0)
2318         return;
2319 
2320     if (!pdev->rom_bar) {
2321         /*
2322          * Load rom via fw_cfg instead of creating a rom bar,
2323          * for 0.11 compatibility.
2324          */
2325         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2326 
2327         /*
2328          * Hot-plugged devices can't use the option ROM
2329          * if the rom bar is disabled.
2330          */
2331         if (DEVICE(pdev)->hotplugged) {
2332             error_setg(errp, "Hot-plugged device without ROM bar"
2333                        " can't have an option ROM");
2334             return;
2335         }
2336 
2337         if (class == 0x0300) {
2338             rom_add_vga(pdev->romfile);
2339         } else {
2340             rom_add_option(pdev->romfile, -1);
2341         }
2342         return;
2343     }
2344 
2345     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2346     if (path == NULL) {
2347         path = g_strdup(pdev->romfile);
2348     }
2349 
2350     size = get_image_size(path);
2351     if (size < 0) {
2352         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2353         g_free(path);
2354         return;
2355     } else if (size == 0) {
2356         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2357         g_free(path);
2358         return;
2359     } else if (size > 2 * GiB) {
2360         error_setg(errp, "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2361                    pdev->romfile);
2362         g_free(path);
2363         return;
2364     }
2365     if (pdev->romsize != -1) {
2366         if (size > pdev->romsize) {
2367             error_setg(errp, "romfile \"%s\" (%u bytes) is too large for ROM size %u",
2368                        pdev->romfile, (uint32_t)size, pdev->romsize);
2369             g_free(path);
2370             return;
2371         }
2372     } else {
2373         pdev->romsize = pow2ceil(size);
2374     }
2375 
2376     vmsd = qdev_get_vmsd(DEVICE(pdev));
2377 
2378     if (vmsd) {
2379         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2380     } else {
2381         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2382     }
2383     pdev->has_rom = true;
2384     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize, &error_fatal);
2385     ptr = memory_region_get_ram_ptr(&pdev->rom);
2386     if (load_image_size(path, ptr, size) < 0) {
2387         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2388         g_free(path);
2389         return;
2390     }
2391     g_free(path);
2392 
2393     if (is_default_rom) {
2394         /* Only the default rom images will be patched (if needed). */
2395         pci_patch_ids(pdev, ptr, size);
2396     }
2397 
2398     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2399 }
2400 
2401 static void pci_del_option_rom(PCIDevice *pdev)
2402 {
2403     if (!pdev->has_rom)
2404         return;
2405 
2406     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2407     pdev->has_rom = false;
2408 }
2409 
2410 /*
2411  * On success, pci_add_capability() returns a positive value
2412  * that the offset of the pci capability.
2413  * On failure, it sets an error and returns a negative error
2414  * code.
2415  */
2416 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2417                        uint8_t offset, uint8_t size,
2418                        Error **errp)
2419 {
2420     uint8_t *config;
2421     int i, overlapping_cap;
2422 
2423     if (!offset) {
2424         offset = pci_find_space(pdev, size);
2425         /* out of PCI config space is programming error */
2426         assert(offset);
2427     } else {
2428         /* Verify that capabilities don't overlap.  Note: device assignment
2429          * depends on this check to verify that the device is not broken.
