xref: /openbmc/qemu/hw/pci/pci.c (revision 4565917b)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu/datadir.h"
27 #include "qemu/units.h"
28 #include "hw/irq.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bridge.h"
31 #include "hw/pci/pci_bus.h"
32 #include "hw/pci/pci_host.h"
33 #include "hw/qdev-properties.h"
34 #include "hw/qdev-properties-system.h"
35 #include "migration/qemu-file-types.h"
36 #include "migration/vmstate.h"
37 #include "net/net.h"
38 #include "sysemu/numa.h"
39 #include "sysemu/runstate.h"
40 #include "sysemu/sysemu.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "qapi/error.h"
50 #include "qemu/cutils.h"
51 #include "pci-internal.h"
52 
53 #include "hw/xen/xen.h"
54 #include "hw/i386/kvm/xen_evtchn.h"
55 
56 //#define DEBUG_PCI
57 #ifdef DEBUG_PCI
58 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
59 #else
60 # define PCI_DPRINTF(format, ...)       do { } while (0)
61 #endif
62 
63 bool pci_available = true;
64 
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset(BusState *qbus);
68 static bool pcie_has_upstream_port(PCIDevice *dev);
69 
70 static Property pci_props[] = {
71     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
72     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
73     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
74     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
75     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
76                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
77     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
78                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
79     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
80                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
81     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
82                        failover_pair_id),
83     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
84     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
85                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
86     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
87                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
88     DEFINE_PROP_END_OF_LIST()
89 };
90 
91 static const VMStateDescription vmstate_pcibus = {
92     .name = "PCIBUS",
93     .version_id = 1,
94     .minimum_version_id = 1,
95     .fields = (VMStateField[]) {
96         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
97         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
98                              nirq, 0, vmstate_info_int32,
99                              int32_t),
100         VMSTATE_END_OF_LIST()
101     }
102 };
103 
104 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
105 {
106     return a - b;
107 }
108 
109 static GSequence *pci_acpi_index_list(void)
110 {
111     static GSequence *used_acpi_index_list;
112 
113     if (!used_acpi_index_list) {
114         used_acpi_index_list = g_sequence_new(NULL);
115     }
116     return used_acpi_index_list;
117 }
118 
119 static void pci_init_bus_master(PCIDevice *pci_dev)
120 {
121     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
122 
123     memory_region_init_alias(&pci_dev->bus_master_enable_region,
124                              OBJECT(pci_dev), "bus master",
125                              dma_as->root, 0, memory_region_size(dma_as->root));
126     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
127     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
128                                 &pci_dev->bus_master_enable_region);
129 }
130 
131 static void pcibus_machine_done(Notifier *notifier, void *data)
132 {
133     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
134     int i;
135 
136     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
137         if (bus->devices[i]) {
138             pci_init_bus_master(bus->devices[i]);
139         }
140     }
141 }
142 
143 static void pci_bus_realize(BusState *qbus, Error **errp)
144 {
145     PCIBus *bus = PCI_BUS(qbus);
146 
147     bus->machine_done.notify = pcibus_machine_done;
148     qemu_add_machine_init_done_notifier(&bus->machine_done);
149 
150     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
151 }
152 
153 static void pcie_bus_realize(BusState *qbus, Error **errp)
154 {
155     PCIBus *bus = PCI_BUS(qbus);
156     Error *local_err = NULL;
157 
158     pci_bus_realize(qbus, &local_err);
159     if (local_err) {
160         error_propagate(errp, local_err);
161         return;
162     }
163 
164     /*
165      * A PCI-E bus can support extended config space if it's the root
166      * bus, or if the bus/bridge above it does as well
167      */
168     if (pci_bus_is_root(bus)) {
169         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
170     } else {
171         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
172 
173         if (pci_bus_allows_extended_config_space(parent_bus)) {
174             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
175         }
176     }
177 }
178 
179 static void pci_bus_unrealize(BusState *qbus)
180 {
181     PCIBus *bus = PCI_BUS(qbus);
182 
183     qemu_remove_machine_init_done_notifier(&bus->machine_done);
184 
185     vmstate_unregister(NULL, &vmstate_pcibus, bus);
186 }
187 
188 static int pcibus_num(PCIBus *bus)
