xref: /openbmc/qemu/hw/pci/pci.c (revision feb58e3b)
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_hold(Object *obj, ResetType type);
68 static bool pcie_has_upstream_port(PCIDevice *dev);
69 
70 static void prop_pci_busnr_get(Object *obj, Visitor *v, const char *name,
71                              void *opaque, Error **errp)
72 {
73     uint8_t busnr = pci_dev_bus_num(PCI_DEVICE(obj));
74 
75     visit_type_uint8(v, name, &busnr, errp);
76 }
77 
78 static const PropertyInfo prop_pci_busnr = {
79     .name = "busnr",
80     .get = prop_pci_busnr_get,
81 };
82 
83 static Property pci_props[] = {
84     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
85     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
86     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, UINT32_MAX),
87     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
88     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
89                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
90     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
91                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
92     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
93                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
94     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
95                        failover_pair_id),
96     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
97     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
98                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
99     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
100                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
101     DEFINE_PROP_SIZE32("x-max-bounce-buffer-size", PCIDevice,
102                      max_bounce_buffer_size, DEFAULT_MAX_BOUNCE_BUFFER_SIZE),
103     { .name = "busnr", .info = &prop_pci_busnr },
104     DEFINE_PROP_END_OF_LIST()
105 };
106 
107 static const VMStateDescription vmstate_pcibus = {
108     .name = "PCIBUS",
109     .version_id = 1,
110     .minimum_version_id = 1,
111     .fields = (const VMStateField[]) {
112         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
113         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
114                              nirq, 0, vmstate_info_int32,
115                              int32_t),
116         VMSTATE_END_OF_LIST()
117     }
118 };
119 
120 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
121 {
122     return a - b;
123 }
124 
125 static GSequence *pci_acpi_index_list(void)
126 {
127     static GSequence *used_acpi_index_list;
128 
129     if (!used_acpi_index_list) {
130         used_acpi_index_list = g_sequence_new(NULL);
131     }
132     return used_acpi_index_list;
133 }
134 
135 static void pci_init_bus_master(PCIDevice *pci_dev)
136 {
137     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
138 
139     memory_region_init_alias(&pci_dev->bus_master_enable_region,
140                              OBJECT(pci_dev), "bus master",
141                              dma_as->root, 0, memory_region_size(dma_as->root));
142     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
143     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
144                                 &pci_dev->bus_master_enable_region);
145 }
146 
147 static void pcibus_machine_done(Notifier *notifier, void *data)
148 {
149     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
150     int i;
151 
152     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
153         if (bus->devices[i]) {
154             pci_init_bus_master(bus->devices[i]);
155         }
156     }
157 }
158 
159 static void pci_bus_realize(BusState *qbus, Error **errp)
160 {
161     PCIBus *bus = PCI_BUS(qbus);
162 
163     bus->machine_done.notify = pcibus_machine_done;
164     qemu_add_machine_init_done_notifier(&bus->machine_done);
165 
166     vmstate_register_any(NULL, &vmstate_pcibus, bus);
167 }
168 
169 static void pcie_bus_realize(BusState *qbus, Error **errp)
170 {
171     PCIBus *bus = PCI_BUS(qbus);
172     Error *local_err = NULL;
173 
174     pci_bus_realize(qbus, &local_err);
175     if (local_err) {
176         error_propagate(errp, local_err);
177         return;
178     }
179 
180     /*
181      * A PCI-E bus can support extended config space if it's the root
182      * bus, or if the bus/bridge above it does as well
183      */
184     if (pci_bus_is_root(bus)) {
185         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
186     } else {
187         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
188 
189         if (pci_bus_allows_extended_config_space(parent_bus)) {
190             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
191         }
192     }
193 }
194 
195 static void pci_bus_unrealize(BusState *qbus)
196 {
197     PCIBus *bus = PCI_BUS(qbus);
198 
199     qemu_remove_machine_init_done_notifier(&bus->machine_done);
200 
201     vmstate_unregister(NULL, &vmstate_pcibus, bus);
202 }
203 
204 static int pcibus_num(PCIBus *bus)
205 {
206     if (pci_bus_is_root(bus)) {
207         return 0; /* pci host bridge */
208     }
209     return bus->parent_dev->config[PCI_SECONDARY_BUS];
210 }
211 
212 static uint16_t pcibus_numa_node(PCIBus *bus)
213 {
214     return NUMA_NODE_UNASSIGNED;
215 }
216 
217 static void pci_bus_class_init(ObjectClass *klass, void *data)
218 {
219     BusClass *k = BUS_CLASS(klass);
220     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
221     ResettableClass *rc = RESETTABLE_CLASS(klass);
222 
223     k->print_dev = pcibus_dev_print;
224     k->get_dev_path = pcibus_get_dev_path;
225     k->get_fw_dev_path = pcibus_get_fw_dev_path;
226     k->realize = pci_bus_realize;
227     k->unrealize = pci_bus_unrealize;
228 
229     rc->phases.hold = pcibus_reset_hold;
230 
231     pbc->bus_num = pcibus_num;
232     pbc->numa_node = pcibus_numa_node;
233 }
234 
235 static const TypeInfo pci_bus_info = {
236     .name = TYPE_PCI_BUS,
237     .parent = TYPE_BUS,
238     .instance_size = sizeof(PCIBus),
239     .class_size = sizeof(PCIBusClass),
240     .class_init = pci_bus_class_init,
241 };
242 
243 static const TypeInfo cxl_interface_info = {
244     .name          = INTERFACE_CXL_DEVICE,
245     .parent        = TYPE_INTERFACE,
246 };
247 
248 static const TypeInfo pcie_interface_info = {
249     .name          = INTERFACE_PCIE_DEVICE,
250     .parent        = TYPE_INTERFACE,
251 };
252 
253 static const TypeInfo conventional_pci_interface_info = {
254     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
255     .parent        = TYPE_INTERFACE,
256 };
257 
258 static void pcie_bus_class_init(ObjectClass *klass, void *data)
259 {
260     BusClass *k = BUS_CLASS(klass);
261 
262     k->realize = pcie_bus_realize;
263 }
264 
265 static const TypeInfo pcie_bus_info = {
266     .name = TYPE_PCIE_BUS,
267     .parent = TYPE_PCI_BUS,
268     .class_init = pcie_bus_class_init,
269 };
270 
271 static const TypeInfo cxl_bus_info = {
272     .name       = TYPE_CXL_BUS,
273     .parent     = TYPE_PCIE_BUS,
274     .class_init = pcie_bus_class_init,
275 };
276 
277 static void pci_update_mappings(PCIDevice *d);
278 static void pci_irq_handler(void *opaque, int irq_num, int level);
279 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
280 static void pci_del_option_rom(PCIDevice *pdev);
281 
282 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
283 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
284 
285 PCIHostStateList pci_host_bridges;
286 
287 int pci_bar(PCIDevice *d, int reg)
288 {
289     uint8_t type;
290 
291     /* PCIe virtual functions do not have their own BARs */
292     assert(!pci_is_vf(d));
293 
294     if (reg != PCI_ROM_SLOT)
295         return PCI_BASE_ADDRESS_0 + reg * 4;
296 
297     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
298     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
299 }
300 
301 static inline int pci_irq_state(PCIDevice *d, int irq_num)
302 {
303         return (d->irq_state >> irq_num) & 0x1;
304 }
305 
306 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
307 {
308         d->irq_state &= ~(0x1 << irq_num);
309         d->irq_state |= level << irq_num;
310 }
311 
312 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
313 {
314     assert(irq_num >= 0);
315     assert(irq_num < bus->nirq);
316     bus->irq_count[irq_num] += change;
317     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
318 }
319 
320 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
321 {
322     PCIBus *bus;
323     for (;;) {
324         int dev_irq = irq_num;
325         bus = pci_get_bus(pci_dev);
326         assert(bus->map_irq);
327         irq_num = bus->map_irq(pci_dev, irq_num);
328         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
329                             pci_bus_is_root(bus) ? "root-complex"
330                                     : DEVICE(bus->parent_dev)->canonical_path);
331         if (bus->set_irq)
332             break;
333         pci_dev = bus->parent_dev;
334     }
335     pci_bus_change_irq_level(bus, irq_num, change);
336 }
337 
338 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
339 {
340     assert(irq_num >= 0);
341     assert(irq_num < bus->nirq);
342     return !!bus->irq_count[irq_num];
343 }
344 
345 /* Update interrupt status bit in config space on interrupt
346  * state change. */
347 static void pci_update_irq_status(PCIDevice *dev)
348 {
349     if (dev->irq_state) {
350         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
351     } else {
352         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
353     }
354 }
355 
356 void pci_device_deassert_intx(PCIDevice *dev)
357 {
358     int i;
359     for (i = 0; i < PCI_NUM_PINS; ++i) {
360         pci_irq_handler(dev, i, 0);
361     }
362 }
363 
364 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
365 {
366     MemTxAttrs attrs = {};
367 
368     /*
369      * Xen uses the high bits of the address to contain some of the bits
370      * of the PIRQ#. Therefore we can't just send the write cycle and
371      * trust that it's caught by the APIC at 0xfee00000 because the
372      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
373      * So we intercept the delivery here instead of in kvm_send_msi().
374      */
375     if (xen_mode == XEN_EMULATE &&
376         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
377         return;
378     }
379     attrs.requester_id = pci_requester_id(dev);
380     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
381                          attrs, NULL);
382 }
383 
384 static void pci_reset_regions(PCIDevice *dev)
385 {
386     int r;
387     if (pci_is_vf(dev)) {
388         return;
389     }
390 
391     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
392         PCIIORegion *region = &dev->io_regions[r];
393         if (!region->size) {
394             continue;
395         }
396 
397         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
398             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
399             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
400         } else {
401             pci_set_long(dev->config + pci_bar(dev, r), region->type);
402         }
403     }
404 }
405 
406 static void pci_do_device_reset(PCIDevice *dev)
407 {
408     pci_device_deassert_intx(dev);
409     assert(dev->irq_state == 0);
410 
411     /* Clear all writable bits */
412     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
413                                  pci_get_word(dev->wmask + PCI_COMMAND) |
414                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
415     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
416                                  pci_get_word(dev->wmask + PCI_STATUS) |
417                                  pci_get_word(dev->w1cmask + PCI_STATUS));
418     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
419     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
420                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
421                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
422     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
423     pci_reset_regions(dev);
424     pci_update_mappings(dev);
425 
426     msi_reset(dev);
427     msix_reset(dev);
428     pcie_sriov_pf_reset(dev);
429 }
430 
431 /*
432  * This function is called on #RST and FLR.
