xref: /qemu/hw/pci/pci.c (revision ec7e5a90fea996f04ea24e81b680a87bc975354a)
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/datadir.h"
27 #include "qemu/units.h"
28 #include "hw/irq.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bridge.h"
31 #include "hw/pci/pci_bus.h"
32 #include "hw/pci/pci_host.h"
33 #include "hw/qdev-properties.h"
34 #include "hw/qdev-properties-system.h"
35 #include "migration/qemu-file-types.h"
36 #include "migration/vmstate.h"
37 #include "net/net.h"
38 #include "system/numa.h"
39 #include "system/runstate.h"
40 #include "system/system.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "hw/nvram/fw_cfg.h"
50 #include "qapi/error.h"
51 #include "qemu/cutils.h"
52 #include "pci-internal.h"
53 
54 #include "hw/xen/xen.h"
55 #include "hw/i386/kvm/xen_evtchn.h"
56 
57 bool pci_available = true;
58 
59 static char *pcibus_get_dev_path(DeviceState *dev);
60 static char *pcibus_get_fw_dev_path(DeviceState *dev);
61 static void pcibus_reset_hold(Object *obj, ResetType type);
62 static bool pcie_has_upstream_port(PCIDevice *dev);
63 
64 static void prop_pci_busnr_get(Object *obj, Visitor *v, const char *name,
65                              void *opaque, Error **errp)
66 {
67     uint8_t busnr = pci_dev_bus_num(PCI_DEVICE(obj));
68 
69     visit_type_uint8(v, name, &busnr, errp);
70 }
71 
72 static const PropertyInfo prop_pci_busnr = {
73     .type = "busnr",
74     .get = prop_pci_busnr_get,
75 };
76 
77 static const Property pci_props[] = {
78     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
79     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
80     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, UINT32_MAX),
81     DEFINE_PROP_INT32("rombar",  PCIDevice, rom_bar, -1),
82     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
83                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
84     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
85                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
86     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
87                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
88     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
89                        failover_pair_id),
90     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
91     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
92                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
93     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
94                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
95     DEFINE_PROP_SIZE32("x-max-bounce-buffer-size", PCIDevice,
96                      max_bounce_buffer_size, DEFAULT_MAX_BOUNCE_BUFFER_SIZE),
97     DEFINE_PROP_BIT("x-pcie-ext-tag", PCIDevice, cap_present,
98                     QEMU_PCIE_EXT_TAG_BITNR, true),
99     { .name = "busnr", .info = &prop_pci_busnr },
100 };
101 
102 static const VMStateDescription vmstate_pcibus = {
103     .name = "PCIBUS",
104     .version_id = 1,
105     .minimum_version_id = 1,
106     .fields = (const VMStateField[]) {
107         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
108         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
109                              nirq, 0, vmstate_info_int32,
110                              int32_t),
111         VMSTATE_END_OF_LIST()
112     }
113 };
114 
115 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
116 {
117     return a - b;
118 }
119 
120 static GSequence *pci_acpi_index_list(void)
121 {
122     static GSequence *used_acpi_index_list;
123 
124     if (!used_acpi_index_list) {
125         used_acpi_index_list = g_sequence_new(NULL);
126     }
127     return used_acpi_index_list;
128 }
129 
130 static void pci_init_bus_master(PCIDevice *pci_dev)
131 {
132     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
133 
134     memory_region_init_alias(&pci_dev->bus_master_enable_region,
135                              OBJECT(pci_dev), "bus master",
136                              dma_as->root, 0, memory_region_size(dma_as->root));
137     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
138     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
139                                 &pci_dev->bus_master_enable_region);
140 }
141 
142 static void pcibus_machine_done(Notifier *notifier, void *data)
143 {
144     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
145     int i;
146 
147     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
148         if (bus->devices[i]) {
149             pci_init_bus_master(bus->devices[i]);
150         }
151     }
152 }
153 
154 static void pci_bus_realize(BusState *qbus, Error **errp)
155 {
156     PCIBus *bus = PCI_BUS(qbus);
157 
158     bus->machine_done.notify = pcibus_machine_done;
159     qemu_add_machine_init_done_notifier(&bus->machine_done);
160 
161     vmstate_register_any(NULL, &vmstate_pcibus, bus);
162 }
163 
164 static void pcie_bus_realize(BusState *qbus, Error **errp)
165 {
166     PCIBus *bus = PCI_BUS(qbus);
167     Error *local_err = NULL;
168 
169     pci_bus_realize(qbus, &local_err);
170     if (local_err) {
171         error_propagate(errp, local_err);
172         return;
173     }
174 
175     /*
176      * A PCI-E bus can support extended config space if it's the root
177      * bus, or if the bus/bridge above it does as well
178      */
179     if (pci_bus_is_root(bus)) {
180         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
181     } else {
182         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
183 
184         if (pci_bus_allows_extended_config_space(parent_bus)) {
185             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
186         }
187     }
188 }
189 
190 static void pci_bus_unrealize(BusState *qbus)
191 {
192     PCIBus *bus = PCI_BUS(qbus);
193 
194     qemu_remove_machine_init_done_notifier(&bus->machine_done);
195 
196     vmstate_unregister(NULL, &vmstate_pcibus, bus);
197 }
198 
199 static int pcibus_num(PCIBus *bus)
200 {
201     if (pci_bus_is_root(bus)) {
202         return 0; /* pci host bridge */
203     }
204     return bus->parent_dev->config[PCI_SECONDARY_BUS];
205 }
206 
207 static uint16_t pcibus_numa_node(PCIBus *bus)
208 {
209     return NUMA_NODE_UNASSIGNED;
210 }
211 
212 bool pci_bus_add_fw_cfg_extra_pci_roots(FWCfgState *fw_cfg,
213                                         PCIBus *bus,
214                                         Error **errp)
215 {
216     Object *obj;
217 
218     if (!bus) {
219         return true;
220     }
221     obj = OBJECT(bus);
222 
223     return fw_cfg_add_file_from_generator(fw_cfg, obj->parent,
224                                           object_get_canonical_path_component(obj),
225                                           "etc/extra-pci-roots", errp);
226 }
227 
228 static GByteArray *pci_bus_fw_cfg_gen_data(Object *obj, Error **errp)
229 {
230     PCIBus *bus = PCI_BUS(obj);
231     GByteArray *byte_array;
232     uint64_t extra_hosts = 0;
233 
234     if (!bus) {
235         return NULL;
236     }
237 
238     QLIST_FOREACH(bus, &bus->child, sibling) {
239         /* look for expander root buses */
240         if (pci_bus_is_root(bus)) {
241             extra_hosts++;
242         }
243     }
244 
245     if (!extra_hosts) {
246         return NULL;
247     }
248     extra_hosts = cpu_to_le64(extra_hosts);
249 
250     byte_array = g_byte_array_new();
251     g_byte_array_append(byte_array,
252                         (const void *)&extra_hosts, sizeof(extra_hosts));
253 
254     return byte_array;
255 }
256 
257 static void pci_bus_class_init(ObjectClass *klass, const void *data)
258 {
259     BusClass *k = BUS_CLASS(klass);
260     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
261     ResettableClass *rc = RESETTABLE_CLASS(klass);
262     FWCfgDataGeneratorClass *fwgc = FW_CFG_DATA_GENERATOR_CLASS(klass);
263 
264     k->print_dev = pcibus_dev_print;
265     k->get_dev_path = pcibus_get_dev_path;
266     k->get_fw_dev_path = pcibus_get_fw_dev_path;
267     k->realize = pci_bus_realize;
268     k->unrealize = pci_bus_unrealize;
269 
270     rc->phases.hold = pcibus_reset_hold;
271 
272     pbc->bus_num = pcibus_num;
273     pbc->numa_node = pcibus_numa_node;
274 
275     fwgc->get_data = pci_bus_fw_cfg_gen_data;
276 }
277 
278 static const TypeInfo pci_bus_info = {
279     .name = TYPE_PCI_BUS,
280     .parent = TYPE_BUS,
281     .instance_size = sizeof(PCIBus),
282     .class_size = sizeof(PCIBusClass),
283     .class_init = pci_bus_class_init,
284     .interfaces = (const InterfaceInfo[]) {
285         { TYPE_FW_CFG_DATA_GENERATOR_INTERFACE },
286         { }
287     }
288 };
289 
290 static const TypeInfo cxl_interface_info = {
291     .name          = INTERFACE_CXL_DEVICE,
292     .parent        = TYPE_INTERFACE,
293 };
294 
295 static const TypeInfo pcie_interface_info = {
296     .name          = INTERFACE_PCIE_DEVICE,
297     .parent        = TYPE_INTERFACE,
298 };
299 
300 static const TypeInfo conventional_pci_interface_info = {
301     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
302     .parent        = TYPE_INTERFACE,
303 };
304 
305 static void pcie_bus_class_init(ObjectClass *klass, const void *data)
306 {
307     BusClass *k = BUS_CLASS(klass);
308 
309     k->realize = pcie_bus_realize;
310 }
311 
312 static const TypeInfo pcie_bus_info = {
313     .name = TYPE_PCIE_BUS,
314     .parent = TYPE_PCI_BUS,
315     .class_init = pcie_bus_class_init,
316 };
317 
318 static const TypeInfo cxl_bus_info = {
319     .name       = TYPE_CXL_BUS,
320     .parent     = TYPE_PCIE_BUS,
321     .class_init = pcie_bus_class_init,
322 };
323 
324 static void pci_update_mappings(PCIDevice *d);
325 static void pci_irq_handler(void *opaque, int irq_num, int level);
326 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
327 static void pci_del_option_rom(PCIDevice *pdev);
328 
329 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
330 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
331 
332 PCIHostStateList pci_host_bridges;
333 
334 int pci_bar(PCIDevice *d, int reg)
335 {
336     uint8_t type;
337 
338     /* PCIe virtual functions do not have their own BARs */
339     assert(!pci_is_vf(d));
340 
341     if (reg != PCI_ROM_SLOT)
342         return PCI_BASE_ADDRESS_0 + reg * 4;
343 
344     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
345     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
346 }
347 
348 static inline int pci_irq_state(PCIDevice *d, int irq_num)
349 {
350         return (d->irq_state >> irq_num) & 0x1;
351 }
352 
353 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
354 {
355         d->irq_state &= ~(0x1 << irq_num);
356         d->irq_state |= level << irq_num;
357 }
358 
359 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
360 {
361     assert(irq_num >= 0);
362     assert(irq_num < bus->nirq);
363     bus->irq_count[irq_num] += change;
364     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
365 }
366 
367 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
368 {
369     PCIBus *bus;
370     for (;;) {
371         int dev_irq = irq_num;
372         bus = pci_get_bus(pci_dev);
373         assert(bus->map_irq);
374         irq_num = bus->map_irq(pci_dev, irq_num);
375         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
376                             pci_bus_is_root(bus) ? "root-complex"
377                                     : DEVICE(bus->parent_dev)->canonical_path);
378         if (bus->set_irq)
379             break;
380         pci_dev = bus->parent_dev;
381     }
382     pci_bus_change_irq_level(bus, irq_num, change);
383 }
384 
385 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
386 {
387     assert(irq_num >= 0);
388     assert(irq_num < bus->nirq);
389     return !!bus->irq_count[irq_num];
390 }
391 
392 /* Update interrupt status bit in config space on interrupt
393  * state change. */
394 static void pci_update_irq_status(PCIDevice *dev)
395 {
396     if (dev->irq_state) {
397         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
398     } else {
399         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
400     }
401 }
402 
403 void pci_device_deassert_intx(PCIDevice *dev)
404 {
405     int i;
406     for (i = 0; i < PCI_NUM_PINS; ++i) {
407         pci_irq_handler(dev, i, 0);
408     }
409 }
410 
411 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
412 {
413     MemTxAttrs attrs = {};
414 
415     /*
416      * Xen uses the high bits of the address to contain some of the bits
417      * of the PIRQ#. Therefore we can't just send the write cycle and
418      * trust that it's caught by the APIC at 0xfee00000 because the
419      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
420      * So we intercept the delivery here instead of in kvm_send_msi().
