xref: /openbmc/qemu/hw/pci/pci.c (revision effd60c8)
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 "sysemu/numa.h"
39 #include "sysemu/runstate.h"
40 #include "sysemu/sysemu.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "qapi/error.h"
50 #include "qemu/cutils.h"
51 #include "pci-internal.h"
52 
53 #include "hw/xen/xen.h"
54 #include "hw/i386/kvm/xen_evtchn.h"
55 
56 //#define DEBUG_PCI
57 #ifdef DEBUG_PCI
58 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
59 #else
60 # define PCI_DPRINTF(format, ...)       do { } while (0)
61 #endif
62 
63 bool pci_available = true;
64 
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset(BusState *qbus);
68 static bool pcie_has_upstream_port(PCIDevice *dev);
69 
70 static Property pci_props[] = {
71     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
72     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
73     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
74     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
75     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
76                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
77     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
78                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
79     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
80                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
81     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
82                        failover_pair_id),
83     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
84     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
85                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
86     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
87                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
88     DEFINE_PROP_END_OF_LIST()
89 };
90 
91 static const VMStateDescription vmstate_pcibus = {
92     .name = "PCIBUS",
93     .version_id = 1,
94     .minimum_version_id = 1,
95     .fields = (const VMStateField[]) {
96         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
97         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
98                              nirq, 0, vmstate_info_int32,
99                              int32_t),
100         VMSTATE_END_OF_LIST()
101     }
102 };
103 
104 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
105 {
106     return a - b;
107 }
108 
109 static GSequence *pci_acpi_index_list(void)
110 {
111     static GSequence *used_acpi_index_list;
112 
113     if (!used_acpi_index_list) {
114         used_acpi_index_list = g_sequence_new(NULL);
115     }
116     return used_acpi_index_list;
117 }
118 
119 static void pci_init_bus_master(PCIDevice *pci_dev)
120 {
121     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
122 
123     memory_region_init_alias(&pci_dev->bus_master_enable_region,
124                              OBJECT(pci_dev), "bus master",
125                              dma_as->root, 0, memory_region_size(dma_as->root));
126     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
127     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
128                                 &pci_dev->bus_master_enable_region);
129 }
130 
131 static void pcibus_machine_done(Notifier *notifier, void *data)
132 {
133     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
134     int i;
135 
136     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
137         if (bus->devices[i]) {
138             pci_init_bus_master(bus->devices[i]);
139         }
140     }
141 }
142 
143 static void pci_bus_realize(BusState *qbus, Error **errp)
144 {
145     PCIBus *bus = PCI_BUS(qbus);
146 
147     bus->machine_done.notify = pcibus_machine_done;
148     qemu_add_machine_init_done_notifier(&bus->machine_done);
149 
150     vmstate_register_any(NULL, &vmstate_pcibus, bus);
151 }
152 
153 static void pcie_bus_realize(BusState *qbus, Error **errp)
154 {
155     PCIBus *bus = PCI_BUS(qbus);
156     Error *local_err = NULL;
157 
158     pci_bus_realize(qbus, &local_err);
159     if (local_err) {
160         error_propagate(errp, local_err);
161         return;
162     }
163 
164     /*
165      * A PCI-E bus can support extended config space if it's the root
166      * bus, or if the bus/bridge above it does as well
167      */
168     if (pci_bus_is_root(bus)) {
169         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
170     } else {
171         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
172 
173         if (pci_bus_allows_extended_config_space(parent_bus)) {
174             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
175         }
176     }
177 }
178 
179 static void pci_bus_unrealize(BusState *qbus)
180 {
181     PCIBus *bus = PCI_BUS(qbus);
182 
183     qemu_remove_machine_init_done_notifier(&bus->machine_done);
184 
185     vmstate_unregister(NULL, &vmstate_pcibus, bus);
186 }
187 
188 static int pcibus_num(PCIBus *bus)
189 {
190     if (pci_bus_is_root(bus)) {
191         return 0; /* pci host bridge */
192     }
193     return bus->parent_dev->config[PCI_SECONDARY_BUS];
194 }
195 
196 static uint16_t pcibus_numa_node(PCIBus *bus)
197 {
198     return NUMA_NODE_UNASSIGNED;
199 }
200 
201 static void pci_bus_class_init(ObjectClass *klass, void *data)
202 {
203     BusClass *k = BUS_CLASS(klass);
204     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
205 
206     k->print_dev = pcibus_dev_print;
207     k->get_dev_path = pcibus_get_dev_path;
208     k->get_fw_dev_path = pcibus_get_fw_dev_path;
209     k->realize = pci_bus_realize;
210     k->unrealize = pci_bus_unrealize;
211     k->reset = pcibus_reset;
212 
213     pbc->bus_num = pcibus_num;
214     pbc->numa_node = pcibus_numa_node;
215 }
216 
217 static const TypeInfo pci_bus_info = {
218     .name = TYPE_PCI_BUS,
219     .parent = TYPE_BUS,
220     .instance_size = sizeof(PCIBus),
221     .class_size = sizeof(PCIBusClass),
222     .class_init = pci_bus_class_init,
223 };
224 
225 static const TypeInfo cxl_interface_info = {
226     .name          = INTERFACE_CXL_DEVICE,
227     .parent        = TYPE_INTERFACE,
228 };
229 
230 static const TypeInfo pcie_interface_info = {
231     .name          = INTERFACE_PCIE_DEVICE,
232     .parent        = TYPE_INTERFACE,
233 };
234 
235 static const TypeInfo conventional_pci_interface_info = {
236     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
237     .parent        = TYPE_INTERFACE,
238 };
239 
240 static void pcie_bus_class_init(ObjectClass *klass, void *data)
241 {
242     BusClass *k = BUS_CLASS(klass);
243 
244     k->realize = pcie_bus_realize;
245 }
246 
247 static const TypeInfo pcie_bus_info = {
248     .name = TYPE_PCIE_BUS,
249     .parent = TYPE_PCI_BUS,
250     .class_init = pcie_bus_class_init,
251 };
252 
253 static const TypeInfo cxl_bus_info = {
254     .name       = TYPE_CXL_BUS,
255     .parent     = TYPE_PCIE_BUS,
256     .class_init = pcie_bus_class_init,
257 };
258 
259 static void pci_update_mappings(PCIDevice *d);
260 static void pci_irq_handler(void *opaque, int irq_num, int level);
261 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
262 static void pci_del_option_rom(PCIDevice *pdev);
263 
264 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
265 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
266 
267 PCIHostStateList pci_host_bridges;
268 
269 int pci_bar(PCIDevice *d, int reg)
270 {
271     uint8_t type;
272 
273     /* PCIe virtual functions do not have their own BARs */
274     assert(!pci_is_vf(d));
275 
276     if (reg != PCI_ROM_SLOT)
277         return PCI_BASE_ADDRESS_0 + reg * 4;
278 
279     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
280     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
281 }
282 
283 static inline int pci_irq_state(PCIDevice *d, int irq_num)
284 {
285         return (d->irq_state >> irq_num) & 0x1;
286 }
287 
288 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
289 {
290         d->irq_state &= ~(0x1 << irq_num);
291         d->irq_state |= level << irq_num;
292 }
293 
294 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
295 {
296     assert(irq_num >= 0);
297     assert(irq_num < bus->nirq);
298     bus->irq_count[irq_num] += change;
299     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
300 }
301 
302 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
303 {
304     PCIBus *bus;
305     for (;;) {
306         int dev_irq = irq_num;
307         bus = pci_get_bus(pci_dev);
308         assert(bus->map_irq);
309         irq_num = bus->map_irq(pci_dev, irq_num);
310         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
311                             pci_bus_is_root(bus) ? "root-complex"
312                                     : DEVICE(bus->parent_dev)->canonical_path);
313         if (bus->set_irq)
314             break;
315         pci_dev = bus->parent_dev;
316     }
317     pci_bus_change_irq_level(bus, irq_num, change);
318 }
319 
320 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
321 {
322     assert(irq_num >= 0);
323     assert(irq_num < bus->nirq);
324     return !!bus->irq_count[irq_num];
325 }
326 
327 /* Update interrupt status bit in config space on interrupt
328  * state change. */
329 static void pci_update_irq_status(PCIDevice *dev)
330 {
331     if (dev->irq_state) {
332         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
333     } else {
334         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
335     }
336 }
337 
338 void pci_device_deassert_intx(PCIDevice *dev)
339 {
340     int i;
341     for (i = 0; i < PCI_NUM_PINS; ++i) {
342         pci_irq_handler(dev, i, 0);
343     }
344 }
345 
346 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
347 {
348     MemTxAttrs attrs = {};
349 
350     /*
351      * Xen uses the high bits of the address to contain some of the bits
352      * of the PIRQ#. Therefore we can't just send the write cycle and
353      * trust that it's caught by the APIC at 0xfee00000 because the
354      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
355      * So we intercept the delivery here instead of in kvm_send_msi().
356      */
357     if (xen_mode == XEN_EMULATE &&
358         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
359         return;
360     }
361     attrs.requester_id = pci_requester_id(dev);
362     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
363                          attrs, NULL);
364 }
365 
366 static void pci_reset_regions(PCIDevice *dev)
367 {
368     int r;
369     if (pci_is_vf(dev)) {
370         return;
371     }
372 
373     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
374         PCIIORegion *region = &dev->io_regions[r];
375         if (!region->size) {
376             continue;
377         }
378 
379         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
380             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
381             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
382         } else {
383             pci_set_long(dev->config + pci_bar(dev, r), region->type);
384         }
385     }
386 }
387 
388 static void pci_do_device_reset(PCIDevice *dev)
389 {
390     pci_device_deassert_intx(dev);
391     assert(dev->irq_state == 0);
392 
393     /* Clear all writable bits */
394     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
395                                  pci_get_word(dev->wmask + PCI_COMMAND) |
396                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
397     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
398                                  pci_get_word(dev->wmask + PCI_STATUS) |
399                                  pci_get_word(dev->w1cmask + PCI_STATUS));
400     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
401     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
402                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
403                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
404     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
405     pci_reset_regions(dev);
406     pci_update_mappings(dev);
407 
408     msi_reset(dev);
409     msix_reset(dev);
410 }
411 
412 /*
413  * This function is called on #RST and FLR.