433  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
434  */
435 void pci_device_reset(PCIDevice *dev)
436 {
437     device_cold_reset(&dev->qdev);
438     pci_do_device_reset(dev);
439 }
440 
441 /*
442  * Trigger pci bus reset under a given bus.
443  * Called via bus_cold_reset on RST# assert, after the devices
444  * have been reset device_cold_reset-ed already.
445  */
446 static void pcibus_reset_hold(Object *obj, ResetType type)
447 {
448     PCIBus *bus = PCI_BUS(obj);
449     int i;
450 
451     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
452         if (bus->devices[i]) {
453             pci_do_device_reset(bus->devices[i]);
454         }
455     }
456 
457     for (i = 0; i < bus->nirq; i++) {
458         assert(bus->irq_count[i] == 0);
459     }
460 }
461 
462 static void pci_host_bus_register(DeviceState *host)
463 {
464     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
465 
466     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
467 }
468 
469 static void pci_host_bus_unregister(DeviceState *host)
470 {
471     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
472 
473     QLIST_REMOVE(host_bridge, next);
474 }
475 
476 PCIBus *pci_device_root_bus(const PCIDevice *d)
477 {
478     PCIBus *bus = pci_get_bus(d);
479 
480     while (!pci_bus_is_root(bus)) {
481         d = bus->parent_dev;
482         assert(d != NULL);
483 
484         bus = pci_get_bus(d);
485     }
486 
487     return bus;
488 }
489 
490 const char *pci_root_bus_path(PCIDevice *dev)
491 {
492     PCIBus *rootbus = pci_device_root_bus(dev);
493     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
494     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
495 
496     assert(host_bridge->bus == rootbus);
497 
498     if (hc->root_bus_path) {
499         return (*hc->root_bus_path)(host_bridge, rootbus);
500     }
501 
502     return rootbus->qbus.name;
503 }
504 
505 bool pci_bus_bypass_iommu(PCIBus *bus)
506 {
507     PCIBus *rootbus = bus;
508     PCIHostState *host_bridge;
509 
510     if (!pci_bus_is_root(bus)) {
511         rootbus = pci_device_root_bus(bus->parent_dev);
512     }
513 
514     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
515 
516     assert(host_bridge->bus == rootbus);
517 
518     return host_bridge->bypass_iommu;
519 }
520 
521 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
522                                        MemoryRegion *mem, MemoryRegion *io,
523                                        uint8_t devfn_min)
524 {
525     assert(PCI_FUNC(devfn_min) == 0);
526     bus->devfn_min = devfn_min;
527     bus->slot_reserved_mask = 0x0;
528     bus->address_space_mem = mem;
529     bus->address_space_io = io;
530     bus->flags |= PCI_BUS_IS_ROOT;
531 
532     /* host bridge */
533     QLIST_INIT(&bus->child);
534 
535     pci_host_bus_register(parent);
536 }
537 
538 static void pci_bus_uninit(PCIBus *bus)
539 {
540     pci_host_bus_unregister(BUS(bus)->parent);
541 }
542 
543 bool pci_bus_is_express(const PCIBus *bus)
544 {
545     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
546 }
547 
548 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
549                        const char *name,
550                        MemoryRegion *mem, MemoryRegion *io,
551                        uint8_t devfn_min, const char *typename)
552 {
553     qbus_init(bus, bus_size, typename, parent, name);
554     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
555 }
556 
557 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
558                          MemoryRegion *mem, MemoryRegion *io,
559                          uint8_t devfn_min, const char *typename)
560 {
561     PCIBus *bus;
562 
563     bus = PCI_BUS(qbus_new(typename, parent, name));
564     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
565     return bus;
566 }
567 
568 void pci_root_bus_cleanup(PCIBus *bus)
569 {
570     pci_bus_uninit(bus);
571     /* the caller of the unplug hotplug handler will delete this device */
572     qbus_unrealize(BUS(bus));
573 }
574 
575 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
576                   void *irq_opaque, int nirq)
577 {
578     bus->set_irq = set_irq;
579     bus->irq_opaque = irq_opaque;
580     bus->nirq = nirq;
581     g_free(bus->irq_count);
582     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
583 }
584 
585 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
586 {
587     bus->map_irq = map_irq;
588 }
589 
590 void pci_bus_irqs_cleanup(PCIBus *bus)
591 {
592     bus->set_irq = NULL;
593     bus->map_irq = NULL;
594     bus->irq_opaque = NULL;
595     bus->nirq = 0;
596     g_free(bus->irq_count);
597     bus->irq_count = NULL;
598 }
599 
600 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
601                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
602                               void *irq_opaque,
603                               MemoryRegion *mem, MemoryRegion *io,
604                               uint8_t devfn_min, int nirq,
605                               const char *typename)
606 {
607     PCIBus *bus;
608 
609     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
610     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
611     pci_bus_map_irqs(bus, map_irq);
612     return bus;
613 }
614 
615 void pci_unregister_root_bus(PCIBus *bus)
616 {
617     pci_bus_irqs_cleanup(bus);
618     pci_root_bus_cleanup(bus);
619 }
620 
621 int pci_bus_num(PCIBus *s)
622 {
623     return PCI_BUS_GET_CLASS(s)->bus_num(s);
624 }
625 
626 /* Returns the min and max bus numbers of a PCI bus hierarchy */
627 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
628 {
629     int i;
630     *min_bus = *max_bus = pci_bus_num(bus);
631 
632     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
633         PCIDevice *dev = bus->devices[i];
634 
635         if (dev && IS_PCI_BRIDGE(dev)) {
636             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
637             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
638         }
639     }
640 }
641 
642 int pci_bus_numa_node(PCIBus *bus)
643 {
644     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
645 }
646 
647 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
648                                  const VMStateField *field)
649 {
650     PCIDevice *s = container_of(pv, PCIDevice, config);
651     uint8_t *config;
652     int i;
653 
654     assert(size == pci_config_size(s));
655     config = g_malloc(size);
656 
657     qemu_get_buffer(f, config, size);
658     for (i = 0; i < size; ++i) {
659         if ((config[i] ^ s->config[i]) &
660             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
661             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
662                          "cmask: %x wmask: %x w1cmask:%x", __func__,
663                          i, config[i], s->config[i],
664                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
665             g_free(config);
666             return -EINVAL;
667         }
668     }
669     memcpy(s->config, config, size);
670 
671     pci_update_mappings(s);
672     if (IS_PCI_BRIDGE(s)) {
673         pci_bridge_update_mappings(PCI_BRIDGE(s));
674     }
675 
676     memory_region_set_enabled(&s->bus_master_enable_region,
677                               pci_get_word(s->config + PCI_COMMAND)
678                               & PCI_COMMAND_MASTER);
679 
680     g_free(config);
681     return 0;
682 }
683 
684 /* just put buffer */
685 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
686                                  const VMStateField *field, JSONWriter *vmdesc)
687 {
688     const uint8_t **v = pv;
689     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
690     qemu_put_buffer(f, *v, size);
691 
692     return 0;
693 }
694 
695 static const VMStateInfo vmstate_info_pci_config = {
696     .name = "pci config",
697     .get  = get_pci_config_device,
698     .put  = put_pci_config_device,
699 };
700 
701 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
702                              const VMStateField *field)
703 {
704     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
705     uint32_t irq_state[PCI_NUM_PINS];
706     int i;
707     for (i = 0; i < PCI_NUM_PINS; ++i) {
708         irq_state[i] = qemu_get_be32(f);
709         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
710             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
711                     irq_state[i]);
712             return -EINVAL;
713         }
714     }
715 
716     for (i = 0; i < PCI_NUM_PINS; ++i) {
717         pci_set_irq_state(s, i, irq_state[i]);
718     }
719 
720     return 0;
721 }
722 
723 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
724                              const VMStateField *field, JSONWriter *vmdesc)
725 {
726     int i;
727     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
728 
729     for (i = 0; i < PCI_NUM_PINS; ++i) {
730         qemu_put_be32(f, pci_irq_state(s, i));
731     }
732 
733     return 0;
734 }
735 
736 static const VMStateInfo vmstate_info_pci_irq_state = {
737     .name = "pci irq state",
738     .get  = get_pci_irq_state,
739     .put  = put_pci_irq_state,
740 };
741 
742 static bool migrate_is_pcie(void *opaque, int version_id)
743 {
744     return pci_is_express((PCIDevice *)opaque);
745 }
746 
747 static bool migrate_is_not_pcie(void *opaque, int version_id)
748 {
749     return !pci_is_express((PCIDevice *)opaque);
750 }
751 
752 const VMStateDescription vmstate_pci_device = {
753     .name = "PCIDevice",
754     .version_id = 2,
755     .minimum_version_id = 1,
756     .fields = (const VMStateField[]) {
757         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
758         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
759                                    migrate_is_not_pcie,
760                                    0, vmstate_info_pci_config,
761                                    PCI_CONFIG_SPACE_SIZE),
762         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
763                                    migrate_is_pcie,
764                                    0, vmstate_info_pci_config,
765                                    PCIE_CONFIG_SPACE_SIZE),
766         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
767                                    vmstate_info_pci_irq_state,
768                                    PCI_NUM_PINS * sizeof(int32_t)),
769         VMSTATE_END_OF_LIST()
770     }
771 };
772 
773 
774 void pci_device_save(PCIDevice *s, QEMUFile *f)
775 {
776     /* Clear interrupt status bit: it is implicit
777      * in irq_state which we are saving.
778      * This makes us compatible with old devices
779      * which never set or clear this bit. */
780     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
781     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
782     /* Restore the interrupt status bit. */
783     pci_update_irq_status(s);
784 }
785 
786 int pci_device_load(PCIDevice *s, QEMUFile *f)
787 {
788     int ret;
789     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
790     /* Restore the interrupt status bit. */
791     pci_update_irq_status(s);
792     return ret;
793 }
794 
795 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
796 {
797     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
798                  pci_default_sub_vendor_id);
799     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
800                  pci_default_sub_device_id);
801 }
802 
803 /*
804  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
805  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
806  */
807 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
808                              unsigned int *slotp, unsigned int *funcp)
809 {
810     const char *p;
811     char *e;
812     unsigned long val;
813     unsigned long dom = 0, bus = 0;
814     unsigned int slot = 0;
815     unsigned int func = 0;
816 
817     p = addr;
818     val = strtoul(p, &e, 16);
819     if (e == p)
820         return -1;
821     if (*e == ':') {
822         bus = val;
823         p = e + 1;
824         val = strtoul(p, &e, 16);
825         if (e == p)
826             return -1;
827         if (*e == ':') {
828             dom = bus;
829             bus = val;
830             p = e + 1;
831             val = strtoul(p, &e, 16);
832             if (e == p)
833                 return -1;
834         }
835     }
836 
837     slot = val;
838 
839     if (funcp != NULL) {
840         if (*e != '.')