421      */
422     if (xen_mode == XEN_EMULATE &&
423         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
424         return;
425     }
426     attrs.requester_id = pci_requester_id(dev);
427     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
428                          attrs, NULL);
429 }
430 
431 /*
432  * Register and track a PM capability.  If wmask is also enabled for the power
433  * state field of the pmcsr register, guest writes may change the device PM
434  * state.  BAR access is only enabled while the device is in the D0 state.
435  * Return the capability offset or negative error code.
436  */
437 int pci_pm_init(PCIDevice *d, uint8_t offset, Error **errp)
438 {
439     int cap = pci_add_capability(d, PCI_CAP_ID_PM, offset, PCI_PM_SIZEOF, errp);
440 
441     if (cap < 0) {
442         return cap;
443     }
444 
445     d->pm_cap = cap;
446     d->cap_present |= QEMU_PCI_CAP_PM;
447 
448     return cap;
449 }
450 
451 static uint8_t pci_pm_state(PCIDevice *d)
452 {
453     uint16_t pmcsr;
454 
455     if (!(d->cap_present & QEMU_PCI_CAP_PM)) {
456         return 0;
457     }
458 
459     pmcsr = pci_get_word(d->config + d->pm_cap + PCI_PM_CTRL);
460 
461     return pmcsr & PCI_PM_CTRL_STATE_MASK;
462 }
463 
464 /*
465  * Update the PM capability state based on the new value stored in config
466  * space respective to the old, pre-write state provided.  If the new value
467  * is rejected (unsupported or invalid transition) restore the old value.
468  * Return the resulting PM state.
469  */
470 static uint8_t pci_pm_update(PCIDevice *d, uint32_t addr, int l, uint8_t old)
471 {
472     uint16_t pmc;
473     uint8_t new;
474 
475     if (!(d->cap_present & QEMU_PCI_CAP_PM) ||
476         !range_covers_byte(addr, l, d->pm_cap + PCI_PM_CTRL)) {
477         return old;
478     }
479 
480     new = pci_pm_state(d);
481     if (new == old) {
482         return old;
483     }
484 
485     pmc = pci_get_word(d->config + d->pm_cap + PCI_PM_PMC);
486 
487     /*
488      * Transitions to D1 & D2 are only allowed if supported.  Devices may
489      * only transition to higher D-states or to D0.
490      */
491     if ((!(pmc & PCI_PM_CAP_D1) && new == 1) ||
492         (!(pmc & PCI_PM_CAP_D2) && new == 2) ||
493         (old && new && new < old)) {
494         pci_word_test_and_clear_mask(d->config + d->pm_cap + PCI_PM_CTRL,
495                                      PCI_PM_CTRL_STATE_MASK);
496         pci_word_test_and_set_mask(d->config + d->pm_cap + PCI_PM_CTRL,
497                                    old);
498         trace_pci_pm_bad_transition(d->name, pci_dev_bus_num(d),
499                                     PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
500                                     old, new);
501         return old;
502     }
503 
504     trace_pci_pm_transition(d->name, pci_dev_bus_num(d), PCI_SLOT(d->devfn),
505                             PCI_FUNC(d->devfn), old, new);
506     return new;
507 }
508 
509 static void pci_reset_regions(PCIDevice *dev)
510 {
511     int r;
512     if (pci_is_vf(dev)) {
513         return;
514     }
515 
516     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
517         PCIIORegion *region = &dev->io_regions[r];
518         if (!region->size) {
519             continue;
520         }
521 
522         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
523             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
524             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
525         } else {
526             pci_set_long(dev->config + pci_bar(dev, r), region->type);
527         }
528     }
529 }
530 
531 static void pci_do_device_reset(PCIDevice *dev)
532 {
533     pci_device_deassert_intx(dev);
534     assert(dev->irq_state == 0);
535 
536     /* Clear all writable bits */
537     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
538                                  pci_get_word(dev->wmask + PCI_COMMAND) |
539                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
540     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
541                                  pci_get_word(dev->wmask + PCI_STATUS) |
542                                  pci_get_word(dev->w1cmask + PCI_STATUS));
543     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
544     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
545                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
546                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
547     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
548     /* Default PM state is D0 */
549     if (dev->cap_present & QEMU_PCI_CAP_PM) {
550         pci_word_test_and_clear_mask(dev->config + dev->pm_cap + PCI_PM_CTRL,
551                                      PCI_PM_CTRL_STATE_MASK);
552     }
553     pci_reset_regions(dev);
554     pci_update_mappings(dev);
555 
556     msi_reset(dev);
557     msix_reset(dev);
558     pcie_sriov_pf_reset(dev);
559 }
560 
561 /*
562  * This function is called on #RST and FLR.
563  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
564  */
565 void pci_device_reset(PCIDevice *dev)
566 {
567     device_cold_reset(&dev->qdev);
568     pci_do_device_reset(dev);
569 }
570 
571 /*
572  * Trigger pci bus reset under a given bus.
573  * Called via bus_cold_reset on RST# assert, after the devices
574  * have been reset device_cold_reset-ed already.
575  */
576 static void pcibus_reset_hold(Object *obj, ResetType type)
577 {
578     PCIBus *bus = PCI_BUS(obj);
579     int i;
580 
581     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
582         if (bus->devices[i]) {
583             pci_do_device_reset(bus->devices[i]);
584         }
585     }
586 
587     for (i = 0; i < bus->nirq; i++) {
588         assert(bus->irq_count[i] == 0);
589     }
590 }
591 
592 static void pci_host_bus_register(DeviceState *host)
593 {
594     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
595 
596     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
597 }
598 
599 static void pci_host_bus_unregister(DeviceState *host)
600 {
601     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
602 
603     QLIST_REMOVE(host_bridge, next);
604 }
605 
606 PCIBus *pci_device_root_bus(const PCIDevice *d)
607 {
608     PCIBus *bus = pci_get_bus(d);
609 
610     while (!pci_bus_is_root(bus)) {
611         d = bus->parent_dev;
612         assert(d != NULL);
613 
614         bus = pci_get_bus(d);
615     }
616 
617     return bus;
618 }
619 
620 const char *pci_root_bus_path(PCIDevice *dev)
621 {
622     PCIBus *rootbus = pci_device_root_bus(dev);
623     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
624     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
625 
626     assert(host_bridge->bus == rootbus);
627 
628     if (hc->root_bus_path) {
629         return (*hc->root_bus_path)(host_bridge, rootbus);
630     }
631 
632     return rootbus->qbus.name;
633 }
634 
635 bool pci_bus_bypass_iommu(PCIBus *bus)
636 {
637     PCIBus *rootbus = bus;
638     PCIHostState *host_bridge;
639 
640     if (!pci_bus_is_root(bus)) {
641         rootbus = pci_device_root_bus(bus->parent_dev);
642     }
643 
644     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
645 
646     assert(host_bridge->bus == rootbus);
647 
648     return host_bridge->bypass_iommu;
649 }
650 
651 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
652                                        MemoryRegion *mem, MemoryRegion *io,
653                                        uint8_t devfn_min)
654 {
655     assert(PCI_FUNC(devfn_min) == 0);
656     bus->devfn_min = devfn_min;
657     bus->slot_reserved_mask = 0x0;
658     bus->address_space_mem = mem;
659     bus->address_space_io = io;
660     bus->flags |= PCI_BUS_IS_ROOT;
661 
662     /* host bridge */
663     QLIST_INIT(&bus->child);
664 
665     pci_host_bus_register(parent);
666 }
667 
668 static void pci_bus_uninit(PCIBus *bus)
669 {
670     pci_host_bus_unregister(BUS(bus)->parent);
671 }
672 
673 bool pci_bus_is_express(const PCIBus *bus)
674 {
675     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
676 }
677 
678 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
679                        const char *name,
680                        MemoryRegion *mem, MemoryRegion *io,
681                        uint8_t devfn_min, const char *typename)
682 {
683     qbus_init(bus, bus_size, typename, parent, name);
684     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
685 }
686 
687 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
688                          MemoryRegion *mem, MemoryRegion *io,
689                          uint8_t devfn_min, const char *typename)
690 {
691     PCIBus *bus;
692 
693     bus = PCI_BUS(qbus_new(typename, parent, name));
694     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
695     return bus;
696 }
697 
698 void pci_root_bus_cleanup(PCIBus *bus)
699 {
700     pci_bus_uninit(bus);
701     /* the caller of the unplug hotplug handler will delete this device */
702     qbus_unrealize(BUS(bus));
703 }
704 
705 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
706                   void *irq_opaque, int nirq)
707 {
708     bus->set_irq = set_irq;
709     bus->irq_opaque = irq_opaque;
710     bus->nirq = nirq;
711     g_free(bus->irq_count);
712     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
713 }
714 
715 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
716 {
717     bus->map_irq = map_irq;
718 }
719 
720 void pci_bus_irqs_cleanup(PCIBus *bus)
721 {
722     bus->set_irq = NULL;
723     bus->map_irq = NULL;
724     bus->irq_opaque = NULL;
725     bus->nirq = 0;
726     g_free(bus->irq_count);
727     bus->irq_count = NULL;
728 }
729 
730 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
731                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
732                               void *irq_opaque,
733                               MemoryRegion *mem, MemoryRegion *io,
734                               uint8_t devfn_min, int nirq,
735                               const char *typename)
736 {
737     PCIBus *bus;
738 
739     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
740     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
741     pci_bus_map_irqs(bus, map_irq);
742     return bus;
743 }
744 
745 void pci_unregister_root_bus(PCIBus *bus)
746 {
747     pci_bus_irqs_cleanup(bus);
748     pci_root_bus_cleanup(bus);
749 }
750 
751 int pci_bus_num(PCIBus *s)
752 {
753     return PCI_BUS_GET_CLASS(s)->bus_num(s);
754 }
755 
756 /* Returns the min and max bus numbers of a PCI bus hierarchy */
757 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
758 {
759     int i;
760     *min_bus = *max_bus = pci_bus_num(bus);
761 
762     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
763         PCIDevice *dev = bus->devices[i];
764 
765         if (dev && IS_PCI_BRIDGE(dev)) {
766             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
767             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
768         }
769     }
770 }
771 
772 int pci_bus_numa_node(PCIBus *bus)
773 {
774     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
775 }
776 
777 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
778                                  const VMStateField *field)
779 {
780     PCIDevice *s = container_of(pv, PCIDevice, config);
781     uint8_t *config;
782     int i;
783 
784     assert(size == pci_config_size(s));
785     config = g_malloc(size);
786 
787     qemu_get_buffer(f, config, size);
788     for (i = 0; i < size; ++i) {
789         if ((config[i] ^ s->config[i]) &
790             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
791             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
792                          "cmask: %x wmask: %x w1cmask:%x", __func__,
793                          i, config[i], s->config[i],
794                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
795             g_free(config);
796             return -EINVAL;
797         }
798     }
799     memcpy(s->config, config, size);
800 
801     pci_update_mappings(s);
802     if (IS_PCI_BRIDGE(s)) {
803         pci_bridge_update_mappings(PCI_BRIDGE(s));
804     }
805 
806     memory_region_set_enabled(&s->bus_master_enable_region,
807                               pci_get_word(s->config + PCI_COMMAND)
808                               & PCI_COMMAND_MASTER);
809 
810     g_free(config);
811     return 0;
812 }
813 
814 /* just put buffer */
815 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
816                                  const VMStateField *field, JSONWriter *vmdesc)
817 {
818     const uint8_t **v = pv;
819     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
820     qemu_put_buffer(f, *v, size);
821 
822     return 0;
823 }
824 
825 static const VMStateInfo vmstate_info_pci_config = {
826     .name = "pci config",
827     .get  = get_pci_config_device,
828     .put  = put_pci_config_device,
829 };
830 
831 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
832                              const VMStateField *field)
833 {
834     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
835     uint32_t irq_state[PCI_NUM_PINS];
836     int i;
837     for (i = 0; i < PCI_NUM_PINS; ++i) {
838         irq_state[i] = qemu_get_be32(f);
839         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
840             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
841                     irq_state[i]);
842             return -EINVAL;
843         }
844     }
845 
846     for (i = 0; i < PCI_NUM_PINS; ++i) {
847         pci_set_irq_state(s, i, irq_state[i]);
848     }
849 
850     return 0;
851 }
852 
853 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
854                              const VMStateField *field, JSONWriter *vmdesc)
855 {
856     int i;
857     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
858 
859     for (i = 0; i < PCI_NUM_PINS; ++i) {
860         qemu_put_be32(f, pci_irq_state(s, i));
861     }
862 
863     return 0;
864 }
865 
866 static const VMStateInfo vmstate_info_pci_irq_state = {
867     .name = "pci irq state",
868     .get  = get_pci_irq_state,
869     .put  = put_pci_irq_state,
870 };
871 
872 static bool migrate_is_pcie(void *opaque, int version_id)
873 {
874     return pci_is_express((PCIDevice *)opaque);
875 }
876 
877 static bool migrate_is_not_pcie(void *opaque, int version_id)
878 {
879     return !pci_is_express((PCIDevice *)opaque);
880 }
881 
882 static int pci_post_load(void *opaque, int version_id)
883 {
884     pcie_sriov_pf_post_load(opaque);
885     return 0;
886 }
887 
888 const VMStateDescription vmstate_pci_device = {
889     .name = "PCIDevice",
890     .version_id = 2,
891     .minimum_version_id = 1,
892     .post_load = pci_post_load,
893     .fields = (const VMStateField[]) {
894         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
895         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
896                                    migrate_is_not_pcie,
897                                    0, vmstate_info_pci_config,
898                                    PCI_CONFIG_SPACE_SIZE),
899         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
900                                    migrate_is_pcie,
901                                    0, vmstate_info_pci_config,
902                                    PCIE_CONFIG_SPACE_SIZE),
903         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
904                                    vmstate_info_pci_irq_state,
905                                    PCI_NUM_PINS * sizeof(int32_t)),
906         VMSTATE_END_OF_LIST()
907     }
908 };
909 
910 
911 void pci_device_save(PCIDevice *s, QEMUFile *f)
912 {
913     /* Clear interrupt status bit: it is implicit
914      * in irq_state which we are saving.