414  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
415  */
416 void pci_device_reset(PCIDevice *dev)
417 {
418     device_cold_reset(&dev->qdev);
419     pci_do_device_reset(dev);
420 }
421 
422 /*
423  * Trigger pci bus reset under a given bus.
424  * Called via bus_cold_reset on RST# assert, after the devices
425  * have been reset device_cold_reset-ed already.
426  */
427 static void pcibus_reset(BusState *qbus)
428 {
429     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
430     int i;
431 
432     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
433         if (bus->devices[i]) {
434             pci_do_device_reset(bus->devices[i]);
435         }
436     }
437 
438     for (i = 0; i < bus->nirq; i++) {
439         assert(bus->irq_count[i] == 0);
440     }
441 }
442 
443 static void pci_host_bus_register(DeviceState *host)
444 {
445     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
446 
447     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
448 }
449 
450 static void pci_host_bus_unregister(DeviceState *host)
451 {
452     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
453 
454     QLIST_REMOVE(host_bridge, next);
455 }
456 
457 PCIBus *pci_device_root_bus(const PCIDevice *d)
458 {
459     PCIBus *bus = pci_get_bus(d);
460 
461     while (!pci_bus_is_root(bus)) {
462         d = bus->parent_dev;
463         assert(d != NULL);
464 
465         bus = pci_get_bus(d);
466     }
467 
468     return bus;
469 }
470 
471 const char *pci_root_bus_path(PCIDevice *dev)
472 {
473     PCIBus *rootbus = pci_device_root_bus(dev);
474     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
475     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
476 
477     assert(host_bridge->bus == rootbus);
478 
479     if (hc->root_bus_path) {
480         return (*hc->root_bus_path)(host_bridge, rootbus);
481     }
482 
483     return rootbus->qbus.name;
484 }
485 
486 bool pci_bus_bypass_iommu(PCIBus *bus)
487 {
488     PCIBus *rootbus = bus;
489     PCIHostState *host_bridge;
490 
491     if (!pci_bus_is_root(bus)) {
492         rootbus = pci_device_root_bus(bus->parent_dev);
493     }
494 
495     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
496 
497     assert(host_bridge->bus == rootbus);
498 
499     return host_bridge->bypass_iommu;
500 }
501 
502 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
503                                        MemoryRegion *mem, MemoryRegion *io,
504                                        uint8_t devfn_min)
505 {
506     assert(PCI_FUNC(devfn_min) == 0);
507     bus->devfn_min = devfn_min;
508     bus->slot_reserved_mask = 0x0;
509     bus->address_space_mem = mem;
510     bus->address_space_io = io;
511     bus->flags |= PCI_BUS_IS_ROOT;
512 
513     /* host bridge */
514     QLIST_INIT(&bus->child);
515 
516     pci_host_bus_register(parent);
517 }
518 
519 static void pci_bus_uninit(PCIBus *bus)
520 {
521     pci_host_bus_unregister(BUS(bus)->parent);
522 }
523 
524 bool pci_bus_is_express(const PCIBus *bus)
525 {
526     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
527 }
528 
529 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
530                        const char *name,
531                        MemoryRegion *mem, MemoryRegion *io,
532                        uint8_t devfn_min, const char *typename)
533 {
534     qbus_init(bus, bus_size, typename, parent, name);
535     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
536 }
537 
538 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
539                          MemoryRegion *mem, MemoryRegion *io,
540                          uint8_t devfn_min, const char *typename)
541 {
542     PCIBus *bus;
543 
544     bus = PCI_BUS(qbus_new(typename, parent, name));
545     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
546     return bus;
547 }
548 
549 void pci_root_bus_cleanup(PCIBus *bus)
550 {
551     pci_bus_uninit(bus);
552     /* the caller of the unplug hotplug handler will delete this device */
553     qbus_unrealize(BUS(bus));
554 }
555 
556 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
557                   void *irq_opaque, int nirq)
558 {
559     bus->set_irq = set_irq;
560     bus->irq_opaque = irq_opaque;
561     bus->nirq = nirq;
562     g_free(bus->irq_count);
563     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
564 }
565 
566 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
567 {
568     bus->map_irq = map_irq;
569 }
570 
571 void pci_bus_irqs_cleanup(PCIBus *bus)
572 {
573     bus->set_irq = NULL;
574     bus->map_irq = NULL;
575     bus->irq_opaque = NULL;
576     bus->nirq = 0;
577     g_free(bus->irq_count);
578     bus->irq_count = NULL;
579 }
580 
581 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
582                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
583                               void *irq_opaque,
584                               MemoryRegion *mem, MemoryRegion *io,
585                               uint8_t devfn_min, int nirq,
586                               const char *typename)
587 {
588     PCIBus *bus;
589 
590     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
591     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
592     pci_bus_map_irqs(bus, map_irq);
593     return bus;
594 }
595 
596 void pci_unregister_root_bus(PCIBus *bus)
597 {
598     pci_bus_irqs_cleanup(bus);
599     pci_root_bus_cleanup(bus);
600 }
601 
602 int pci_bus_num(PCIBus *s)
603 {
604     return PCI_BUS_GET_CLASS(s)->bus_num(s);
605 }
606 
607 /* Returns the min and max bus numbers of a PCI bus hierarchy */
608 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
609 {
610     int i;
611     *min_bus = *max_bus = pci_bus_num(bus);
612 
613     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
614         PCIDevice *dev = bus->devices[i];
615 
616         if (dev && IS_PCI_BRIDGE(dev)) {
617             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
618             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
619         }
620     }
621 }
622 
623 int pci_bus_numa_node(PCIBus *bus)
624 {
625     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
626 }
627 
628 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
629                                  const VMStateField *field)
630 {
631     PCIDevice *s = container_of(pv, PCIDevice, config);
632     uint8_t *config;
633     int i;
634 
635     assert(size == pci_config_size(s));
636     config = g_malloc(size);
637 
638     qemu_get_buffer(f, config, size);
639     for (i = 0; i < size; ++i) {
640         if ((config[i] ^ s->config[i]) &
641             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
642             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
643                          "cmask: %x wmask: %x w1cmask:%x", __func__,
644                          i, config[i], s->config[i],
645                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
646             g_free(config);
647             return -EINVAL;
648         }
649     }
650     memcpy(s->config, config, size);
651 
652     pci_update_mappings(s);
653     if (IS_PCI_BRIDGE(s)) {
654         pci_bridge_update_mappings(PCI_BRIDGE(s));
655     }
656 
657     memory_region_set_enabled(&s->bus_master_enable_region,
658                               pci_get_word(s->config + PCI_COMMAND)
659                               & PCI_COMMAND_MASTER);
660 
661     g_free(config);
662     return 0;
663 }
664 
665 /* just put buffer */
666 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
667                                  const VMStateField *field, JSONWriter *vmdesc)
668 {
669     const uint8_t **v = pv;
670     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
671     qemu_put_buffer(f, *v, size);
672 
673     return 0;
674 }
675 
676 static const VMStateInfo vmstate_info_pci_config = {
677     .name = "pci config",
678     .get  = get_pci_config_device,
679     .put  = put_pci_config_device,
680 };
681 
682 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
683                              const VMStateField *field)
684 {
685     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
686     uint32_t irq_state[PCI_NUM_PINS];
687     int i;
688     for (i = 0; i < PCI_NUM_PINS; ++i) {
689         irq_state[i] = qemu_get_be32(f);
690         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
691             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
692                     irq_state[i]);
693             return -EINVAL;
694         }
695     }
696 
697     for (i = 0; i < PCI_NUM_PINS; ++i) {
698         pci_set_irq_state(s, i, irq_state[i]);
699     }
700 
701     return 0;
702 }
703 
704 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
705                              const VMStateField *field, JSONWriter *vmdesc)
706 {
707     int i;
708     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
709 
710     for (i = 0; i < PCI_NUM_PINS; ++i) {
711         qemu_put_be32(f, pci_irq_state(s, i));
712     }
713 
714     return 0;
715 }
716 
717 static const VMStateInfo vmstate_info_pci_irq_state = {
718     .name = "pci irq state",
719     .get  = get_pci_irq_state,
720     .put  = put_pci_irq_state,
721 };
722 
723 static bool migrate_is_pcie(void *opaque, int version_id)
724 {
725     return pci_is_express((PCIDevice *)opaque);
726 }
727 
728 static bool migrate_is_not_pcie(void *opaque, int version_id)
729 {
730     return !pci_is_express((PCIDevice *)opaque);
731 }
732 
733 const VMStateDescription vmstate_pci_device = {
734     .name = "PCIDevice",
735     .version_id = 2,
736     .minimum_version_id = 1,
737     .fields = (const VMStateField[]) {
738         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
739         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
740                                    migrate_is_not_pcie,
741                                    0, vmstate_info_pci_config,
742                                    PCI_CONFIG_SPACE_SIZE),
743         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
744                                    migrate_is_pcie,
745                                    0, vmstate_info_pci_config,
746                                    PCIE_CONFIG_SPACE_SIZE),
747         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
748                                    vmstate_info_pci_irq_state,
749                                    PCI_NUM_PINS * sizeof(int32_t)),
750         VMSTATE_END_OF_LIST()
751     }
752 };
753 
754 
755 void pci_device_save(PCIDevice *s, QEMUFile *f)
756 {
757     /* Clear interrupt status bit: it is implicit
758      * in irq_state which we are saving.
759      * This makes us compatible with old devices
760      * which never set or clear this bit. */
761     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
762     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
763     /* Restore the interrupt status bit. */
764     pci_update_irq_status(s);
765 }
766 
767 int pci_device_load(PCIDevice *s, QEMUFile *f)
768 {
769     int ret;
770     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
771     /* Restore the interrupt status bit. */
772     pci_update_irq_status(s);
773     return ret;
774 }
775 
776 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
777 {
778     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
779                  pci_default_sub_vendor_id);
780     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
781                  pci_default_sub_device_id);
782 }
783 
784 /*
785  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
786  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
787  */
788 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
789                              unsigned int *slotp, unsigned int *funcp)
790 {
791     const char *p;
792     char *e;
793     unsigned long val;
794     unsigned long dom = 0, bus = 0;
795     unsigned int slot = 0;
796     unsigned int func = 0;
797 
798     p = addr;
799     val = strtoul(p, &e, 16);
800     if (e == p)
801         return -1;
802     if (*e == ':') {
803         bus = val;
804         p = e + 1;
805         val = strtoul(p, &e, 16);
806         if (e == p)
807             return -1;
808         if (*e == ':') {
809             dom = bus;
810             bus = val;
811             p = e + 1;
812             val = strtoul(p, &e, 16);
813             if (e == p)
814                 return -1;
815         }
816     }
817 
818     slot = val;
819 
820     if (funcp != NULL) {
821         if (*e != '.')