841             return -1;
842 
843         p = e + 1;
844         val = strtoul(p, &e, 16);
845         if (e == p)
846             return -1;
847 
848         func = val;
849     }
850 
851     /* if funcp == NULL func is 0 */
852     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
853         return -1;
854 
855     if (*e)
856         return -1;
857 
858     *domp = dom;
859     *busp = bus;
860     *slotp = slot;
861     if (funcp != NULL)
862         *funcp = func;
863     return 0;
864 }
865 
866 static void pci_init_cmask(PCIDevice *dev)
867 {
868     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
869     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
870     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
871     dev->cmask[PCI_REVISION_ID] = 0xff;
872     dev->cmask[PCI_CLASS_PROG] = 0xff;
873     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
874     dev->cmask[PCI_HEADER_TYPE] = 0xff;
875     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
876 }
877 
878 static void pci_init_wmask(PCIDevice *dev)
879 {
880     int config_size = pci_config_size(dev);
881 
882     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
883     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
884     pci_set_word(dev->wmask + PCI_COMMAND,
885                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
886                  PCI_COMMAND_INTX_DISABLE);
887     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
888 
889     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
890            config_size - PCI_CONFIG_HEADER_SIZE);
891 }
892 
893 static void pci_init_w1cmask(PCIDevice *dev)
894 {
895     /*
896      * Note: It's okay to set w1cmask even for readonly bits as
897      * long as their value is hardwired to 0.
898      */
899     pci_set_word(dev->w1cmask + PCI_STATUS,
900                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
901                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
902                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
903 }
904 
905 static void pci_init_mask_bridge(PCIDevice *d)
906 {
907     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
908        PCI_SEC_LATENCY_TIMER */
909     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
910 
911     /* base and limit */
912     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
913     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
914     pci_set_word(d->wmask + PCI_MEMORY_BASE,
915                  PCI_MEMORY_RANGE_MASK & 0xffff);
916     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
917                  PCI_MEMORY_RANGE_MASK & 0xffff);
918     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
919                  PCI_PREF_RANGE_MASK & 0xffff);
920     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
921                  PCI_PREF_RANGE_MASK & 0xffff);
922 
923     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
924     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
925 
926     /* Supported memory and i/o types */
927     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
928     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
929     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
930                                PCI_PREF_RANGE_TYPE_64);
931     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
932                                PCI_PREF_RANGE_TYPE_64);
933 
934     /*
935      * TODO: Bridges default to 10-bit VGA decoding but we currently only
936      * implement 16-bit decoding (no alias support).
937      */
938     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
939                  PCI_BRIDGE_CTL_PARITY |
940                  PCI_BRIDGE_CTL_SERR |
941                  PCI_BRIDGE_CTL_ISA |
942                  PCI_BRIDGE_CTL_VGA |
943                  PCI_BRIDGE_CTL_VGA_16BIT |
944                  PCI_BRIDGE_CTL_MASTER_ABORT |
945                  PCI_BRIDGE_CTL_BUS_RESET |
946                  PCI_BRIDGE_CTL_FAST_BACK |
947                  PCI_BRIDGE_CTL_DISCARD |
948                  PCI_BRIDGE_CTL_SEC_DISCARD |
949                  PCI_BRIDGE_CTL_DISCARD_SERR);
950     /* Below does not do anything as we never set this bit, put here for
951      * completeness. */
952     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
953                  PCI_BRIDGE_CTL_DISCARD_STATUS);
954     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
955     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
956     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
957                                PCI_PREF_RANGE_TYPE_MASK);
958     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
959                                PCI_PREF_RANGE_TYPE_MASK);
960 }
961 
962 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
963 {
964     uint8_t slot = PCI_SLOT(dev->devfn);
965     uint8_t func;
966 
967     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
968         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
969     }
970 
971     /*
972      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
973      * device, as it may just be a VF that ended up with function 0 in
974      * the legacy PCI interpretation. Avoid failing in such cases:
975      */
976     if (pci_is_vf(dev) &&
977         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
978         return;
979     }
980 
981     /*
982      * multifunction bit is interpreted in two ways as follows.
983      *   - all functions must set the bit to 1.
984      *     Example: Intel X53
985      *   - function 0 must set the bit, but the rest function (> 0)
986      *     is allowed to leave the bit to 0.
987      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
988      *
989      * So OS (at least Linux) checks the bit of only function 0,
990      * and doesn't see the bit of function > 0.
991      *
992      * The below check allows both interpretation.
993      */
994     if (PCI_FUNC(dev->devfn)) {
995         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
996         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
997             /* function 0 should set multifunction bit */
998             error_setg(errp, "PCI: single function device can't be populated "
999                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
1000             return;
1001         }
1002         return;
1003     }
1004 
1005     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
1006         return;
1007     }
1008     /* function 0 indicates single function, so function > 0 must be NULL */
1009     for (func = 1; func < PCI_FUNC_MAX; ++func) {
1010         if (bus->devices[PCI_DEVFN(slot, func)]) {
1011             error_setg(errp, "PCI: %x.0 indicates single function, "
1012                        "but %x.%x is already populated.",
1013                        slot, slot, func);
1014             return;
1015         }
1016     }
1017 }
1018 
1019 static void pci_config_alloc(PCIDevice *pci_dev)
1020 {
1021     int config_size = pci_config_size(pci_dev);
1022 
1023     pci_dev->config = g_malloc0(config_size);
1024     pci_dev->cmask = g_malloc0(config_size);
1025     pci_dev->wmask = g_malloc0(config_size);
1026     pci_dev->w1cmask = g_malloc0(config_size);
1027     pci_dev->used = g_malloc0(config_size);
1028 }
1029 
1030 static void pci_config_free(PCIDevice *pci_dev)
1031 {
1032     g_free(pci_dev->config);
1033     g_free(pci_dev->cmask);
1034     g_free(pci_dev->wmask);
1035     g_free(pci_dev->w1cmask);
1036     g_free(pci_dev->used);
1037 }
1038 
1039 static void do_pci_unregister_device(PCIDevice *pci_dev)
1040 {
1041     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1042     pci_config_free(pci_dev);
1043 
1044     if (xen_mode == XEN_EMULATE) {
1045         xen_evtchn_remove_pci_device(pci_dev);
1046     }
1047     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1048         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1049                                     &pci_dev->bus_master_enable_region);
1050     }
1051     address_space_destroy(&pci_dev->bus_master_as);
1052 }
1053 
1054 /* Extract PCIReqIDCache into BDF format */
1055 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1056 {
1057     uint8_t bus_n;
1058     uint16_t result;
1059 
1060     switch (cache->type) {
1061     case PCI_REQ_ID_BDF:
1062         result = pci_get_bdf(cache->dev);
1063         break;
1064     case PCI_REQ_ID_SECONDARY_BUS:
1065         bus_n = pci_dev_bus_num(cache->dev);
1066         result = PCI_BUILD_BDF(bus_n, 0);
1067         break;
1068     default:
1069         error_report("Invalid PCI requester ID cache type: %d",
1070                      cache->type);
1071         exit(1);
1072         break;
1073     }
1074 
1075     return result;
1076 }
1077 
1078 /* Parse bridges up to the root complex and return requester ID
1079  * cache for specific device.  For full PCIe topology, the cache
1080  * result would be exactly the same as getting BDF of the device.
1081  * However, several tricks are required when system mixed up with
1082  * legacy PCI devices and PCIe-to-PCI bridges.
1083  *
1084  * Here we cache the proxy device (and type) not requester ID since
1085  * bus number might change from time to time.
1086  */
1087 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1088 {
1089     PCIDevice *parent;
1090     PCIReqIDCache cache = {
1091         .dev = dev,
1092         .type = PCI_REQ_ID_BDF,
1093     };
1094 
1095     while (!pci_bus_is_root(pci_get_bus(dev))) {
1096         /* We are under PCI/PCIe bridges */
1097         parent = pci_get_bus(dev)->parent_dev;
1098         if (pci_is_express(parent)) {
1099             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1100                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1101                  * override the requester ID using secondary bus
1102                  * number of parent bridge with zeroed devfn
1103                  * (pcie-to-pci bridge spec chap 2.3). */
1104                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1105                 cache.dev = dev;
1106             }
1107         } else {
1108             /* Legacy PCI, override requester ID with the bridge's
1109              * BDF upstream.  When the root complex connects to
1110              * legacy PCI devices (including buses), it can only
1111              * obtain requester ID info from directly attached
1112              * devices.  If devices are attached under bridges, only
1113              * the requester ID of the bridge that is directly
1114              * attached to the root complex can be recognized. */
1115             cache.type = PCI_REQ_ID_BDF;
1116             cache.dev = parent;
1117         }
1118         dev = parent;
1119     }
1120 
1121     return cache;
1122 }
1123 
1124 uint16_t pci_requester_id(PCIDevice *dev)
1125 {
1126     return pci_req_id_cache_extract(&dev->requester_id_cache);
1127 }
1128 
1129 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1130 {
1131     return !(bus->devices[devfn]);
1132 }
1133 
1134 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1135 {
1136     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1137 }
1138 
1139 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1140 {
1141     return bus->slot_reserved_mask;
1142 }
1143 
1144 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1145 {
1146     bus->slot_reserved_mask |= mask;
1147 }
1148 
1149 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1150 {
1151     bus->slot_reserved_mask &= ~mask;
1152 }
1153 
1154 /* -1 for devfn means auto assign */
1155 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1156                                          const char *name, int devfn,
1157                                          Error **errp)
1158 {
1159     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1160     PCIConfigReadFunc *config_read = pc->config_read;
1161     PCIConfigWriteFunc *config_write = pc->config_write;
1162     Error *local_err = NULL;
1163     DeviceState *dev = DEVICE(pci_dev);
1164     PCIBus *bus = pci_get_bus(pci_dev);
1165     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1166 
1167     /* Only pci bridges can be attached to extra PCI root buses */
1168     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1169         error_setg(errp,
1170                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1171                     bus->parent_dev->name);
1172         return NULL;
1173     }
1174 
1175     if (devfn < 0) {
1176         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1177             devfn += PCI_FUNC_MAX) {
1178             if (pci_bus_devfn_available(bus, devfn) &&
1179                    !pci_bus_devfn_reserved(bus, devfn)) {
1180                 goto found;
1181             }
1182         }
1183         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1184                    "or reserved", name);
1185         return NULL;
1186     found: ;
1187     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1188         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1189                    " reserved",
1190                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1191         return NULL;
1192     } else if (!pci_bus_devfn_available(bus, devfn)) {
1193         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1194                    " in use by %s,id=%s",
1195                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1196                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1197         return NULL;
1198     }
1199 
1200     /*
1201      * Populating function 0 triggers a scan from the guest that
1202      * exposes other non-zero functions. Hence we need to ensure that
1203      * function 0 wasn't added yet.
