xref: /openbmc/qemu/hw/pci/pci.c (revision e69b2c67)
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 Property pci_props[] = {
71     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
72     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
73     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, UINT32_MAX),
74     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
75     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
76                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
77     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
78                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
79     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
80                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
81     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
82                        failover_pair_id),
83     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
84     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
85                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
86     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
87                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
88     DEFINE_PROP_SIZE32("x-max-bounce-buffer-size", PCIDevice,
89                      max_bounce_buffer_size, DEFAULT_MAX_BOUNCE_BUFFER_SIZE),
90     DEFINE_PROP_END_OF_LIST()
91 };
92 
93 static const VMStateDescription vmstate_pcibus = {
94     .name = "PCIBUS",
95     .version_id = 1,
96     .minimum_version_id = 1,
97     .fields = (const VMStateField[]) {
98         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
99         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
100                              nirq, 0, vmstate_info_int32,
101                              int32_t),
102         VMSTATE_END_OF_LIST()
103     }
104 };
105 
106 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
107 {
108     return a - b;
109 }
110 
111 static GSequence *pci_acpi_index_list(void)
112 {
113     static GSequence *used_acpi_index_list;
114 
115     if (!used_acpi_index_list) {
116         used_acpi_index_list = g_sequence_new(NULL);
117     }
118     return used_acpi_index_list;
119 }
120 
121 static void pci_init_bus_master(PCIDevice *pci_dev)
122 {
123     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
124 
125     memory_region_init_alias(&pci_dev->bus_master_enable_region,
126                              OBJECT(pci_dev), "bus master",
127                              dma_as->root, 0, memory_region_size(dma_as->root));
128     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
129     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
130                                 &pci_dev->bus_master_enable_region);
131 }
132 
133 static void pcibus_machine_done(Notifier *notifier, void *data)
134 {
135     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
136     int i;
137 
138     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
139         if (bus->devices[i]) {
140             pci_init_bus_master(bus->devices[i]);
141         }
142     }
143 }
144 
145 static void pci_bus_realize(BusState *qbus, Error **errp)
146 {
147     PCIBus *bus = PCI_BUS(qbus);
148 
149     bus->machine_done.notify = pcibus_machine_done;
150     qemu_add_machine_init_done_notifier(&bus->machine_done);
151 
152     vmstate_register_any(NULL, &vmstate_pcibus, bus);
153 }
154 
155 static void pcie_bus_realize(BusState *qbus, Error **errp)
156 {
157     PCIBus *bus = PCI_BUS(qbus);
158     Error *local_err = NULL;
159 
160     pci_bus_realize(qbus, &local_err);
161     if (local_err) {
162         error_propagate(errp, local_err);
163         return;
164     }
165 
166     /*
167      * A PCI-E bus can support extended config space if it's the root
168      * bus, or if the bus/bridge above it does as well
169      */
170     if (pci_bus_is_root(bus)) {
171         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
172     } else {
173         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
174 
175         if (pci_bus_allows_extended_config_space(parent_bus)) {
176             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
177         }
178     }
179 }
180 
181 static void pci_bus_unrealize(BusState *qbus)
182 {
183     PCIBus *bus = PCI_BUS(qbus);
184 
185     qemu_remove_machine_init_done_notifier(&bus->machine_done);
186 
187     vmstate_unregister(NULL, &vmstate_pcibus, bus);
188 }
189 
190 static int pcibus_num(PCIBus *bus)
191 {
192     if (pci_bus_is_root(bus)) {
193         return 0; /* pci host bridge */
194     }
195     return bus->parent_dev->config[PCI_SECONDARY_BUS];
196 }
197 
198 static uint16_t pcibus_numa_node(PCIBus *bus)
199 {
200     return NUMA_NODE_UNASSIGNED;
201 }
202 
203 static void pci_bus_class_init(ObjectClass *klass, void *data)
204 {
205     BusClass *k = BUS_CLASS(klass);
206     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
207     ResettableClass *rc = RESETTABLE_CLASS(klass);
208 
209     k->print_dev = pcibus_dev_print;
210     k->get_dev_path = pcibus_get_dev_path;
211     k->get_fw_dev_path = pcibus_get_fw_dev_path;
212     k->realize = pci_bus_realize;
213     k->unrealize = pci_bus_unrealize;
214 
215     rc->phases.hold = pcibus_reset_hold;
216 
217     pbc->bus_num = pcibus_num;
218     pbc->numa_node = pcibus_numa_node;
219 }
220 
221 static const TypeInfo pci_bus_info = {
222     .name = TYPE_PCI_BUS,
223     .parent = TYPE_BUS,
224     .instance_size = sizeof(PCIBus),
225     .class_size = sizeof(PCIBusClass),
226     .class_init = pci_bus_class_init,
227 };
228 
229 static const TypeInfo cxl_interface_info = {
230     .name          = INTERFACE_CXL_DEVICE,
231     .parent        = TYPE_INTERFACE,
232 };
233 
234 static const TypeInfo pcie_interface_info = {
235     .name          = INTERFACE_PCIE_DEVICE,
236     .parent        = TYPE_INTERFACE,
237 };
238 
239 static const TypeInfo conventional_pci_interface_info = {
240     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
241     .parent        = TYPE_INTERFACE,
242 };
243 
244 static void pcie_bus_class_init(ObjectClass *klass, void *data)
245 {
246     BusClass *k = BUS_CLASS(klass);
247 
248     k->realize = pcie_bus_realize;
249 }
250 
251 static const TypeInfo pcie_bus_info = {
252     .name = TYPE_PCIE_BUS,
253     .parent = TYPE_PCI_BUS,
254     .class_init = pcie_bus_class_init,
255 };
256 
257 static const TypeInfo cxl_bus_info = {
258     .name       = TYPE_CXL_BUS,
259     .parent     = TYPE_PCIE_BUS,
260     .class_init = pcie_bus_class_init,
261 };
262 
263 static void pci_update_mappings(PCIDevice *d);
264 static void pci_irq_handler(void *opaque, int irq_num, int level);
265 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
266 static void pci_del_option_rom(PCIDevice *pdev);
267 
268 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
269 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
270 
271 PCIHostStateList pci_host_bridges;
272 
273 int pci_bar(PCIDevice *d, int reg)
274 {
275     uint8_t type;
276 
277     /* PCIe virtual functions do not have their own BARs */
278     assert(!pci_is_vf(d));
279 
280     if (reg != PCI_ROM_SLOT)
281         return PCI_BASE_ADDRESS_0 + reg * 4;
282 
283     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
284     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
285 }
286 
287 static inline int pci_irq_state(PCIDevice *d, int irq_num)
288 {
289         return (d->irq_state >> irq_num) & 0x1;
290 }
291 
292 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
293 {
294         d->irq_state &= ~(0x1 << irq_num);
295         d->irq_state |= level << irq_num;
296 }
297 
298 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
299 {
300     assert(irq_num >= 0);
301     assert(irq_num < bus->nirq);
302     bus->irq_count[irq_num] += change;
303     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
304 }
305 
306 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
307 {
308     PCIBus *bus;
309     for (;;) {
310         int dev_irq = irq_num;
311         bus = pci_get_bus(pci_dev);
312         assert(bus->map_irq);
313         irq_num = bus->map_irq(pci_dev, irq_num);
314         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
315                             pci_bus_is_root(bus) ? "root-complex"
316                                     : DEVICE(bus->parent_dev)->canonical_path);
317         if (bus->set_irq)
318             break;
319         pci_dev = bus->parent_dev;
320     }
321     pci_bus_change_irq_level(bus, irq_num, change);
322 }
323 
324 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
325 {
326     assert(irq_num >= 0);
327     assert(irq_num < bus->nirq);
328     return !!bus->irq_count[irq_num];
329 }
330 
331 /* Update interrupt status bit in config space on interrupt
332  * state change. */
333 static void pci_update_irq_status(PCIDevice *dev)
334 {
335     if (dev->irq_state) {
336         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
337     } else {
338         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
339     }
340 }
341 
342 void pci_device_deassert_intx(PCIDevice *dev)
343 {
344     int i;
345     for (i = 0; i < PCI_NUM_PINS; ++i) {
346         pci_irq_handler(dev, i, 0);
347     }
348 }
349 
350 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
351 {
352     MemTxAttrs attrs = {};
353 
354     /*
355      * Xen uses the high bits of the address to contain some of the bits
356      * of the PIRQ#. Therefore we can't just send the write cycle and
357      * trust that it's caught by the APIC at 0xfee00000 because the
358      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
359      * So we intercept the delivery here instead of in kvm_send_msi().
360      */
361     if (xen_mode == XEN_EMULATE &&
362         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
363         return;
364     }
365     attrs.requester_id = pci_requester_id(dev);
366     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
367                          attrs, NULL);
368 }
369 
370 static void pci_reset_regions(PCIDevice *dev)
371 {
372     int r;
373     if (pci_is_vf(dev)) {
374         return;
375     }
376 
377     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
378         PCIIORegion *region = &dev->io_regions[r];
379         if (!region->size) {
380             continue;
381         }
382 
383         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
384             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
385             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
386         } else {
387             pci_set_long(dev->config + pci_bar(dev, r), region->type);
388         }
389     }
390 }
391 
392 static void pci_do_device_reset(PCIDevice *dev)
393 {
394     pci_device_deassert_intx(dev);
395     assert(dev->irq_state == 0);
396 
397     /* Clear all writable bits */
398     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
399                                  pci_get_word(dev->wmask + PCI_COMMAND) |
400                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
401     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
402                                  pci_get_word(dev->wmask + PCI_STATUS) |
403                                  pci_get_word(dev->w1cmask + PCI_STATUS));
404     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
405     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
406                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
407                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
408     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
409     pci_reset_regions(dev);
410     pci_update_mappings(dev);
411 
412     msi_reset(dev);
413     msix_reset(dev);
414     pcie_sriov_pf_reset(dev);
415 }
416 
417 /*
418  * This function is called on #RST and FLR.
419  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
420  */
421 void pci_device_reset(PCIDevice *dev)
422 {
423     device_cold_reset(&dev->qdev);
424     pci_do_device_reset(dev);
425 }
426 
427 /*
428  * Trigger pci bus reset under a given bus.
429  * Called via bus_cold_reset on RST# assert, after the devices
430  * have been reset device_cold_reset-ed already.