915      * This makes us compatible with old devices
916      * which never set or clear this bit. */
917     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
918     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
919     /* Restore the interrupt status bit. */
920     pci_update_irq_status(s);
921 }
922 
923 int pci_device_load(PCIDevice *s, QEMUFile *f)
924 {
925     int ret;
926     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
927     /* Restore the interrupt status bit. */
928     pci_update_irq_status(s);
929     return ret;
930 }
931 
932 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
933 {
934     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
935                  pci_default_sub_vendor_id);
936     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
937                  pci_default_sub_device_id);
938 }
939 
940 /*
941  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
942  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
943  */
944 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
945                              unsigned int *slotp, unsigned int *funcp)
946 {
947     const char *p;
948     char *e;
949     unsigned long val;
950     unsigned long dom = 0, bus = 0;
951     unsigned int slot = 0;
952     unsigned int func = 0;
953 
954     p = addr;
955     val = strtoul(p, &e, 16);
956     if (e == p)
957         return -1;
958     if (*e == ':') {
959         bus = val;
960         p = e + 1;
961         val = strtoul(p, &e, 16);
962         if (e == p)
963             return -1;
964         if (*e == ':') {
965             dom = bus;
966             bus = val;
967             p = e + 1;
968             val = strtoul(p, &e, 16);
969             if (e == p)
970                 return -1;
971         }
972     }
973 
974     slot = val;
975 
976     if (funcp != NULL) {
977         if (*e != '.')
978             return -1;
979 
980         p = e + 1;
981         val = strtoul(p, &e, 16);
982         if (e == p)
983             return -1;
984 
985         func = val;
986     }
987 
988     /* if funcp == NULL func is 0 */
989     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
990         return -1;
991 
992     if (*e)
993         return -1;
994 
995     *domp = dom;
996     *busp = bus;
997     *slotp = slot;
998     if (funcp != NULL)
999         *funcp = func;
1000     return 0;
1001 }
1002 
1003 static void pci_init_cmask(PCIDevice *dev)
1004 {
1005     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
1006     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
1007     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
1008     dev->cmask[PCI_REVISION_ID] = 0xff;
1009     dev->cmask[PCI_CLASS_PROG] = 0xff;
1010     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
1011     dev->cmask[PCI_HEADER_TYPE] = 0xff;
1012     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
1013 }
1014 
1015 static void pci_init_wmask(PCIDevice *dev)
1016 {
1017     int config_size = pci_config_size(dev);
1018 
1019     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
1020     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
1021     pci_set_word(dev->wmask + PCI_COMMAND,
1022                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
1023                  PCI_COMMAND_INTX_DISABLE);
1024     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
1025 
1026     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
1027            config_size - PCI_CONFIG_HEADER_SIZE);
1028 }
1029 
1030 static void pci_init_w1cmask(PCIDevice *dev)
1031 {
1032     /*
1033      * Note: It's okay to set w1cmask even for readonly bits as
1034      * long as their value is hardwired to 0.
1035      */
1036     pci_set_word(dev->w1cmask + PCI_STATUS,
1037                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
1038                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
1039                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
1040 }
1041 
1042 static void pci_init_mask_bridge(PCIDevice *d)
1043 {
1044     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
1045        PCI_SEC_LATENCY_TIMER */
1046     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
1047 
1048     /* base and limit */
1049     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
1050     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
1051     pci_set_word(d->wmask + PCI_MEMORY_BASE,
1052                  PCI_MEMORY_RANGE_MASK & 0xffff);
1053     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
1054                  PCI_MEMORY_RANGE_MASK & 0xffff);
1055     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
1056                  PCI_PREF_RANGE_MASK & 0xffff);
1057     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
1058                  PCI_PREF_RANGE_MASK & 0xffff);
1059 
1060     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
1061     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
1062 
1063     /* Supported memory and i/o types */
1064     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
1065     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
1066     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
1067                                PCI_PREF_RANGE_TYPE_64);
1068     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
1069                                PCI_PREF_RANGE_TYPE_64);
1070 
1071     /*
1072      * TODO: Bridges default to 10-bit VGA decoding but we currently only
1073      * implement 16-bit decoding (no alias support).
1074      */
1075     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
1076                  PCI_BRIDGE_CTL_PARITY |
1077                  PCI_BRIDGE_CTL_SERR |
1078                  PCI_BRIDGE_CTL_ISA |
1079                  PCI_BRIDGE_CTL_VGA |
1080                  PCI_BRIDGE_CTL_VGA_16BIT |
1081                  PCI_BRIDGE_CTL_MASTER_ABORT |
1082                  PCI_BRIDGE_CTL_BUS_RESET |
1083                  PCI_BRIDGE_CTL_FAST_BACK |
1084                  PCI_BRIDGE_CTL_DISCARD |
1085                  PCI_BRIDGE_CTL_SEC_DISCARD |
1086                  PCI_BRIDGE_CTL_DISCARD_SERR);
1087     /* Below does not do anything as we never set this bit, put here for
1088      * completeness. */
1089     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
1090                  PCI_BRIDGE_CTL_DISCARD_STATUS);
1091     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
1092     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
1093     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
1094                                PCI_PREF_RANGE_TYPE_MASK);
1095     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
1096                                PCI_PREF_RANGE_TYPE_MASK);
1097 }
1098 
1099 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
1100 {
1101     uint8_t slot = PCI_SLOT(dev->devfn);
1102     uint8_t func;
1103 
1104     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
1105         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
1106     }
1107 
1108     /*
1109      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
1110      * device, as it may just be a VF that ended up with function 0 in
1111      * the legacy PCI interpretation. Avoid failing in such cases:
1112      */
1113     if (pci_is_vf(dev) &&
1114         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
1115         return;
1116     }
1117 
1118     /*
1119      * multifunction bit is interpreted in two ways as follows.
1120      *   - all functions must set the bit to 1.
1121      *     Example: Intel X53
1122      *   - function 0 must set the bit, but the rest function (> 0)
1123      *     is allowed to leave the bit to 0.
1124      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
1125      *
1126      * So OS (at least Linux) checks the bit of only function 0,
1127      * and doesn't see the bit of function > 0.
1128      *
1129      * The below check allows both interpretation.
1130      */
1131     if (PCI_FUNC(dev->devfn)) {
1132         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
1133         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
1134             /* function 0 should set multifunction bit */
1135             error_setg(errp, "PCI: single function device can't be populated "
1136                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
1137             return;
1138         }
1139         return;
1140     }
1141 
1142     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
1143         return;
1144     }
1145     /* function 0 indicates single function, so function > 0 must be NULL */
1146     for (func = 1; func < PCI_FUNC_MAX; ++func) {
1147         if (bus->devices[PCI_DEVFN(slot, func)]) {
1148             error_setg(errp, "PCI: %x.0 indicates single function, "
1149                        "but %x.%x is already populated.",
1150                        slot, slot, func);
1151             return;
1152         }
1153     }
1154 }
1155 
1156 static void pci_config_alloc(PCIDevice *pci_dev)
1157 {
1158     int config_size = pci_config_size(pci_dev);
1159 
1160     pci_dev->config = g_malloc0(config_size);
1161     pci_dev->cmask = g_malloc0(config_size);
1162     pci_dev->wmask = g_malloc0(config_size);
1163     pci_dev->w1cmask = g_malloc0(config_size);
1164     pci_dev->used = g_malloc0(config_size);
1165 }
1166 
1167 static void pci_config_free(PCIDevice *pci_dev)
1168 {
1169     g_free(pci_dev->config);
1170     g_free(pci_dev->cmask);
1171     g_free(pci_dev->wmask);
1172     g_free(pci_dev->w1cmask);
1173     g_free(pci_dev->used);
1174 }
1175 
1176 static void do_pci_unregister_device(PCIDevice *pci_dev)
1177 {
1178     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1179     pci_config_free(pci_dev);
1180 
1181     if (xen_mode == XEN_EMULATE) {
1182         xen_evtchn_remove_pci_device(pci_dev);
1183     }
1184     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1185         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1186                                     &pci_dev->bus_master_enable_region);
1187     }
1188     address_space_destroy(&pci_dev->bus_master_as);
1189 }
1190 
1191 /* Extract PCIReqIDCache into BDF format */
1192 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1193 {
1194     uint8_t bus_n;
1195     uint16_t result;
1196 
1197     switch (cache->type) {
1198     case PCI_REQ_ID_BDF:
1199         result = pci_get_bdf(cache->dev);
1200         break;
1201     case PCI_REQ_ID_SECONDARY_BUS:
1202         bus_n = pci_dev_bus_num(cache->dev);
1203         result = PCI_BUILD_BDF(bus_n, 0);
1204         break;
1205     default:
1206         error_report("Invalid PCI requester ID cache type: %d",
1207                      cache->type);
1208         exit(1);
1209         break;
1210     }
1211 
1212     return result;
1213 }
1214 
1215 /* Parse bridges up to the root complex and return requester ID
1216  * cache for specific device.  For full PCIe topology, the cache
1217  * result would be exactly the same as getting BDF of the device.