1204      */
1205     if (dev->hotplugged && !pci_is_vf(pci_dev) &&
1206         pci_get_function_0(pci_dev)) {
1207         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1208                    " new func %s cannot be exposed to guest.",
1209                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1210                    pci_get_function_0(pci_dev)->name,
1211                    name);
1212 
1213        return NULL;
1214     }
1215 
1216     pci_dev->devfn = devfn;
1217     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1218     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1219 
1220     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1221                        "bus master container", UINT64_MAX);
1222     address_space_init(&pci_dev->bus_master_as,
1223                        &pci_dev->bus_master_container_region, pci_dev->name);
1224     pci_dev->bus_master_as.max_bounce_buffer_size =
1225         pci_dev->max_bounce_buffer_size;
1226 
1227     if (phase_check(PHASE_MACHINE_READY)) {
1228         pci_init_bus_master(pci_dev);
1229     }
1230     pci_dev->irq_state = 0;
1231     pci_config_alloc(pci_dev);
1232 
1233     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1234     pci_config_set_device_id(pci_dev->config, pc->device_id);
1235     pci_config_set_revision(pci_dev->config, pc->revision);
1236     pci_config_set_class(pci_dev->config, pc->class_id);
1237 
1238     if (!is_bridge) {
1239         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1240             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1241                          pc->subsystem_vendor_id);
1242             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1243                          pc->subsystem_id);
1244         } else {
1245             pci_set_default_subsystem_id(pci_dev);
1246         }
1247     } else {
1248         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1249         assert(!pc->subsystem_vendor_id);
1250         assert(!pc->subsystem_id);
1251     }
1252     pci_init_cmask(pci_dev);
1253     pci_init_wmask(pci_dev);
1254     pci_init_w1cmask(pci_dev);
1255     if (is_bridge) {
1256         pci_init_mask_bridge(pci_dev);
1257     }
1258     pci_init_multifunction(bus, pci_dev, &local_err);
1259     if (local_err) {
1260         error_propagate(errp, local_err);
1261         do_pci_unregister_device(pci_dev);
1262         return NULL;
1263     }
1264 
1265     if (!config_read)
1266         config_read = pci_default_read_config;
1267     if (!config_write)
1268         config_write = pci_default_write_config;
1269     pci_dev->config_read = config_read;
1270     pci_dev->config_write = config_write;
1271     bus->devices[devfn] = pci_dev;
1272     pci_dev->version_id = 2; /* Current pci device vmstate version */
1273     return pci_dev;
1274 }
1275 
1276 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1277 {
1278     PCIIORegion *r;
1279     int i;
1280 
1281     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1282         r = &pci_dev->io_regions[i];
1283         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1284             continue;
1285         memory_region_del_subregion(r->address_space, r->memory);
1286     }
1287 
1288     pci_unregister_vga(pci_dev);
1289 }
1290 
1291 static void pci_qdev_unrealize(DeviceState *dev)
1292 {
1293     PCIDevice *pci_dev = PCI_DEVICE(dev);
1294     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1295 
1296     pci_unregister_io_regions(pci_dev);
1297     pci_del_option_rom(pci_dev);
1298 
1299     if (pc->exit) {
1300         pc->exit(pci_dev);
1301     }
1302 
1303     pci_device_deassert_intx(pci_dev);
1304     do_pci_unregister_device(pci_dev);
1305 
1306     pci_dev->msi_trigger = NULL;
1307 
1308     /*
1309      * clean up acpi-index so it could reused by another device
1310      */
1311     if (pci_dev->acpi_index) {
1312         GSequence *used_indexes = pci_acpi_index_list();
1313 
1314         g_sequence_remove(g_sequence_lookup(used_indexes,
1315                           GINT_TO_POINTER(pci_dev->acpi_index),
1316                           g_cmp_uint32, NULL));
1317     }
1318 }
1319 
1320 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1321                       uint8_t type, MemoryRegion *memory)
1322 {
1323     PCIIORegion *r;
1324     uint32_t addr; /* offset in pci config space */
1325     uint64_t wmask;
1326     pcibus_t size = memory_region_size(memory);
1327     uint8_t hdr_type;
1328 
1329     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1330     assert(region_num >= 0);
1331     assert(region_num < PCI_NUM_REGIONS);
1332     assert(is_power_of_2(size));
1333 
1334     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1335     hdr_type =
1336         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1337     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1338 
1339     r = &pci_dev->io_regions[region_num];
1340     r->addr = PCI_BAR_UNMAPPED;
1341     r->size = size;
1342     r->type = type;
1343     r->memory = memory;
1344     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1345                         ? pci_get_bus(pci_dev)->address_space_io
1346                         : pci_get_bus(pci_dev)->address_space_mem;
1347 
1348     wmask = ~(size - 1);
1349     if (region_num == PCI_ROM_SLOT) {
1350         /* ROM enable bit is writable */
1351         wmask |= PCI_ROM_ADDRESS_ENABLE;
1352     }
1353 
1354     addr = pci_bar(pci_dev, region_num);
1355     pci_set_long(pci_dev->config + addr, type);
1356 
1357     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1358         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1359         pci_set_quad(pci_dev->wmask + addr, wmask);
1360         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1361     } else {
1362         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1363         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1364     }
1365 }
1366 
1367 static void pci_update_vga(PCIDevice *pci_dev)
1368 {
1369     uint16_t cmd;
1370 
1371     if (!pci_dev->has_vga) {
1372         return;
1373     }
1374 
1375     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1376 
1377     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1378                               cmd & PCI_COMMAND_MEMORY);
1379     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1380                               cmd & PCI_COMMAND_IO);
1381     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1382                               cmd & PCI_COMMAND_IO);
1383 }
1384 
1385 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1386                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1387 {
1388     PCIBus *bus = pci_get_bus(pci_dev);
1389 
1390     assert(!pci_dev->has_vga);
1391 
1392     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1393     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1394     memory_region_add_subregion_overlap(bus->address_space_mem,
1395                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1396 
1397     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1398     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1399     memory_region_add_subregion_overlap(bus->address_space_io,
1400                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1401 
1402     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1403     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1404     memory_region_add_subregion_overlap(bus->address_space_io,
1405                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1406     pci_dev->has_vga = true;
1407 
1408     pci_update_vga(pci_dev);
1409 }
1410 
1411 void pci_unregister_vga(PCIDevice *pci_dev)
1412 {
1413     PCIBus *bus = pci_get_bus(pci_dev);
1414 
1415     if (!pci_dev->has_vga) {
1416         return;
1417     }
1418 
1419     memory_region_del_subregion(bus->address_space_mem,
1420                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1421     memory_region_del_subregion(bus->address_space_io,
1422                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1423     memory_region_del_subregion(bus->address_space_io,
1424                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1425     pci_dev->has_vga = false;
1426 }
1427 
1428 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1429 {
1430     return pci_dev->io_regions[region_num].addr;
1431 }
1432 
1433 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1434                                         uint8_t type, pcibus_t size)
1435 {
1436     pcibus_t new_addr;
1437     if (!pci_is_vf(d)) {
1438         int bar = pci_bar(d, reg);
1439         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1440             new_addr = pci_get_quad(d->config + bar);
1441         } else {
1442             new_addr = pci_get_long(d->config + bar);
1443         }
1444     } else {
1445         PCIDevice *pf = d->exp.sriov_vf.pf;
1446         uint16_t sriov_cap = pf->exp.sriov_cap;
1447         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1448         uint16_t vf_offset =
1449             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1450         uint16_t vf_stride =
1451             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1452         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1453 
1454         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1455             new_addr = pci_get_quad(pf->config + bar);
1456         } else {
1457             new_addr = pci_get_long(pf->config + bar);
1458         }
1459         new_addr += vf_num * size;
1460     }
1461     /* The ROM slot has a specific enable bit, keep it intact */
1462     if (reg != PCI_ROM_SLOT) {
1463         new_addr &= ~(size - 1);
1464     }
1465     return new_addr;
1466 }
1467 
1468 pcibus_t pci_bar_address(PCIDevice *d,
1469                          int reg, uint8_t type, pcibus_t size)
1470 {
1471     pcibus_t new_addr, last_addr;
1472     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1473     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1474     bool allow_0_address = mc->pci_allow_0_address;
1475 
1476     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1477         if (!(cmd & PCI_COMMAND_IO)) {
1478             return PCI_BAR_UNMAPPED;
1479         }
1480         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1481         last_addr = new_addr + size - 1;
1482         /* Check if 32 bit BAR wraps around explicitly.
1483          * TODO: make priorities correct and remove this work around.
1484          */
1485         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1486             (!allow_0_address && new_addr == 0)) {
1487             return PCI_BAR_UNMAPPED;
1488         }
1489         return new_addr;
1490     }
1491 
1492     if (!(cmd & PCI_COMMAND_MEMORY)) {
1493         return PCI_BAR_UNMAPPED;
1494     }
1495     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1496     /* the ROM slot has a specific enable bit */
1497     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1498         return PCI_BAR_UNMAPPED;
1499     }
1500     new_addr &= ~(size - 1);
1501     last_addr = new_addr + size - 1;
1502     /* NOTE: we do not support wrapping */
1503     /* XXX: as we cannot support really dynamic
1504        mappings, we handle specific values as invalid
1505        mappings. */
1506     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1507         (!allow_0_address && new_addr == 0)) {
1508         return PCI_BAR_UNMAPPED;
1509     }
1510 
1511     /* Now pcibus_t is 64bit.
1512      * Check if 32 bit BAR wraps around explicitly.
1513      * Without this, PC ide doesn't work well.
1514      * TODO: remove this work around.
1515      */
1516     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1517         return PCI_BAR_UNMAPPED;
1518     }
1519 
1520     /*
1521      * OS is allowed to set BAR beyond its addressable
1522      * bits. For example, 32 bit OS can set 64bit bar
1523      * to >4G. Check it. TODO: we might need to support
1524      * it in the future for e.g. PAE.