431  */
432 static void pcibus_reset_hold(Object *obj, ResetType type)
433 {
434     PCIBus *bus = PCI_BUS(obj);
435     int i;
436 
437     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
438         if (bus->devices[i]) {
439             pci_do_device_reset(bus->devices[i]);
440         }
441     }
442 
443     for (i = 0; i < bus->nirq; i++) {
444         assert(bus->irq_count[i] == 0);
445     }
446 }
447 
448 static void pci_host_bus_register(DeviceState *host)
449 {
450     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
451 
452     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
453 }
454 
455 static void pci_host_bus_unregister(DeviceState *host)
456 {
457     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
458 
459     QLIST_REMOVE(host_bridge, next);
460 }
461 
462 PCIBus *pci_device_root_bus(const PCIDevice *d)
463 {
464     PCIBus *bus = pci_get_bus(d);
465 
466     while (!pci_bus_is_root(bus)) {
467         d = bus->parent_dev;
468         assert(d != NULL);
469 
470         bus = pci_get_bus(d);
471     }
472 
473     return bus;
474 }
475 
476 const char *pci_root_bus_path(PCIDevice *dev)
477 {
478     PCIBus *rootbus = pci_device_root_bus(dev);
479     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
480     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
481 
482     assert(host_bridge->bus == rootbus);
483 
484     if (hc->root_bus_path) {
485         return (*hc->root_bus_path)(host_bridge, rootbus);
486     }
487 
488     return rootbus->qbus.name;
489 }
490 
491 bool pci_bus_bypass_iommu(PCIBus *bus)
492 {
493     PCIBus *rootbus = bus;
494     PCIHostState *host_bridge;
495 
496     if (!pci_bus_is_root(bus)) {
497         rootbus = pci_device_root_bus(bus->parent_dev);
498     }
499 
500     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
501 
502     assert(host_bridge->bus == rootbus);
503 
504     return host_bridge->bypass_iommu;
505 }
506 
507 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
508                                        MemoryRegion *mem, MemoryRegion *io,
509                                        uint8_t devfn_min)
510 {
511     assert(PCI_FUNC(devfn_min) == 0);
512     bus->devfn_min = devfn_min;
513     bus->slot_reserved_mask = 0x0;
514     bus->address_space_mem = mem;
515     bus->address_space_io = io;
516     bus->flags |= PCI_BUS_IS_ROOT;
517 
518     /* host bridge */
519     QLIST_INIT(&bus->child);
520 
521     pci_host_bus_register(parent);
522 }
523 
524 static void pci_bus_uninit(PCIBus *bus)
525 {
526     pci_host_bus_unregister(BUS(bus)->parent);
527 }
528 
529 bool pci_bus_is_express(const PCIBus *bus)
530 {
531     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
532 }
533 
534 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
535                        const char *name,
536                        MemoryRegion *mem, MemoryRegion *io,
537                        uint8_t devfn_min, const char *typename)
538 {
539     qbus_init(bus, bus_size, typename, parent, name);
540     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
541 }
542 
543 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
544                          MemoryRegion *mem, MemoryRegion *io,
545                          uint8_t devfn_min, const char *typename)
546 {
547     PCIBus *bus;
548 
549     bus = PCI_BUS(qbus_new(typename, parent, name));
550     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
551     return bus;
552 }
553 
554 void pci_root_bus_cleanup(PCIBus *bus)
555 {
556     pci_bus_uninit(bus);
557     /* the caller of the unplug hotplug handler will delete this device */
558     qbus_unrealize(BUS(bus));
559 }
560 
561 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
562                   void *irq_opaque, int nirq)
563 {
564     bus->set_irq = set_irq;
565     bus->irq_opaque = irq_opaque;
566     bus->nirq = nirq;
567     g_free(bus->irq_count);
568     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
569 }
570 
571 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
572 {
573     bus->map_irq = map_irq;
574 }
575 
576 void pci_bus_irqs_cleanup(PCIBus *bus)
577 {
578     bus->set_irq = NULL;
579     bus->map_irq = NULL;
580     bus->irq_opaque = NULL;
581     bus->nirq = 0;
582     g_free(bus->irq_count);
583     bus->irq_count = NULL;
584 }
585 
586 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
587                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
588                               void *irq_opaque,
589                               MemoryRegion *mem, MemoryRegion *io,
590                               uint8_t devfn_min, int nirq,
591                               const char *typename)
592 {
593     PCIBus *bus;
594 
595     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
596     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
597     pci_bus_map_irqs(bus, map_irq);
598     return bus;
599 }
600 
601 void pci_unregister_root_bus(PCIBus *bus)
602 {
603     pci_bus_irqs_cleanup(bus);
604     pci_root_bus_cleanup(bus);
605 }
606 
607 int pci_bus_num(PCIBus *s)
608 {
609     return PCI_BUS_GET_CLASS(s)->bus_num(s);
610 }
611 
612 /* Returns the min and max bus numbers of a PCI bus hierarchy */
613 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
614 {
615     int i;
616     *min_bus = *max_bus = pci_bus_num(bus);
617 
618     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
619         PCIDevice *dev = bus->devices[i];
620 
621         if (dev && IS_PCI_BRIDGE(dev)) {
622             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
623             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
624         }
625     }
626 }
627 
628 int pci_bus_numa_node(PCIBus *bus)
629 {
630     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
631 }
632 
633 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
634                                  const VMStateField *field)
635 {
636     PCIDevice *s = container_of(pv, PCIDevice, config);
637     uint8_t *config;
638     int i;
639 
640     assert(size == pci_config_size(s));
641     config = g_malloc(size);
642 
643     qemu_get_buffer(f, config, size);
644     for (i = 0; i < size; ++i) {
645         if ((config[i] ^ s->config[i]) &
646             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
647             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
648                          "cmask: %x wmask: %x w1cmask:%x", __func__,
649                          i, config[i], s->config[i],
650                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
651             g_free(config);
652             return -EINVAL;
653         }
654     }
655     memcpy(s->config, config, size);
656 
657     pci_update_mappings(s);
658     if (IS_PCI_BRIDGE(s)) {
659         pci_bridge_update_mappings(PCI_BRIDGE(s));
660     }
661 
662     memory_region_set_enabled(&s->bus_master_enable_region,
663                               pci_get_word(s->config + PCI_COMMAND)
664                               & PCI_COMMAND_MASTER);
665 
666     g_free(config);
667     return 0;
668 }
669 
670 /* just put buffer */
671 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
672                                  const VMStateField *field, JSONWriter *vmdesc)
673 {
674     const uint8_t **v = pv;
675     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
676     qemu_put_buffer(f, *v, size);
677 
678     return 0;
679 }
680 
681 static const VMStateInfo vmstate_info_pci_config = {
682     .name = "pci config",
683     .get  = get_pci_config_device,
684     .put  = put_pci_config_device,
685 };
686 
687 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
688                              const VMStateField *field)
689 {
690     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
691     uint32_t irq_state[PCI_NUM_PINS];
692     int i;
693     for (i = 0; i < PCI_NUM_PINS; ++i) {
694         irq_state[i] = qemu_get_be32(f);
695         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
696             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
697                     irq_state[i]);
698             return -EINVAL;
699         }
700     }
701 
702     for (i = 0; i < PCI_NUM_PINS; ++i) {
703         pci_set_irq_state(s, i, irq_state[i]);
704     }
705 
706     return 0;
707 }
708 
709 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
710                              const VMStateField *field, JSONWriter *vmdesc)
711 {
712     int i;
713     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
714 
715     for (i = 0; i < PCI_NUM_PINS; ++i) {
716         qemu_put_be32(f, pci_irq_state(s, i));
717     }
718 
719     return 0;
720 }
721 
722 static const VMStateInfo vmstate_info_pci_irq_state = {
723     .name = "pci irq state",
724     .get  = get_pci_irq_state,
725     .put  = put_pci_irq_state,
726 };
727 
728 static bool migrate_is_pcie(void *opaque, int version_id)
729 {
730     return pci_is_express((PCIDevice *)opaque);
731 }
732 
733 static bool migrate_is_not_pcie(void *opaque, int version_id)
734 {
735     return !pci_is_express((PCIDevice *)opaque);
736 }
737 
738 const VMStateDescription vmstate_pci_device = {
739     .name = "PCIDevice",
740     .version_id = 2,
741     .minimum_version_id = 1,
742     .fields = (const VMStateField[]) {
743         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
744         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
745                                    migrate_is_not_pcie,
746                                    0, vmstate_info_pci_config,
747                                    PCI_CONFIG_SPACE_SIZE),
748         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
749                                    migrate_is_pcie,
750                                    0, vmstate_info_pci_config,
751                                    PCIE_CONFIG_SPACE_SIZE),
752         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
753                                    vmstate_info_pci_irq_state,
754                                    PCI_NUM_PINS * sizeof(int32_t)),
755         VMSTATE_END_OF_LIST()
756     }
757 };
758 
759 
760 void pci_device_save(PCIDevice *s, QEMUFile *f)
761 {
762     /* Clear interrupt status bit: it is implicit
763      * in irq_state which we are saving.
764      * This makes us compatible with old devices
765      * which never set or clear this bit. */
766     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
767     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
768     /* Restore the interrupt status bit. */
769     pci_update_irq_status(s);
770 }
771 
772 int pci_device_load(PCIDevice *s, QEMUFile *f)
773 {
774     int ret;
775     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
776     /* Restore the interrupt status bit. */
777     pci_update_irq_status(s);
778     return ret;
779 }
780 
781 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
782 {
783     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
784                  pci_default_sub_vendor_id);
785     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
786                  pci_default_sub_device_id);
787 }
788 
789 /*
790  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
791  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
792  */
793 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
794                              unsigned int *slotp, unsigned int *funcp)
795 {
796     const char *p;
797     char *e;
798     unsigned long val;
799     unsigned long dom = 0, bus = 0;
800     unsigned int slot = 0;
801     unsigned int func = 0;
802 
803     p = addr;
804     val = strtoul(p, &e, 16);
805     if (e == p)
806         return -1;
807     if (*e == ':') {
808         bus = val;
809         p = e + 1;
810         val = strtoul(p, &e, 16);
811         if (e == p)
812             return -1;
813         if (*e == ':') {
814             dom = bus;
815             bus = val;
816             p = e + 1;
817             val = strtoul(p, &e, 16);
818             if (e == p)
819                 return -1;
820         }
821     }
822 
823     slot = val;
824 
825     if (funcp != NULL) {
826         if (*e != '.')