822             return -1;
823 
824         p = e + 1;
825         val = strtoul(p, &e, 16);
826         if (e == p)
827             return -1;
828 
829         func = val;
830     }
831 
832     /* if funcp == NULL func is 0 */
833     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
834         return -1;
835 
836     if (*e)
837         return -1;
838 
839     *domp = dom;
840     *busp = bus;
841     *slotp = slot;
842     if (funcp != NULL)
843         *funcp = func;
844     return 0;
845 }
846 
847 static void pci_init_cmask(PCIDevice *dev)
848 {
849     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
850     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
851     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
852     dev->cmask[PCI_REVISION_ID] = 0xff;
853     dev->cmask[PCI_CLASS_PROG] = 0xff;
854     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
855     dev->cmask[PCI_HEADER_TYPE] = 0xff;
856     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
857 }
858 
859 static void pci_init_wmask(PCIDevice *dev)
860 {
861     int config_size = pci_config_size(dev);
862 
863     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
864     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
865     pci_set_word(dev->wmask + PCI_COMMAND,
866                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
867                  PCI_COMMAND_INTX_DISABLE);
868     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
869 
870     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
871            config_size - PCI_CONFIG_HEADER_SIZE);
872 }
873 
874 static void pci_init_w1cmask(PCIDevice *dev)
875 {
876     /*
877      * Note: It's okay to set w1cmask even for readonly bits as
878      * long as their value is hardwired to 0.
879      */
880     pci_set_word(dev->w1cmask + PCI_STATUS,
881                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
882                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
883                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
884 }
885 
886 static void pci_init_mask_bridge(PCIDevice *d)
887 {
888     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
889        PCI_SEC_LATENCY_TIMER */
890     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
891 
892     /* base and limit */
893     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
894     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
895     pci_set_word(d->wmask + PCI_MEMORY_BASE,
896                  PCI_MEMORY_RANGE_MASK & 0xffff);
897     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
898                  PCI_MEMORY_RANGE_MASK & 0xffff);
899     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
900                  PCI_PREF_RANGE_MASK & 0xffff);
901     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
902                  PCI_PREF_RANGE_MASK & 0xffff);
903 
904     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
905     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
906 
907     /* Supported memory and i/o types */
908     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
909     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
910     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
911                                PCI_PREF_RANGE_TYPE_64);
912     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
913                                PCI_PREF_RANGE_TYPE_64);
914 
915     /*
916      * TODO: Bridges default to 10-bit VGA decoding but we currently only
917      * implement 16-bit decoding (no alias support).
918      */
919     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
920                  PCI_BRIDGE_CTL_PARITY |
921                  PCI_BRIDGE_CTL_SERR |
922                  PCI_BRIDGE_CTL_ISA |
923                  PCI_BRIDGE_CTL_VGA |
924                  PCI_BRIDGE_CTL_VGA_16BIT |
925                  PCI_BRIDGE_CTL_MASTER_ABORT |
926                  PCI_BRIDGE_CTL_BUS_RESET |
927                  PCI_BRIDGE_CTL_FAST_BACK |
928                  PCI_BRIDGE_CTL_DISCARD |
929                  PCI_BRIDGE_CTL_SEC_DISCARD |
930                  PCI_BRIDGE_CTL_DISCARD_SERR);
931     /* Below does not do anything as we never set this bit, put here for
932      * completeness. */
933     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
934                  PCI_BRIDGE_CTL_DISCARD_STATUS);
935     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
936     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
937     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
938                                PCI_PREF_RANGE_TYPE_MASK);
939     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
940                                PCI_PREF_RANGE_TYPE_MASK);
941 }
942 
943 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
944 {
945     uint8_t slot = PCI_SLOT(dev->devfn);
946     uint8_t func;
947 
948     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
949         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
950     }
951 
952     /*
953      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
954      * device, as it may just be a VF that ended up with function 0 in
955      * the legacy PCI interpretation. Avoid failing in such cases:
956      */
957     if (pci_is_vf(dev) &&
958         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
959         return;
960     }
961 
962     /*
963      * multifunction bit is interpreted in two ways as follows.
964      *   - all functions must set the bit to 1.
965      *     Example: Intel X53
966      *   - function 0 must set the bit, but the rest function (> 0)
967      *     is allowed to leave the bit to 0.
968      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
969      *
970      * So OS (at least Linux) checks the bit of only function 0,
971      * and doesn't see the bit of function > 0.
972      *
973      * The below check allows both interpretation.
974      */
975     if (PCI_FUNC(dev->devfn)) {
976         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
977         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
978             /* function 0 should set multifunction bit */
979             error_setg(errp, "PCI: single function device can't be populated "
980                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
981             return;
982         }
983         return;
984     }
985 
986     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
987         return;
988     }
989     /* function 0 indicates single function, so function > 0 must be NULL */
990     for (func = 1; func < PCI_FUNC_MAX; ++func) {
991         if (bus->devices[PCI_DEVFN(slot, func)]) {
992             error_setg(errp, "PCI: %x.0 indicates single function, "
993                        "but %x.%x is already populated.",
994                        slot, slot, func);
995             return;
996         }
997     }
998 }
999 
1000 static void pci_config_alloc(PCIDevice *pci_dev)
1001 {
1002     int config_size = pci_config_size(pci_dev);
1003 
1004     pci_dev->config = g_malloc0(config_size);
1005     pci_dev->cmask = g_malloc0(config_size);
1006     pci_dev->wmask = g_malloc0(config_size);
1007     pci_dev->w1cmask = g_malloc0(config_size);
1008     pci_dev->used = g_malloc0(config_size);
1009 }
1010 
1011 static void pci_config_free(PCIDevice *pci_dev)
1012 {
1013     g_free(pci_dev->config);
1014     g_free(pci_dev->cmask);
1015     g_free(pci_dev->wmask);
1016     g_free(pci_dev->w1cmask);
1017     g_free(pci_dev->used);
1018 }
1019 
1020 static void do_pci_unregister_device(PCIDevice *pci_dev)
1021 {
1022     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1023     pci_config_free(pci_dev);
1024 
1025     if (xen_mode == XEN_EMULATE) {
1026         xen_evtchn_remove_pci_device(pci_dev);
1027     }
1028     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1029         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1030                                     &pci_dev->bus_master_enable_region);
1031     }
1032     address_space_destroy(&pci_dev->bus_master_as);
1033 }
1034 
1035 /* Extract PCIReqIDCache into BDF format */
1036 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1037 {
1038     uint8_t bus_n;
1039     uint16_t result;
1040 
1041     switch (cache->type) {
1042     case PCI_REQ_ID_BDF:
1043         result = pci_get_bdf(cache->dev);
1044         break;
1045     case PCI_REQ_ID_SECONDARY_BUS:
1046         bus_n = pci_dev_bus_num(cache->dev);
1047         result = PCI_BUILD_BDF(bus_n, 0);
1048         break;
1049     default:
1050         error_report("Invalid PCI requester ID cache type: %d",
1051                      cache->type);
1052         exit(1);
1053         break;
1054     }
1055 
1056     return result;
1057 }
1058 
1059 /* Parse bridges up to the root complex and return requester ID
1060  * cache for specific device.  For full PCIe topology, the cache
1061  * result would be exactly the same as getting BDF of the device.
1062  * However, several tricks are required when system mixed up with
1063  * legacy PCI devices and PCIe-to-PCI bridges.
1064  *
1065  * Here we cache the proxy device (and type) not requester ID since
1066  * bus number might change from time to time.
1067  */
1068 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1069 {
1070     PCIDevice *parent;
1071     PCIReqIDCache cache = {
1072         .dev = dev,
1073         .type = PCI_REQ_ID_BDF,
1074     };
1075 
1076     while (!pci_bus_is_root(pci_get_bus(dev))) {
1077         /* We are under PCI/PCIe bridges */
1078         parent = pci_get_bus(dev)->parent_dev;
1079         if (pci_is_express(parent)) {
1080             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1081                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1082                  * override the requester ID using secondary bus
1083                  * number of parent bridge with zeroed devfn
1084                  * (pcie-to-pci bridge spec chap 2.3). */
1085                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1086                 cache.dev = dev;
1087             }
1088         } else {
1089             /* Legacy PCI, override requester ID with the bridge's
1090              * BDF upstream.  When the root complex connects to
1091              * legacy PCI devices (including buses), it can only
1092              * obtain requester ID info from directly attached
1093              * devices.  If devices are attached under bridges, only
1094              * the requester ID of the bridge that is directly
1095              * attached to the root complex can be recognized. */
1096             cache.type = PCI_REQ_ID_BDF;
1097             cache.dev = parent;
1098         }
1099         dev = parent;
1100     }
1101 
1102     return cache;
1103 }
1104 
1105 uint16_t pci_requester_id(PCIDevice *dev)
1106 {
1107     return pci_req_id_cache_extract(&dev->requester_id_cache);
1108 }
1109 
1110 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1111 {
1112     return !(bus->devices[devfn]);
1113 }
1114 
1115 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1116 {
1117     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1118 }
1119 
1120 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1121 {
1122     return bus->slot_reserved_mask;
1123 }
1124 
1125 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1126 {
1127     bus->slot_reserved_mask |= mask;
1128 }
1129 
1130 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1131 {
1132     bus->slot_reserved_mask &= ~mask;
1133 }
1134 
1135 /* -1 for devfn means auto assign */
1136 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1137                                          const char *name, int devfn,
1138                                          Error **errp)
1139 {
1140     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1141     PCIConfigReadFunc *config_read = pc->config_read;
1142     PCIConfigWriteFunc *config_write = pc->config_write;
1143     Error *local_err = NULL;
1144     DeviceState *dev = DEVICE(pci_dev);
1145     PCIBus *bus = pci_get_bus(pci_dev);
1146     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1147 
1148     /* Only pci bridges can be attached to extra PCI root buses */
1149     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1150         error_setg(errp,
1151                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1152                     bus->parent_dev->name);
1153         return NULL;
1154     }
1155 
1156     if (devfn < 0) {
1157         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1158             devfn += PCI_FUNC_MAX) {
1159             if (pci_bus_devfn_available(bus, devfn) &&
1160                    !pci_bus_devfn_reserved(bus, devfn)) {
1161                 goto found;
1162             }
1163         }
1164         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1165                    "or reserved", name);
1166         return NULL;
1167     found: ;
1168     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1169         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1170                    " reserved",
1171                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1172         return NULL;
1173     } else if (!pci_bus_devfn_available(bus, devfn)) {
1174         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1175                    " in use by %s,id=%s",
1176                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1177                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1178         return NULL;
1179     } /*
1180        * Populating function 0 triggers a scan from the guest that
1181        * exposes other non-zero functions. Hence we need to ensure that
1182        * function 0 wasn't added yet.