1525      */
1526     if (last_addr >= HWADDR_MAX) {
1527         return PCI_BAR_UNMAPPED;
1528     }
1529 
1530     return new_addr;
1531 }
1532 
1533 static void pci_update_mappings(PCIDevice *d)
1534 {
1535     PCIIORegion *r;
1536     int i;
1537     pcibus_t new_addr;
1538 
1539     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1540         r = &d->io_regions[i];
1541 
1542         /* this region isn't registered */
1543         if (!r->size)
1544             continue;
1545 
1546         new_addr = pci_bar_address(d, i, r->type, r->size);
1547         if (!d->has_power) {
1548             new_addr = PCI_BAR_UNMAPPED;
1549         }
1550 
1551         /* This bar isn't changed */
1552         if (new_addr == r->addr)
1553             continue;
1554 
1555         /* now do the real mapping */
1556         if (r->addr != PCI_BAR_UNMAPPED) {
1557             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1558                                           PCI_SLOT(d->devfn),
1559                                           PCI_FUNC(d->devfn),
1560                                           i, r->addr, r->size);
1561             memory_region_del_subregion(r->address_space, r->memory);
1562         }
1563         r->addr = new_addr;
1564         if (r->addr != PCI_BAR_UNMAPPED) {
1565             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1566                                           PCI_SLOT(d->devfn),
1567                                           PCI_FUNC(d->devfn),
1568                                           i, r->addr, r->size);
1569             memory_region_add_subregion_overlap(r->address_space,
1570                                                 r->addr, r->memory, 1);
1571         }
1572     }
1573 
1574     pci_update_vga(d);
1575 }
1576 
1577 static inline int pci_irq_disabled(PCIDevice *d)
1578 {
1579     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1580 }
1581 
1582 /* Called after interrupt disabled field update in config space,
1583  * assert/deassert interrupts if necessary.
1584  * Gets original interrupt disable bit value (before update). */
1585 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1586 {
1587     int i, disabled = pci_irq_disabled(d);
1588     if (disabled == was_irq_disabled)
1589         return;
1590     for (i = 0; i < PCI_NUM_PINS; ++i) {
1591         int state = pci_irq_state(d, i);
1592         pci_change_irq_level(d, i, disabled ? -state : state);
1593     }
1594 }
1595 
1596 uint32_t pci_default_read_config(PCIDevice *d,
1597                                  uint32_t address, int len)
1598 {
1599     uint32_t val = 0;
1600 
1601     assert(address + len <= pci_config_size(d));
1602 
1603     if (pci_is_express_downstream_port(d) &&
1604         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1605         pcie_sync_bridge_lnk(d);
1606     }
1607     memcpy(&val, d->config + address, len);
1608     return le32_to_cpu(val);
1609 }
1610 
1611 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1612 {
1613     int i, was_irq_disabled = pci_irq_disabled(d);
1614     uint32_t val = val_in;
1615 
1616     assert(addr + l <= pci_config_size(d));
1617 
1618     for (i = 0; i < l; val >>= 8, ++i) {
1619         uint8_t wmask = d->wmask[addr + i];
1620         uint8_t w1cmask = d->w1cmask[addr + i];
1621         assert(!(wmask & w1cmask));
1622         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1623         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1624     }
1625     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1626         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1627         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1628         range_covers_byte(addr, l, PCI_COMMAND))
1629         pci_update_mappings(d);
1630 
1631     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1632         pci_update_irq_disabled(d, was_irq_disabled);
1633         memory_region_set_enabled(&d->bus_master_enable_region,
1634                                   (pci_get_word(d->config + PCI_COMMAND)
1635                                    & PCI_COMMAND_MASTER) && d->has_power);
1636     }
1637 
1638     msi_write_config(d, addr, val_in, l);
1639     msix_write_config(d, addr, val_in, l);
1640     pcie_sriov_config_write(d, addr, val_in, l);
1641 }
1642 
1643 /***********************************************************/
1644 /* generic PCI irq support */
1645 
1646 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1647 static void pci_irq_handler(void *opaque, int irq_num, int level)
1648 {
1649     PCIDevice *pci_dev = opaque;
1650     int change;
1651 
1652     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1653     assert(level == 0 || level == 1);
1654     change = level - pci_irq_state(pci_dev, irq_num);
1655     if (!change)
1656         return;
1657 
1658     pci_set_irq_state(pci_dev, irq_num, level);
1659     pci_update_irq_status(pci_dev);
1660     if (pci_irq_disabled(pci_dev))
1661         return;
1662     pci_change_irq_level(pci_dev, irq_num, change);
1663 }
1664 
1665 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1666 {
1667     int intx = pci_intx(pci_dev);
1668     assert(0 <= intx && intx < PCI_NUM_PINS);
1669 
1670     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1671 }
1672 
1673 void pci_set_irq(PCIDevice *pci_dev, int level)
1674 {
1675     int intx = pci_intx(pci_dev);
1676     pci_irq_handler(pci_dev, intx, level);
1677 }
1678 
1679 /* Special hooks used by device assignment */
1680 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1681 {
1682     assert(pci_bus_is_root(bus));
1683     bus->route_intx_to_irq = route_intx_to_irq;
1684 }
1685 
1686 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1687 {
1688     PCIBus *bus;
1689 
1690     do {
1691         int dev_irq = pin;
1692         bus = pci_get_bus(dev);
1693         pin = bus->map_irq(dev, pin);
1694         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1695                             pci_bus_is_root(bus) ? "root-complex"
1696                                     : DEVICE(bus->parent_dev)->canonical_path);
1697         dev = bus->parent_dev;
1698     } while (dev);
1699 
1700     if (!bus->route_intx_to_irq) {
1701         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1702                      object_get_typename(OBJECT(bus->qbus.parent)));
1703         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1704     }
1705 
1706     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1707 }
1708 
1709 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1710 {
1711     return old->mode != new->mode || old->irq != new->irq;
1712 }
1713 
1714 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1715 {
1716     PCIDevice *dev;
1717     PCIBus *sec;
1718     int i;
1719 
1720     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1721         dev = bus->devices[i];
1722         if (dev && dev->intx_routing_notifier) {
1723             dev->intx_routing_notifier(dev);
1724         }
1725     }
1726 
1727     QLIST_FOREACH(sec, &bus->child, sibling) {
1728         pci_bus_fire_intx_routing_notifier(sec);
1729     }
1730 }
1731 
1732 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1733                                           PCIINTxRoutingNotifier notifier)
1734 {
1735     dev->intx_routing_notifier = notifier;
1736 }
1737 
1738 /*
1739  * PCI-to-PCI bridge specification
1740  * 9.1: Interrupt routing. Table 9-1
1741  *
1742  * the PCI Express Base Specification, Revision 2.1
1743  * 2.2.8.1: INTx interrupt signaling - Rules
1744  *          the Implementation Note
1745  *          Table 2-20
1746  */
1747 /*
1748  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1749  * 0-origin unlike PCI interrupt pin register.
1750  */
1751 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1752 {
1753     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1754 }
1755 
1756 /***********************************************************/
1757 /* monitor info on PCI */
1758 
1759 static const pci_class_desc pci_class_descriptions[] =
1760 {
1761     { 0x0001, "VGA controller", "display"},
1762     { 0x0100, "SCSI controller", "scsi"},
1763     { 0x0101, "IDE controller", "ide"},
1764     { 0x0102, "Floppy controller", "fdc"},
1765     { 0x0103, "IPI controller", "ipi"},
1766     { 0x0104, "RAID controller", "raid"},
1767     { 0x0106, "SATA controller"},
1768     { 0x0107, "SAS controller"},
1769     { 0x0180, "Storage controller"},
1770     { 0x0200, "Ethernet controller", "ethernet"},
1771     { 0x0201, "Token Ring controller", "token-ring"},
1772     { 0x0202, "FDDI controller", "fddi"},
1773     { 0x0203, "ATM controller", "atm"},
1774     { 0x0280, "Network controller"},
1775     { 0x0300, "VGA controller", "display", 0x00ff},
1776     { 0x0301, "XGA controller"},
1777     { 0x0302, "3D controller"},
1778     { 0x0380, "Display controller"},
1779     { 0x0400, "Video controller", "video"},
1780     { 0x0401, "Audio controller", "sound"},
1781     { 0x0402, "Phone"},
1782     { 0x0403, "Audio controller", "sound"},
1783     { 0x0480, "Multimedia controller"},
1784     { 0x0500, "RAM controller", "memory"},
1785     { 0x0501, "Flash controller", "flash"},
1786     { 0x0580, "Memory controller"},
1787     { 0x0600, "Host bridge", "host"},
1788     { 0x0601, "ISA bridge", "isa"},
1789     { 0x0602, "EISA bridge", "eisa"},
1790     { 0x0603, "MC bridge", "mca"},
1791     { 0x0604, "PCI bridge", "pci-bridge"},
1792     { 0x0605, "PCMCIA bridge", "pcmcia"},
1793     { 0x0606, "NUBUS bridge", "nubus"},
1794     { 0x0607, "CARDBUS bridge", "cardbus"},
1795     { 0x0608, "RACEWAY bridge"},
1796     { 0x0680, "Bridge"},
1797     { 0x0700, "Serial port", "serial"},
1798     { 0x0701, "Parallel port", "parallel"},
1799     { 0x0800, "Interrupt controller", "interrupt-controller"},
1800     { 0x0801, "DMA controller", "dma-controller"},
1801     { 0x0802, "Timer", "timer"},
1802     { 0x0803, "RTC", "rtc"},
1803     { 0x0900, "Keyboard", "keyboard"},
1804     { 0x0901, "Pen", "pen"},
1805     { 0x0902, "Mouse", "mouse"},
1806     { 0x0A00, "Dock station", "dock", 0x00ff},
1807     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1808     { 0x0c00, "Firewire controller", "firewire"},
1809     { 0x0c01, "Access bus controller", "access-bus"},
1810     { 0x0c02, "SSA controller", "ssa"},
1811     { 0x0c03, "USB controller", "usb"},
1812     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1813     { 0x0c05, "SMBus"},
1814     { 0, NULL}
1815 };
1816 
1817 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1818                                            pci_bus_dev_fn fn,
1819                                            void *opaque)
1820 {
1821     PCIDevice *d;
1822     int devfn;
1823 
1824     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1825         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1826         if (d) {
1827             fn(bus, d, opaque);
1828         }
1829     }
1830 }
1831 
1832 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1833                                  pci_bus_dev_fn fn, void *opaque)
1834 {
1835     bus = pci_find_bus_nr(bus, bus_num);
1836 
1837     if (bus) {
1838         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1839     }
1840 }
1841 
1842 void pci_for_each_device_under_bus(PCIBus *bus,
1843                                    pci_bus_dev_fn fn, void *opaque)
1844 {
1845     PCIDevice *d;
1846     int devfn;
1847 
1848     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1849         d = bus->devices[devfn];
1850         if (d) {
1851             fn(bus, d, opaque);
1852         }
1853     }
1854 }
1855 
1856 void pci_for_each_device(PCIBus *bus, int bus_num,
1857                          pci_bus_dev_fn fn, void *opaque)
1858 {
1859     bus = pci_find_bus_nr(bus, bus_num);
1860 
1861     if (bus) {
1862         pci_for_each_device_under_bus(bus, fn, opaque);
1863     }
1864 }
1865 
1866 const pci_class_desc *get_class_desc(int class)
1867 {
1868     const pci_class_desc *desc;
1869 
1870     desc = pci_class_descriptions;
1871     while (desc->desc && class != desc->class) {
1872         desc++;
1873     }
1874 
1875     return desc;
1876 }
1877 
1878 void pci_init_nic_devices(PCIBus *bus, const char *default_model)
1879 {
1880     qemu_create_nic_bus_devices(&bus->qbus, TYPE_PCI_DEVICE, default_model,
1881                                 "virtio", "virtio-net-pci");
1882 }
1883 
1884 bool pci_init_nic_in_slot(PCIBus *rootbus, const char *model,
1885                           const char *alias, const char *devaddr)
1886 {
1887     NICInfo *nd = qemu_find_nic_info(model, true, alias);
1888     int dom, busnr, devfn;
1889     PCIDevice *pci_dev;
1890     unsigned slot;
1891     PCIBus *bus;
1892 
1893     if (!nd) {
1894         return false;
1895     }
1896 
1897     if (!devaddr || pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1898         error_report("Invalid PCI device address %s for device %s",
1899                      devaddr, model);
1900         exit(1);
1901     }
1902 
1903     if (dom != 0) {
1904         error_report("No support for non-zero PCI domains");
1905         exit(1);
1906     }
1907 
1908     devfn = PCI_DEVFN(slot, 0);
1909 
1910     bus = pci_find_bus_nr(rootbus, busnr);
1911     if (!bus) {
1912         error_report("Invalid PCI device address %s for device %s",
1913                      devaddr, model);
1914         exit(1);
1915     }
1916 
1917     pci_dev = pci_new(devfn, model);
1918     qdev_set_nic_properties(&pci_dev->qdev, nd);
1919     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1920     return true;
1921 }
1922 
1923 PCIDevice *pci_vga_init(PCIBus *bus)
1924 {
1925     vga_interface_created = true;
1926     switch (vga_interface_type) {
1927     case VGA_CIRRUS:
1928         return pci_create_simple(bus, -1, "cirrus-vga");
1929     case VGA_QXL:
1930         return pci_create_simple(bus, -1, "qxl-vga");
1931     case VGA_STD:
1932         return pci_create_simple(bus, -1, "VGA");
1933     case VGA_VMWARE:
1934         return pci_create_simple(bus, -1, "vmware-svga");