827             return -1;
828 
829         p = e + 1;
830         val = strtoul(p, &e, 16);
831         if (e == p)
832             return -1;
833 
834         func = val;
835     }
836 
837     /* if funcp == NULL func is 0 */
838     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
839         return -1;
840 
841     if (*e)
842         return -1;
843 
844     *domp = dom;
845     *busp = bus;
846     *slotp = slot;
847     if (funcp != NULL)
848         *funcp = func;
849     return 0;
850 }
851 
852 static void pci_init_cmask(PCIDevice *dev)
853 {
854     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
855     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
856     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
857     dev->cmask[PCI_REVISION_ID] = 0xff;
858     dev->cmask[PCI_CLASS_PROG] = 0xff;
859     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
860     dev->cmask[PCI_HEADER_TYPE] = 0xff;
861     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
862 }
863 
864 static void pci_init_wmask(PCIDevice *dev)
865 {
866     int config_size = pci_config_size(dev);
867 
868     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
869     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
870     pci_set_word(dev->wmask + PCI_COMMAND,
871                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
872                  PCI_COMMAND_INTX_DISABLE);
873     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
874 
875     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
876            config_size - PCI_CONFIG_HEADER_SIZE);
877 }
878 
879 static void pci_init_w1cmask(PCIDevice *dev)
880 {
881     /*
882      * Note: It's okay to set w1cmask even for readonly bits as
883      * long as their value is hardwired to 0.
884      */
885     pci_set_word(dev->w1cmask + PCI_STATUS,
886                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
887                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
888                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
889 }
890 
891 static void pci_init_mask_bridge(PCIDevice *d)
892 {
893     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
894        PCI_SEC_LATENCY_TIMER */
895     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
896 
897     /* base and limit */
898     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
899     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
900     pci_set_word(d->wmask + PCI_MEMORY_BASE,
901                  PCI_MEMORY_RANGE_MASK & 0xffff);
902     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
903                  PCI_MEMORY_RANGE_MASK & 0xffff);
904     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
905                  PCI_PREF_RANGE_MASK & 0xffff);
906     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
907                  PCI_PREF_RANGE_MASK & 0xffff);
908 
909     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
910     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
911 
912     /* Supported memory and i/o types */
913     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
914     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
915     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
916                                PCI_PREF_RANGE_TYPE_64);
917     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
918                                PCI_PREF_RANGE_TYPE_64);
919 
920     /*
921      * TODO: Bridges default to 10-bit VGA decoding but we currently only
922      * implement 16-bit decoding (no alias support).
923      */
924     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
925                  PCI_BRIDGE_CTL_PARITY |
926                  PCI_BRIDGE_CTL_SERR |
927                  PCI_BRIDGE_CTL_ISA |
928                  PCI_BRIDGE_CTL_VGA |
929                  PCI_BRIDGE_CTL_VGA_16BIT |
930                  PCI_BRIDGE_CTL_MASTER_ABORT |
931                  PCI_BRIDGE_CTL_BUS_RESET |
932                  PCI_BRIDGE_CTL_FAST_BACK |
933                  PCI_BRIDGE_CTL_DISCARD |
934                  PCI_BRIDGE_CTL_SEC_DISCARD |
935                  PCI_BRIDGE_CTL_DISCARD_SERR);
936     /* Below does not do anything as we never set this bit, put here for
937      * completeness. */
938     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
939                  PCI_BRIDGE_CTL_DISCARD_STATUS);
940     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
941     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
942     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
943                                PCI_PREF_RANGE_TYPE_MASK);
944     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
945                                PCI_PREF_RANGE_TYPE_MASK);
946 }
947 
948 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
949 {
950     uint8_t slot = PCI_SLOT(dev->devfn);
951     uint8_t func;
952 
953     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
954         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
955     }
956 
957     /*
958      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
959      * device, as it may just be a VF that ended up with function 0 in
960      * the legacy PCI interpretation. Avoid failing in such cases:
961      */
962     if (pci_is_vf(dev) &&
963         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
964         return;
965     }
966 
967     /*
968      * multifunction bit is interpreted in two ways as follows.
969      *   - all functions must set the bit to 1.
970      *     Example: Intel X53
971      *   - function 0 must set the bit, but the rest function (> 0)
972      *     is allowed to leave the bit to 0.
973      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
974      *
975      * So OS (at least Linux) checks the bit of only function 0,
976      * and doesn't see the bit of function > 0.
977      *
978      * The below check allows both interpretation.
979      */
980     if (PCI_FUNC(dev->devfn)) {
981         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
982         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
983             /* function 0 should set multifunction bit */
984             error_setg(errp, "PCI: single function device can't be populated "
985                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
986             return;
987         }
988         return;
989     }
990 
991     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
992         return;
993     }
994     /* function 0 indicates single function, so function > 0 must be NULL */
995     for (func = 1; func < PCI_FUNC_MAX; ++func) {
996         if (bus->devices[PCI_DEVFN(slot, func)]) {
997             error_setg(errp, "PCI: %x.0 indicates single function, "
998                        "but %x.%x is already populated.",
999                        slot, slot, func);
1000             return;
1001         }
1002     }
1003 }
1004 
1005 static void pci_config_alloc(PCIDevice *pci_dev)
1006 {
1007     int config_size = pci_config_size(pci_dev);
1008 
1009     pci_dev->config = g_malloc0(config_size);
1010     pci_dev->cmask = g_malloc0(config_size);
1011     pci_dev->wmask = g_malloc0(config_size);
1012     pci_dev->w1cmask = g_malloc0(config_size);
1013     pci_dev->used = g_malloc0(config_size);
1014 }
1015 
1016 static void pci_config_free(PCIDevice *pci_dev)
1017 {
1018     g_free(pci_dev->config);
1019     g_free(pci_dev->cmask);
1020     g_free(pci_dev->wmask);
1021     g_free(pci_dev->w1cmask);
1022     g_free(pci_dev->used);
1023 }
1024 
1025 static void do_pci_unregister_device(PCIDevice *pci_dev)
1026 {
1027     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1028     pci_config_free(pci_dev);
1029 
1030     if (xen_mode == XEN_EMULATE) {
1031         xen_evtchn_remove_pci_device(pci_dev);
1032     }
1033     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1034         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1035                                     &pci_dev->bus_master_enable_region);
1036     }
1037     address_space_destroy(&pci_dev->bus_master_as);
1038 }
1039 
1040 /* Extract PCIReqIDCache into BDF format */
1041 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1042 {
1043     uint8_t bus_n;
1044     uint16_t result;
1045 
1046     switch (cache->type) {
1047     case PCI_REQ_ID_BDF:
1048         result = pci_get_bdf(cache->dev);
1049         break;
1050     case PCI_REQ_ID_SECONDARY_BUS:
1051         bus_n = pci_dev_bus_num(cache->dev);
1052         result = PCI_BUILD_BDF(bus_n, 0);
1053         break;
1054     default:
1055         error_report("Invalid PCI requester ID cache type: %d",
1056                      cache->type);
1057         exit(1);
1058         break;
1059     }
1060 
1061     return result;
1062 }
1063 
1064 /* Parse bridges up to the root complex and return requester ID
1065  * cache for specific device.  For full PCIe topology, the cache
1066  * result would be exactly the same as getting BDF of the device.
1067  * However, several tricks are required when system mixed up with
1068  * legacy PCI devices and PCIe-to-PCI bridges.
1069  *
1070  * Here we cache the proxy device (and type) not requester ID since
1071  * bus number might change from time to time.
1072  */
1073 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1074 {
1075     PCIDevice *parent;
1076     PCIReqIDCache cache = {
1077         .dev = dev,
1078         .type = PCI_REQ_ID_BDF,
1079     };
1080 
1081     while (!pci_bus_is_root(pci_get_bus(dev))) {
1082         /* We are under PCI/PCIe bridges */
1083         parent = pci_get_bus(dev)->parent_dev;
1084         if (pci_is_express(parent)) {
1085             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1086                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1087                  * override the requester ID using secondary bus
1088                  * number of parent bridge with zeroed devfn
1089                  * (pcie-to-pci bridge spec chap 2.3). */
1090                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1091                 cache.dev = dev;
1092             }
1093         } else {
1094             /* Legacy PCI, override requester ID with the bridge's
1095              * BDF upstream.  When the root complex connects to
1096              * legacy PCI devices (including buses), it can only
1097              * obtain requester ID info from directly attached
1098              * devices.  If devices are attached under bridges, only
1099              * the requester ID of the bridge that is directly
1100              * attached to the root complex can be recognized. */
1101             cache.type = PCI_REQ_ID_BDF;
1102             cache.dev = parent;
1103         }
1104         dev = parent;
1105     }
1106 
1107     return cache;
1108 }
1109 
1110 uint16_t pci_requester_id(PCIDevice *dev)
1111 {
1112     return pci_req_id_cache_extract(&dev->requester_id_cache);
1113 }
1114 
1115 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1116 {
1117     return !(bus->devices[devfn]);
1118 }
1119 
1120 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1121 {
1122     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1123 }
1124 
1125 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1126 {
1127     return bus->slot_reserved_mask;
1128 }
1129 
1130 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1131 {
1132     bus->slot_reserved_mask |= mask;
1133 }
1134 
1135 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1136 {
1137     bus->slot_reserved_mask &= ~mask;
1138 }
1139 
1140 /* -1 for devfn means auto assign */
1141 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1142                                          const char *name, int devfn,
1143                                          Error **errp)
1144 {
1145     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1146     PCIConfigReadFunc *config_read = pc->config_read;
1147     PCIConfigWriteFunc *config_write = pc->config_write;
1148     Error *local_err = NULL;
1149     DeviceState *dev = DEVICE(pci_dev);
1150     PCIBus *bus = pci_get_bus(pci_dev);
1151     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1152 
1153     /* Only pci bridges can be attached to extra PCI root buses */
1154     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1155         error_setg(errp,
1156                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1157                     bus->parent_dev->name);
1158         return NULL;
1159     }
1160 
1161     if (devfn < 0) {
1162         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1163             devfn += PCI_FUNC_MAX) {
1164             if (pci_bus_devfn_available(bus, devfn) &&
1165                    !pci_bus_devfn_reserved(bus, devfn)) {
1166                 goto found;
1167             }
1168         }
1169         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1170                    "or reserved", name);
1171         return NULL;
1172     found: ;
1173     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1174         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1175                    " reserved",
1176                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1177         return NULL;
1178     } else if (!pci_bus_devfn_available(bus, devfn)) {
1179         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1180                    " in use by %s,id=%s",
1181                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1182                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1183         return NULL;
1184     } /*
1185        * Populating function 0 triggers a scan from the guest that
1186        * exposes other non-zero functions. Hence we need to ensure that
1187        * function 0 wasn't added yet.