1183        */
1184     else if (dev->hotplugged &&
1185              !pci_is_vf(pci_dev) &&
1186              pci_get_function_0(pci_dev)) {
1187         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1188                    " new func %s cannot be exposed to guest.",
1189                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1190                    pci_get_function_0(pci_dev)->name,
1191                    name);
1192 
1193        return NULL;
1194     }
1195 
1196     pci_dev->devfn = devfn;
1197     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1198     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1199 
1200     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1201                        "bus master container", UINT64_MAX);
1202     address_space_init(&pci_dev->bus_master_as,
1203                        &pci_dev->bus_master_container_region, pci_dev->name);
1204 
1205     if (phase_check(PHASE_MACHINE_READY)) {
1206         pci_init_bus_master(pci_dev);
1207     }
1208     pci_dev->irq_state = 0;
1209     pci_config_alloc(pci_dev);
1210 
1211     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1212     pci_config_set_device_id(pci_dev->config, pc->device_id);
1213     pci_config_set_revision(pci_dev->config, pc->revision);
1214     pci_config_set_class(pci_dev->config, pc->class_id);
1215 
1216     if (!is_bridge) {
1217         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1218             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1219                          pc->subsystem_vendor_id);
1220             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1221                          pc->subsystem_id);
1222         } else {
1223             pci_set_default_subsystem_id(pci_dev);
1224         }
1225     } else {
1226         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1227         assert(!pc->subsystem_vendor_id);
1228         assert(!pc->subsystem_id);
1229     }
1230     pci_init_cmask(pci_dev);
1231     pci_init_wmask(pci_dev);
1232     pci_init_w1cmask(pci_dev);
1233     if (is_bridge) {
1234         pci_init_mask_bridge(pci_dev);
1235     }
1236     pci_init_multifunction(bus, pci_dev, &local_err);
1237     if (local_err) {
1238         error_propagate(errp, local_err);
1239         do_pci_unregister_device(pci_dev);
1240         return NULL;
1241     }
1242 
1243     if (!config_read)
1244         config_read = pci_default_read_config;
1245     if (!config_write)
1246         config_write = pci_default_write_config;
1247     pci_dev->config_read = config_read;
1248     pci_dev->config_write = config_write;
1249     bus->devices[devfn] = pci_dev;
1250     pci_dev->version_id = 2; /* Current pci device vmstate version */
1251     return pci_dev;
1252 }
1253 
1254 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1255 {
1256     PCIIORegion *r;
1257     int i;
1258 
1259     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1260         r = &pci_dev->io_regions[i];
1261         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1262             continue;
1263         memory_region_del_subregion(r->address_space, r->memory);
1264     }
1265 
1266     pci_unregister_vga(pci_dev);
1267 }
1268 
1269 static void pci_qdev_unrealize(DeviceState *dev)
1270 {
1271     PCIDevice *pci_dev = PCI_DEVICE(dev);
1272     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1273 
1274     pci_unregister_io_regions(pci_dev);
1275     pci_del_option_rom(pci_dev);
1276 
1277     if (pc->exit) {
1278         pc->exit(pci_dev);
1279     }
1280 
1281     pci_device_deassert_intx(pci_dev);
1282     do_pci_unregister_device(pci_dev);
1283 
1284     pci_dev->msi_trigger = NULL;
1285 
1286     /*
1287      * clean up acpi-index so it could reused by another device
1288      */
1289     if (pci_dev->acpi_index) {
1290         GSequence *used_indexes = pci_acpi_index_list();
1291 
1292         g_sequence_remove(g_sequence_lookup(used_indexes,
1293                           GINT_TO_POINTER(pci_dev->acpi_index),
1294                           g_cmp_uint32, NULL));
1295     }
1296 }
1297 
1298 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1299                       uint8_t type, MemoryRegion *memory)
1300 {
1301     PCIIORegion *r;
1302     uint32_t addr; /* offset in pci config space */
1303     uint64_t wmask;
1304     pcibus_t size = memory_region_size(memory);
1305     uint8_t hdr_type;
1306 
1307     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1308     assert(region_num >= 0);
1309     assert(region_num < PCI_NUM_REGIONS);
1310     assert(is_power_of_2(size));
1311 
1312     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1313     hdr_type =
1314         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1315     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1316 
1317     r = &pci_dev->io_regions[region_num];
1318     r->addr = PCI_BAR_UNMAPPED;
1319     r->size = size;
1320     r->type = type;
1321     r->memory = memory;
1322     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1323                         ? pci_get_bus(pci_dev)->address_space_io
1324                         : pci_get_bus(pci_dev)->address_space_mem;
1325 
1326     wmask = ~(size - 1);
1327     if (region_num == PCI_ROM_SLOT) {
1328         /* ROM enable bit is writable */
1329         wmask |= PCI_ROM_ADDRESS_ENABLE;
1330     }
1331 
1332     addr = pci_bar(pci_dev, region_num);
1333     pci_set_long(pci_dev->config + addr, type);
1334 
1335     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1336         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1337         pci_set_quad(pci_dev->wmask + addr, wmask);
1338         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1339     } else {
1340         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1341         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1342     }
1343 }
1344 
1345 static void pci_update_vga(PCIDevice *pci_dev)
1346 {
1347     uint16_t cmd;
1348 
1349     if (!pci_dev->has_vga) {
1350         return;
1351     }
1352 
1353     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1354 
1355     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1356                               cmd & PCI_COMMAND_MEMORY);
1357     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1358                               cmd & PCI_COMMAND_IO);
1359     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1360                               cmd & PCI_COMMAND_IO);
1361 }
1362 
1363 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1364                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1365 {
1366     PCIBus *bus = pci_get_bus(pci_dev);
1367 
1368     assert(!pci_dev->has_vga);
1369 
1370     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1371     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1372     memory_region_add_subregion_overlap(bus->address_space_mem,
1373                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1374 
1375     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1376     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1377     memory_region_add_subregion_overlap(bus->address_space_io,
1378                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1379 
1380     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1381     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1382     memory_region_add_subregion_overlap(bus->address_space_io,
1383                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1384     pci_dev->has_vga = true;
1385 
1386     pci_update_vga(pci_dev);
1387 }
1388 
1389 void pci_unregister_vga(PCIDevice *pci_dev)
1390 {
1391     PCIBus *bus = pci_get_bus(pci_dev);
1392 
1393     if (!pci_dev->has_vga) {
1394         return;
1395     }
1396 
1397     memory_region_del_subregion(bus->address_space_mem,
1398                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1399     memory_region_del_subregion(bus->address_space_io,
1400                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1401     memory_region_del_subregion(bus->address_space_io,
1402                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1403     pci_dev->has_vga = false;
1404 }
1405 
1406 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1407 {
1408     return pci_dev->io_regions[region_num].addr;
1409 }
1410 
1411 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1412                                         uint8_t type, pcibus_t size)
1413 {
1414     pcibus_t new_addr;
1415     if (!pci_is_vf(d)) {
1416         int bar = pci_bar(d, reg);
1417         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1418             new_addr = pci_get_quad(d->config + bar);
1419         } else {
1420             new_addr = pci_get_long(d->config + bar);
1421         }
1422     } else {
1423         PCIDevice *pf = d->exp.sriov_vf.pf;
1424         uint16_t sriov_cap = pf->exp.sriov_cap;
1425         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1426         uint16_t vf_offset =
1427             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1428         uint16_t vf_stride =
1429             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1430         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1431 
1432         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1433             new_addr = pci_get_quad(pf->config + bar);
1434         } else {
1435             new_addr = pci_get_long(pf->config + bar);
1436         }
1437         new_addr += vf_num * size;
1438     }
1439     /* The ROM slot has a specific enable bit, keep it intact */
1440     if (reg != PCI_ROM_SLOT) {
1441         new_addr &= ~(size - 1);
1442     }
1443     return new_addr;
1444 }
1445 
1446 pcibus_t pci_bar_address(PCIDevice *d,
1447                          int reg, uint8_t type, pcibus_t size)
1448 {
1449     pcibus_t new_addr, last_addr;
1450     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1451     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1452     bool allow_0_address = mc->pci_allow_0_address;
1453 
1454     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1455         if (!(cmd & PCI_COMMAND_IO)) {
1456             return PCI_BAR_UNMAPPED;
1457         }
1458         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1459         last_addr = new_addr + size - 1;
1460         /* Check if 32 bit BAR wraps around explicitly.
1461          * TODO: make priorities correct and remove this work around.
1462          */
1463         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1464             (!allow_0_address && new_addr == 0)) {
1465             return PCI_BAR_UNMAPPED;
1466         }
1467         return new_addr;
1468     }
1469 
1470     if (!(cmd & PCI_COMMAND_MEMORY)) {
1471         return PCI_BAR_UNMAPPED;
1472     }
1473     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1474     /* the ROM slot has a specific enable bit */
1475     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1476         return PCI_BAR_UNMAPPED;
1477     }
1478     new_addr &= ~(size - 1);
1479     last_addr = new_addr + size - 1;
1480     /* NOTE: we do not support wrapping */
1481     /* XXX: as we cannot support really dynamic
1482        mappings, we handle specific values as invalid
1483        mappings. */
1484     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1485         (!allow_0_address && new_addr == 0)) {
1486         return PCI_BAR_UNMAPPED;
1487     }
1488 
1489     /* Now pcibus_t is 64bit.