2430          * Should never trigger for emulated devices, but it's helpful
2431          * for debugging these. */
2432         for (i = offset; i < offset + size; i++) {
2433             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2434             if (overlapping_cap) {
2435                 error_setg(errp, "%s:%02x:%02x.%x "
2436                            "Attempt to add PCI capability %x at offset "
2437                            "%x overlaps existing capability %x at offset %x",
2438                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2439                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2440                            cap_id, offset, overlapping_cap, i);
2441                 return -EINVAL;
2442             }
2443         }
2444     }
2445 
2446     config = pdev->config + offset;
2447     config[PCI_CAP_LIST_ID] = cap_id;
2448     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2449     pdev->config[PCI_CAPABILITY_LIST] = offset;
2450     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2451     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2452     /* Make capability read-only by default */
2453     memset(pdev->wmask + offset, 0, size);
2454     /* Check capability by default */
2455     memset(pdev->cmask + offset, 0xFF, size);
2456     return offset;
2457 }
2458 
2459 /* Unlink capability from the pci config space. */
2460 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2461 {
2462     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2463     if (!offset)
2464         return;
2465     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2466     /* Make capability writable again */
2467     memset(pdev->wmask + offset, 0xff, size);
2468     memset(pdev->w1cmask + offset, 0, size);
2469     /* Clear cmask as device-specific registers can't be checked */
2470     memset(pdev->cmask + offset, 0, size);
2471     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2472 
2473     if (!pdev->config[PCI_CAPABILITY_LIST])
2474         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2475 }
2476 
2477 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2478 {
2479     return pci_find_capability_list(pdev, cap_id, NULL);
2480 }
2481 
2482 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2483 {
2484     PCIDevice *d = (PCIDevice *)dev;
2485     const pci_class_desc *desc;
2486     char ctxt[64];
2487     PCIIORegion *r;
2488     int i, class;
2489 
2490     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2491     desc = pci_class_descriptions;
2492     while (desc->desc && class != desc->class)
2493         desc++;
2494     if (desc->desc) {
2495         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2496     } else {
2497         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2498     }
2499 
2500     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2501                    "pci id %04x:%04x (sub %04x:%04x)\n",
2502                    indent, "", ctxt, pci_dev_bus_num(d),
2503                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2504                    pci_get_word(d->config + PCI_VENDOR_ID),
2505                    pci_get_word(d->config + PCI_DEVICE_ID),
2506                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2507                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2508     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2509         r = &d->io_regions[i];
2510         if (!r->size)
2511             continue;
2512         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2513                        " [0x%"FMT_PCIBUS"]\n",
2514                        indent, "",
2515                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2516                        r->addr, r->addr + r->size - 1);
2517     }
2518 }
2519 
2520 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2521 {
2522     PCIDevice *d = (PCIDevice *)dev;
2523     const char *name = NULL;
2524     const pci_class_desc *desc =  pci_class_descriptions;
2525     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2526 
2527     while (desc->desc &&
2528           (class & ~desc->fw_ign_bits) !=
2529           (desc->class & ~desc->fw_ign_bits)) {
2530         desc++;
2531     }
2532 
2533     if (desc->desc) {
2534         name = desc->fw_name;
2535     }
2536 
2537     if (name) {
2538         pstrcpy(buf, len, name);
2539     } else {
2540         snprintf(buf, len, "pci%04x,%04x",
2541                  pci_get_word(d->config + PCI_VENDOR_ID),
2542                  pci_get_word(d->config + PCI_DEVICE_ID));
2543     }
2544 
2545     return buf;
2546 }
2547 
2548 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2549 {
2550     PCIDevice *d = (PCIDevice *)dev;
2551     char path[50], name[33];
2552     int off;
2553 
2554     off = snprintf(path, sizeof(path), "%s@%x",
2555                    pci_dev_fw_name(dev, name, sizeof name),
2556                    PCI_SLOT(d->devfn));
2557     if (PCI_FUNC(d->devfn))
2558         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2559     return g_strdup(path);
2560 }
2561 
2562 static char *pcibus_get_dev_path(DeviceState *dev)
2563 {
2564     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2565     PCIDevice *t;
2566     int slot_depth;