1188        */
1189     else if (dev->hotplugged &&
1190              !pci_is_vf(pci_dev) &&
1191              pci_get_function_0(pci_dev)) {
1192         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1193                    " new func %s cannot be exposed to guest.",
1194                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1195                    pci_get_function_0(pci_dev)->name,
1196                    name);
1197 
1198        return NULL;
1199     }
1200 
1201     pci_dev->devfn = devfn;
1202     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1203     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1204 
1205     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1206                        "bus master container", UINT64_MAX);
1207     address_space_init(&pci_dev->bus_master_as,
1208                        &pci_dev->bus_master_container_region, pci_dev->name);
1209     pci_dev->bus_master_as.max_bounce_buffer_size =
1210         pci_dev->max_bounce_buffer_size;
1211 
1212     if (phase_check(PHASE_MACHINE_READY)) {
1213         pci_init_bus_master(pci_dev);
1214     }
1215     pci_dev->irq_state = 0;
1216     pci_config_alloc(pci_dev);
1217 
1218     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1219     pci_config_set_device_id(pci_dev->config, pc->device_id);
1220     pci_config_set_revision(pci_dev->config, pc->revision);
1221     pci_config_set_class(pci_dev->config, pc->class_id);
1222 
1223     if (!is_bridge) {
1224         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1225             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1226                          pc->subsystem_vendor_id);
1227             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1228                          pc->subsystem_id);
1229         } else {
1230             pci_set_default_subsystem_id(pci_dev);
1231         }
1232     } else {
1233         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1234         assert(!pc->subsystem_vendor_id);
1235         assert(!pc->subsystem_id);
1236     }
1237     pci_init_cmask(pci_dev);
1238     pci_init_wmask(pci_dev);
1239     pci_init_w1cmask(pci_dev);
1240     if (is_bridge) {
1241         pci_init_mask_bridge(pci_dev);
1242     }
1243     pci_init_multifunction(bus, pci_dev, &local_err);
1244     if (local_err) {
1245         error_propagate(errp, local_err);
1246         do_pci_unregister_device(pci_dev);
1247         return NULL;
1248     }
1249 
1250     if (!config_read)
1251         config_read = pci_default_read_config;
1252     if (!config_write)
1253         config_write = pci_default_write_config;
1254     pci_dev->config_read = config_read;
1255     pci_dev->config_write = config_write;
1256     bus->devices[devfn] = pci_dev;
1257     pci_dev->version_id = 2; /* Current pci device vmstate version */
1258     return pci_dev;
1259 }
1260 
1261 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1262 {
1263     PCIIORegion *r;
1264     int i;
1265 
1266     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1267         r = &pci_dev->io_regions[i];
1268         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1269             continue;
1270         memory_region_del_subregion(r->address_space, r->memory);
1271     }
1272 
1273     pci_unregister_vga(pci_dev);
1274 }
1275 
1276 static void pci_qdev_unrealize(DeviceState *dev)
1277 {
1278     PCIDevice *pci_dev = PCI_DEVICE(dev);
1279     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1280 
1281     pci_unregister_io_regions(pci_dev);
1282     pci_del_option_rom(pci_dev);
1283 
1284     if (pc->exit) {
1285         pc->exit(pci_dev);
1286     }
1287 
1288     pci_device_deassert_intx(pci_dev);
1289     do_pci_unregister_device(pci_dev);
1290 
1291     pci_dev->msi_trigger = NULL;
1292 
1293     /*
1294      * clean up acpi-index so it could reused by another device
1295      */
1296     if (pci_dev->acpi_index) {
1297         GSequence *used_indexes = pci_acpi_index_list();
1298 
1299         g_sequence_remove(g_sequence_lookup(used_indexes,
1300                           GINT_TO_POINTER(pci_dev->acpi_index),
1301                           g_cmp_uint32, NULL));
1302     }
1303 }
1304 
1305 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1306                       uint8_t type, MemoryRegion *memory)
1307 {
1308     PCIIORegion *r;
1309     uint32_t addr; /* offset in pci config space */
1310     uint64_t wmask;
1311     pcibus_t size = memory_region_size(memory);
1312     uint8_t hdr_type;
1313 
1314     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1315     assert(region_num >= 0);
1316     assert(region_num < PCI_NUM_REGIONS);
1317     assert(is_power_of_2(size));
1318 
1319     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1320     hdr_type =
1321         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1322     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1323 
1324     r = &pci_dev->io_regions[region_num];
1325     r->addr = PCI_BAR_UNMAPPED;
1326     r->size = size;
1327     r->type = type;
1328     r->memory = memory;
1329     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1330                         ? pci_get_bus(pci_dev)->address_space_io
1331                         : pci_get_bus(pci_dev)->address_space_mem;
1332 
1333     wmask = ~(size - 1);
1334     if (region_num == PCI_ROM_SLOT) {
1335         /* ROM enable bit is writable */
1336         wmask |= PCI_ROM_ADDRESS_ENABLE;
1337     }
1338 
1339     addr = pci_bar(pci_dev, region_num);
1340     pci_set_long(pci_dev->config + addr, type);
1341 
1342     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1343         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1344         pci_set_quad(pci_dev->wmask + addr, wmask);
1345         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1346     } else {
1347         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1348         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1349     }
1350 }
1351 
1352 static void pci_update_vga(PCIDevice *pci_dev)
1353 {
1354     uint16_t cmd;
1355 
1356     if (!pci_dev->has_vga) {
1357         return;
1358     }
1359 
1360     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1361 
1362     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1363                               cmd & PCI_COMMAND_MEMORY);
1364     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1365                               cmd & PCI_COMMAND_IO);
1366     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1367                               cmd & PCI_COMMAND_IO);
1368 }
1369 
1370 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1371                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1372 {
1373     PCIBus *bus = pci_get_bus(pci_dev);
1374 
1375     assert(!pci_dev->has_vga);
1376 
1377     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1378     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1379     memory_region_add_subregion_overlap(bus->address_space_mem,
1380                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1381 
1382     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1383     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1384     memory_region_add_subregion_overlap(bus->address_space_io,
1385                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1386 
1387     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1388     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1389     memory_region_add_subregion_overlap(bus->address_space_io,
1390                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1391     pci_dev->has_vga = true;
1392 
1393     pci_update_vga(pci_dev);
1394 }
1395 
1396 void pci_unregister_vga(PCIDevice *pci_dev)
1397 {
1398     PCIBus *bus = pci_get_bus(pci_dev);
1399 
1400     if (!pci_dev->has_vga) {
1401         return;
1402     }
1403 
1404     memory_region_del_subregion(bus->address_space_mem,
1405                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1406     memory_region_del_subregion(bus->address_space_io,
1407                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1408     memory_region_del_subregion(bus->address_space_io,
1409                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1410     pci_dev->has_vga = false;
1411 }
1412 
1413 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1414 {
1415     return pci_dev->io_regions[region_num].addr;
1416 }
1417 
1418 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1419                                         uint8_t type, pcibus_t size)
1420 {
1421     pcibus_t new_addr;
1422     if (!pci_is_vf(d)) {
1423         int bar = pci_bar(d, reg);
1424         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1425             new_addr = pci_get_quad(d->config + bar);
1426         } else {
1427             new_addr = pci_get_long(d->config + bar);
1428         }
1429     } else {
1430         PCIDevice *pf = d->exp.sriov_vf.pf;
1431         uint16_t sriov_cap = pf->exp.sriov_cap;
1432         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1433         uint16_t vf_offset =
1434             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1435         uint16_t vf_stride =
1436             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1437         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1438 
1439         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1440             new_addr = pci_get_quad(pf->config + bar);
1441         } else {
1442             new_addr = pci_get_long(pf->config + bar);
1443         }
1444         new_addr += vf_num * size;
1445     }
1446     /* The ROM slot has a specific enable bit, keep it intact */
1447     if (reg != PCI_ROM_SLOT) {
1448         new_addr &= ~(size - 1);
1449     }
1450     return new_addr;
1451 }
1452 
1453 pcibus_t pci_bar_address(PCIDevice *d,
1454                          int reg, uint8_t type, pcibus_t size)
1455 {
1456     pcibus_t new_addr, last_addr;
1457     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1458     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1459     bool allow_0_address = mc->pci_allow_0_address;
1460 
1461     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1462         if (!(cmd & PCI_COMMAND_IO)) {
1463             return PCI_BAR_UNMAPPED;
1464         }
1465         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1466         last_addr = new_addr + size - 1;
1467         /* Check if 32 bit BAR wraps around explicitly.
1468          * TODO: make priorities correct and remove this work around.
1469          */
1470         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1471             (!allow_0_address && new_addr == 0)) {
1472             return PCI_BAR_UNMAPPED;
1473         }
1474         return new_addr;
1475     }
1476 
1477     if (!(cmd & PCI_COMMAND_MEMORY)) {
1478         return PCI_BAR_UNMAPPED;
1479     }
1480     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1481     /* the ROM slot has a specific enable bit */
1482     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1483         return PCI_BAR_UNMAPPED;
1484     }
1485     new_addr &= ~(size - 1);
1486     last_addr = new_addr + size - 1;
1487     /* NOTE: we do not support wrapping */
1488     /* XXX: as we cannot support really dynamic
1489        mappings, we handle specific values as invalid
1490        mappings. */
1491     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1492         (!allow_0_address && new_addr == 0)) {
1493         return PCI_BAR_UNMAPPED;
1494     }
1495 
1496     /* Now pcibus_t is 64bit.
1497      * Check if 32 bit BAR wraps around explicitly.
1498      * Without this, PC ide doesn't work well.
1499      * TODO: remove this work around.
1500      */
1501     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1502         return PCI_BAR_UNMAPPED;
1503     }
1504 
1505     /*
1506      * OS is allowed to set BAR beyond its addressable
1507      * bits. For example, 32 bit OS can set 64bit bar
1508      * to >4G. Check it. TODO: we might need to support
1509      * it in the future for e.g. PAE.