1490      * Check if 32 bit BAR wraps around explicitly.
1491      * Without this, PC ide doesn't work well.
1492      * TODO: remove this work around.
1493      */
1494     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1495         return PCI_BAR_UNMAPPED;
1496     }
1497 
1498     /*
1499      * OS is allowed to set BAR beyond its addressable
1500      * bits. For example, 32 bit OS can set 64bit bar
1501      * to >4G. Check it. TODO: we might need to support
1502      * it in the future for e.g. PAE.
1503      */
1504     if (last_addr >= HWADDR_MAX) {
1505         return PCI_BAR_UNMAPPED;
1506     }
1507 
1508     return new_addr;
1509 }
1510 
1511 static void pci_update_mappings(PCIDevice *d)
1512 {
1513     PCIIORegion *r;
1514     int i;
1515     pcibus_t new_addr;
1516 
1517     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1518         r = &d->io_regions[i];
1519 
1520         /* this region isn't registered */
1521         if (!r->size)
1522             continue;
1523 
1524         new_addr = pci_bar_address(d, i, r->type, r->size);
1525         if (!d->has_power) {
1526             new_addr = PCI_BAR_UNMAPPED;
1527         }
1528 
1529         /* This bar isn't changed */
1530         if (new_addr == r->addr)
1531             continue;
1532 
1533         /* now do the real mapping */
1534         if (r->addr != PCI_BAR_UNMAPPED) {
1535             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1536                                           PCI_SLOT(d->devfn),
1537                                           PCI_FUNC(d->devfn),
1538                                           i, r->addr, r->size);
1539             memory_region_del_subregion(r->address_space, r->memory);
1540         }
1541         r->addr = new_addr;
1542         if (r->addr != PCI_BAR_UNMAPPED) {
1543             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1544                                           PCI_SLOT(d->devfn),
1545                                           PCI_FUNC(d->devfn),
1546                                           i, r->addr, r->size);
1547             memory_region_add_subregion_overlap(r->address_space,
1548                                                 r->addr, r->memory, 1);
1549         }
1550     }
1551 
1552     pci_update_vga(d);
1553 }
1554 
1555 static inline int pci_irq_disabled(PCIDevice *d)
1556 {
1557     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1558 }
1559 
1560 /* Called after interrupt disabled field update in config space,
1561  * assert/deassert interrupts if necessary.
1562  * Gets original interrupt disable bit value (before update). */
1563 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1564 {
1565     int i, disabled = pci_irq_disabled(d);
1566     if (disabled == was_irq_disabled)
1567         return;
1568     for (i = 0; i < PCI_NUM_PINS; ++i) {
1569         int state = pci_irq_state(d, i);
1570         pci_change_irq_level(d, i, disabled ? -state : state);
1571     }
1572 }
1573 
1574 uint32_t pci_default_read_config(PCIDevice *d,
1575                                  uint32_t address, int len)
1576 {
1577     uint32_t val = 0;
1578 
1579     assert(address + len <= pci_config_size(d));
1580 
1581     if (pci_is_express_downstream_port(d) &&
1582         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1583         pcie_sync_bridge_lnk(d);
1584     }
1585     memcpy(&val, d->config + address, len);
1586     return le32_to_cpu(val);
1587 }
1588 
1589 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1590 {
1591     int i, was_irq_disabled = pci_irq_disabled(d);
1592     uint32_t val = val_in;
1593 
1594     assert(addr + l <= pci_config_size(d));
1595 
1596     for (i = 0; i < l; val >>= 8, ++i) {
1597         uint8_t wmask = d->wmask[addr + i];
1598         uint8_t w1cmask = d->w1cmask[addr + i];
1599         assert(!(wmask & w1cmask));
1600         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1601         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1602     }
1603     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1604         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1605         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1606         range_covers_byte(addr, l, PCI_COMMAND))
1607         pci_update_mappings(d);
1608 
1609     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1610         pci_update_irq_disabled(d, was_irq_disabled);
1611         memory_region_set_enabled(&d->bus_master_enable_region,
1612                                   (pci_get_word(d->config + PCI_COMMAND)
1613                                    & PCI_COMMAND_MASTER) && d->has_power);
1614     }
1615 
1616     msi_write_config(d, addr, val_in, l);
1617     msix_write_config(d, addr, val_in, l);
1618     pcie_sriov_config_write(d, addr, val_in, l);
1619 }
1620 
1621 /***********************************************************/
1622 /* generic PCI irq support */
1623 
1624 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1625 static void pci_irq_handler(void *opaque, int irq_num, int level)
1626 {
1627     PCIDevice *pci_dev = opaque;
1628     int change;
1629 
1630     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1631     assert(level == 0 || level == 1);
1632     change = level - pci_irq_state(pci_dev, irq_num);
1633     if (!change)
1634         return;
1635 
1636     pci_set_irq_state(pci_dev, irq_num, level);
1637     pci_update_irq_status(pci_dev);
1638     if (pci_irq_disabled(pci_dev))
1639         return;
1640     pci_change_irq_level(pci_dev, irq_num, change);
1641 }
1642 
1643 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1644 {
1645     int intx = pci_intx(pci_dev);
1646     assert(0 <= intx && intx < PCI_NUM_PINS);
1647 
1648     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1649 }
1650 
1651 void pci_set_irq(PCIDevice *pci_dev, int level)
1652 {
1653     int intx = pci_intx(pci_dev);
1654     pci_irq_handler(pci_dev, intx, level);
1655 }
1656 
1657 /* Special hooks used by device assignment */
1658 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1659 {
1660     assert(pci_bus_is_root(bus));
1661     bus->route_intx_to_irq = route_intx_to_irq;
1662 }
1663 
1664 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1665 {
1666     PCIBus *bus;
1667 
1668     do {
1669         int dev_irq = pin;
1670         bus = pci_get_bus(dev);
1671         pin = bus->map_irq(dev, pin);
1672         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1673                             pci_bus_is_root(bus) ? "root-complex"
1674                                     : DEVICE(bus->parent_dev)->canonical_path);
1675         dev = bus->parent_dev;
1676     } while (dev);
1677 
1678     if (!bus->route_intx_to_irq) {
1679         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1680                      object_get_typename(OBJECT(bus->qbus.parent)));
1681         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1682     }
1683 
1684     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1685 }
1686 
1687 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1688 {
1689     return old->mode != new->mode || old->irq != new->irq;
1690 }
1691 
1692 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1693 {
1694     PCIDevice *dev;
1695     PCIBus *sec;
1696     int i;
1697 
1698     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1699         dev = bus->devices[i];
1700         if (dev && dev->intx_routing_notifier) {
1701             dev->intx_routing_notifier(dev);
1702         }
1703     }
1704 
1705     QLIST_FOREACH(sec, &bus->child, sibling) {
1706         pci_bus_fire_intx_routing_notifier(sec);
1707     }
1708 }
1709 
1710 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1711                                           PCIINTxRoutingNotifier notifier)
1712 {
1713     dev->intx_routing_notifier = notifier;
1714 }
1715 
1716 /*
1717  * PCI-to-PCI bridge specification
1718  * 9.1: Interrupt routing. Table 9-1
1719  *
1720  * the PCI Express Base Specification, Revision 2.1
1721  * 2.2.8.1: INTx interrupt signaling - Rules
1722  *          the Implementation Note
1723  *          Table 2-20
1724  */
1725 /*
1726  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1727  * 0-origin unlike PCI interrupt pin register.