1935     case VGA_VIRTIO:
1936         return pci_create_simple(bus, -1, "virtio-vga");
1937     case VGA_NONE:
1938     default: /* Other non-PCI types. Checking for unsupported types is already
1939                 done in vl.c. */
1940         return NULL;
1941     }
1942 }
1943 
1944 /* Whether a given bus number is in range of the secondary
1945  * bus of the given bridge device. */
1946 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1947 {
1948     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1949              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1950         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1951         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1952 }
1953 
1954 /* Whether a given bus number is in a range of a root bus */
1955 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1956 {
1957     int i;
1958 
1959     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1960         PCIDevice *dev = bus->devices[i];
1961 
1962         if (dev && IS_PCI_BRIDGE(dev)) {
1963             if (pci_secondary_bus_in_range(dev, bus_num)) {
1964                 return true;
1965             }
1966         }
1967     }
1968 
1969     return false;
1970 }
1971 
1972 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1973 {
1974     PCIBus *sec;
1975 
1976     if (!bus) {
1977         return NULL;
1978     }
1979 
1980     if (pci_bus_num(bus) == bus_num) {
1981         return bus;
1982     }
1983 
1984     /* Consider all bus numbers in range for the host pci bridge. */
1985     if (!pci_bus_is_root(bus) &&
1986         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1987         return NULL;
1988     }
1989 
1990     /* try child bus */
1991     for (; bus; bus = sec) {
1992         QLIST_FOREACH(sec, &bus->child, sibling) {
1993             if (pci_bus_num(sec) == bus_num) {
1994                 return sec;
1995             }
1996             /* PXB buses assumed to be children of bus 0 */
1997             if (pci_bus_is_root(sec)) {
1998                 if (pci_root_bus_in_range(sec, bus_num)) {
1999                     break;
2000                 }
2001             } else {
2002                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2003                     break;
2004                 }
2005             }
2006         }
2007     }
2008 
2009     return NULL;
2010 }
2011 
2012 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
2013                                   pci_bus_fn end, void *parent_state)
2014 {
2015     PCIBus *sec;
2016     void *state;
2017 
2018     if (!bus) {
2019         return;
2020     }
2021 
2022     if (begin) {
2023         state = begin(bus, parent_state);
2024     } else {
2025         state = parent_state;
2026     }
2027 
2028     QLIST_FOREACH(sec, &bus->child, sibling) {
2029         pci_for_each_bus_depth_first(sec, begin, end, state);
2030     }
2031 
2032     if (end) {
2033         end(bus, state);
2034     }
2035 }
2036 
2037 
2038 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2039 {
2040     bus = pci_find_bus_nr(bus, bus_num);
2041 
2042     if (!bus)
2043         return NULL;
2044 
2045     return bus->devices[devfn];
2046 }
2047 
2048 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2049 
2050 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2051 {
2052     PCIDevice *pci_dev = (PCIDevice *)qdev;
2053     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2054     ObjectClass *klass = OBJECT_CLASS(pc);
2055     Error *local_err = NULL;
2056     bool is_default_rom;
2057     uint16_t class_id;
2058 
2059     /*
2060      * capped by systemd (see: udev-builtin-net_id.c)
2061      * as it's the only known user honor it to avoid users
2062      * misconfigure QEMU and then wonder why acpi-index doesn't work
2063      */
2064     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2065         error_setg(errp, "acpi-index should be less or equal to %u",
2066                    ONBOARD_INDEX_MAX);
2067         return;
2068     }
2069 
2070     /*
2071      * make sure that acpi-index is unique across all present PCI devices
2072      */
2073     if (pci_dev->acpi_index) {
2074         GSequence *used_indexes = pci_acpi_index_list();
2075 
2076         if (g_sequence_lookup(used_indexes,
2077                               GINT_TO_POINTER(pci_dev->acpi_index),
2078                               g_cmp_uint32, NULL)) {
2079             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2080                        " already exist", pci_dev->acpi_index);
2081             return;
2082         }
2083         g_sequence_insert_sorted(used_indexes,
2084                                  GINT_TO_POINTER(pci_dev->acpi_index),
2085                                  g_cmp_uint32, NULL);
2086     }
2087 
2088     if (pci_dev->romsize != UINT32_MAX && !is_power_of_2(pci_dev->romsize)) {
2089         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2090         return;
2091     }
2092 
2093     /* initialize cap_present for pci_is_express() and pci_config_size(),
2094      * Note that hybrid PCIs are not set automatically and need to manage
2095      * QEMU_PCI_CAP_EXPRESS manually */
2096     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2097        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2098         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2099     }
2100 
2101     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2102         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2103     }
2104 
2105     pci_dev = do_pci_register_device(pci_dev,
2106                                      object_get_typename(OBJECT(qdev)),
2107                                      pci_dev->devfn, errp);
2108     if (pci_dev == NULL)
2109         return;
2110 
2111     if (pc->realize) {
2112         pc->realize(pci_dev, &local_err);
2113         if (local_err) {
2114             error_propagate(errp, local_err);
2115             do_pci_unregister_device(pci_dev);
2116             return;
2117         }
2118     }
2119 
2120     /*
2121      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2122      * associate only Device 0 with the device attached to the bus
2123      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2124      * sec 7.3.1).
2125      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2126      * 5-bit Device Number and 3-bit Function Number fields in its associated
2127      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2128      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2129      */
2130     if (pci_is_express(pci_dev) &&
2131         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2132         pcie_has_upstream_port(pci_dev) &&
2133         PCI_SLOT(pci_dev->devfn)) {
2134         warn_report("PCI: slot %d is not valid for %s,"
2135                     " parent device only allows plugging into slot 0.",
2136                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2137     }
2138 
2139     if (pci_dev->failover_pair_id) {
2140         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2141             error_setg(errp, "failover primary device must be on "
2142                              "PCIExpress bus");
2143             pci_qdev_unrealize(DEVICE(pci_dev));
2144             return;
2145         }
2146         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2147         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2148             error_setg(errp, "failover primary device is not an "
2149                              "Ethernet device");
2150             pci_qdev_unrealize(DEVICE(pci_dev));
2151             return;
2152         }
2153         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2154             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2155             error_setg(errp, "failover: primary device must be in its own "
2156                               "PCI slot");
2157             pci_qdev_unrealize(DEVICE(pci_dev));
2158             return;
2159         }
2160         qdev->allow_unplug_during_migration = true;
2161     }
2162 
2163     /* rom loading */
2164     is_default_rom = false;
2165     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2166         pci_dev->romfile = g_strdup(pc->romfile);
2167         is_default_rom = true;
2168     }
2169 
2170     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2171     if (local_err) {
2172         error_propagate(errp, local_err);
2173         pci_qdev_unrealize(DEVICE(pci_dev));
2174         return;
2175     }
2176 
2177     pci_set_power(pci_dev, true);
2178 
2179     pci_dev->msi_trigger = pci_msi_trigger;
2180 }
2181 
2182 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2183                                    const char *name)
2184 {
2185     DeviceState *dev;
2186 
2187     dev = qdev_new(name);
2188     qdev_prop_set_int32(dev, "addr", devfn);
2189     qdev_prop_set_bit(dev, "multifunction", multifunction);
2190     return PCI_DEVICE(dev);
2191 }
2192 
2193 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2194 {
2195     return pci_new_internal(devfn, true, name);
2196 }
2197 
2198 PCIDevice *pci_new(int devfn, const char *name)
2199 {
2200     return pci_new_internal(devfn, false, name);
2201 }
2202 
2203 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2204 {
2205     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2206 }
2207 
2208 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2209                                            const char *name)
2210 {
2211     PCIDevice *dev = pci_new_multifunction(devfn, name);
2212     pci_realize_and_unref(dev, bus, &error_fatal);
2213     return dev;
2214 }
2215 
2216 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2217 {
2218     PCIDevice *dev = pci_new(devfn, name);
2219     pci_realize_and_unref(dev, bus, &error_fatal);
2220     return dev;
2221 }
2222 
2223 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2224 {
2225     int offset = PCI_CONFIG_HEADER_SIZE;
2226     int i;
2227     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2228         if (pdev->used[i])
2229             offset = i + 1;
2230         else if (i - offset + 1 == size)
2231             return offset;
2232     }
2233     return 0;
2234 }
2235 
2236 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2237                                         uint8_t *prev_p)
2238 {
2239     uint8_t next, prev;
2240 
2241     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2242         return 0;
2243 
2244     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2245          prev = next + PCI_CAP_LIST_NEXT)
2246         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2247             break;
2248 
2249     if (prev_p)
2250         *prev_p = prev;
2251     return next;
2252 }
2253 
2254 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2255 {
2256     uint8_t next, prev, found = 0;
2257 
2258     if (!(pdev->used[offset])) {
2259         return 0;
2260     }
2261 
2262     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2263 
2264     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2265          prev = next + PCI_CAP_LIST_NEXT) {
2266         if (next <= offset && next > found) {
2267             found = next;
2268         }
2269     }
2270     return found;
2271 }
2272 
2273 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2274    This is needed for an option rom which is used for more than one device. */
2275 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2276 {
2277     uint16_t vendor_id;
2278     uint16_t device_id;
2279     uint16_t rom_vendor_id;
2280     uint16_t rom_device_id;
2281     uint16_t rom_magic;
2282     uint16_t pcir_offset;
2283     uint8_t checksum;
2284 
2285     /* Words in rom data are little endian (like in PCI configuration),
2286        so they can be read / written with pci_get_word / pci_set_word. */
2287 
2288     /* Only a valid rom will be patched. */
2289     rom_magic = pci_get_word(ptr);
2290     if (rom_magic != 0xaa55) {
2291         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2292         return;
2293     }
2294     pcir_offset = pci_get_word(ptr + 0x18);
2295     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2296         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2297         return;
2298     }
2299 
2300     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2301     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2302     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2303     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2304 
2305     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2306                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2307 
2308     checksum = ptr[6];
2309 
2310     if (vendor_id != rom_vendor_id) {
2311         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2312         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2313         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2314         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2315         ptr[6] = checksum;
2316         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2317     }
2318 
2319     if (device_id != rom_device_id) {
2320         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2321         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2322         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2323         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2324         ptr[6] = checksum;
2325         pci_set_word(ptr + pcir_offset + 6, device_id);
2326     }
2327 }
2328 
2329 /* Add an option rom for the device */
2330 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2331                                Error **errp)
2332 {
2333     int64_t size = 0;
2334     g_autofree char *path = NULL;
2335     char name[32];
2336     const VMStateDescription *vmsd;
2337 
2338     /*
2339      * In case of incoming migration ROM will come with migration stream, no
2340      * reason to load the file.  Neither we want to fail if local ROM file
2341      * mismatches with specified romsize.