2567     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2568      * 00 is added here to make this format compatible with
2569      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2570      * Slot.Function list specifies the slot and function numbers for all
2571      * devices on the path from root to the specific device. */
2572     const char *root_bus_path;
2573     int root_bus_len;
2574     char slot[] = ":SS.F";
2575     int slot_len = sizeof slot - 1 /* For '\0' */;
2576     int path_len;
2577     char *path, *p;
2578     int s;
2579 
2580     root_bus_path = pci_root_bus_path(d);
2581     root_bus_len = strlen(root_bus_path);
2582 
2583     /* Calculate # of slots on path between device and root. */;
2584     slot_depth = 0;
2585     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2586         ++slot_depth;
2587     }
2588 
2589     path_len = root_bus_len + slot_len * slot_depth;
2590 
2591     /* Allocate memory, fill in the terminating null byte. */
2592     path = g_malloc(path_len + 1 /* For '\0' */);
2593     path[path_len] = '\0';
2594 
2595     memcpy(path, root_bus_path, root_bus_len);
2596 
2597     /* Fill in slot numbers. We walk up from device to root, so need to print
2598      * them in the reverse order, last to first. */
2599     p = path + path_len;
2600     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2601         p -= slot_len;
2602         s = snprintf(slot, sizeof slot, ":%02x.%x",
2603                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2604         assert(s == slot_len);
2605         memcpy(p, slot, slot_len);
2606     }
2607 
2608     return path;
2609 }
2610 
2611 static int pci_qdev_find_recursive(PCIBus *bus,
2612                                    const char *id, PCIDevice **pdev)
2613 {
2614     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2615     if (!qdev) {
2616         return -ENODEV;
2617     }
2618 
2619     /* roughly check if given qdev is pci device */
2620     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2621         *pdev = PCI_DEVICE(qdev);
2622         return 0;
2623     }
2624     return -EINVAL;
2625 }
2626 
2627 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2628 {
2629     PCIHostState *host_bridge;
2630     int rc = -ENODEV;
2631 
2632     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2633         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2634         if (!tmp) {
2635             rc = 0;
2636             break;
2637         }
2638         if (tmp != -ENODEV) {
2639             rc = tmp;
2640         }
2641     }
2642 
2643     return rc;
2644 }
2645 
2646 MemoryRegion *pci_address_space(PCIDevice *dev)
2647 {
2648     return pci_get_bus(dev)->address_space_mem;
2649 }
2650 
2651 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2652 {
2653     return pci_get_bus(dev)->address_space_io;
2654 }
2655 
2656 static void pci_device_class_init(ObjectClass *klass, void *data)
2657 {
2658     DeviceClass *k = DEVICE_CLASS(klass);
2659 
2660     k->realize = pci_qdev_realize;
2661     k->unrealize = pci_qdev_unrealize;
2662     k->bus_type = TYPE_PCI_BUS;
2663     device_class_set_props(k, pci_props);
2664 }
2665 
2666 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2667 {
2668     if (!object_class_is_abstract(klass)) {
2669         ObjectClass *conventional =
2670             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2671         ObjectClass *pcie =
2672             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2673         assert(conventional || pcie);
2674     }
2675 }
2676 
2677 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2678 {
2679     PCIBus *bus = pci_get_bus(dev);
2680     PCIBus *iommu_bus = bus;
2681     uint8_t devfn = dev->devfn;
2682 
2683     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2684         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2685 
2686         /*
2687          * The requester ID of the provided device may be aliased, as seen from
2688          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2689          * requester ID to provide a unique AddressSpace for devices, but
2690          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2691          * to-PCI bridge creates and accepts transactions on behalf of down-
2692          * stream devices.  When doing so, all downstream devices are masked
2693          * (aliased) behind a single requester ID.  The requester ID used
2694          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2695          * bridges, with a PCIe capability indicating such, follow the
2696          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2697          * where the bridge uses the seconary bus as the bridge portion of the
2698          * requester ID and devfn of 00.0.  For other bridges, typically those
2699          * found on the root complex such as the dmi-to-pci-bridge, we follow
2700          * the convention of typical bare-metal hardware, which uses the
2701          * requester ID of the bridge itself.  There are device specific
2702          * exceptions to these rules, but these are the defaults that the
2703          * Linux kernel uses when determining DMA aliases itself and believed
2704          * to be true for the bare metal equivalents of the devices emulated
2705          * in QEMU.