1510      */
1511     if (last_addr >= HWADDR_MAX) {
1512         return PCI_BAR_UNMAPPED;
1513     }
1514 
1515     return new_addr;
1516 }
1517 
1518 static void pci_update_mappings(PCIDevice *d)
1519 {
1520     PCIIORegion *r;
1521     int i;
1522     pcibus_t new_addr;
1523 
1524     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1525         r = &d->io_regions[i];
1526 
1527         /* this region isn't registered */
1528         if (!r->size)
1529             continue;
1530 
1531         new_addr = pci_bar_address(d, i, r->type, r->size);
1532         if (!d->has_power) {
1533             new_addr = PCI_BAR_UNMAPPED;
1534         }
1535 
1536         /* This bar isn't changed */
1537         if (new_addr == r->addr)
1538             continue;
1539 
1540         /* now do the real mapping */
1541         if (r->addr != PCI_BAR_UNMAPPED) {
1542             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1543                                           PCI_SLOT(d->devfn),
1544                                           PCI_FUNC(d->devfn),
1545                                           i, r->addr, r->size);
1546             memory_region_del_subregion(r->address_space, r->memory);
1547         }
1548         r->addr = new_addr;
1549         if (r->addr != PCI_BAR_UNMAPPED) {
1550             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1551                                           PCI_SLOT(d->devfn),
1552                                           PCI_FUNC(d->devfn),
1553                                           i, r->addr, r->size);
1554             memory_region_add_subregion_overlap(r->address_space,
1555                                                 r->addr, r->memory, 1);
1556         }
1557     }
1558 
1559     pci_update_vga(d);
1560 }
1561 
1562 static inline int pci_irq_disabled(PCIDevice *d)
1563 {
1564     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1565 }
1566 
1567 /* Called after interrupt disabled field update in config space,
1568  * assert/deassert interrupts if necessary.
1569  * Gets original interrupt disable bit value (before update). */
1570 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1571 {
1572     int i, disabled = pci_irq_disabled(d);
1573     if (disabled == was_irq_disabled)
1574         return;
1575     for (i = 0; i < PCI_NUM_PINS; ++i) {
1576         int state = pci_irq_state(d, i);
1577         pci_change_irq_level(d, i, disabled ? -state : state);
1578     }
1579 }
1580 
1581 uint32_t pci_default_read_config(PCIDevice *d,
1582                                  uint32_t address, int len)
1583 {
1584     uint32_t val = 0;
1585 
1586     assert(address + len <= pci_config_size(d));
1587 
1588     if (pci_is_express_downstream_port(d) &&
1589         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1590         pcie_sync_bridge_lnk(d);
1591     }
1592     memcpy(&val, d->config + address, len);
1593     return le32_to_cpu(val);
1594 }
1595 
1596 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1597 {
1598     int i, was_irq_disabled = pci_irq_disabled(d);
1599     uint32_t val = val_in;
1600 
1601     assert(addr + l <= pci_config_size(d));
1602 
1603     for (i = 0; i < l; val >>= 8, ++i) {
1604         uint8_t wmask = d->wmask[addr + i];
1605         uint8_t w1cmask = d->w1cmask[addr + i];
1606         assert(!(wmask & w1cmask));
1607         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1608         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1609     }
1610     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1611         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1612         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1613         range_covers_byte(addr, l, PCI_COMMAND))
1614         pci_update_mappings(d);
1615 
1616     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1617         pci_update_irq_disabled(d, was_irq_disabled);
1618         memory_region_set_enabled(&d->bus_master_enable_region,
1619                                   (pci_get_word(d->config + PCI_COMMAND)
1620                                    & PCI_COMMAND_MASTER) && d->has_power);
1621     }
1622 
1623     msi_write_config(d, addr, val_in, l);
1624     msix_write_config(d, addr, val_in, l);
1625     pcie_sriov_config_write(d, addr, val_in, l);
1626 }
1627 
1628 /***********************************************************/
1629 /* generic PCI irq support */
1630 
1631 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1632 static void pci_irq_handler(void *opaque, int irq_num, int level)
1633 {
1634     PCIDevice *pci_dev = opaque;
1635     int change;
1636 
1637     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1638     assert(level == 0 || level == 1);
1639     change = level - pci_irq_state(pci_dev, irq_num);
1640     if (!change)
1641         return;
1642 
1643     pci_set_irq_state(pci_dev, irq_num, level);
1644     pci_update_irq_status(pci_dev);
1645     if (pci_irq_disabled(pci_dev))
1646         return;
1647     pci_change_irq_level(pci_dev, irq_num, change);
1648 }
1649 
1650 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1651 {
1652     int intx = pci_intx(pci_dev);
1653     assert(0 <= intx && intx < PCI_NUM_PINS);
1654 
1655     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1656 }
1657 
1658 void pci_set_irq(PCIDevice *pci_dev, int level)
1659 {
1660     int intx = pci_intx(pci_dev);
1661     pci_irq_handler(pci_dev, intx, level);
1662 }
1663 
1664 /* Special hooks used by device assignment */
1665 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1666 {
1667     assert(pci_bus_is_root(bus));
1668     bus->route_intx_to_irq = route_intx_to_irq;
1669 }
1670 
1671 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1672 {
1673     PCIBus *bus;
1674 
1675     do {
1676         int dev_irq = pin;
1677         bus = pci_get_bus(dev);
1678         pin = bus->map_irq(dev, pin);
1679         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1680                             pci_bus_is_root(bus) ? "root-complex"
1681                                     : DEVICE(bus->parent_dev)->canonical_path);
1682         dev = bus->parent_dev;
1683     } while (dev);
1684 
1685     if (!bus->route_intx_to_irq) {
1686         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1687                      object_get_typename(OBJECT(bus->qbus.parent)));
1688         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1689     }
1690 
1691     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1692 }
1693 
1694 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1695 {
1696     return old->mode != new->mode || old->irq != new->irq;
1697 }
1698 
1699 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1700 {
1701     PCIDevice *dev;
1702     PCIBus *sec;
1703     int i;
1704 
1705     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1706         dev = bus->devices[i];
1707         if (dev && dev->intx_routing_notifier) {
1708             dev->intx_routing_notifier(dev);
1709         }
1710     }
1711 
1712     QLIST_FOREACH(sec, &bus->child, sibling) {
1713         pci_bus_fire_intx_routing_notifier(sec);
1714     }
1715 }
1716 
1717 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1718                                           PCIINTxRoutingNotifier notifier)
1719 {
1720     dev->intx_routing_notifier = notifier;
1721 }
1722 
1723 /*
1724  * PCI-to-PCI bridge specification
1725  * 9.1: Interrupt routing. Table 9-1
1726  *
1727  * the PCI Express Base Specification, Revision 2.1
1728  * 2.2.8.1: INTx interrupt signaling - Rules
1729  *          the Implementation Note
1730  *          Table 2-20
1731  */
1732 /*
1733  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1734  * 0-origin unlike PCI interrupt pin register.
1735  */
1736 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1737 {
1738     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1739 }
1740 
1741 /***********************************************************/
1742 /* monitor info on PCI */
1743 
1744 static const pci_class_desc pci_class_descriptions[] =
1745 {
1746     { 0x0001, "VGA controller", "display"},
1747     { 0x0100, "SCSI controller", "scsi"},
1748     { 0x0101, "IDE controller", "ide"},
1749     { 0x0102, "Floppy controller", "fdc"},
1750     { 0x0103, "IPI controller", "ipi"},
1751     { 0x0104, "RAID controller", "raid"},
1752     { 0x0106, "SATA controller"},
1753     { 0x0107, "SAS controller"},
1754     { 0x0180, "Storage controller"},
1755     { 0x0200, "Ethernet controller", "ethernet"},
1756     { 0x0201, "Token Ring controller", "token-ring"},
1757     { 0x0202, "FDDI controller", "fddi"},
1758     { 0x0203, "ATM controller", "atm"},
1759     { 0x0280, "Network controller"},
1760     { 0x0300, "VGA controller", "display", 0x00ff},
1761     { 0x0301, "XGA controller"},
1762     { 0x0302, "3D controller"},
1763     { 0x0380, "Display controller"},
1764     { 0x0400, "Video controller", "video"},
1765     { 0x0401, "Audio controller", "sound"},
1766     { 0x0402, "Phone"},
1767     { 0x0403, "Audio controller", "sound"},
1768     { 0x0480, "Multimedia controller"},
1769     { 0x0500, "RAM controller", "memory"},
1770     { 0x0501, "Flash controller", "flash"},
1771     { 0x0580, "Memory controller"},
1772     { 0x0600, "Host bridge", "host"},
1773     { 0x0601, "ISA bridge", "isa"},
1774     { 0x0602, "EISA bridge", "eisa"},
1775     { 0x0603, "MC bridge", "mca"},
1776     { 0x0604, "PCI bridge", "pci-bridge"},
1777     { 0x0605, "PCMCIA bridge", "pcmcia"},
1778     { 0x0606, "NUBUS bridge", "nubus"},
1779     { 0x0607, "CARDBUS bridge", "cardbus"},
1780     { 0x0608, "RACEWAY bridge"},
1781     { 0x0680, "Bridge"},
1782     { 0x0700, "Serial port", "serial"},
1783     { 0x0701, "Parallel port", "parallel"},
1784     { 0x0800, "Interrupt controller", "interrupt-controller"},
1785     { 0x0801, "DMA controller", "dma-controller"},
1786     { 0x0802, "Timer", "timer"},
1787     { 0x0803, "RTC", "rtc"},
1788     { 0x0900, "Keyboard", "keyboard"},
1789     { 0x0901, "Pen", "pen"},
1790     { 0x0902, "Mouse", "mouse"},
1791     { 0x0A00, "Dock station", "dock", 0x00ff},
1792     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1793     { 0x0c00, "Firewire controller", "firewire"},
1794     { 0x0c01, "Access bus controller", "access-bus"},
1795     { 0x0c02, "SSA controller", "ssa"},
1796     { 0x0c03, "USB controller", "usb"},
1797     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1798     { 0x0c05, "SMBus"},
1799     { 0, NULL}
1800 };
1801 
1802 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1803                                            pci_bus_dev_fn fn,
1804                                            void *opaque)
1805 {
1806     PCIDevice *d;
1807     int devfn;
1808 
1809     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1810         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1811         if (d) {
1812             fn(bus, d, opaque);
1813         }
1814     }
1815 }
1816 
1817 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1818                                  pci_bus_dev_fn fn, void *opaque)
1819 {
1820     bus = pci_find_bus_nr(bus, bus_num);
1821 
1822     if (bus) {
1823         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1824     }
1825 }
1826 
1827 void pci_for_each_device_under_bus(PCIBus *bus,
1828                                    pci_bus_dev_fn fn, void *opaque)
1829 {
1830     PCIDevice *d;
1831     int devfn;
1832 
1833     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1834         d = bus->devices[devfn];
1835         if (d) {
1836             fn(bus, d, opaque);
1837         }
1838     }
1839 }
1840 
1841 void pci_for_each_device(PCIBus *bus, int bus_num,
1842                          pci_bus_dev_fn fn, void *opaque)
1843 {
1844     bus = pci_find_bus_nr(bus, bus_num);
1845 
1846     if (bus) {
1847         pci_for_each_device_under_bus(bus, fn, opaque);
1848     }
1849 }
1850 
1851 const pci_class_desc *get_class_desc(int class)
1852 {
1853     const pci_class_desc *desc;
1854 
1855     desc = pci_class_descriptions;
1856     while (desc->desc && class != desc->class) {
1857         desc++;
1858     }
1859 
1860     return desc;
1861 }
1862 
1863 void pci_init_nic_devices(PCIBus *bus, const char *default_model)
1864 {
1865     qemu_create_nic_bus_devices(&bus->qbus, TYPE_PCI_DEVICE, default_model,
1866                                 "virtio", "virtio-net-pci");
1867 }
1868 
1869 bool pci_init_nic_in_slot(PCIBus *rootbus, const char *model,
1870                           const char *alias, const char *devaddr)
1871 {
1872     NICInfo *nd = qemu_find_nic_info(model, true, alias);
1873     int dom, busnr, devfn;
1874     PCIDevice *pci_dev;
1875     unsigned slot;
1876     PCIBus *bus;
1877 
1878     if (!nd) {
1879         return false;
1880     }
1881 
1882     if (!devaddr || pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1883         error_report("Invalid PCI device address %s for device %s",
1884                      devaddr, model);
1885         exit(1);
1886     }
1887 
1888     if (dom != 0) {
1889         error_report("No support for non-zero PCI domains");
1890         exit(1);
1891     }
1892 
1893     devfn = PCI_DEVFN(slot, 0);
1894 
1895     bus = pci_find_bus_nr(rootbus, busnr);
1896     if (!bus) {
1897         error_report("Invalid PCI device address %s for device %s",
1898                      devaddr, model);
1899         exit(1);
1900     }
1901 
1902     pci_dev = pci_new(devfn, model);
1903     qdev_set_nic_properties(&pci_dev->qdev, nd);
1904     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1905     return true;
1906 }
1907 
1908 PCIDevice *pci_vga_init(PCIBus *bus)
1909 {
1910     vga_interface_created = true;
1911     switch (vga_interface_type) {
1912     case VGA_CIRRUS:
1913         return pci_create_simple(bus, -1, "cirrus-vga");
1914     case VGA_QXL:
1915         return pci_create_simple(bus, -1, "qxl-vga");
1916     case VGA_STD:
1917         return pci_create_simple(bus, -1, "VGA");
1918     case VGA_VMWARE:
1919         return pci_create_simple(bus, -1, "vmware-svga");