1218  * However, several tricks are required when system mixed up with
1219  * legacy PCI devices and PCIe-to-PCI bridges.
1220  *
1221  * Here we cache the proxy device (and type) not requester ID since
1222  * bus number might change from time to time.
1223  */
1224 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1225 {
1226     PCIDevice *parent;
1227     PCIReqIDCache cache = {
1228         .dev = dev,
1229         .type = PCI_REQ_ID_BDF,
1230     };
1231 
1232     while (!pci_bus_is_root(pci_get_bus(dev))) {
1233         /* We are under PCI/PCIe bridges */
1234         parent = pci_get_bus(dev)->parent_dev;
1235         if (pci_is_express(parent)) {
1236             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1237                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1238                  * override the requester ID using secondary bus
1239                  * number of parent bridge with zeroed devfn
1240                  * (pcie-to-pci bridge spec chap 2.3). */
1241                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1242                 cache.dev = dev;
1243             }
1244         } else {
1245             /* Legacy PCI, override requester ID with the bridge's
1246              * BDF upstream.  When the root complex connects to
1247              * legacy PCI devices (including buses), it can only
1248              * obtain requester ID info from directly attached
1249              * devices.  If devices are attached under bridges, only
1250              * the requester ID of the bridge that is directly
1251              * attached to the root complex can be recognized. */
1252             cache.type = PCI_REQ_ID_BDF;
1253             cache.dev = parent;
1254         }
1255         dev = parent;
1256     }
1257 
1258     return cache;
1259 }
1260 
1261 uint16_t pci_requester_id(PCIDevice *dev)
1262 {
1263     return pci_req_id_cache_extract(&dev->requester_id_cache);
1264 }
1265 
1266 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1267 {
1268     return !(bus->devices[devfn]);
1269 }
1270 
1271 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1272 {
1273     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1274 }
1275 
1276 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1277 {
1278     return bus->slot_reserved_mask;
1279 }
1280 
1281 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1282 {
1283     bus->slot_reserved_mask |= mask;
1284 }
1285 
1286 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1287 {
1288     bus->slot_reserved_mask &= ~mask;
1289 }
1290 
1291 /* -1 for devfn means auto assign */
1292 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1293                                          const char *name, int devfn,
1294                                          Error **errp)
1295 {
1296     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1297     PCIConfigReadFunc *config_read = pc->config_read;
1298     PCIConfigWriteFunc *config_write = pc->config_write;
1299     Error *local_err = NULL;
1300     DeviceState *dev = DEVICE(pci_dev);
1301     PCIBus *bus = pci_get_bus(pci_dev);
1302     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1303 
1304     /* Only pci bridges can be attached to extra PCI root buses */
1305     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1306         error_setg(errp,
1307                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1308                     bus->parent_dev->name);
1309         return NULL;
1310     }
1311 
1312     if (devfn < 0) {
1313         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1314             devfn += PCI_FUNC_MAX) {
1315             if (pci_bus_devfn_available(bus, devfn) &&
1316                    !pci_bus_devfn_reserved(bus, devfn)) {
1317                 goto found;
1318             }
1319         }
1320         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1321                    "or reserved", name);
1322         return NULL;
1323     found: ;
1324     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1325         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1326                    " reserved",
1327                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1328         return NULL;
1329     } else if (!pci_bus_devfn_available(bus, devfn)) {
1330         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1331                    " in use by %s,id=%s",
1332                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1333                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1334         return NULL;
1335     }
1336 
1337     /*
1338      * Populating function 0 triggers a scan from the guest that
1339      * exposes other non-zero functions. Hence we need to ensure that
1340      * function 0 wasn't added yet.
1341      */
1342     if (dev->hotplugged && !pci_is_vf(pci_dev) &&
1343         pci_get_function_0(pci_dev)) {
1344         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1345                    " new func %s cannot be exposed to guest.",
1346                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1347                    pci_get_function_0(pci_dev)->name,
1348                    name);
1349 
1350        return NULL;
1351     }
1352 
1353     pci_dev->devfn = devfn;
1354     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1355     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1356 
1357     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1358                        "bus master container", UINT64_MAX);
1359     address_space_init(&pci_dev->bus_master_as,
1360                        &pci_dev->bus_master_container_region, pci_dev->name);
1361     pci_dev->bus_master_as.max_bounce_buffer_size =
1362         pci_dev->max_bounce_buffer_size;
1363 
1364     if (phase_check(PHASE_MACHINE_READY)) {
1365         pci_init_bus_master(pci_dev);
1366     }
1367     pci_dev->irq_state = 0;
1368     pci_config_alloc(pci_dev);
1369 
1370     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1371     pci_config_set_device_id(pci_dev->config, pc->device_id);
1372     pci_config_set_revision(pci_dev->config, pc->revision);
1373     pci_config_set_class(pci_dev->config, pc->class_id);
1374 
1375     if (!is_bridge) {
1376         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1377             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1378                          pc->subsystem_vendor_id);
1379             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1380                          pc->subsystem_id);
1381         } else {
1382             pci_set_default_subsystem_id(pci_dev);
1383         }
1384     } else {
1385         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1386         assert(!pc->subsystem_vendor_id);
1387         assert(!pc->subsystem_id);
1388     }
1389     pci_init_cmask(pci_dev);
1390     pci_init_wmask(pci_dev);
1391     pci_init_w1cmask(pci_dev);
1392     if (is_bridge) {
1393         pci_init_mask_bridge(pci_dev);
1394     }
1395     pci_init_multifunction(bus, pci_dev, &local_err);
1396     if (local_err) {
1397         error_propagate(errp, local_err);
1398         do_pci_unregister_device(pci_dev);
1399         return NULL;
1400     }
1401 
1402     if (!config_read)
1403         config_read = pci_default_read_config;
1404     if (!config_write)
1405         config_write = pci_default_write_config;
1406     pci_dev->config_read = config_read;
1407     pci_dev->config_write = config_write;
1408     bus->devices[devfn] = pci_dev;
1409     pci_dev->version_id = 2; /* Current pci device vmstate version */
1410     return pci_dev;
1411 }
1412 
1413 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1414 {
1415     PCIIORegion *r;
1416     int i;
1417 
1418     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1419         r = &pci_dev->io_regions[i];
1420         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1421             continue;
1422         memory_region_del_subregion(r->address_space, r->memory);
1423     }
1424 
1425     pci_unregister_vga(pci_dev);
1426 }
1427 
1428 static void pci_qdev_unrealize(DeviceState *dev)
1429 {
1430     PCIDevice *pci_dev = PCI_DEVICE(dev);
1431     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1432 
1433     pci_unregister_io_regions(pci_dev);
1434     pci_del_option_rom(pci_dev);
1435 
1436     if (pc->exit) {
1437         pc->exit(pci_dev);
1438     }
1439 
1440     pci_device_deassert_intx(pci_dev);
1441     do_pci_unregister_device(pci_dev);
1442 
1443     pci_dev->msi_trigger = NULL;
1444 
1445     /*
1446      * clean up acpi-index so it could reused by another device
1447      */
1448     if (pci_dev->acpi_index) {
1449         GSequence *used_indexes = pci_acpi_index_list();
1450 
1451         g_sequence_remove(g_sequence_lookup(used_indexes,
1452                           GINT_TO_POINTER(pci_dev->acpi_index),
1453                           g_cmp_uint32, NULL));
1454     }
1455 }
1456 
1457 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1458                       uint8_t type, MemoryRegion *memory)
1459 {
1460     PCIIORegion *r;
1461     uint32_t addr; /* offset in pci config space */
1462     uint64_t wmask;
1463     pcibus_t size = memory_region_size(memory);
1464     uint8_t hdr_type;
1465 
1466     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1467     assert(region_num >= 0);
1468     assert(region_num < PCI_NUM_REGIONS);
1469     assert(is_power_of_2(size));
1470 
1471     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1472     hdr_type =
1473         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1474     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1475 
1476     r = &pci_dev->io_regions[region_num];
1477     assert(!r->size);
1478     r->addr = PCI_BAR_UNMAPPED;
1479     r->size = size;
1480     r->type = type;
1481     r->memory = memory;
1482     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1483                         ? pci_get_bus(pci_dev)->address_space_io
1484                         : pci_get_bus(pci_dev)->address_space_mem;
1485 
1486     wmask = ~(size - 1);
1487     if (region_num == PCI_ROM_SLOT) {
1488         /* ROM enable bit is writable */
1489         wmask |= PCI_ROM_ADDRESS_ENABLE;
1490     }
1491 
1492     addr = pci_bar(pci_dev, region_num);
1493     pci_set_long(pci_dev->config + addr, type);
1494 
1495     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1496         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1497         pci_set_quad(pci_dev->wmask + addr, wmask);
1498         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1499     } else {
1500         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1501         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1502     }
1503 }
1504 
1505 static void pci_update_vga(PCIDevice *pci_dev)
1506 {
1507     uint16_t cmd;
1508 
1509     if (!pci_dev->has_vga) {
1510         return;
1511     }
1512 
1513     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1514 
1515     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1516                               cmd & PCI_COMMAND_MEMORY);
1517     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1518                               cmd & PCI_COMMAND_IO);
1519     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1520                               cmd & PCI_COMMAND_IO);
1521 }
1522 
1523 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1524                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1525 {
1526     PCIBus *bus = pci_get_bus(pci_dev);
1527 
1528     assert(!pci_dev->has_vga);
1529 
1530     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1531     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1532     memory_region_add_subregion_overlap(bus->address_space_mem,
1533                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1534 
1535     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1536     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1537     memory_region_add_subregion_overlap(bus->address_space_io,
1538                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1539 
1540     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1541     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1542     memory_region_add_subregion_overlap(bus->address_space_io,
1543                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1544     pci_dev->has_vga = true;
1545 
1546     pci_update_vga(pci_dev);
1547 }
1548 
1549 void pci_unregister_vga(PCIDevice *pci_dev)
1550 {
1551     PCIBus *bus = pci_get_bus(pci_dev);
1552 
1553     if (!pci_dev->has_vga) {
1554         return;
1555     }
1556 
1557     memory_region_del_subregion(bus->address_space_mem,
1558                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1559     memory_region_del_subregion(bus->address_space_io,
1560                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1561     memory_region_del_subregion(bus->address_space_io,
1562                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1563     pci_dev->has_vga = false;
1564 }
1565 
1566 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1567 {
1568     return pci_dev->io_regions[region_num].addr;
1569 }
1570 
1571 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1572                                         uint8_t type, pcibus_t size)
1573 {
1574     pcibus_t new_addr;
1575     if (!pci_is_vf(d)) {
1576         int bar = pci_bar(d, reg);
1577         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1578             new_addr = pci_get_quad(d->config + bar);
1579         } else {
1580             new_addr = pci_get_long(d->config + bar);
1581         }
1582     } else {
1583         PCIDevice *pf = d->exp.sriov_vf.pf;
1584         uint16_t sriov_cap = pf->exp.sriov_cap;
1585         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1586         uint16_t vf_offset =
1587             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1588         uint16_t vf_stride =
1589             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1590         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1591 
1592         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1593             new_addr = pci_get_quad(pf->config + bar);
1594         } else {
1595             new_addr = pci_get_long(pf->config + bar);
1596         }
1597         new_addr += vf_num * size;
1598     }
1599     /* The ROM slot has a specific enable bit, keep it intact */
1600     if (reg != PCI_ROM_SLOT) {
1601         new_addr &= ~(size - 1);
1602     }
1603     return new_addr;
1604 }
1605 
1606 pcibus_t pci_bar_address(PCIDevice *d,
1607                          int reg, uint8_t type, pcibus_t size)
1608 {
1609     pcibus_t new_addr, last_addr;
1610     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1611     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1612     bool allow_0_address = mc->pci_allow_0_address;
1613 
1614     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1615         if (!(cmd & PCI_COMMAND_IO)) {
1616             return PCI_BAR_UNMAPPED;
1617         }
1618         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1619         last_addr = new_addr + size - 1;
1620         /* Check if 32 bit BAR wraps around explicitly.