1728  */
1729 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1730 {
1731     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1732 }
1733 
1734 /***********************************************************/
1735 /* monitor info on PCI */
1736 
1737 static const pci_class_desc pci_class_descriptions[] =
1738 {
1739     { 0x0001, "VGA controller", "display"},
1740     { 0x0100, "SCSI controller", "scsi"},
1741     { 0x0101, "IDE controller", "ide"},
1742     { 0x0102, "Floppy controller", "fdc"},
1743     { 0x0103, "IPI controller", "ipi"},
1744     { 0x0104, "RAID controller", "raid"},
1745     { 0x0106, "SATA controller"},
1746     { 0x0107, "SAS controller"},
1747     { 0x0180, "Storage controller"},
1748     { 0x0200, "Ethernet controller", "ethernet"},
1749     { 0x0201, "Token Ring controller", "token-ring"},
1750     { 0x0202, "FDDI controller", "fddi"},
1751     { 0x0203, "ATM controller", "atm"},
1752     { 0x0280, "Network controller"},
1753     { 0x0300, "VGA controller", "display", 0x00ff},
1754     { 0x0301, "XGA controller"},
1755     { 0x0302, "3D controller"},
1756     { 0x0380, "Display controller"},
1757     { 0x0400, "Video controller", "video"},
1758     { 0x0401, "Audio controller", "sound"},
1759     { 0x0402, "Phone"},
1760     { 0x0403, "Audio controller", "sound"},
1761     { 0x0480, "Multimedia controller"},
1762     { 0x0500, "RAM controller", "memory"},
1763     { 0x0501, "Flash controller", "flash"},
1764     { 0x0580, "Memory controller"},
1765     { 0x0600, "Host bridge", "host"},
1766     { 0x0601, "ISA bridge", "isa"},
1767     { 0x0602, "EISA bridge", "eisa"},
1768     { 0x0603, "MC bridge", "mca"},
1769     { 0x0604, "PCI bridge", "pci-bridge"},
1770     { 0x0605, "PCMCIA bridge", "pcmcia"},
1771     { 0x0606, "NUBUS bridge", "nubus"},
1772     { 0x0607, "CARDBUS bridge", "cardbus"},
1773     { 0x0608, "RACEWAY bridge"},
1774     { 0x0680, "Bridge"},
1775     { 0x0700, "Serial port", "serial"},
1776     { 0x0701, "Parallel port", "parallel"},
1777     { 0x0800, "Interrupt controller", "interrupt-controller"},
1778     { 0x0801, "DMA controller", "dma-controller"},
1779     { 0x0802, "Timer", "timer"},
1780     { 0x0803, "RTC", "rtc"},
1781     { 0x0900, "Keyboard", "keyboard"},
1782     { 0x0901, "Pen", "pen"},
1783     { 0x0902, "Mouse", "mouse"},
1784     { 0x0A00, "Dock station", "dock", 0x00ff},
1785     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1786     { 0x0c00, "Firewire controller", "firewire"},
1787     { 0x0c01, "Access bus controller", "access-bus"},
1788     { 0x0c02, "SSA controller", "ssa"},
1789     { 0x0c03, "USB controller", "usb"},
1790     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1791     { 0x0c05, "SMBus"},
1792     { 0, NULL}
1793 };
1794 
1795 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1796                                            pci_bus_dev_fn fn,
1797                                            void *opaque)
1798 {
1799     PCIDevice *d;
1800     int devfn;
1801 
1802     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1803         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1804         if (d) {
1805             fn(bus, d, opaque);
1806         }
1807     }
1808 }
1809 
1810 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1811                                  pci_bus_dev_fn fn, void *opaque)
1812 {
1813     bus = pci_find_bus_nr(bus, bus_num);
1814 
1815     if (bus) {
1816         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1817     }
1818 }
1819 
1820 void pci_for_each_device_under_bus(PCIBus *bus,
1821                                    pci_bus_dev_fn fn, void *opaque)
1822 {
1823     PCIDevice *d;
1824     int devfn;
1825 
1826     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1827         d = bus->devices[devfn];
1828         if (d) {
1829             fn(bus, d, opaque);
1830         }
1831     }
1832 }
1833 
1834 void pci_for_each_device(PCIBus *bus, int bus_num,
1835                          pci_bus_dev_fn fn, void *opaque)
1836 {
1837     bus = pci_find_bus_nr(bus, bus_num);
1838 
1839     if (bus) {
1840         pci_for_each_device_under_bus(bus, fn, opaque);
1841     }
1842 }
1843 
1844 const pci_class_desc *get_class_desc(int class)
1845 {
1846     const pci_class_desc *desc;
1847 
1848     desc = pci_class_descriptions;
1849     while (desc->desc && class != desc->class) {
1850         desc++;
1851     }
1852 
1853     return desc;
1854 }
1855 
1856 /* Initialize a PCI NIC.  */
1857 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1858                                const char *default_model,
1859                                const char *default_devaddr)
1860 {
1861     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1862     GPtrArray *pci_nic_models;
1863     PCIBus *bus;
1864     PCIDevice *pci_dev;
1865     DeviceState *dev;
1866     int devfn;
1867     int i;
1868     int dom, busnr;
1869     unsigned slot;
1870 
1871     if (nd->model && !strcmp(nd->model, "virtio")) {
1872         g_free(nd->model);
1873         nd->model = g_strdup("virtio-net-pci");
1874     }
1875 
1876     pci_nic_models = qemu_get_nic_models(TYPE_PCI_DEVICE);
1877 
1878     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1879         exit(0);
1880     }
1881 
1882     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1883                             default_model);
1884     if (i < 0) {
1885         exit(1);
1886     }
1887 
1888     if (!rootbus) {
1889         error_report("No primary PCI bus");
1890         exit(1);
1891     }
1892 
1893     assert(!rootbus->parent_dev);
1894 
1895     if (!devaddr) {
1896         devfn = -1;
1897         busnr = 0;
1898     } else {
1899         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1900             error_report("Invalid PCI device address %s for device %s",
1901                          devaddr, nd->model);
1902             exit(1);
1903         }
1904 
1905         if (dom != 0) {
1906             error_report("No support for non-zero PCI domains");
1907             exit(1);
1908         }
1909 
1910         devfn = PCI_DEVFN(slot, 0);
1911     }
1912 
1913     bus = pci_find_bus_nr(rootbus, busnr);
1914     if (!bus) {
1915         error_report("Invalid PCI device address %s for device %s",
1916                      devaddr, nd->model);
1917         exit(1);
1918     }
1919 
1920     pci_dev = pci_new(devfn, nd->model);
1921     dev = &pci_dev->qdev;
1922     qdev_set_nic_properties(dev, nd);
1923     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1924     g_ptr_array_free(pci_nic_models, true);
1925     return pci_dev;
1926 }
1927 
1928 PCIDevice *pci_vga_init(PCIBus *bus)
1929 {
1930     vga_interface_created = true;
1931     switch (vga_interface_type) {
1932     case VGA_CIRRUS:
1933         return pci_create_simple(bus, -1, "cirrus-vga");
1934     case VGA_QXL:
1935         return pci_create_simple(bus, -1, "qxl-vga");
1936     case VGA_STD:
1937         return pci_create_simple(bus, -1, "VGA");
1938     case VGA_VMWARE:
1939         return pci_create_simple(bus, -1, "vmware-svga");
1940     case VGA_VIRTIO:
1941         return pci_create_simple(bus, -1, "virtio-vga");
1942     case VGA_NONE:
1943     default: /* Other non-PCI types. Checking for unsupported types is already
1944                 done in vl.c. */
1945         return NULL;
1946     }
1947 }
1948 
1949 /* Whether a given bus number is in range of the secondary
1950  * bus of the given bridge device. */
1951 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1952 {
1953     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1954              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1955         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1956         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1957 }
1958 
1959 /* Whether a given bus number is in a range of a root bus */
1960 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1961 {
1962     int i;
1963 
1964     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1965         PCIDevice *dev = bus->devices[i];
1966 
1967         if (dev && IS_PCI_BRIDGE(dev)) {
1968             if (pci_secondary_bus_in_range(dev, bus_num)) {
1969                 return true;
1970             }
1971         }
1972     }
1973 
1974     return false;
1975 }
1976 
1977 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1978 {
1979     PCIBus *sec;
1980 
1981     if (!bus) {
1982         return NULL;
1983     }
1984 
1985     if (pci_bus_num(bus) == bus_num) {
1986         return bus;
1987     }
1988 
1989     /* Consider all bus numbers in range for the host pci bridge. */
1990     if (!pci_bus_is_root(bus) &&
1991         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1992         return NULL;
1993     }
1994 
1995     /* try child bus */
1996     for (; bus; bus = sec) {
1997         QLIST_FOREACH(sec, &bus->child, sibling) {
1998             if (pci_bus_num(sec) == bus_num) {
1999                 return sec;
2000             }
2001             /* PXB buses assumed to be children of bus 0 */
2002             if (pci_bus_is_root(sec)) {
2003                 if (pci_root_bus_in_range(sec, bus_num)) {
2004                     break;
2005                 }
2006             } else {
2007                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2008                     break;
2009                 }
2010             }
2011         }
2012     }
2013 
2014     return NULL;
2015 }
2016 
2017 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
2018                                   pci_bus_fn end, void *parent_state)
2019 {
2020     PCIBus *sec;
2021     void *state;
2022 
2023     if (!bus) {
2024         return;
2025     }
2026 
2027     if (begin) {
2028         state = begin(bus, parent_state);
2029     } else {
2030         state = parent_state;
2031     }
2032 
2033     QLIST_FOREACH(sec, &bus->child, sibling) {
2034         pci_for_each_bus_depth_first(sec, begin, end, state);
2035     }
2036 
2037     if (end) {
2038         end(bus, state);
2039     }
2040 }
2041 
2042 
2043 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2044 {
2045     bus = pci_find_bus_nr(bus, bus_num);
2046 
2047     if (!bus)
2048         return NULL;
2049 
2050     return bus->devices[devfn];
2051 }
2052 
2053 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2054 
2055 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2056 {
2057     PCIDevice *pci_dev = (PCIDevice *)qdev;
2058     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2059     ObjectClass *klass = OBJECT_CLASS(pc);
2060     Error *local_err = NULL;
2061     bool is_default_rom;
2062     uint16_t class_id;
2063 
2064     /*
2065      * capped by systemd (see: udev-builtin-net_id.c)
2066      * as it's the only known user honor it to avoid users
2067      * misconfigure QEMU and then wonder why acpi-index doesn't work
2068      */
2069     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2070         error_setg(errp, "acpi-index should be less or equal to %u",
2071                    ONBOARD_INDEX_MAX);
2072         return;
2073     }
2074 
2075     /*
2076      * make sure that acpi-index is unique across all present PCI devices
2077      */
2078     if (pci_dev->acpi_index) {
2079         GSequence *used_indexes = pci_acpi_index_list();
2080 
2081         if (g_sequence_lookup(used_indexes,
2082                               GINT_TO_POINTER(pci_dev->acpi_index),
2083                               g_cmp_uint32, NULL)) {
2084             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2085                        " already exist", pci_dev->acpi_index);
2086             return;
2087         }
2088         g_sequence_insert_sorted(used_indexes,
2089                                  GINT_TO_POINTER(pci_dev->acpi_index),
2090                                  g_cmp_uint32, NULL);
2091     }
2092 
2093     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2094         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2095         return;
2096     }
2097 
2098     /* initialize cap_present for pci_is_express() and pci_config_size(),
2099      * Note that hybrid PCIs are not set automatically and need to manage
2100      * QEMU_PCI_CAP_EXPRESS manually */
2101     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2102        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2103         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2104     }
2105 
2106     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2107         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2108     }
2109 
2110     pci_dev = do_pci_register_device(pci_dev,
2111                                      object_get_typename(OBJECT(qdev)),
2112                                      pci_dev->devfn, errp);
2113     if (pci_dev == NULL)
2114         return;
2115 
2116     if (pc->realize) {
2117         pc->realize(pci_dev, &local_err);
2118         if (local_err) {
2119             error_propagate(errp, local_err);
2120             do_pci_unregister_device(pci_dev);
2121             return;
2122         }
2123     }
2124 
2125     /*
2126      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2127      * associate only Device 0 with the device attached to the bus
2128      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2129      * sec 7.3.1).