2706          */
2707         if (!pci_bus_is_express(iommu_bus)) {
2708             PCIDevice *parent = iommu_bus->parent_dev;
2709 
2710             if (pci_is_express(parent) &&
2711                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2712                 devfn = PCI_DEVFN(0, 0);
2713                 bus = iommu_bus;
2714             } else {
2715                 devfn = parent->devfn;
2716                 bus = parent_bus;
2717             }
2718         }
2719 
2720         iommu_bus = parent_bus;
2721     }
2722     if (iommu_bus && iommu_bus->iommu_fn) {
2723         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2724     }
2725     return &address_space_memory;
2726 }
2727 
2728 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2729 {
2730     bus->iommu_fn = fn;
2731     bus->iommu_opaque = opaque;
2732 }
2733 
2734 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2735 {
2736     Range *range = opaque;
2737     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2738     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2739     int i;
2740 
2741     if (!(cmd & PCI_COMMAND_MEMORY)) {
2742         return;
2743     }
2744 
2745     if (pc->is_bridge) {
2746         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2747         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2748 
2749         base = MAX(base, 0x1ULL << 32);
2750 
2751         if (limit >= base) {
2752             Range pref_range;
2753             range_set_bounds(&pref_range, base, limit);
2754             range_extend(range, &pref_range);
2755         }
2756     }
2757     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2758         PCIIORegion *r = &dev->io_regions[i];
2759         pcibus_t lob, upb;
2760         Range region_range;
2761 
2762         if (!r->size ||
2763             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2764             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2765             continue;
2766         }
2767 
2768         lob = pci_bar_address(dev, i, r->type, r->size);
2769         upb = lob + r->size - 1;
2770         if (lob == PCI_BAR_UNMAPPED) {
2771             continue;
2772         }
2773 
2774         lob = MAX(lob, 0x1ULL << 32);
2775 
2776         if (upb >= lob) {
2777             range_set_bounds(&region_range, lob, upb);
2778             range_extend(range, &region_range);
2779         }
2780     }
2781 }
2782 
2783 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2784 {
2785     range_make_empty(range);
2786     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2787 }
2788 
2789 static bool pcie_has_upstream_port(PCIDevice *dev)
2790 {
2791     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2792 
2793     /* Device associated with an upstream port.
2794      * As there are several types of these, it's easier to check the
2795      * parent device: upstream ports are always connected to
2796      * root or downstream ports.
2797      */
2798     return parent_dev &&
2799         pci_is_express(parent_dev) &&
2800         parent_dev->exp.exp_cap &&
2801         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2802          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2803 }
2804 
2805 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2806 {
2807     PCIBus *bus = pci_get_bus(pci_dev);
2808 
2809     if(pcie_has_upstream_port(pci_dev)) {
2810         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2811         return bus->devices[0];
2812     } else {
2813         /* Other bus types might support multiple devices at slots 0-31 */
2814         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2815     }
2816 }
2817 
2818 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2819 {
2820     MSIMessage msg;
2821     if (msix_enabled(dev)) {
2822         msg = msix_get_message(dev, vector);
2823     } else if (msi_enabled(dev)) {
2824         msg = msi_get_message(dev, vector);
2825     } else {
2826         /* Should never happen */
2827         error_report("%s: unknown interrupt type", __func__);
2828         abort();
2829     }
2830     return msg;
2831 }
2832 
2833 static const TypeInfo pci_device_type_info = {
2834     .name = TYPE_PCI_DEVICE,
2835     .parent = TYPE_DEVICE,
2836     .instance_size = sizeof(PCIDevice),
2837     .abstract = true,
2838     .class_size = sizeof(PCIDeviceClass),
2839     .class_init = pci_device_class_init,
2840     .class_base_init = pci_device_class_base_init,
2841 };
2842 
2843 static void pci_register_types(void)
2844 {
2845     type_register_static(&pci_bus_info);
2846     type_register_static(&pcie_bus_info);
2847     type_register_static(&conventional_pci_interface_info);
2848     type_register_static(&pcie_interface_info);
2849     type_register_static(&pci_device_type_info);
2850 }
2851 
2852 type_init(pci_register_types)
2853