1621          * TODO: make priorities correct and remove this work around.
1622          */
1623         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1624             (!allow_0_address && new_addr == 0)) {
1625             return PCI_BAR_UNMAPPED;
1626         }
1627         return new_addr;
1628     }
1629 
1630     if (!(cmd & PCI_COMMAND_MEMORY)) {
1631         return PCI_BAR_UNMAPPED;
1632     }
1633     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1634     /* the ROM slot has a specific enable bit */
1635     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1636         return PCI_BAR_UNMAPPED;
1637     }
1638     new_addr &= ~(size - 1);
1639     last_addr = new_addr + size - 1;
1640     /* NOTE: we do not support wrapping */
1641     /* XXX: as we cannot support really dynamic
1642        mappings, we handle specific values as invalid
1643        mappings. */
1644     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1645         (!allow_0_address && new_addr == 0)) {
1646         return PCI_BAR_UNMAPPED;
1647     }
1648 
1649     /* Now pcibus_t is 64bit.
1650      * Check if 32 bit BAR wraps around explicitly.
1651      * Without this, PC ide doesn't work well.
1652      * TODO: remove this work around.
1653      */
1654     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1655         return PCI_BAR_UNMAPPED;
1656     }
1657 
1658     /*
1659      * OS is allowed to set BAR beyond its addressable
1660      * bits. For example, 32 bit OS can set 64bit bar
1661      * to >4G. Check it. TODO: we might need to support
1662      * it in the future for e.g. PAE.
1663      */
1664     if (last_addr >= HWADDR_MAX) {
1665         return PCI_BAR_UNMAPPED;
1666     }
1667 
1668     return new_addr;
1669 }
1670 
1671 static void pci_update_mappings(PCIDevice *d)
1672 {
1673     PCIIORegion *r;
1674     int i;
1675     pcibus_t new_addr;
1676 
1677     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1678         r = &d->io_regions[i];
1679 
1680         /* this region isn't registered */
1681         if (!r->size)
1682             continue;
1683 
1684         new_addr = pci_bar_address(d, i, r->type, r->size);
1685         if (!d->enabled || pci_pm_state(d)) {
1686             new_addr = PCI_BAR_UNMAPPED;
1687         }
1688 
1689         /* This bar isn't changed */
1690         if (new_addr == r->addr)
1691             continue;
1692 
1693         /* now do the real mapping */
1694         if (r->addr != PCI_BAR_UNMAPPED) {
1695             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1696                                           PCI_SLOT(d->devfn),
1697                                           PCI_FUNC(d->devfn),
1698                                           i, r->addr, r->size);
1699             memory_region_del_subregion(r->address_space, r->memory);
1700         }
1701         r->addr = new_addr;
1702         if (r->addr != PCI_BAR_UNMAPPED) {
1703             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1704                                           PCI_SLOT(d->devfn),
1705                                           PCI_FUNC(d->devfn),
1706                                           i, r->addr, r->size);
1707             memory_region_add_subregion_overlap(r->address_space,
1708                                                 r->addr, r->memory, 1);
1709         }
1710     }
1711 
1712     pci_update_vga(d);
1713 }
1714 
1715 static inline int pci_irq_disabled(PCIDevice *d)
1716 {
1717     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1718 }
1719 
1720 /* Called after interrupt disabled field update in config space,
1721  * assert/deassert interrupts if necessary.
1722  * Gets original interrupt disable bit value (before update). */
1723 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1724 {
1725     int i, disabled = pci_irq_disabled(d);
1726     if (disabled == was_irq_disabled)
1727         return;
1728     for (i = 0; i < PCI_NUM_PINS; ++i) {
1729         int state = pci_irq_state(d, i);
1730         pci_change_irq_level(d, i, disabled ? -state : state);
1731     }
1732 }
1733 
1734 uint32_t pci_default_read_config(PCIDevice *d,
1735                                  uint32_t address, int len)
1736 {
1737     uint32_t val = 0;
1738 
1739     assert(address + len <= pci_config_size(d));
1740 
1741     if (pci_is_express_downstream_port(d) &&
1742         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1743         pcie_sync_bridge_lnk(d);
1744     }
1745     memcpy(&val, d->config + address, len);
1746     return le32_to_cpu(val);
1747 }
1748 
1749 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1750 {
1751     uint8_t new_pm_state, old_pm_state = pci_pm_state(d);
1752     int i, was_irq_disabled = pci_irq_disabled(d);
1753     uint32_t val = val_in;
1754 
1755     assert(addr + l <= pci_config_size(d));
1756 
1757     for (i = 0; i < l; val >>= 8, ++i) {
1758         uint8_t wmask = d->wmask[addr + i];
1759         uint8_t w1cmask = d->w1cmask[addr + i];
1760         assert(!(wmask & w1cmask));
1761         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1762         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1763     }
1764 
1765     new_pm_state = pci_pm_update(d, addr, l, old_pm_state);
1766 
1767     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1768         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1769         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1770         range_covers_byte(addr, l, PCI_COMMAND) ||
1771         !!new_pm_state != !!old_pm_state) {
1772         pci_update_mappings(d);
1773     }
1774 
1775     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1776         pci_update_irq_disabled(d, was_irq_disabled);
1777         memory_region_set_enabled(&d->bus_master_enable_region,
1778                                   (pci_get_word(d->config + PCI_COMMAND)
1779                                    & PCI_COMMAND_MASTER) && d->enabled);
1780     }
1781 
1782     msi_write_config(d, addr, val_in, l);
1783     msix_write_config(d, addr, val_in, l);
1784     pcie_sriov_config_write(d, addr, val_in, l);
1785 }
1786 
1787 /***********************************************************/
1788 /* generic PCI irq support */
1789 
1790 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1791 static void pci_irq_handler(void *opaque, int irq_num, int level)
1792 {
1793     PCIDevice *pci_dev = opaque;
1794     int change;
1795 
1796     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1797     assert(level == 0 || level == 1);
1798     change = level - pci_irq_state(pci_dev, irq_num);
1799     if (!change)
1800         return;
1801 
1802     pci_set_irq_state(pci_dev, irq_num, level);
1803     pci_update_irq_status(pci_dev);
1804     if (pci_irq_disabled(pci_dev))
1805         return;
1806     pci_change_irq_level(pci_dev, irq_num, change);
1807 }
1808 
1809 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1810 {
1811     int intx = pci_intx(pci_dev);
1812     assert(0 <= intx && intx < PCI_NUM_PINS);
1813 
1814     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1815 }
1816 
1817 void pci_set_irq(PCIDevice *pci_dev, int level)
1818 {
1819     int intx = pci_intx(pci_dev);
1820     pci_irq_handler(pci_dev, intx, level);
1821 }
1822 
1823 /* Special hooks used by device assignment */
1824 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1825 {
1826     assert(pci_bus_is_root(bus));
1827     bus->route_intx_to_irq = route_intx_to_irq;
1828 }
1829 
1830 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1831 {
1832     PCIBus *bus;
1833 
1834     do {
1835         int dev_irq = pin;
1836         bus = pci_get_bus(dev);
1837         pin = bus->map_irq(dev, pin);
1838         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1839                             pci_bus_is_root(bus) ? "root-complex"
1840                                     : DEVICE(bus->parent_dev)->canonical_path);
1841         dev = bus->parent_dev;
1842     } while (dev);
1843 
1844     if (!bus->route_intx_to_irq) {
1845         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1846                      object_get_typename(OBJECT(bus->qbus.parent)));
1847         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1848     }
1849 
1850     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1851 }
1852 
1853 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1854 {
1855     return old->mode != new->mode || old->irq != new->irq;
1856 }
1857 
1858 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1859 {
1860     PCIDevice *dev;
1861     PCIBus *sec;
1862     int i;
1863 
1864     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1865         dev = bus->devices[i];
1866         if (dev && dev->intx_routing_notifier) {
1867             dev->intx_routing_notifier(dev);
1868         }
1869     }
1870 
1871     QLIST_FOREACH(sec, &bus->child, sibling) {
1872         pci_bus_fire_intx_routing_notifier(sec);
1873     }
1874 }
1875 
1876 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1877                                           PCIINTxRoutingNotifier notifier)
1878 {
1879     dev->intx_routing_notifier = notifier;
1880 }
1881 
1882 /*
1883  * PCI-to-PCI bridge specification
1884  * 9.1: Interrupt routing. Table 9-1
1885  *
1886  * the PCI Express Base Specification, Revision 2.1
1887  * 2.2.8.1: INTx interrupt signaling - Rules
1888  *          the Implementation Note
1889  *          Table 2-20
1890  */
1891 /*
1892  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1893  * 0-origin unlike PCI interrupt pin register.