2130      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2131      * 5-bit Device Number and 3-bit Function Number fields in its associated
2132      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2133      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2134      */
2135     if (pci_is_express(pci_dev) &&
2136         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2137         pcie_has_upstream_port(pci_dev) &&
2138         PCI_SLOT(pci_dev->devfn)) {
2139         warn_report("PCI: slot %d is not valid for %s,"
2140                     " parent device only allows plugging into slot 0.",
2141                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2142     }
2143 
2144     if (pci_dev->failover_pair_id) {
2145         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2146             error_setg(errp, "failover primary device must be on "
2147                              "PCIExpress bus");
2148             pci_qdev_unrealize(DEVICE(pci_dev));
2149             return;
2150         }
2151         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2152         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2153             error_setg(errp, "failover primary device is not an "
2154                              "Ethernet device");
2155             pci_qdev_unrealize(DEVICE(pci_dev));
2156             return;
2157         }
2158         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2159             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2160             error_setg(errp, "failover: primary device must be in its own "
2161                               "PCI slot");
2162             pci_qdev_unrealize(DEVICE(pci_dev));
2163             return;
2164         }
2165         qdev->allow_unplug_during_migration = true;
2166     }
2167 
2168     /* rom loading */
2169     is_default_rom = false;
2170     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2171         pci_dev->romfile = g_strdup(pc->romfile);
2172         is_default_rom = true;
2173     }
2174 
2175     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2176     if (local_err) {
2177         error_propagate(errp, local_err);
2178         pci_qdev_unrealize(DEVICE(pci_dev));
2179         return;
2180     }
2181 
2182     pci_set_power(pci_dev, true);
2183 
2184     pci_dev->msi_trigger = pci_msi_trigger;
2185 }
2186 
2187 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2188                                    const char *name)
2189 {
2190     DeviceState *dev;
2191 
2192     dev = qdev_new(name);
2193     qdev_prop_set_int32(dev, "addr", devfn);
2194     qdev_prop_set_bit(dev, "multifunction", multifunction);
2195     return PCI_DEVICE(dev);
2196 }
2197 
2198 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2199 {
2200     return pci_new_internal(devfn, true, name);
2201 }
2202 
2203 PCIDevice *pci_new(int devfn, const char *name)
2204 {
2205     return pci_new_internal(devfn, false, name);
2206 }
2207 
2208 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2209 {
2210     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2211 }
2212 
2213 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2214                                            const char *name)
2215 {
2216     PCIDevice *dev = pci_new_multifunction(devfn, name);
2217     pci_realize_and_unref(dev, bus, &error_fatal);
2218     return dev;
2219 }
2220 
2221 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2222 {
2223     PCIDevice *dev = pci_new(devfn, name);
2224     pci_realize_and_unref(dev, bus, &error_fatal);
2225     return dev;
2226 }
2227 
2228 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2229 {
2230     int offset = PCI_CONFIG_HEADER_SIZE;
2231     int i;
2232     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2233         if (pdev->used[i])
2234             offset = i + 1;
2235         else if (i - offset + 1 == size)
2236             return offset;
2237     }
2238     return 0;
2239 }
2240 
2241 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2242                                         uint8_t *prev_p)
2243 {
2244     uint8_t next, prev;
2245 
2246     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2247         return 0;
2248 
2249     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2250          prev = next + PCI_CAP_LIST_NEXT)
2251         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2252             break;
2253 
2254     if (prev_p)
2255         *prev_p = prev;
2256     return next;
2257 }
2258 
2259 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2260 {
2261     uint8_t next, prev, found = 0;
2262 
2263     if (!(pdev->used[offset])) {
2264         return 0;
2265     }
2266 
2267     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2268 
2269     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2270          prev = next + PCI_CAP_LIST_NEXT) {
2271         if (next <= offset && next > found) {
2272             found = next;
2273         }
2274     }
2275     return found;
2276 }
2277 
2278 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2279    This is needed for an option rom which is used for more than one device. */
2280 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2281 {
2282     uint16_t vendor_id;
2283     uint16_t device_id;
2284     uint16_t rom_vendor_id;
2285     uint16_t rom_device_id;
2286     uint16_t rom_magic;
2287     uint16_t pcir_offset;
2288     uint8_t checksum;
2289 
2290     /* Words in rom data are little endian (like in PCI configuration),
2291        so they can be read / written with pci_get_word / pci_set_word. */
2292 
2293     /* Only a valid rom will be patched. */
2294     rom_magic = pci_get_word(ptr);
2295     if (rom_magic != 0xaa55) {
2296         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2297         return;
2298     }
2299     pcir_offset = pci_get_word(ptr + 0x18);
2300     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2301         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2302         return;
2303     }
2304 
2305     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2306     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2307     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2308     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2309 
2310     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2311                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2312 
2313     checksum = ptr[6];
2314 
2315     if (vendor_id != rom_vendor_id) {
2316         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2317         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2318         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2319         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2320         ptr[6] = checksum;
2321         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2322     }
2323 
2324     if (device_id != rom_device_id) {
2325         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2326         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2327         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2328         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2329         ptr[6] = checksum;
2330         pci_set_word(ptr + pcir_offset + 6, device_id);
2331     }
2332 }
2333 
2334 /* Add an option rom for the device */
2335 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2336                                Error **errp)
2337 {
2338     int64_t size = 0;
2339     g_autofree char *path = NULL;
2340     char name[32];
2341     const VMStateDescription *vmsd;
2342 
2343     /*
2344      * In case of incoming migration ROM will come with migration stream, no
2345      * reason to load the file.  Neither we want to fail if local ROM file
2346      * mismatches with specified romsize.
2347      */
2348     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2349 
2350     if (!pdev->romfile || !strlen(pdev->romfile)) {
2351         return;
2352     }
2353 
2354     if (!pdev->rom_bar) {
2355         /*
2356          * Load rom via fw_cfg instead of creating a rom bar,
2357          * for 0.11 compatibility.
2358          */
2359         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2360 
2361         /*
2362          * Hot-plugged devices can't use the option ROM
2363          * if the rom bar is disabled.
2364          */
2365         if (DEVICE(pdev)->hotplugged) {
2366             error_setg(errp, "Hot-plugged device without ROM bar"
2367                        " can't have an option ROM");
2368             return;
2369         }
2370 
2371         if (class == 0x0300) {
2372             rom_add_vga(pdev->romfile);
2373         } else {
2374             rom_add_option(pdev->romfile, -1);
2375         }
2376         return;
2377     }
2378 
2379     if (load_file || pdev->romsize == -1) {
2380         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2381         if (path == NULL) {
2382             path = g_strdup(pdev->romfile);
2383         }
2384 
2385         size = get_image_size(path);
2386         if (size < 0) {
2387             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2388             return;
2389         } else if (size == 0) {
2390             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2391             return;
2392         } else if (size > 2 * GiB) {
2393             error_setg(errp,
2394                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2395                        pdev->romfile);
2396             return;
2397         }
2398         if (pdev->romsize != -1) {
2399             if (size > pdev->romsize) {
2400                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2401                            "is too large for ROM size %u",
2402                            pdev->romfile, (uint32_t)size, pdev->romsize);
2403                 return;
2404             }
2405         } else {
2406             pdev->romsize = pow2ceil(size);
2407         }
2408     }
2409 
2410     vmsd = qdev_get_vmsd(DEVICE(pdev));
2411     snprintf(name, sizeof(name), "%s.rom",
2412              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2413 
2414     pdev->has_rom = true;
2415     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2416                            &error_fatal);
2417 
2418     if (load_file) {
2419         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2420 
2421         if (load_image_size(path, ptr, size) < 0) {
2422             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2423             return;
2424         }
2425 
2426         if (is_default_rom) {
2427             /* Only the default rom images will be patched (if needed). */
2428             pci_patch_ids(pdev, ptr, size);
2429         }
2430     }
2431 
2432     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2433 }
2434 
2435 static void pci_del_option_rom(PCIDevice *pdev)
2436 {
2437     if (!pdev->has_rom)
2438         return;
2439 
2440     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2441     pdev->has_rom = false;
2442 }
2443 
2444 /*
2445  * On success, pci_add_capability() returns a positive value
2446  * that the offset of the pci capability.
2447  * On failure, it sets an error and returns a negative error
2448  * code.
2449  */
2450 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2451                        uint8_t offset, uint8_t size,
2452                        Error **errp)
2453 {
2454     uint8_t *config;
2455     int i, overlapping_cap;
2456 
2457     if (!offset) {
2458         offset = pci_find_space(pdev, size);
2459         /* out of PCI config space is programming error */
2460         assert(offset);
2461     } else {
2462         /* Verify that capabilities don't overlap.  Note: device assignment
2463          * depends on this check to verify that the device is not broken.