2342      */
2343     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2344 
2345     if (!pdev->romfile || !strlen(pdev->romfile)) {
2346         return;
2347     }
2348 
2349     if (!pdev->rom_bar) {
2350         /*
2351          * Load rom via fw_cfg instead of creating a rom bar,
2352          * for 0.11 compatibility.
2353          */
2354         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2355 
2356         /*
2357          * Hot-plugged devices can't use the option ROM
2358          * if the rom bar is disabled.
2359          */
2360         if (DEVICE(pdev)->hotplugged) {
2361             error_setg(errp, "Hot-plugged device without ROM bar"
2362                        " can't have an option ROM");
2363             return;
2364         }
2365 
2366         if (class == 0x0300) {
2367             rom_add_vga(pdev->romfile);
2368         } else {
2369             rom_add_option(pdev->romfile, -1);
2370         }
2371         return;
2372     }
2373 
2374     if (load_file || pdev->romsize == UINT32_MAX) {
2375         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2376         if (path == NULL) {
2377             path = g_strdup(pdev->romfile);
2378         }
2379 
2380         size = get_image_size(path);
2381         if (size < 0) {
2382             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2383             return;
2384         } else if (size == 0) {
2385             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2386             return;
2387         } else if (size > 2 * GiB) {
2388             error_setg(errp,
2389                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2390                        pdev->romfile);
2391             return;
2392         }
2393         if (pdev->romsize != UINT_MAX) {
2394             if (size > pdev->romsize) {
2395                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2396                            "is too large for ROM size %u",
2397                            pdev->romfile, (uint32_t)size, pdev->romsize);
2398                 return;
2399             }
2400         } else {
2401             pdev->romsize = pow2ceil(size);
2402         }
2403     }
2404 
2405     vmsd = qdev_get_vmsd(DEVICE(pdev));
2406     snprintf(name, sizeof(name), "%s.rom",
2407              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2408 
2409     pdev->has_rom = true;
2410     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2411                            &error_fatal);
2412 
2413     if (load_file) {
2414         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2415 
2416         if (load_image_size(path, ptr, size) < 0) {
2417             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2418             return;
2419         }
2420 
2421         if (is_default_rom) {
2422             /* Only the default rom images will be patched (if needed). */
2423             pci_patch_ids(pdev, ptr, size);
2424         }
2425     }
2426 
2427     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2428 }
2429 
2430 static void pci_del_option_rom(PCIDevice *pdev)
2431 {
2432     if (!pdev->has_rom)
2433         return;
2434 
2435     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2436     pdev->has_rom = false;
2437 }
2438 
2439 /*
2440  * On success, pci_add_capability() returns a positive value
2441  * that the offset of the pci capability.
2442  * On failure, it sets an error and returns a negative error
2443  * code.
2444  */
2445 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2446                        uint8_t offset, uint8_t size,
2447                        Error **errp)
2448 {
2449     uint8_t *config;
2450     int i, overlapping_cap;
2451 
2452     if (!offset) {
2453         offset = pci_find_space(pdev, size);
2454         /* out of PCI config space is programming error */
2455         assert(offset);
2456     } else {
2457         /* Verify that capabilities don't overlap.  Note: device assignment
2458          * depends on this check to verify that the device is not broken.
2459          * Should never trigger for emulated devices, but it's helpful
2460          * for debugging these. */
2461         for (i = offset; i < offset + size; i++) {
2462             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2463             if (overlapping_cap) {
2464                 error_setg(errp, "%s:%02x:%02x.%x "
2465                            "Attempt to add PCI capability %x at offset "
2466                            "%x overlaps existing capability %x at offset %x",
2467                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2468                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2469                            cap_id, offset, overlapping_cap, i);
2470                 return -EINVAL;
2471             }
2472         }
2473     }
2474 
2475     config = pdev->config + offset;
2476     config[PCI_CAP_LIST_ID] = cap_id;
2477     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2478     pdev->config[PCI_CAPABILITY_LIST] = offset;
2479     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2480     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2481     /* Make capability read-only by default */
2482     memset(pdev->wmask + offset, 0, size);
2483     /* Check capability by default */
2484     memset(pdev->cmask + offset, 0xFF, size);
2485     return offset;
2486 }
2487 
2488 /* Unlink capability from the pci config space. */
2489 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2490 {
2491     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2492     if (!offset)
2493         return;
2494     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2495     /* Make capability writable again */
2496     memset(pdev->wmask + offset, 0xff, size);
2497     memset(pdev->w1cmask + offset, 0, size);
2498     /* Clear cmask as device-specific registers can't be checked */
2499     memset(pdev->cmask + offset, 0, size);
2500     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2501 
2502     if (!pdev->config[PCI_CAPABILITY_LIST])
2503         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2504 }
2505 
2506 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2507 {
2508     return pci_find_capability_list(pdev, cap_id, NULL);
2509 }
2510 
2511 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2512 {
2513     PCIDevice *d = (PCIDevice *)dev;
2514     const char *name = NULL;
2515     const pci_class_desc *desc =  pci_class_descriptions;
2516     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2517 
2518     while (desc->desc &&
2519           (class & ~desc->fw_ign_bits) !=
2520           (desc->class & ~desc->fw_ign_bits)) {
2521         desc++;
2522     }
2523 
2524     if (desc->desc) {
2525         name = desc->fw_name;
2526     }
2527 
2528     if (name) {
2529         pstrcpy(buf, len, name);
2530     } else {
2531         snprintf(buf, len, "pci%04x,%04x",
2532                  pci_get_word(d->config + PCI_VENDOR_ID),
2533                  pci_get_word(d->config + PCI_DEVICE_ID));
2534     }
2535 
2536     return buf;
2537 }
2538 
2539 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2540 {
2541     PCIDevice *d = (PCIDevice *)dev;
2542     char name[33];
2543     int has_func = !!PCI_FUNC(d->devfn);
2544 
2545     return g_strdup_printf("%s@%x%s%.*x",
2546                            pci_dev_fw_name(dev, name, sizeof(name)),
2547                            PCI_SLOT(d->devfn),
2548                            has_func ? "," : "",
2549                            has_func,
2550                            PCI_FUNC(d->devfn));
2551 }
2552 
2553 static char *pcibus_get_dev_path(DeviceState *dev)
2554 {
2555     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2556     PCIDevice *t;
2557     int slot_depth;
2558     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2559      * 00 is added here to make this format compatible with
2560      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2561      * Slot.Function list specifies the slot and function numbers for all
2562      * devices on the path from root to the specific device. */
2563     const char *root_bus_path;
2564     int root_bus_len;
2565     char slot[] = ":SS.F";
2566     int slot_len = sizeof slot - 1 /* For '\0' */;
2567     int path_len;
2568     char *path, *p;
2569     int s;
2570 
2571     root_bus_path = pci_root_bus_path(d);
2572     root_bus_len = strlen(root_bus_path);
2573 
2574     /* Calculate # of slots on path between device and root. */;
2575     slot_depth = 0;
2576     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2577         ++slot_depth;
2578     }
2579 
2580     path_len = root_bus_len + slot_len * slot_depth;
2581 
2582     /* Allocate memory, fill in the terminating null byte. */
2583     path = g_malloc(path_len + 1 /* For '\0' */);
2584     path[path_len] = '\0';
2585 
2586     memcpy(path, root_bus_path, root_bus_len);
2587 
2588     /* Fill in slot numbers. We walk up from device to root, so need to print
2589      * them in the reverse order, last to first. */
2590     p = path + path_len;
2591     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2592         p -= slot_len;
2593         s = snprintf(slot, sizeof slot, ":%02x.%x",
2594                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2595         assert(s == slot_len);
2596         memcpy(p, slot, slot_len);
2597     }
2598 
2599     return path;
2600 }
2601 
2602 static int pci_qdev_find_recursive(PCIBus *bus,
2603                                    const char *id, PCIDevice **pdev)
2604 {
2605     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2606     if (!qdev) {
2607         return -ENODEV;
2608     }
2609 
2610     /* roughly check if given qdev is pci device */
2611     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2612         *pdev = PCI_DEVICE(qdev);
2613         return 0;
2614     }
2615     return -EINVAL;
2616 }
2617 
2618 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2619 {
2620     PCIHostState *host_bridge;
2621     int rc = -ENODEV;
2622 
2623     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2624         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2625         if (!tmp) {
2626             rc = 0;
2627             break;
2628         }
2629         if (tmp != -ENODEV) {
2630             rc = tmp;
2631         }
2632     }
2633 
2634     return rc;
2635 }
2636 
2637 MemoryRegion *pci_address_space(PCIDevice *dev)
2638 {
2639     return pci_get_bus(dev)->address_space_mem;
2640 }
2641 
2642 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2643 {
2644     return pci_get_bus(dev)->address_space_io;
2645 }
2646 
2647 static void pci_device_class_init(ObjectClass *klass, void *data)
2648 {
2649     DeviceClass *k = DEVICE_CLASS(klass);
2650 
2651     k->realize = pci_qdev_realize;
2652     k->unrealize = pci_qdev_unrealize;
2653     k->bus_type = TYPE_PCI_BUS;
2654     device_class_set_props(k, pci_props);
2655     object_class_property_set_description(
2656         klass, "x-max-bounce-buffer-size",
2657         "Maximum buffer size allocated for bounce buffers used for mapped "
2658         "access to indirect DMA memory");
2659 }
2660 
2661 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2662 {
2663     if (!object_class_is_abstract(klass)) {
2664         ObjectClass *conventional =
2665             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2666         ObjectClass *pcie =
2667             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2668         ObjectClass *cxl =
2669             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2670         assert(conventional || pcie || cxl);
2671     }
2672 }
2673 
2674 /*
2675  * Get IOMMU root bus, aliased bus and devfn of a PCI device
2676  *
2677  * IOMMU root bus is needed by all call sites to call into iommu_ops.