1920     case VGA_VIRTIO:
1921         return pci_create_simple(bus, -1, "virtio-vga");
1922     case VGA_NONE:
1923     default: /* Other non-PCI types. Checking for unsupported types is already
1924                 done in vl.c. */
1925         return NULL;
1926     }
1927 }
1928 
1929 /* Whether a given bus number is in range of the secondary
1930  * bus of the given bridge device. */
1931 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1932 {
1933     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1934              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1935         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1936         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1937 }
1938 
1939 /* Whether a given bus number is in a range of a root bus */
1940 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1941 {
1942     int i;
1943 
1944     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1945         PCIDevice *dev = bus->devices[i];
1946 
1947         if (dev && IS_PCI_BRIDGE(dev)) {
1948             if (pci_secondary_bus_in_range(dev, bus_num)) {
1949                 return true;
1950             }
1951         }
1952     }
1953 
1954     return false;
1955 }
1956 
1957 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1958 {
1959     PCIBus *sec;
1960 
1961     if (!bus) {
1962         return NULL;
1963     }
1964 
1965     if (pci_bus_num(bus) == bus_num) {
1966         return bus;
1967     }
1968 
1969     /* Consider all bus numbers in range for the host pci bridge. */
1970     if (!pci_bus_is_root(bus) &&
1971         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1972         return NULL;
1973     }
1974 
1975     /* try child bus */
1976     for (; bus; bus = sec) {
1977         QLIST_FOREACH(sec, &bus->child, sibling) {
1978             if (pci_bus_num(sec) == bus_num) {
1979                 return sec;
1980             }
1981             /* PXB buses assumed to be children of bus 0 */
1982             if (pci_bus_is_root(sec)) {
1983                 if (pci_root_bus_in_range(sec, bus_num)) {
1984                     break;
1985                 }
1986             } else {
1987                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1988                     break;
1989                 }
1990             }
1991         }
1992     }
1993 
1994     return NULL;
1995 }
1996 
1997 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
1998                                   pci_bus_fn end, void *parent_state)
1999 {
2000     PCIBus *sec;
2001     void *state;
2002 
2003     if (!bus) {
2004         return;
2005     }
2006 
2007     if (begin) {
2008         state = begin(bus, parent_state);
2009     } else {
2010         state = parent_state;
2011     }
2012 
2013     QLIST_FOREACH(sec, &bus->child, sibling) {
2014         pci_for_each_bus_depth_first(sec, begin, end, state);
2015     }
2016 
2017     if (end) {
2018         end(bus, state);
2019     }
2020 }
2021 
2022 
2023 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2024 {
2025     bus = pci_find_bus_nr(bus, bus_num);
2026 
2027     if (!bus)
2028         return NULL;
2029 
2030     return bus->devices[devfn];
2031 }
2032 
2033 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2034 
2035 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2036 {
2037     PCIDevice *pci_dev = (PCIDevice *)qdev;
2038     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2039     ObjectClass *klass = OBJECT_CLASS(pc);
2040     Error *local_err = NULL;
2041     bool is_default_rom;
2042     uint16_t class_id;
2043 
2044     /*
2045      * capped by systemd (see: udev-builtin-net_id.c)
2046      * as it's the only known user honor it to avoid users
2047      * misconfigure QEMU and then wonder why acpi-index doesn't work
2048      */
2049     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2050         error_setg(errp, "acpi-index should be less or equal to %u",
2051                    ONBOARD_INDEX_MAX);
2052         return;
2053     }
2054 
2055     /*
2056      * make sure that acpi-index is unique across all present PCI devices
2057      */
2058     if (pci_dev->acpi_index) {
2059         GSequence *used_indexes = pci_acpi_index_list();
2060 
2061         if (g_sequence_lookup(used_indexes,
2062                               GINT_TO_POINTER(pci_dev->acpi_index),
2063                               g_cmp_uint32, NULL)) {
2064             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2065                        " already exist", pci_dev->acpi_index);
2066             return;
2067         }
2068         g_sequence_insert_sorted(used_indexes,
2069                                  GINT_TO_POINTER(pci_dev->acpi_index),
2070                                  g_cmp_uint32, NULL);
2071     }
2072 
2073     if (pci_dev->romsize != UINT32_MAX && !is_power_of_2(pci_dev->romsize)) {
2074         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2075         return;
2076     }
2077 
2078     /* initialize cap_present for pci_is_express() and pci_config_size(),
2079      * Note that hybrid PCIs are not set automatically and need to manage
2080      * QEMU_PCI_CAP_EXPRESS manually */
2081     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2082        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2083         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2084     }
2085 
2086     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2087         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2088     }
2089 
2090     pci_dev = do_pci_register_device(pci_dev,
2091                                      object_get_typename(OBJECT(qdev)),
2092                                      pci_dev->devfn, errp);
2093     if (pci_dev == NULL)
2094         return;
2095 
2096     if (pc->realize) {
2097         pc->realize(pci_dev, &local_err);
2098         if (local_err) {
2099             error_propagate(errp, local_err);
2100             do_pci_unregister_device(pci_dev);
2101             return;
2102         }
2103     }
2104 
2105     /*
2106      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2107      * associate only Device 0 with the device attached to the bus
2108      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2109      * sec 7.3.1).
2110      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2111      * 5-bit Device Number and 3-bit Function Number fields in its associated
2112      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2113      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2114      */
2115     if (pci_is_express(pci_dev) &&
2116         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2117         pcie_has_upstream_port(pci_dev) &&
2118         PCI_SLOT(pci_dev->devfn)) {
2119         warn_report("PCI: slot %d is not valid for %s,"
2120                     " parent device only allows plugging into slot 0.",
2121                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2122     }
2123 
2124     if (pci_dev->failover_pair_id) {
2125         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2126             error_setg(errp, "failover primary device must be on "
2127                              "PCIExpress bus");
2128             pci_qdev_unrealize(DEVICE(pci_dev));
2129             return;
2130         }
2131         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2132         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2133             error_setg(errp, "failover primary device is not an "
2134                              "Ethernet device");
2135             pci_qdev_unrealize(DEVICE(pci_dev));
2136             return;
2137         }
2138         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2139             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2140             error_setg(errp, "failover: primary device must be in its own "
2141                               "PCI slot");
2142             pci_qdev_unrealize(DEVICE(pci_dev));
2143             return;
2144         }
2145         qdev->allow_unplug_during_migration = true;
2146     }
2147 
2148     /* rom loading */
2149     is_default_rom = false;
2150     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2151         pci_dev->romfile = g_strdup(pc->romfile);
2152         is_default_rom = true;
2153     }
2154 
2155     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2156     if (local_err) {
2157         error_propagate(errp, local_err);
2158         pci_qdev_unrealize(DEVICE(pci_dev));
2159         return;
2160     }
2161 
2162     pci_set_power(pci_dev, true);
2163 
2164     pci_dev->msi_trigger = pci_msi_trigger;
2165 }
2166 
2167 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2168                                    const char *name)
2169 {
2170     DeviceState *dev;
2171 
2172     dev = qdev_new(name);
2173     qdev_prop_set_int32(dev, "addr", devfn);
2174     qdev_prop_set_bit(dev, "multifunction", multifunction);
2175     return PCI_DEVICE(dev);
2176 }
2177 
2178 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2179 {
2180     return pci_new_internal(devfn, true, name);
2181 }
2182 
2183 PCIDevice *pci_new(int devfn, const char *name)
2184 {
2185     return pci_new_internal(devfn, false, name);
2186 }
2187 
2188 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2189 {
2190     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2191 }
2192 
2193 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2194                                            const char *name)
2195 {
2196     PCIDevice *dev = pci_new_multifunction(devfn, name);
2197     pci_realize_and_unref(dev, bus, &error_fatal);
2198     return dev;
2199 }
2200 
2201 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2202 {
2203     PCIDevice *dev = pci_new(devfn, name);
2204     pci_realize_and_unref(dev, bus, &error_fatal);
2205     return dev;
2206 }
2207 
2208 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2209 {
2210     int offset = PCI_CONFIG_HEADER_SIZE;
2211     int i;
2212     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2213         if (pdev->used[i])
2214             offset = i + 1;
2215         else if (i - offset + 1 == size)
2216             return offset;
2217     }
2218     return 0;
2219 }
2220 
2221 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2222                                         uint8_t *prev_p)
2223 {
2224     uint8_t next, prev;
2225 
2226     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2227         return 0;
2228 
2229     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2230          prev = next + PCI_CAP_LIST_NEXT)
2231         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2232             break;
2233 
2234     if (prev_p)
2235         *prev_p = prev;
2236     return next;
2237 }
2238 
2239 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2240 {
2241     uint8_t next, prev, found = 0;
2242 
2243     if (!(pdev->used[offset])) {
2244         return 0;
2245     }
2246 
2247     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2248 
2249     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2250          prev = next + PCI_CAP_LIST_NEXT) {
2251         if (next <= offset && next > found) {
2252             found = next;
2253         }
2254     }
2255     return found;
2256 }
2257 
2258 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2259    This is needed for an option rom which is used for more than one device. */
2260 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2261 {
2262     uint16_t vendor_id;
2263     uint16_t device_id;
2264     uint16_t rom_vendor_id;
2265     uint16_t rom_device_id;
2266     uint16_t rom_magic;
2267     uint16_t pcir_offset;
2268     uint8_t checksum;
2269 
2270     /* Words in rom data are little endian (like in PCI configuration),
2271        so they can be read / written with pci_get_word / pci_set_word. */
2272 
2273     /* Only a valid rom will be patched. */
2274     rom_magic = pci_get_word(ptr);
2275     if (rom_magic != 0xaa55) {
2276         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2277         return;
2278     }
2279     pcir_offset = pci_get_word(ptr + 0x18);
2280     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2281         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2282         return;
2283     }
2284 
2285     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2286     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2287     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2288     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2289 
2290     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2291                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2292 
2293     checksum = ptr[6];
2294 
2295     if (vendor_id != rom_vendor_id) {
2296         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2297         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2298         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2299         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2300         ptr[6] = checksum;
2301         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2302     }
2303 
2304     if (device_id != rom_device_id) {
2305         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2306         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2307         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2308         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2309         ptr[6] = checksum;
2310         pci_set_word(ptr + pcir_offset + 6, device_id);
2311     }
2312 }
2313 
2314 /* Add an option rom for the device */
2315 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2316                                Error **errp)
2317 {
2318     int64_t size = 0;
2319     g_autofree char *path = NULL;
2320     char name[32];
2321     const VMStateDescription *vmsd;
2322 
2323     /*
2324      * In case of incoming migration ROM will come with migration stream, no
2325      * reason to load the file.  Neither we want to fail if local ROM file
2326      * mismatches with specified romsize.