1894  */
1895 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1896 {
1897     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1898 }
1899 
1900 /***********************************************************/
1901 /* monitor info on PCI */
1902 
1903 static const pci_class_desc pci_class_descriptions[] =
1904 {
1905     { 0x0001, "VGA controller", "display"},
1906     { 0x0100, "SCSI controller", "scsi"},
1907     { 0x0101, "IDE controller", "ide"},
1908     { 0x0102, "Floppy controller", "fdc"},
1909     { 0x0103, "IPI controller", "ipi"},
1910     { 0x0104, "RAID controller", "raid"},
1911     { 0x0106, "SATA controller"},
1912     { 0x0107, "SAS controller"},
1913     { 0x0180, "Storage controller"},
1914     { 0x0200, "Ethernet controller", "ethernet"},
1915     { 0x0201, "Token Ring controller", "token-ring"},
1916     { 0x0202, "FDDI controller", "fddi"},
1917     { 0x0203, "ATM controller", "atm"},
1918     { 0x0280, "Network controller"},
1919     { 0x0300, "VGA controller", "display", 0x00ff},
1920     { 0x0301, "XGA controller"},
1921     { 0x0302, "3D controller"},
1922     { 0x0380, "Display controller"},
1923     { 0x0400, "Video controller", "video"},
1924     { 0x0401, "Audio controller", "sound"},
1925     { 0x0402, "Phone"},
1926     { 0x0403, "Audio controller", "sound"},
1927     { 0x0480, "Multimedia controller"},
1928     { 0x0500, "RAM controller", "memory"},
1929     { 0x0501, "Flash controller", "flash"},
1930     { 0x0580, "Memory controller"},
1931     { 0x0600, "Host bridge", "host"},
1932     { 0x0601, "ISA bridge", "isa"},
1933     { 0x0602, "EISA bridge", "eisa"},
1934     { 0x0603, "MC bridge", "mca"},
1935     { 0x0604, "PCI bridge", "pci-bridge"},
1936     { 0x0605, "PCMCIA bridge", "pcmcia"},
1937     { 0x0606, "NUBUS bridge", "nubus"},
1938     { 0x0607, "CARDBUS bridge", "cardbus"},
1939     { 0x0608, "RACEWAY bridge"},
1940     { 0x0680, "Bridge"},
1941     { 0x0700, "Serial port", "serial"},
1942     { 0x0701, "Parallel port", "parallel"},
1943     { 0x0800, "Interrupt controller", "interrupt-controller"},
1944     { 0x0801, "DMA controller", "dma-controller"},
1945     { 0x0802, "Timer", "timer"},
1946     { 0x0803, "RTC", "rtc"},
1947     { 0x0900, "Keyboard", "keyboard"},
1948     { 0x0901, "Pen", "pen"},
1949     { 0x0902, "Mouse", "mouse"},
1950     { 0x0A00, "Dock station", "dock", 0x00ff},
1951     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1952     { 0x0c00, "Firewire controller", "firewire"},
1953     { 0x0c01, "Access bus controller", "access-bus"},
1954     { 0x0c02, "SSA controller", "ssa"},
1955     { 0x0c03, "USB controller", "usb"},
1956     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1957     { 0x0c05, "SMBus"},
1958     { 0, NULL}
1959 };
1960 
1961 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1962                                            pci_bus_dev_fn fn,
1963                                            void *opaque)
1964 {
1965     PCIDevice *d;
1966     int devfn;
1967 
1968     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1969         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1970         if (d) {
1971             fn(bus, d, opaque);
1972         }
1973     }
1974 }
1975 
1976 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1977                                  pci_bus_dev_fn fn, void *opaque)
1978 {
1979     bus = pci_find_bus_nr(bus, bus_num);
1980 
1981     if (bus) {
1982         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1983     }
1984 }
1985 
1986 void pci_for_each_device_under_bus(PCIBus *bus,
1987                                    pci_bus_dev_fn fn, void *opaque)
1988 {
1989     PCIDevice *d;
1990     int devfn;
1991 
1992     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1993         d = bus->devices[devfn];
1994         if (d) {
1995             fn(bus, d, opaque);
1996         }
1997     }
1998 }
1999 
2000 void pci_for_each_device(PCIBus *bus, int bus_num,
2001                          pci_bus_dev_fn fn, void *opaque)
2002 {
2003     bus = pci_find_bus_nr(bus, bus_num);
2004 
2005     if (bus) {
2006         pci_for_each_device_under_bus(bus, fn, opaque);
2007     }
2008 }
2009 
2010 const pci_class_desc *get_class_desc(int class)
2011 {
2012     const pci_class_desc *desc;
2013 
2014     desc = pci_class_descriptions;
2015     while (desc->desc && class != desc->class) {
2016         desc++;
2017     }
2018 
2019     return desc;
2020 }
2021 
2022 void pci_init_nic_devices(PCIBus *bus, const char *default_model)
2023 {
2024     qemu_create_nic_bus_devices(&bus->qbus, TYPE_PCI_DEVICE, default_model,
2025                                 "virtio", "virtio-net-pci");
2026 }
2027 
2028 bool pci_init_nic_in_slot(PCIBus *rootbus, const char *model,
2029                           const char *alias, const char *devaddr)
2030 {
2031     NICInfo *nd = qemu_find_nic_info(model, true, alias);
2032     int dom, busnr, devfn;
2033     PCIDevice *pci_dev;
2034     unsigned slot;
2035     PCIBus *bus;
2036 
2037     if (!nd) {
2038         return false;
2039     }
2040 
2041     if (!devaddr || pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
2042         error_report("Invalid PCI device address %s for device %s",
2043                      devaddr, model);
2044         exit(1);
2045     }
2046 
2047     if (dom != 0) {
2048         error_report("No support for non-zero PCI domains");
2049         exit(1);
2050     }
2051 
2052     devfn = PCI_DEVFN(slot, 0);
2053 
2054     bus = pci_find_bus_nr(rootbus, busnr);
2055     if (!bus) {
2056         error_report("Invalid PCI device address %s for device %s",
2057                      devaddr, model);
2058         exit(1);
2059     }
2060 
2061     pci_dev = pci_new(devfn, model);
2062     qdev_set_nic_properties(&pci_dev->qdev, nd);
2063     pci_realize_and_unref(pci_dev, bus, &error_fatal);
2064     return true;
2065 }
2066 
2067 PCIDevice *pci_vga_init(PCIBus *bus)
2068 {
2069     vga_interface_created = true;
2070     switch (vga_interface_type) {
2071     case VGA_CIRRUS:
2072         return pci_create_simple(bus, -1, "cirrus-vga");
2073     case VGA_QXL:
2074         return pci_create_simple(bus, -1, "qxl-vga");
2075     case VGA_STD:
2076         return pci_create_simple(bus, -1, "VGA");
2077     case VGA_VMWARE:
2078         return pci_create_simple(bus, -1, "vmware-svga");
2079     case VGA_VIRTIO:
2080         return pci_create_simple(bus, -1, "virtio-vga");
2081     case VGA_NONE:
2082     default: /* Other non-PCI types. Checking for unsupported types is already
2083                 done in vl.c. */
2084         return NULL;
2085     }
2086 }
2087 
2088 /* Whether a given bus number is in range of the secondary
2089  * bus of the given bridge device. */
2090 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
2091 {
2092     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
2093              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
2094         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
2095         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
2096 }
2097 
2098 /* Whether a given bus number is in a range of a root bus */
2099 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
2100 {
2101     int i;
2102 
2103     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
2104         PCIDevice *dev = bus->devices[i];
2105 
2106         if (dev && IS_PCI_BRIDGE(dev)) {
2107             if (pci_secondary_bus_in_range(dev, bus_num)) {
2108                 return true;
2109             }
2110         }
2111     }
2112 
2113     return false;
2114 }
2115 
2116 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2117 {
2118     PCIBus *sec;
2119 
2120     if (!bus) {
2121         return NULL;
2122     }
2123 
2124     if (pci_bus_num(bus) == bus_num) {
2125         return bus;
2126     }
2127 
2128     /* Consider all bus numbers in range for the host pci bridge. */
2129     if (!pci_bus_is_root(bus) &&
2130         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2131         return NULL;
2132     }
2133 
2134     /* try child bus */
2135     for (; bus; bus = sec) {
2136         QLIST_FOREACH(sec, &bus->child, sibling) {
2137             if (pci_bus_num(sec) == bus_num) {
2138                 return sec;
2139             }
2140             /* PXB buses assumed to be children of bus 0 */
2141             if (pci_bus_is_root(sec)) {
2142                 if (pci_root_bus_in_range(sec, bus_num)) {
2143                     break;
2144                 }
2145             } else {
2146                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2147                     break;
2148                 }
2149             }
2150         }
2151     }
2152 
2153     return NULL;
2154 }
2155 
2156 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
2157                                   pci_bus_fn end, void *parent_state)
2158 {
2159     PCIBus *sec;
2160     void *state;
2161 
2162     if (!bus) {
2163         return;
2164     }
2165 
2166     if (begin) {
2167         state = begin(bus, parent_state);
2168     } else {
2169         state = parent_state;
2170     }
2171 
2172     QLIST_FOREACH(sec, &bus->child, sibling) {
2173         pci_for_each_bus_depth_first(sec, begin, end, state);
2174     }
2175 
2176     if (end) {
2177         end(bus, state);
2178     }
2179 }
2180 
2181 
2182 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2183 {
2184     bus = pci_find_bus_nr(bus, bus_num);
2185 
2186     if (!bus)
2187         return NULL;
2188 
2189     return bus->devices[devfn];
2190 }
2191 
2192 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2193 
2194 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2195 {
2196     PCIDevice *pci_dev = (PCIDevice *)qdev;
2197     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2198     ObjectClass *klass = OBJECT_CLASS(pc);
2199     Error *local_err = NULL;
2200     bool is_default_rom;
2201     uint16_t class_id;
2202 
2203     /*
2204      * capped by systemd (see: udev-builtin-net_id.c)
2205      * as it's the only known user honor it to avoid users
2206      * misconfigure QEMU and then wonder why acpi-index doesn't work
2207      */
2208     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2209         error_setg(errp, "acpi-index should be less or equal to %u",
2210                    ONBOARD_INDEX_MAX);
2211         return;
2212     }
2213 
2214     /*
2215      * make sure that acpi-index is unique across all present PCI devices
2216      */
2217     if (pci_dev->acpi_index) {
2218         GSequence *used_indexes = pci_acpi_index_list();
2219 
2220         if (g_sequence_lookup(used_indexes,
2221                               GINT_TO_POINTER(pci_dev->acpi_index),
2222                               g_cmp_uint32, NULL)) {
2223             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2224                        " already exist", pci_dev->acpi_index);
2225             return;
2226         }
2227         g_sequence_insert_sorted(used_indexes,
2228                                  GINT_TO_POINTER(pci_dev->acpi_index),
2229                                  g_cmp_uint32, NULL);
2230     }
2231 
2232     if (pci_dev->romsize != UINT32_MAX && !is_power_of_2(pci_dev->romsize)) {
2233         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2234         return;
2235     }
2236 
2237     /* initialize cap_present for pci_is_express() and pci_config_size(),
2238      * Note that hybrid PCIs are not set automatically and need to manage
2239      * QEMU_PCI_CAP_EXPRESS manually */
2240     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2241        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2242         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2243     }
2244 
2245     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2246         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2247     }
2248 
2249     pci_dev = do_pci_register_device(pci_dev,
2250                                      object_get_typename(OBJECT(qdev)),
2251                                      pci_dev->devfn, errp);
2252     if (pci_dev == NULL)
2253         return;
2254 
2255     if (pc->realize) {
2256         pc->realize(pci_dev, &local_err);
2257         if (local_err) {
2258             error_propagate(errp, local_err);
2259             do_pci_unregister_device(pci_dev);
2260             return;
2261         }
2262     }
2263 
2264     /*
2265      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2266      * associate only Device 0 with the device attached to the bus
2267      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2268      * sec 7.3.1).