2678  * For call sites which don't need aliased BDF, passing NULL to
2679  * aliased_[bus|devfn] is allowed.
2680  *
2681  * @piommu_bus: return root #PCIBus backed by an IOMMU for the PCI device.
2682  *
2683  * @aliased_bus: return aliased #PCIBus of the PCI device, optional.
2684  *
2685  * @aliased_devfn: return aliased devfn of the PCI device, optional.
2686  */
2687 static void pci_device_get_iommu_bus_devfn(PCIDevice *dev,
2688                                            PCIBus **piommu_bus,
2689                                            PCIBus **aliased_bus,
2690                                            int *aliased_devfn)
2691 {
2692     PCIBus *bus = pci_get_bus(dev);
2693     PCIBus *iommu_bus = bus;
2694     int devfn = dev->devfn;
2695 
2696     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2697         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2698 
2699         /*
2700          * The requester ID of the provided device may be aliased, as seen from
2701          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2702          * requester ID to provide a unique AddressSpace for devices, but
2703          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2704          * to-PCI bridge creates and accepts transactions on behalf of down-
2705          * stream devices.  When doing so, all downstream devices are masked
2706          * (aliased) behind a single requester ID.  The requester ID used
2707          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2708          * bridges, with a PCIe capability indicating such, follow the
2709          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2710          * where the bridge uses the seconary bus as the bridge portion of the
2711          * requester ID and devfn of 00.0.  For other bridges, typically those
2712          * found on the root complex such as the dmi-to-pci-bridge, we follow
2713          * the convention of typical bare-metal hardware, which uses the
2714          * requester ID of the bridge itself.  There are device specific
2715          * exceptions to these rules, but these are the defaults that the
2716          * Linux kernel uses when determining DMA aliases itself and believed
2717          * to be true for the bare metal equivalents of the devices emulated
2718          * in QEMU.
2719          */
2720         if (!pci_bus_is_express(iommu_bus)) {
2721             PCIDevice *parent = iommu_bus->parent_dev;
2722 
2723             if (pci_is_express(parent) &&
2724                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2725                 devfn = PCI_DEVFN(0, 0);
2726                 bus = iommu_bus;
2727             } else {
2728                 devfn = parent->devfn;
2729                 bus = parent_bus;
2730             }
2731         }
2732 
2733         iommu_bus = parent_bus;
2734     }
2735 
2736     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
2737     assert(iommu_bus);
2738 
2739     if (pci_bus_bypass_iommu(bus) || !iommu_bus->iommu_ops) {
2740         iommu_bus = NULL;
2741     }
2742 
2743     *piommu_bus = iommu_bus;
2744 
2745     if (aliased_bus) {
2746         *aliased_bus = bus;
2747     }
2748 
2749     if (aliased_devfn) {
2750         *aliased_devfn = devfn;
2751     }
2752 }
2753 
2754 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2755 {
2756     PCIBus *bus;
2757     PCIBus *iommu_bus;
2758     int devfn;
2759 
2760     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn);
2761     if (iommu_bus) {
2762         return iommu_bus->iommu_ops->get_address_space(bus,
2763                                  iommu_bus->iommu_opaque, devfn);
2764     }
2765     return &address_space_memory;
2766 }
2767 
2768 bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod,
2769                                  Error **errp)
2770 {
2771     PCIBus *iommu_bus, *aliased_bus;
2772     int aliased_devfn;
2773 
2774     /* set_iommu_device requires device's direct BDF instead of aliased BDF */
2775     pci_device_get_iommu_bus_devfn(dev, &iommu_bus,
2776                                    &aliased_bus, &aliased_devfn);
2777     if (iommu_bus && iommu_bus->iommu_ops->set_iommu_device) {
2778         hiod->aliased_bus = aliased_bus;
2779         hiod->aliased_devfn = aliased_devfn;
2780         return iommu_bus->iommu_ops->set_iommu_device(pci_get_bus(dev),
2781                                                       iommu_bus->iommu_opaque,
2782                                                       dev->devfn, hiod, errp);
2783     }
2784     return true;
2785 }
2786 
2787 void pci_device_unset_iommu_device(PCIDevice *dev)
2788 {
2789     PCIBus *iommu_bus;
2790 
2791     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, NULL, NULL);
2792     if (iommu_bus && iommu_bus->iommu_ops->unset_iommu_device) {
2793         return iommu_bus->iommu_ops->unset_iommu_device(pci_get_bus(dev),
2794                                                         iommu_bus->iommu_opaque,
2795                                                         dev->devfn);
2796     }
2797 }
2798 
2799 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2800 {
2801     /*
2802      * If called, pci_setup_iommu() should provide a minimum set of
2803      * useful callbacks for the bus.
2804      */
2805     assert(ops);
2806     assert(ops->get_address_space);
2807 
2808     bus->iommu_ops = ops;
2809     bus->iommu_opaque = opaque;
2810 }
2811 
2812 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2813 {
2814     Range *range = opaque;
2815     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2816     int i;
2817 
2818     if (!(cmd & PCI_COMMAND_MEMORY)) {
2819         return;
2820     }
2821 
2822     if (IS_PCI_BRIDGE(dev)) {
2823         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2824         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2825 
2826         base = MAX(base, 0x1ULL << 32);
2827 
2828         if (limit >= base) {
2829             Range pref_range;
2830             range_set_bounds(&pref_range, base, limit);
2831             range_extend(range, &pref_range);
2832         }
2833     }
2834     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2835         PCIIORegion *r = &dev->io_regions[i];
2836         pcibus_t lob, upb;
2837         Range region_range;
2838 
2839         if (!r->size ||
2840             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2841             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2842             continue;
2843         }
2844 
2845         lob = pci_bar_address(dev, i, r->type, r->size);
2846         upb = lob + r->size - 1;
2847         if (lob == PCI_BAR_UNMAPPED) {
2848             continue;
2849         }
2850 
2851         lob = MAX(lob, 0x1ULL << 32);
2852 
2853         if (upb >= lob) {
2854             range_set_bounds(&region_range, lob, upb);
2855             range_extend(range, &region_range);
2856         }
2857     }
2858 }
2859 
2860 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2861 {
2862     range_make_empty(range);
2863     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2864 }
2865 
2866 static bool pcie_has_upstream_port(PCIDevice *dev)
2867 {
2868     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2869 
2870     /* Device associated with an upstream port.
2871      * As there are several types of these, it's easier to check the
2872      * parent device: upstream ports are always connected to
2873      * root or downstream ports.
2874      */
2875     return parent_dev &&
2876         pci_is_express(parent_dev) &&
2877         parent_dev->exp.exp_cap &&
2878         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2879          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2880 }
2881 
2882 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2883 {
2884     PCIBus *bus = pci_get_bus(pci_dev);
2885 
2886     if(pcie_has_upstream_port(pci_dev)) {
2887         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2888         return bus->devices[0];
2889     } else {
2890         /* Other bus types might support multiple devices at slots 0-31 */
2891         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2892     }
2893 }
2894 
2895 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2896 {
2897     MSIMessage msg;
2898     if (msix_enabled(dev)) {
2899         msg = msix_get_message(dev, vector);
2900     } else if (msi_enabled(dev)) {
2901         msg = msi_get_message(dev, vector);
2902     } else {
2903         /* Should never happen */
2904         error_report("%s: unknown interrupt type", __func__);
2905         abort();
2906     }
2907     return msg;
2908 }
2909 
2910 void pci_set_power(PCIDevice *d, bool state)
2911 {
2912     if (d->has_power == state) {
2913         return;
2914     }
2915 
2916     d->has_power = state;
2917     pci_update_mappings(d);
2918     memory_region_set_enabled(&d->bus_master_enable_region,
2919                               (pci_get_word(d->config + PCI_COMMAND)
2920                                & PCI_COMMAND_MASTER) && d->has_power);
2921     if (!d->has_power) {
2922         pci_device_reset(d);
2923     }
2924 }
2925 
2926 static const TypeInfo pci_device_type_info = {
2927     .name = TYPE_PCI_DEVICE,
2928     .parent = TYPE_DEVICE,
2929     .instance_size = sizeof(PCIDevice),
2930     .abstract = true,
2931     .class_size = sizeof(PCIDeviceClass),
2932     .class_init = pci_device_class_init,
2933     .class_base_init = pci_device_class_base_init,
2934 };
2935 
2936 static void pci_register_types(void)
2937 {
2938     type_register_static(&pci_bus_info);
2939     type_register_static(&pcie_bus_info);
2940     type_register_static(&cxl_bus_info);
2941     type_register_static(&conventional_pci_interface_info);
2942     type_register_static(&cxl_interface_info);
2943     type_register_static(&pcie_interface_info);
2944     type_register_static(&pci_device_type_info);
2945 }
2946 
2947 type_init(pci_register_types)
2948