2327      */
2328     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2329 
2330     if (!pdev->romfile || !strlen(pdev->romfile)) {
2331         return;
2332     }
2333 
2334     if (!pdev->rom_bar) {
2335         /*
2336          * Load rom via fw_cfg instead of creating a rom bar,
2337          * for 0.11 compatibility.
2338          */
2339         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2340 
2341         /*
2342          * Hot-plugged devices can't use the option ROM
2343          * if the rom bar is disabled.
2344          */
2345         if (DEVICE(pdev)->hotplugged) {
2346             error_setg(errp, "Hot-plugged device without ROM bar"
2347                        " can't have an option ROM");
2348             return;
2349         }
2350 
2351         if (class == 0x0300) {
2352             rom_add_vga(pdev->romfile);
2353         } else {
2354             rom_add_option(pdev->romfile, -1);
2355         }
2356         return;
2357     }
2358 
2359     if (load_file || pdev->romsize == UINT32_MAX) {
2360         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2361         if (path == NULL) {
2362             path = g_strdup(pdev->romfile);
2363         }
2364 
2365         size = get_image_size(path);
2366         if (size < 0) {
2367             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2368             return;
2369         } else if (size == 0) {
2370             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2371             return;
2372         } else if (size > 2 * GiB) {
2373             error_setg(errp,
2374                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2375                        pdev->romfile);
2376             return;
2377         }
2378         if (pdev->romsize != UINT_MAX) {
2379             if (size > pdev->romsize) {
2380                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2381                            "is too large for ROM size %u",
2382                            pdev->romfile, (uint32_t)size, pdev->romsize);
2383                 return;
2384             }
2385         } else {
2386             pdev->romsize = pow2ceil(size);
2387         }
2388     }
2389 
2390     vmsd = qdev_get_vmsd(DEVICE(pdev));
2391     snprintf(name, sizeof(name), "%s.rom",
2392              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2393 
2394     pdev->has_rom = true;
2395     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2396                            &error_fatal);
2397 
2398     if (load_file) {
2399         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2400 
2401         if (load_image_size(path, ptr, size) < 0) {
2402             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2403             return;
2404         }
2405 
2406         if (is_default_rom) {
2407             /* Only the default rom images will be patched (if needed). */
2408             pci_patch_ids(pdev, ptr, size);
2409         }
2410     }
2411 
2412     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2413 }
2414 
2415 static void pci_del_option_rom(PCIDevice *pdev)
2416 {
2417     if (!pdev->has_rom)
2418         return;
2419 
2420     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2421     pdev->has_rom = false;
2422 }
2423 
2424 /*
2425  * On success, pci_add_capability() returns a positive value
2426  * that the offset of the pci capability.
2427  * On failure, it sets an error and returns a negative error
2428  * code.
2429  */
2430 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2431                        uint8_t offset, uint8_t size,
2432                        Error **errp)
2433 {
2434     uint8_t *config;
2435     int i, overlapping_cap;
2436 
2437     if (!offset) {
2438         offset = pci_find_space(pdev, size);
2439         /* out of PCI config space is programming error */
2440         assert(offset);
2441     } else {
2442         /* Verify that capabilities don't overlap.  Note: device assignment
2443          * depends on this check to verify that the device is not broken.
2444          * Should never trigger for emulated devices, but it's helpful
2445          * for debugging these. */
2446         for (i = offset; i < offset + size; i++) {
2447             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2448             if (overlapping_cap) {
2449                 error_setg(errp, "%s:%02x:%02x.%x "
2450                            "Attempt to add PCI capability %x at offset "
2451                            "%x overlaps existing capability %x at offset %x",
2452                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2453                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2454                            cap_id, offset, overlapping_cap, i);
2455                 return -EINVAL;
2456             }
2457         }
2458     }
2459 
2460     config = pdev->config + offset;
2461     config[PCI_CAP_LIST_ID] = cap_id;
2462     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2463     pdev->config[PCI_CAPABILITY_LIST] = offset;
2464     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2465     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2466     /* Make capability read-only by default */
2467     memset(pdev->wmask + offset, 0, size);
2468     /* Check capability by default */
2469     memset(pdev->cmask + offset, 0xFF, size);
2470     return offset;
2471 }
2472 
2473 /* Unlink capability from the pci config space. */
2474 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2475 {
2476     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2477     if (!offset)
2478         return;
2479     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2480     /* Make capability writable again */
2481     memset(pdev->wmask + offset, 0xff, size);
2482     memset(pdev->w1cmask + offset, 0, size);
2483     /* Clear cmask as device-specific registers can't be checked */
2484     memset(pdev->cmask + offset, 0, size);
2485     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2486 
2487     if (!pdev->config[PCI_CAPABILITY_LIST])
2488         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2489 }
2490 
2491 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2492 {
2493     return pci_find_capability_list(pdev, cap_id, NULL);
2494 }
2495 
2496 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2497 {
2498     PCIDevice *d = (PCIDevice *)dev;
2499     const char *name = NULL;
2500     const pci_class_desc *desc =  pci_class_descriptions;
2501     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2502 
2503     while (desc->desc &&
2504           (class & ~desc->fw_ign_bits) !=
2505           (desc->class & ~desc->fw_ign_bits)) {
2506         desc++;
2507     }
2508 
2509     if (desc->desc) {
2510         name = desc->fw_name;
2511     }
2512 
2513     if (name) {
2514         pstrcpy(buf, len, name);
2515     } else {
2516         snprintf(buf, len, "pci%04x,%04x",
2517                  pci_get_word(d->config + PCI_VENDOR_ID),
2518                  pci_get_word(d->config + PCI_DEVICE_ID));
2519     }
2520 
2521     return buf;
2522 }
2523 
2524 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2525 {
2526     PCIDevice *d = (PCIDevice *)dev;
2527     char name[33];
2528     int has_func = !!PCI_FUNC(d->devfn);
2529 
2530     return g_strdup_printf("%s@%x%s%.*x",
2531                            pci_dev_fw_name(dev, name, sizeof(name)),
2532                            PCI_SLOT(d->devfn),
2533                            has_func ? "," : "",
2534                            has_func,
2535                            PCI_FUNC(d->devfn));
2536 }
2537 
2538 static char *pcibus_get_dev_path(DeviceState *dev)
2539 {
2540     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2541     PCIDevice *t;
2542     int slot_depth;
2543     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2544      * 00 is added here to make this format compatible with
2545      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2546      * Slot.Function list specifies the slot and function numbers for all
2547      * devices on the path from root to the specific device. */
2548     const char *root_bus_path;
2549     int root_bus_len;
2550     char slot[] = ":SS.F";
2551     int slot_len = sizeof slot - 1 /* For '\0' */;
2552     int path_len;
2553     char *path, *p;
2554     int s;
2555 
2556     root_bus_path = pci_root_bus_path(d);
2557     root_bus_len = strlen(root_bus_path);
2558 
2559     /* Calculate # of slots on path between device and root. */;
2560     slot_depth = 0;
2561     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2562         ++slot_depth;
2563     }
2564 
2565     path_len = root_bus_len + slot_len * slot_depth;
2566 
2567     /* Allocate memory, fill in the terminating null byte. */
2568     path = g_malloc(path_len + 1 /* For '\0' */);
2569     path[path_len] = '\0';
2570 
2571     memcpy(path, root_bus_path, root_bus_len);
2572 
2573     /* Fill in slot numbers. We walk up from device to root, so need to print
2574      * them in the reverse order, last to first. */
2575     p = path + path_len;
2576     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2577         p -= slot_len;
2578         s = snprintf(slot, sizeof slot, ":%02x.%x",
2579                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2580         assert(s == slot_len);
2581         memcpy(p, slot, slot_len);
2582     }
2583 
2584     return path;
2585 }
2586 
2587 static int pci_qdev_find_recursive(PCIBus *bus,
2588                                    const char *id, PCIDevice **pdev)
2589 {
2590     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2591     if (!qdev) {
2592         return -ENODEV;
2593     }
2594 
2595     /* roughly check if given qdev is pci device */
2596     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2597         *pdev = PCI_DEVICE(qdev);
2598         return 0;
2599     }
2600     return -EINVAL;
2601 }
2602 
2603 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2604 {
2605     PCIHostState *host_bridge;
2606     int rc = -ENODEV;
2607 
2608     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2609         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2610         if (!tmp) {
2611             rc = 0;
2612             break;
2613         }
2614         if (tmp != -ENODEV) {
2615             rc = tmp;
2616         }
2617     }
2618 
2619     return rc;
2620 }
2621 
2622 MemoryRegion *pci_address_space(PCIDevice *dev)
2623 {
2624     return pci_get_bus(dev)->address_space_mem;
2625 }
2626 
2627 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2628 {
2629     return pci_get_bus(dev)->address_space_io;
2630 }
2631 
2632 static void pci_device_class_init(ObjectClass *klass, void *data)
2633 {
2634     DeviceClass *k = DEVICE_CLASS(klass);
2635 
2636     k->realize = pci_qdev_realize;
2637     k->unrealize = pci_qdev_unrealize;
2638     k->bus_type = TYPE_PCI_BUS;
2639     device_class_set_props(k, pci_props);
2640     object_class_property_set_description(
2641         klass, "x-max-bounce-buffer-size",
2642         "Maximum buffer size allocated for bounce buffers used for mapped "
2643         "access to indirect DMA memory");
2644 }
2645 
2646 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2647 {
2648     if (!object_class_is_abstract(klass)) {
2649         ObjectClass *conventional =
2650             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2651         ObjectClass *pcie =
2652             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2653         ObjectClass *cxl =
2654             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2655         assert(conventional || pcie || cxl);
2656     }
2657 }
2658 
2659 /*
2660  * Get IOMMU root bus, aliased bus and devfn of a PCI device
2661  *
2662  * IOMMU root bus is needed by all call sites to call into iommu_ops.