2269      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2270      * 5-bit Device Number and 3-bit Function Number fields in its associated
2271      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2272      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2273      */
2274     if (pci_is_express(pci_dev) &&
2275         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2276         pcie_has_upstream_port(pci_dev) &&
2277         PCI_SLOT(pci_dev->devfn)) {
2278         warn_report("PCI: slot %d is not valid for %s,"
2279                     " parent device only allows plugging into slot 0.",
2280                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2281     }
2282 
2283     if (pci_dev->failover_pair_id) {
2284         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2285             error_setg(errp, "failover primary device must be on "
2286                              "PCIExpress bus");
2287             pci_qdev_unrealize(DEVICE(pci_dev));
2288             return;
2289         }
2290         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2291         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2292             error_setg(errp, "failover primary device is not an "
2293                              "Ethernet device");
2294             pci_qdev_unrealize(DEVICE(pci_dev));
2295             return;
2296         }
2297         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2298             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2299             error_setg(errp, "failover: primary device must be in its own "
2300                               "PCI slot");
2301             pci_qdev_unrealize(DEVICE(pci_dev));
2302             return;
2303         }
2304         qdev->allow_unplug_during_migration = true;
2305     }
2306 
2307     /* rom loading */
2308     is_default_rom = false;
2309     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2310         pci_dev->romfile = g_strdup(pc->romfile);
2311         is_default_rom = true;
2312     }
2313 
2314     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2315     if (local_err) {
2316         error_propagate(errp, local_err);
2317         pci_qdev_unrealize(DEVICE(pci_dev));
2318         return;
2319     }
2320 
2321     pci_set_power(pci_dev, true);
2322 
2323     pci_dev->msi_trigger = pci_msi_trigger;
2324 }
2325 
2326 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2327                                    const char *name)
2328 {
2329     DeviceState *dev;
2330 
2331     dev = qdev_new(name);
2332     qdev_prop_set_int32(dev, "addr", devfn);
2333     qdev_prop_set_bit(dev, "multifunction", multifunction);
2334     return PCI_DEVICE(dev);
2335 }
2336 
2337 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2338 {
2339     return pci_new_internal(devfn, true, name);
2340 }
2341 
2342 PCIDevice *pci_new(int devfn, const char *name)
2343 {
2344     return pci_new_internal(devfn, false, name);
2345 }
2346 
2347 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2348 {
2349     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2350 }
2351 
2352 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2353                                            const char *name)
2354 {
2355     PCIDevice *dev = pci_new_multifunction(devfn, name);
2356     pci_realize_and_unref(dev, bus, &error_fatal);
2357     return dev;
2358 }
2359 
2360 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2361 {
2362     PCIDevice *dev = pci_new(devfn, name);
2363     pci_realize_and_unref(dev, bus, &error_fatal);
2364     return dev;
2365 }
2366 
2367 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2368 {
2369     int offset = PCI_CONFIG_HEADER_SIZE;
2370     int i;
2371     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2372         if (pdev->used[i])
2373             offset = i + 1;
2374         else if (i - offset + 1 == size)
2375             return offset;
2376     }
2377     return 0;
2378 }
2379 
2380 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2381                                         uint8_t *prev_p)
2382 {
2383     uint8_t next, prev;
2384 
2385     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2386         return 0;
2387 
2388     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2389          prev = next + PCI_CAP_LIST_NEXT)
2390         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2391             break;
2392 
2393     if (prev_p)
2394         *prev_p = prev;
2395     return next;
2396 }
2397 
2398 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2399 {
2400     uint8_t next, prev, found = 0;
2401 
2402     if (!(pdev->used[offset])) {
2403         return 0;
2404     }
2405 
2406     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2407 
2408     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2409          prev = next + PCI_CAP_LIST_NEXT) {
2410         if (next <= offset && next > found) {
2411             found = next;
2412         }
2413     }
2414     return found;
2415 }
2416 
2417 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2418    This is needed for an option rom which is used for more than one device. */
2419 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2420 {
2421     uint16_t vendor_id;
2422     uint16_t device_id;
2423     uint16_t rom_vendor_id;
2424     uint16_t rom_device_id;
2425     uint16_t rom_magic;
2426     uint16_t pcir_offset;
2427     uint8_t checksum;
2428 
2429     /* Words in rom data are little endian (like in PCI configuration),
2430        so they can be read / written with pci_get_word / pci_set_word. */
2431 
2432     /* Only a valid rom will be patched. */
2433     rom_magic = pci_get_word(ptr);
2434     if (rom_magic != 0xaa55) {
2435         trace_pci_bad_rom_magic(rom_magic, 0xaa55);
2436         return;
2437     }
2438     pcir_offset = pci_get_word(ptr + 0x18);
2439     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2440         trace_pci_bad_pcir_offset(pcir_offset);
2441         return;
2442     }
2443 
2444     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2445     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2446     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2447     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2448 
2449     trace_pci_rom_and_pci_ids(pdev->romfile, vendor_id, device_id,
2450                               rom_vendor_id, rom_device_id);
2451 
2452     checksum = ptr[6];
2453 
2454     if (vendor_id != rom_vendor_id) {
2455         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2456         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2457         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2458         trace_pci_rom_checksum_change(ptr[6], checksum);
2459         ptr[6] = checksum;
2460         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2461     }
2462 
2463     if (device_id != rom_device_id) {
2464         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2465         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2466         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2467         trace_pci_rom_checksum_change(ptr[6], checksum);
2468         ptr[6] = checksum;
2469         pci_set_word(ptr + pcir_offset + 6, device_id);
2470     }
2471 }
2472 
2473 /* Add an option rom for the device */
2474 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2475                                Error **errp)
2476 {
2477     int64_t size = 0;
2478     g_autofree char *path = NULL;
2479     char name[32];
2480     const VMStateDescription *vmsd;
2481 
2482     /*
2483      * In case of incoming migration ROM will come with migration stream, no
2484      * reason to load the file.  Neither we want to fail if local ROM file
2485      * mismatches with specified romsize.
2486      */
2487     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2488 
2489     if (!pdev->romfile || !strlen(pdev->romfile)) {
2490         return;
2491     }
2492 
2493     if (!pdev->rom_bar) {
2494         /*
2495          * Load rom via fw_cfg instead of creating a rom bar,
2496          * for 0.11 compatibility.
2497          */
2498         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2499 
2500         /*
2501          * Hot-plugged devices can't use the option ROM
2502          * if the rom bar is disabled.
2503          */
2504         if (DEVICE(pdev)->hotplugged) {
2505             error_setg(errp, "Hot-plugged device without ROM bar"
2506                        " can't have an option ROM");
2507             return;
2508         }
2509 
2510         if (class == 0x0300) {
2511             rom_add_vga(pdev->romfile);
2512         } else {
2513             rom_add_option(pdev->romfile, -1);
2514         }
2515         return;
2516     }
2517 
2518     if (load_file || pdev->romsize == UINT32_MAX) {
2519         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2520         if (path == NULL) {
2521             path = g_strdup(pdev->romfile);
2522         }
2523 
2524         size = get_image_size(path);
2525         if (size < 0) {
2526             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2527             return;
2528         } else if (size == 0) {
2529             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2530             return;
2531         } else if (size > 2 * GiB) {
2532             error_setg(errp,
2533                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2534                        pdev->romfile);
2535             return;
2536         }
2537         if (pdev->romsize != UINT_MAX) {
2538             if (size > pdev->romsize) {
2539                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2540                            "is too large for ROM size %u",
2541                            pdev->romfile, (uint32_t)size, pdev->romsize);
2542                 return;
2543             }
2544         } else {
2545             pdev->romsize = pow2ceil(size);
2546         }
2547     }
2548 
2549     vmsd = qdev_get_vmsd(DEVICE(pdev));
2550     snprintf(name, sizeof(name), "%s.rom",
2551              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2552 
2553     pdev->has_rom = true;
2554     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2555                            &error_fatal);
2556 
2557     if (load_file) {
2558         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2559 
2560         if (load_image_size(path, ptr, size) < 0) {
2561             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2562             return;
2563         }
2564 
2565         if (is_default_rom) {
2566             /* Only the default rom images will be patched (if needed). */
2567             pci_patch_ids(pdev, ptr, size);
2568         }
2569     }
2570 
2571     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2572 }
2573 
2574 static void pci_del_option_rom(PCIDevice *pdev)
2575 {
2576     if (!pdev->has_rom)
2577         return;
2578 
2579     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2580     pdev->has_rom = false;
2581 }
2582 
2583 /*
2584  * On success, pci_add_capability() returns a positive value
2585  * that the offset of the pci capability.
2586  * On failure, it sets an error and returns a negative error
2587  * code.
2588  */
2589 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2590                        uint8_t offset, uint8_t size,
2591                        Error **errp)
2592 {
2593     uint8_t *config;
2594     int i, overlapping_cap;
2595 
2596     if (!offset) {
2597         offset = pci_find_space(pdev, size);
2598         /* out of PCI config space is programming error */
2599         assert(offset);
2600     } else {
2601         /* Verify that capabilities don't overlap.  Note: device assignment
2602          * depends on this check to verify that the device is not broken.
2603          * Should never trigger for emulated devices, but it's helpful
2604          * for debugging these. */
2605         for (i = offset; i < offset + size; i++) {
2606             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2607             if (overlapping_cap) {
2608                 error_setg(errp, "%s:%02x:%02x.%x "
2609                            "Attempt to add PCI capability %x at offset "
2610                            "%x overlaps existing capability %x at offset %x",
2611                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2612                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2613                            cap_id, offset, overlapping_cap, i);
2614                 return -EINVAL;
2615             }
2616         }
2617     }
2618 
2619     config = pdev->config + offset;
2620     config[PCI_CAP_LIST_ID] = cap_id;
2621     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2622     pdev->config[PCI_CAPABILITY_LIST] = offset;
2623     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2624     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2625     /* Make capability read-only by default */
2626     memset(pdev->wmask + offset, 0, size);
2627     /* Check capability by default */
2628     memset(pdev->cmask + offset, 0xFF, size);
2629     return offset;
2630 }
2631 
2632 /* Unlink capability from the pci config space. */
2633 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2634 {
2635     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2636     if (!offset)
2637         return;
2638     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2639     /* Make capability writable again */
2640     memset(pdev->wmask + offset, 0xff, size);
2641     memset(pdev->w1cmask + offset, 0, size);
2642     /* Clear cmask as device-specific registers can't be checked */
2643     memset(pdev->cmask + offset, 0, size);
2644     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2645 
2646     if (!pdev->config[PCI_CAPABILITY_LIST])
2647         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2648 }
2649 
2650 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2651 {
2652     return pci_find_capability_list(pdev, cap_id, NULL);
2653 }
2654 
2655 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2656 {
2657     PCIDevice *d = (PCIDevice *)dev;
2658     const char *name = NULL;
2659     const pci_class_desc *desc =  pci_class_descriptions;
2660     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2661 
2662     while (desc->desc &&
2663           (class & ~desc->fw_ign_bits) !=
2664           (desc->class & ~desc->fw_ign_bits)) {
2665         desc++;
2666     }
2667 
2668     if (desc->desc) {
2669         name = desc->fw_name;
2670     }
2671 
2672     if (name) {
2673         pstrcpy(buf, len, name);
2674     } else {
2675         snprintf(buf, len, "pci%04x,%04x",
2676                  pci_get_word(d->config + PCI_VENDOR_ID),
2677                  pci_get_word(d->config + PCI_DEVICE_ID));
2678     }
2679 
2680     return buf;
2681 }
2682 
2683 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2684 {
2685     PCIDevice *d = (PCIDevice *)dev;
2686     char name[33];
2687     int has_func = !!PCI_FUNC(d->devfn);
2688 
2689     return g_strdup_printf("%s@%x%s%.*x",
2690                            pci_dev_fw_name(dev, name, sizeof(name)),
2691                            PCI_SLOT(d->devfn),
2692                            has_func ? "," : "",
2693                            has_func,
2694                            PCI_FUNC(d->devfn));
2695 }
2696 
2697 static char *pcibus_get_dev_path(DeviceState *dev)
2698 {
2699     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2700     PCIDevice *t;
2701     int slot_depth;
2702     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2703      * 00 is added here to make this format compatible with
2704      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2705      * Slot.Function list specifies the slot and function numbers for all
2706      * devices on the path from root to the specific device. */
2707     const char *root_bus_path;
2708     int root_bus_len;
2709     char slot[] = ":SS.F";
2710     int slot_len = sizeof slot - 1 /* For '\0' */;
2711     int path_len;
2712     char *path, *p;
2713     int s;
2714 
2715     root_bus_path = pci_root_bus_path(d);
2716     root_bus_len = strlen(root_bus_path);
2717 
2718     /* Calculate # of slots on path between device and root. */;
2719     slot_depth = 0;
2720     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2721         ++slot_depth;
2722     }
2723 
2724     path_len = root_bus_len + slot_len * slot_depth;
2725 
2726     /* Allocate memory, fill in the terminating null byte. */
2727     path = g_malloc(path_len + 1 /* For '\0' */);
2728     path[path_len] = '\0';
2729 
2730     memcpy(path, root_bus_path, root_bus_len);
2731 
2732     /* Fill in slot numbers. We walk up from device to root, so need to print
2733      * them in the reverse order, last to first. */
2734     p = path + path_len;
2735     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2736         p -= slot_len;
2737         s = snprintf(slot, sizeof slot, ":%02x.%x",
2738                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2739         assert(s == slot_len);
2740         memcpy(p, slot, slot_len);
2741     }
2742 
2743     return path;
2744 }
2745 
2746 static int pci_qdev_find_recursive(PCIBus *bus,
2747                                    const char *id, PCIDevice **pdev)
2748 {
2749     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2750     if (!qdev) {
2751         return -ENODEV;
2752     }
2753 
2754     /* roughly check if given qdev is pci device */
2755     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2756         *pdev = PCI_DEVICE(qdev);
2757         return 0;
2758     }
2759     return -EINVAL;
2760 }
2761 
2762 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2763 {
2764     PCIHostState *host_bridge;
2765     int rc = -ENODEV;
2766 
2767     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2768         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2769         if (!tmp) {
2770             rc = 0;
2771             break;
2772         }
2773         if (tmp != -ENODEV) {
2774             rc = tmp;
2775         }
2776     }
2777 
2778     return rc;
2779 }
2780 
2781 MemoryRegion *pci_address_space(PCIDevice *dev)
2782 {
2783     return pci_get_bus(dev)->address_space_mem;
2784 }
2785 
2786 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2787 {
2788     return pci_get_bus(dev)->address_space_io;
2789 }
2790 
2791 static void pci_device_class_init(ObjectClass *klass, const void *data)
2792 {
2793     DeviceClass *k = DEVICE_CLASS(klass);
2794 
2795     k->realize = pci_qdev_realize;
2796     k->unrealize = pci_qdev_unrealize;
2797     k->bus_type = TYPE_PCI_BUS;
2798     device_class_set_props(k, pci_props);
2799     object_class_property_set_description(
2800         klass, "x-max-bounce-buffer-size",
2801         "Maximum buffer size allocated for bounce buffers used for mapped "
2802         "access to indirect DMA memory");
2803 }
2804 
2805 static void pci_device_class_base_init(ObjectClass *klass, const void *data)
2806 {
2807     if (!object_class_is_abstract(klass)) {
2808         ObjectClass *conventional =
2809             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2810         ObjectClass *pcie =
2811             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2812         ObjectClass *cxl =
2813             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2814         assert(conventional || pcie || cxl);
2815     }
2816 }
2817 
2818 /*
2819  * Get IOMMU root bus, aliased bus and devfn of a PCI device
2820  *
2821  * IOMMU root bus is needed by all call sites to call into iommu_ops.