2464          * Should never trigger for emulated devices, but it's helpful
2465          * for debugging these. */
2466         for (i = offset; i < offset + size; i++) {
2467             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2468             if (overlapping_cap) {
2469                 error_setg(errp, "%s:%02x:%02x.%x "
2470                            "Attempt to add PCI capability %x at offset "
2471                            "%x overlaps existing capability %x at offset %x",
2472                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2473                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2474                            cap_id, offset, overlapping_cap, i);
2475                 return -EINVAL;
2476             }
2477         }
2478     }
2479 
2480     config = pdev->config + offset;
2481     config[PCI_CAP_LIST_ID] = cap_id;
2482     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2483     pdev->config[PCI_CAPABILITY_LIST] = offset;
2484     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2485     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2486     /* Make capability read-only by default */
2487     memset(pdev->wmask + offset, 0, size);
2488     /* Check capability by default */
2489     memset(pdev->cmask + offset, 0xFF, size);
2490     return offset;
2491 }
2492 
2493 /* Unlink capability from the pci config space. */
2494 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2495 {
2496     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2497     if (!offset)
2498         return;
2499     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2500     /* Make capability writable again */
2501     memset(pdev->wmask + offset, 0xff, size);
2502     memset(pdev->w1cmask + offset, 0, size);
2503     /* Clear cmask as device-specific registers can't be checked */
2504     memset(pdev->cmask + offset, 0, size);
2505     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2506 
2507     if (!pdev->config[PCI_CAPABILITY_LIST])
2508         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2509 }
2510 
2511 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2512 {
2513     return pci_find_capability_list(pdev, cap_id, NULL);
2514 }
2515 
2516 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2517 {
2518     PCIDevice *d = (PCIDevice *)dev;
2519     const char *name = NULL;
2520     const pci_class_desc *desc =  pci_class_descriptions;
2521     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2522 
2523     while (desc->desc &&
2524           (class & ~desc->fw_ign_bits) !=
2525           (desc->class & ~desc->fw_ign_bits)) {
2526         desc++;
2527     }
2528 
2529     if (desc->desc) {
2530         name = desc->fw_name;
2531     }
2532 
2533     if (name) {
2534         pstrcpy(buf, len, name);
2535     } else {
2536         snprintf(buf, len, "pci%04x,%04x",
2537                  pci_get_word(d->config + PCI_VENDOR_ID),
2538                  pci_get_word(d->config + PCI_DEVICE_ID));
2539     }
2540 
2541     return buf;
2542 }
2543 
2544 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2545 {
2546     PCIDevice *d = (PCIDevice *)dev;
2547     char name[33];
2548     int has_func = !!PCI_FUNC(d->devfn);
2549 
2550     return g_strdup_printf("%s@%x%s%.*x",
2551                            pci_dev_fw_name(dev, name, sizeof(name)),
2552                            PCI_SLOT(d->devfn),
2553                            has_func ? "," : "",
2554                            has_func,
2555                            PCI_FUNC(d->devfn));
2556 }
2557 
2558 static char *pcibus_get_dev_path(DeviceState *dev)
2559 {
2560     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2561     PCIDevice *t;
2562     int slot_depth;
2563     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2564      * 00 is added here to make this format compatible with
2565      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2566      * Slot.Function list specifies the slot and function numbers for all
2567      * devices on the path from root to the specific device. */
2568     const char *root_bus_path;
2569     int root_bus_len;
2570     char slot[] = ":SS.F";
2571     int slot_len = sizeof slot - 1 /* For '\0' */;
2572     int path_len;
2573     char *path, *p;
2574     int s;
2575 
2576     root_bus_path = pci_root_bus_path(d);
2577     root_bus_len = strlen(root_bus_path);
2578 
2579     /* Calculate # of slots on path between device and root. */;
2580     slot_depth = 0;
2581     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2582         ++slot_depth;
2583     }
2584 
2585     path_len = root_bus_len + slot_len * slot_depth;
2586 
2587     /* Allocate memory, fill in the terminating null byte. */
2588     path = g_malloc(path_len + 1 /* For '\0' */);
2589     path[path_len] = '\0';
2590 
2591     memcpy(path, root_bus_path, root_bus_len);
2592 
2593     /* Fill in slot numbers. We walk up from device to root, so need to print
2594      * them in the reverse order, last to first. */
2595     p = path + path_len;
2596     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2597         p -= slot_len;
2598         s = snprintf(slot, sizeof slot, ":%02x.%x",
2599                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2600         assert(s == slot_len);
2601         memcpy(p, slot, slot_len);
2602     }
2603 
2604     return path;
2605 }
2606 
2607 static int pci_qdev_find_recursive(PCIBus *bus,
2608                                    const char *id, PCIDevice **pdev)
2609 {
2610     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2611     if (!qdev) {
2612         return -ENODEV;
2613     }
2614 
2615     /* roughly check if given qdev is pci device */
2616     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2617         *pdev = PCI_DEVICE(qdev);
2618         return 0;
2619     }
2620     return -EINVAL;
2621 }
2622 
2623 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2624 {
2625     PCIHostState *host_bridge;
2626     int rc = -ENODEV;
2627 
2628     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2629         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2630         if (!tmp) {
2631             rc = 0;
2632             break;
2633         }
2634         if (tmp != -ENODEV) {
2635             rc = tmp;
2636         }
2637     }
2638 
2639     return rc;
2640 }
2641 
2642 MemoryRegion *pci_address_space(PCIDevice *dev)
2643 {
2644     return pci_get_bus(dev)->address_space_mem;
2645 }
2646 
2647 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2648 {
2649     return pci_get_bus(dev)->address_space_io;
2650 }
2651 
2652 static void pci_device_class_init(ObjectClass *klass, void *data)
2653 {
2654     DeviceClass *k = DEVICE_CLASS(klass);
2655 
2656     k->realize = pci_qdev_realize;
2657     k->unrealize = pci_qdev_unrealize;
2658     k->bus_type = TYPE_PCI_BUS;
2659     device_class_set_props(k, pci_props);
2660 }
2661 
2662 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2663 {
2664     if (!object_class_is_abstract(klass)) {
2665         ObjectClass *conventional =
2666             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2667         ObjectClass *pcie =
2668             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2669         ObjectClass *cxl =
2670             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2671         assert(conventional || pcie || cxl);
2672     }
2673 }
2674 
2675 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2676 {
2677     PCIBus *bus = pci_get_bus(dev);
2678     PCIBus *iommu_bus = bus;
2679     uint8_t devfn = dev->devfn;
2680 
2681     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2682         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2683 
2684         /*
2685          * The requester ID of the provided device may be aliased, as seen from
2686          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2687          * requester ID to provide a unique AddressSpace for devices, but
2688          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2689          * to-PCI bridge creates and accepts transactions on behalf of down-
2690          * stream devices.  When doing so, all downstream devices are masked
2691          * (aliased) behind a single requester ID.  The requester ID used
2692          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2693          * bridges, with a PCIe capability indicating such, follow the
2694          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2695          * where the bridge uses the seconary bus as the bridge portion of the
2696          * requester ID and devfn of 00.0.  For other bridges, typically those
2697          * found on the root complex such as the dmi-to-pci-bridge, we follow
2698          * the convention of typical bare-metal hardware, which uses the
2699          * requester ID of the bridge itself.  There are device specific
2700          * exceptions to these rules, but these are the defaults that the
2701          * Linux kernel uses when determining DMA aliases itself and believed
2702          * to be true for the bare metal equivalents of the devices emulated
2703          * in QEMU.
2704          */
2705         if (!pci_bus_is_express(iommu_bus)) {
2706             PCIDevice *parent = iommu_bus->parent_dev;
2707 
2708             if (pci_is_express(parent) &&
2709                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2710                 devfn = PCI_DEVFN(0, 0);
2711                 bus = iommu_bus;
2712             } else {
2713                 devfn = parent->devfn;
2714                 bus = parent_bus;
2715             }
2716         }
2717 
2718         iommu_bus = parent_bus;
2719     }
2720     if (!pci_bus_bypass_iommu(bus) && iommu_bus->iommu_ops) {
2721         return iommu_bus->iommu_ops->get_address_space(bus,
2722                                  iommu_bus->iommu_opaque, devfn);
2723     }
2724     return &address_space_memory;
2725 }
2726 
2727 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2728 {
2729     /*
2730      * If called, pci_setup_iommu() should provide a minimum set of
2731      * useful callbacks for the bus.
2732      */
2733     assert(ops);
2734     assert(ops->get_address_space);
2735 
2736     bus->iommu_ops = ops;
2737     bus->iommu_opaque = opaque;
2738 }
2739 
2740 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2741 {
2742     Range *range = opaque;
2743     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2744     int i;
2745 
2746     if (!(cmd & PCI_COMMAND_MEMORY)) {
2747         return;
2748     }
2749 
2750     if (IS_PCI_BRIDGE(dev)) {
2751         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2752         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2753 
2754         base = MAX(base, 0x1ULL << 32);
2755 
2756         if (limit >= base) {
2757             Range pref_range;
2758             range_set_bounds(&pref_range, base, limit);
2759             range_extend(range, &pref_range);
2760         }
2761     }
2762     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2763         PCIIORegion *r = &dev->io_regions[i];
2764         pcibus_t lob, upb;
2765         Range region_range;
2766 
2767         if (!r->size ||
2768             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2769             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2770             continue;
2771         }
2772 
2773         lob = pci_bar_address(dev, i, r->type, r->size);
2774         upb = lob + r->size - 1;
2775         if (lob == PCI_BAR_UNMAPPED) {
2776             continue;
2777         }
2778 
2779         lob = MAX(lob, 0x1ULL << 32);
2780 
2781         if (upb >= lob) {
2782             range_set_bounds(&region_range, lob, upb);
2783             range_extend(range, &region_range);
2784         }
2785     }
2786 }
2787 
2788 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2789 {
2790     range_make_empty(range);
2791     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2792 }
2793 
2794 static bool pcie_has_upstream_port(PCIDevice *dev)
2795 {
2796     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2797 
2798     /* Device associated with an upstream port.
2799      * As there are several types of these, it's easier to check the
2800      * parent device: upstream ports are always connected to
2801      * root or downstream ports.
2802      */
2803     return parent_dev &&
2804         pci_is_express(parent_dev) &&
2805         parent_dev->exp.exp_cap &&
2806         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2807          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2808 }
2809 
2810 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2811 {
2812     PCIBus *bus = pci_get_bus(pci_dev);
2813 
2814     if(pcie_has_upstream_port(pci_dev)) {
2815         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2816         return bus->devices[0];
2817     } else {
2818         /* Other bus types might support multiple devices at slots 0-31 */
2819         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2820     }
2821 }
2822 
2823 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2824 {
2825     MSIMessage msg;
2826     if (msix_enabled(dev)) {
2827         msg = msix_get_message(dev, vector);
2828     } else if (msi_enabled(dev)) {
2829         msg = msi_get_message(dev, vector);
2830     } else {
2831         /* Should never happen */
2832         error_report("%s: unknown interrupt type", __func__);
2833         abort();
2834     }
2835     return msg;
2836 }
2837 
2838 void pci_set_power(PCIDevice *d, bool state)
2839 {
2840     if (d->has_power == state) {
2841         return;
2842     }
2843 
2844     d->has_power = state;
2845     pci_update_mappings(d);
2846     memory_region_set_enabled(&d->bus_master_enable_region,
2847                               (pci_get_word(d->config + PCI_COMMAND)
2848                                & PCI_COMMAND_MASTER) && d->has_power);
2849     if (!d->has_power) {
2850         pci_device_reset(d);
2851     }
2852 }
2853 
2854 static const TypeInfo pci_device_type_info = {
2855     .name = TYPE_PCI_DEVICE,
2856     .parent = TYPE_DEVICE,
2857     .instance_size = sizeof(PCIDevice),
2858     .abstract = true,
2859     .class_size = sizeof(PCIDeviceClass),
2860     .class_init = pci_device_class_init,
2861     .class_base_init = pci_device_class_base_init,
2862 };
2863 
2864 static void pci_register_types(void)
2865 {
2866     type_register_static(&pci_bus_info);
2867     type_register_static(&pcie_bus_info);
2868     type_register_static(&cxl_bus_info);
2869     type_register_static(&conventional_pci_interface_info);
2870     type_register_static(&cxl_interface_info);
2871     type_register_static(&pcie_interface_info);
2872     type_register_static(&pci_device_type_info);
2873 }
2874 
2875 type_init(pci_register_types)
2876