2663  * For call sites which don't need aliased BDF, passing NULL to
2664  * aliased_[bus|devfn] is allowed.
2665  *
2666  * @piommu_bus: return root #PCIBus backed by an IOMMU for the PCI device.
2667  *
2668  * @aliased_bus: return aliased #PCIBus of the PCI device, optional.
2669  *
2670  * @aliased_devfn: return aliased devfn of the PCI device, optional.
2671  */
2672 static void pci_device_get_iommu_bus_devfn(PCIDevice *dev,
2673                                            PCIBus **piommu_bus,
2674                                            PCIBus **aliased_bus,
2675                                            int *aliased_devfn)
2676 {
2677     PCIBus *bus = pci_get_bus(dev);
2678     PCIBus *iommu_bus = bus;
2679     int devfn = dev->devfn;
2680 
2681     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2682         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2683 
2684         /*
2685          * The requester ID of the provided device may be aliased, as seen from
2686          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2687          * requester ID to provide a unique AddressSpace for devices, but
2688          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2689          * to-PCI bridge creates and accepts transactions on behalf of down-
2690          * stream devices.  When doing so, all downstream devices are masked
2691          * (aliased) behind a single requester ID.  The requester ID used
2692          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2693          * bridges, with a PCIe capability indicating such, follow the
2694          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2695          * where the bridge uses the seconary bus as the bridge portion of the
2696          * requester ID and devfn of 00.0.  For other bridges, typically those
2697          * found on the root complex such as the dmi-to-pci-bridge, we follow
2698          * the convention of typical bare-metal hardware, which uses the
2699          * requester ID of the bridge itself.  There are device specific
2700          * exceptions to these rules, but these are the defaults that the
2701          * Linux kernel uses when determining DMA aliases itself and believed
2702          * to be true for the bare metal equivalents of the devices emulated
2703          * in QEMU.
2704          */
2705         if (!pci_bus_is_express(iommu_bus)) {
2706             PCIDevice *parent = iommu_bus->parent_dev;
2707 
2708             if (pci_is_express(parent) &&
2709                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2710                 devfn = PCI_DEVFN(0, 0);
2711                 bus = iommu_bus;
2712             } else {
2713                 devfn = parent->devfn;
2714                 bus = parent_bus;
2715             }
2716         }
2717 
2718         iommu_bus = parent_bus;
2719     }
2720 
2721     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
2722     assert(iommu_bus);
2723 
2724     if (pci_bus_bypass_iommu(bus) || !iommu_bus->iommu_ops) {
2725         iommu_bus = NULL;
2726     }
2727 
2728     *piommu_bus = iommu_bus;
2729 
2730     if (aliased_bus) {
2731         *aliased_bus = bus;
2732     }
2733 
2734     if (aliased_devfn) {
2735         *aliased_devfn = devfn;
2736     }
2737 }
2738 
2739 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2740 {
2741     PCIBus *bus;
2742     PCIBus *iommu_bus;
2743     int devfn;
2744 
2745     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn);
2746     if (iommu_bus) {
2747         return iommu_bus->iommu_ops->get_address_space(bus,
2748                                  iommu_bus->iommu_opaque, devfn);
2749     }
2750     return &address_space_memory;
2751 }
2752 
2753 bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod,
2754                                  Error **errp)
2755 {
2756     PCIBus *iommu_bus, *aliased_bus;
2757     int aliased_devfn;
2758 
2759     /* set_iommu_device requires device's direct BDF instead of aliased BDF */
2760     pci_device_get_iommu_bus_devfn(dev, &iommu_bus,
2761                                    &aliased_bus, &aliased_devfn);
2762     if (iommu_bus && iommu_bus->iommu_ops->set_iommu_device) {
2763         hiod->aliased_bus = aliased_bus;
2764         hiod->aliased_devfn = aliased_devfn;
2765         return iommu_bus->iommu_ops->set_iommu_device(pci_get_bus(dev),
2766                                                       iommu_bus->iommu_opaque,
2767                                                       dev->devfn, hiod, errp);
2768     }
2769     return true;
2770 }
2771 
2772 void pci_device_unset_iommu_device(PCIDevice *dev)
2773 {
2774     PCIBus *iommu_bus;
2775 
2776     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, NULL, NULL);
2777     if (iommu_bus && iommu_bus->iommu_ops->unset_iommu_device) {
2778         return iommu_bus->iommu_ops->unset_iommu_device(pci_get_bus(dev),
2779                                                         iommu_bus->iommu_opaque,
2780                                                         dev->devfn);
2781     }
2782 }
2783 
2784 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2785 {
2786     /*
2787      * If called, pci_setup_iommu() should provide a minimum set of
2788      * useful callbacks for the bus.
2789      */
2790     assert(ops);
2791     assert(ops->get_address_space);
2792 
2793     bus->iommu_ops = ops;
2794     bus->iommu_opaque = opaque;
2795 }
2796 
2797 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2798 {
2799     Range *range = opaque;
2800     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2801     int i;
2802 
2803     if (!(cmd & PCI_COMMAND_MEMORY)) {
2804         return;
2805     }
2806 
2807     if (IS_PCI_BRIDGE(dev)) {
2808         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2809         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2810 
2811         base = MAX(base, 0x1ULL << 32);
2812 
2813         if (limit >= base) {
2814             Range pref_range;
2815             range_set_bounds(&pref_range, base, limit);
2816             range_extend(range, &pref_range);
2817         }
2818     }
2819     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2820         PCIIORegion *r = &dev->io_regions[i];
2821         pcibus_t lob, upb;
2822         Range region_range;
2823 
2824         if (!r->size ||
2825             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2826             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2827             continue;
2828         }
2829 
2830         lob = pci_bar_address(dev, i, r->type, r->size);
2831         upb = lob + r->size - 1;
2832         if (lob == PCI_BAR_UNMAPPED) {
2833             continue;
2834         }
2835 
2836         lob = MAX(lob, 0x1ULL << 32);
2837 
2838         if (upb >= lob) {
2839             range_set_bounds(&region_range, lob, upb);
2840             range_extend(range, &region_range);
2841         }
2842     }
2843 }
2844 
2845 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2846 {
2847     range_make_empty(range);
2848     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2849 }
2850 
2851 static bool pcie_has_upstream_port(PCIDevice *dev)
2852 {
2853     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2854 
2855     /* Device associated with an upstream port.
2856      * As there are several types of these, it's easier to check the
2857      * parent device: upstream ports are always connected to
2858      * root or downstream ports.
2859      */
2860     return parent_dev &&
2861         pci_is_express(parent_dev) &&
2862         parent_dev->exp.exp_cap &&
2863         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2864          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2865 }
2866 
2867 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2868 {
2869     PCIBus *bus = pci_get_bus(pci_dev);
2870 
2871     if(pcie_has_upstream_port(pci_dev)) {
2872         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2873         return bus->devices[0];
2874     } else {
2875         /* Other bus types might support multiple devices at slots 0-31 */
2876         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2877     }
2878 }
2879 
2880 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2881 {
2882     MSIMessage msg;
2883     if (msix_enabled(dev)) {
2884         msg = msix_get_message(dev, vector);
2885     } else if (msi_enabled(dev)) {
2886         msg = msi_get_message(dev, vector);
2887     } else {
2888         /* Should never happen */
2889         error_report("%s: unknown interrupt type", __func__);
2890         abort();
2891     }
2892     return msg;
2893 }
2894 
2895 void pci_set_power(PCIDevice *d, bool state)
2896 {
2897     if (d->has_power == state) {
2898         return;
2899     }
2900 
2901     d->has_power = state;
2902     pci_update_mappings(d);
2903     memory_region_set_enabled(&d->bus_master_enable_region,
2904                               (pci_get_word(d->config + PCI_COMMAND)
2905                                & PCI_COMMAND_MASTER) && d->has_power);
2906     if (!d->has_power) {
2907         pci_device_reset(d);
2908     }
2909 }
2910 
2911 static const TypeInfo pci_device_type_info = {
2912     .name = TYPE_PCI_DEVICE,
2913     .parent = TYPE_DEVICE,
2914     .instance_size = sizeof(PCIDevice),
2915     .abstract = true,
2916     .class_size = sizeof(PCIDeviceClass),
2917     .class_init = pci_device_class_init,
2918     .class_base_init = pci_device_class_base_init,
2919 };
2920 
2921 static void pci_register_types(void)
2922 {
2923     type_register_static(&pci_bus_info);
2924     type_register_static(&pcie_bus_info);
2925     type_register_static(&cxl_bus_info);
2926     type_register_static(&conventional_pci_interface_info);
2927     type_register_static(&cxl_interface_info);
2928     type_register_static(&pcie_interface_info);
2929     type_register_static(&pci_device_type_info);
2930 }
2931 
2932 type_init(pci_register_types)
2933