2822  * For call sites which don't need aliased BDF, passing NULL to
2823  * aliased_[bus|devfn] is allowed.
2824  *
2825  * @piommu_bus: return root #PCIBus backed by an IOMMU for the PCI device.
2826  *
2827  * @aliased_bus: return aliased #PCIBus of the PCI device, optional.
2828  *
2829  * @aliased_devfn: return aliased devfn of the PCI device, optional.
2830  */
2831 static void pci_device_get_iommu_bus_devfn(PCIDevice *dev,
2832                                            PCIBus **piommu_bus,
2833                                            PCIBus **aliased_bus,
2834                                            int *aliased_devfn)
2835 {
2836     PCIBus *bus = pci_get_bus(dev);
2837     PCIBus *iommu_bus = bus;
2838     int devfn = dev->devfn;
2839 
2840     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2841         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2842 
2843         /*
2844          * The requester ID of the provided device may be aliased, as seen from
2845          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2846          * requester ID to provide a unique AddressSpace for devices, but
2847          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2848          * to-PCI bridge creates and accepts transactions on behalf of down-
2849          * stream devices.  When doing so, all downstream devices are masked
2850          * (aliased) behind a single requester ID.  The requester ID used
2851          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2852          * bridges, with a PCIe capability indicating such, follow the
2853          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2854          * where the bridge uses the seconary bus as the bridge portion of the
2855          * requester ID and devfn of 00.0.  For other bridges, typically those
2856          * found on the root complex such as the dmi-to-pci-bridge, we follow
2857          * the convention of typical bare-metal hardware, which uses the
2858          * requester ID of the bridge itself.  There are device specific
2859          * exceptions to these rules, but these are the defaults that the
2860          * Linux kernel uses when determining DMA aliases itself and believed
2861          * to be true for the bare metal equivalents of the devices emulated
2862          * in QEMU.
2863          */
2864         if (!pci_bus_is_express(iommu_bus)) {
2865             PCIDevice *parent = iommu_bus->parent_dev;
2866 
2867             if (pci_is_express(parent) &&
2868                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2869                 devfn = PCI_DEVFN(0, 0);
2870                 bus = iommu_bus;
2871             } else {
2872                 devfn = parent->devfn;
2873                 bus = parent_bus;
2874             }
2875         }
2876 
2877         iommu_bus = parent_bus;
2878     }
2879 
2880     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
2881     assert(iommu_bus);
2882 
2883     if (pci_bus_bypass_iommu(bus) || !iommu_bus->iommu_ops) {
2884         iommu_bus = NULL;
2885     }
2886 
2887     *piommu_bus = iommu_bus;
2888 
2889     if (aliased_bus) {
2890         *aliased_bus = bus;
2891     }
2892 
2893     if (aliased_devfn) {
2894         *aliased_devfn = devfn;
2895     }
2896 }
2897 
2898 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2899 {
2900     PCIBus *bus;
2901     PCIBus *iommu_bus;
2902     int devfn;
2903 
2904     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn);
2905     if (iommu_bus) {
2906         return iommu_bus->iommu_ops->get_address_space(bus,
2907                                  iommu_bus->iommu_opaque, devfn);
2908     }
2909     return &address_space_memory;
2910 }
2911 
2912 bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod,
2913                                  Error **errp)
2914 {
2915     PCIBus *iommu_bus, *aliased_bus;
2916     int aliased_devfn;
2917 
2918     /* set_iommu_device requires device's direct BDF instead of aliased BDF */
2919     pci_device_get_iommu_bus_devfn(dev, &iommu_bus,
2920                                    &aliased_bus, &aliased_devfn);
2921     if (iommu_bus && iommu_bus->iommu_ops->set_iommu_device) {
2922         hiod->aliased_bus = aliased_bus;
2923         hiod->aliased_devfn = aliased_devfn;
2924         return iommu_bus->iommu_ops->set_iommu_device(pci_get_bus(dev),
2925                                                       iommu_bus->iommu_opaque,
2926                                                       dev->devfn, hiod, errp);
2927     }
2928     return true;
2929 }
2930 
2931 void pci_device_unset_iommu_device(PCIDevice *dev)
2932 {
2933     PCIBus *iommu_bus;
2934 
2935     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, NULL, NULL);
2936     if (iommu_bus && iommu_bus->iommu_ops->unset_iommu_device) {
2937         return iommu_bus->iommu_ops->unset_iommu_device(pci_get_bus(dev),
2938                                                         iommu_bus->iommu_opaque,
2939                                                         dev->devfn);
2940     }
2941 }
2942 
2943 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2944 {
2945     /*
2946      * If called, pci_setup_iommu() should provide a minimum set of
2947      * useful callbacks for the bus.
2948      */
2949     assert(ops);
2950     assert(ops->get_address_space);
2951 
2952     bus->iommu_ops = ops;
2953     bus->iommu_opaque = opaque;
2954 }
2955 
2956 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2957 {
2958     Range *range = opaque;
2959     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2960     int i;
2961 
2962     if (!(cmd & PCI_COMMAND_MEMORY)) {
2963         return;
2964     }
2965 
2966     if (IS_PCI_BRIDGE(dev)) {
2967         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2968         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2969 
2970         base = MAX(base, 0x1ULL << 32);
2971 
2972         if (limit >= base) {
2973             Range pref_range;
2974             range_set_bounds(&pref_range, base, limit);
2975             range_extend(range, &pref_range);
2976         }
2977     }
2978     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2979         PCIIORegion *r = &dev->io_regions[i];
2980         pcibus_t lob, upb;
2981         Range region_range;
2982 
2983         if (!r->size ||
2984             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2985             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2986             continue;
2987         }
2988 
2989         lob = pci_bar_address(dev, i, r->type, r->size);
2990         upb = lob + r->size - 1;
2991         if (lob == PCI_BAR_UNMAPPED) {
2992             continue;
2993         }
2994 
2995         lob = MAX(lob, 0x1ULL << 32);
2996 
2997         if (upb >= lob) {
2998             range_set_bounds(&region_range, lob, upb);
2999             range_extend(range, &region_range);
3000         }
3001     }
3002 }
3003 
3004 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
3005 {
3006     range_make_empty(range);
3007     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
3008 }
3009 
3010 static bool pcie_has_upstream_port(PCIDevice *dev)
3011 {
3012     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
3013 
3014     /* Device associated with an upstream port.
3015      * As there are several types of these, it's easier to check the
3016      * parent device: upstream ports are always connected to
3017      * root or downstream ports.
3018      */
3019     return parent_dev &&
3020         pci_is_express(parent_dev) &&
3021         parent_dev->exp.exp_cap &&
3022         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
3023          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
3024 }
3025 
3026 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
3027 {
3028     PCIBus *bus = pci_get_bus(pci_dev);
3029 
3030     if(pcie_has_upstream_port(pci_dev)) {
3031         /* With an upstream PCIe port, we only support 1 device at slot 0 */
3032         return bus->devices[0];
3033     } else {
3034         /* Other bus types might support multiple devices at slots 0-31 */
3035         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
3036     }
3037 }
3038 
3039 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
3040 {
3041     MSIMessage msg;
3042     if (msix_enabled(dev)) {
3043         msg = msix_get_message(dev, vector);
3044     } else if (msi_enabled(dev)) {
3045         msg = msi_get_message(dev, vector);
3046     } else {
3047         /* Should never happen */
3048         error_report("%s: unknown interrupt type", __func__);
3049         abort();
3050     }
3051     return msg;
3052 }
3053 
3054 void pci_set_power(PCIDevice *d, bool state)
3055 {
3056     /*
3057      * Don't change the enabled state of VFs when powering on/off the device.
3058      *
3059      * When powering on, VFs must not be enabled immediately but they must
3060      * wait until the guest configures SR-IOV.
3061      * When powering off, their corresponding PFs will be reset and disable
3062      * VFs.
3063      */
3064     if (!pci_is_vf(d)) {
3065         pci_set_enabled(d, state);
3066     }
3067 }
3068 
3069 void pci_set_enabled(PCIDevice *d, bool state)
3070 {
3071     if (d->enabled == state) {
3072         return;
3073     }
3074 
3075     d->enabled = state;
3076     pci_update_mappings(d);
3077     memory_region_set_enabled(&d->bus_master_enable_region,
3078                               (pci_get_word(d->config + PCI_COMMAND)
3079                                & PCI_COMMAND_MASTER) && d->enabled);
3080     if (qdev_is_realized(&d->qdev)) {
3081         pci_device_reset(d);
3082     }
3083 }
3084 
3085 static const TypeInfo pci_device_type_info = {
3086     .name = TYPE_PCI_DEVICE,
3087     .parent = TYPE_DEVICE,
3088     .instance_size = sizeof(PCIDevice),
3089     .abstract = true,
3090     .class_size = sizeof(PCIDeviceClass),
3091     .class_init = pci_device_class_init,
3092     .class_base_init = pci_device_class_base_init,
3093 };
3094 
3095 static void pci_register_types(void)
3096 {
3097     type_register_static(&pci_bus_info);
3098     type_register_static(&pcie_bus_info);
3099     type_register_static(&cxl_bus_info);
3100     type_register_static(&conventional_pci_interface_info);
3101     type_register_static(&cxl_interface_info);
3102     type_register_static(&pcie_interface_info);
3103     type_register_static(&pci_device_type_info);
3104 }
3105 
3106 type_init(pci_register_types)
3107