189 {
190     if (pci_bus_is_root(bus)) {
191         return 0; /* pci host bridge */
192     }
193     return bus->parent_dev->config[PCI_SECONDARY_BUS];
194 }
195 
196 static uint16_t pcibus_numa_node(PCIBus *bus)
197 {
198     return NUMA_NODE_UNASSIGNED;
199 }
200 
201 static void pci_bus_class_init(ObjectClass *klass, void *data)
202 {
203     BusClass *k = BUS_CLASS(klass);
204     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
205 
206     k->print_dev = pcibus_dev_print;
207     k->get_dev_path = pcibus_get_dev_path;
208     k->get_fw_dev_path = pcibus_get_fw_dev_path;
209     k->realize = pci_bus_realize;
210     k->unrealize = pci_bus_unrealize;
211     k->reset = pcibus_reset;
212 
213     pbc->bus_num = pcibus_num;
214     pbc->numa_node = pcibus_numa_node;
215 }
216 
217 static const TypeInfo pci_bus_info = {
218     .name = TYPE_PCI_BUS,
219     .parent = TYPE_BUS,
220     .instance_size = sizeof(PCIBus),
221     .class_size = sizeof(PCIBusClass),
222     .class_init = pci_bus_class_init,
223 };
224 
225 static const TypeInfo cxl_interface_info = {
226     .name          = INTERFACE_CXL_DEVICE,
227     .parent        = TYPE_INTERFACE,
228 };
229 
230 static const TypeInfo pcie_interface_info = {
231     .name          = INTERFACE_PCIE_DEVICE,
232     .parent        = TYPE_INTERFACE,
233 };
234 
235 static const TypeInfo conventional_pci_interface_info = {
236     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
237     .parent        = TYPE_INTERFACE,
238 };
239 
240 static void pcie_bus_class_init(ObjectClass *klass, void *data)
241 {
242     BusClass *k = BUS_CLASS(klass);
243 
244     k->realize = pcie_bus_realize;
245 }
246 
247 static const TypeInfo pcie_bus_info = {
248     .name = TYPE_PCIE_BUS,
249     .parent = TYPE_PCI_BUS,
250     .class_init = pcie_bus_class_init,
251 };
252 
253 static const TypeInfo cxl_bus_info = {
254     .name       = TYPE_CXL_BUS,
255     .parent     = TYPE_PCIE_BUS,
256     .class_init = pcie_bus_class_init,
257 };
258 
259 static void pci_update_mappings(PCIDevice *d);
260 static void pci_irq_handler(void *opaque, int irq_num, int level);
261 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
262 static void pci_del_option_rom(PCIDevice *pdev);
263 
264 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
265 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
266 
267 PCIHostStateList pci_host_bridges;
268 
269 int pci_bar(PCIDevice *d, int reg)
270 {
271     uint8_t type;
272 
273     /* PCIe virtual functions do not have their own BARs */
274     assert(!pci_is_vf(d));
275 
276     if (reg != PCI_ROM_SLOT)
277         return PCI_BASE_ADDRESS_0 + reg * 4;
278 
279     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
280     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
281 }
282 
283 static inline int pci_irq_state(PCIDevice *d, int irq_num)
284 {
285         return (d->irq_state >> irq_num) & 0x1;
286 }
287 
288 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
289 {
290         d->irq_state &= ~(0x1 << irq_num);
291         d->irq_state |= level << irq_num;
292 }
293 
294 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
295 {
296     assert(irq_num >= 0);
297     assert(irq_num < bus->nirq);
298     bus->irq_count[irq_num] += change;
299     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
300 }
301 
302 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
303 {
304     PCIBus *bus;
305     for (;;) {
306         int dev_irq = irq_num;
307         bus = pci_get_bus(pci_dev);
308         assert(bus->map_irq);
309         irq_num = bus->map_irq(pci_dev, irq_num);
310         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
311                             pci_bus_is_root(bus) ? "root-complex"
312                                     : DEVICE(bus->parent_dev)->canonical_path);
313         if (bus->set_irq)
314             break;
315         pci_dev = bus->parent_dev;
316     }
317     pci_bus_change_irq_level(bus, irq_num, change);
318 }
319 
320 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
321 {
322     assert(irq_num >= 0);
323     assert(irq_num < bus->nirq);
324     return !!bus->irq_count[irq_num];
325 }
326 
327 /* Update interrupt status bit in config space on interrupt
328  * state change. */
329 static void pci_update_irq_status(PCIDevice *dev)
330 {
331     if (dev->irq_state) {
332         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
333     } else {
334         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
335     }
336 }
337 
338 void pci_device_deassert_intx(PCIDevice *dev)
339 {
340     int i;
341     for (i = 0; i < PCI_NUM_PINS; ++i) {
342         pci_irq_handler(dev, i, 0);
343     }
344 }
345 
346 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
347 {
348     MemTxAttrs attrs = {};
349 
350     /*
351      * Xen uses the high bits of the address to contain some of the bits
352      * of the PIRQ#. Therefore we can't just send the write cycle and
353      * trust that it's caught by the APIC at 0xfee00000 because the
354      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
355      * So we intercept the delivery here instead of in kvm_send_msi().
356      */
357     if (xen_mode == XEN_EMULATE &&
358         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
359         return;
360     }
361     attrs.requester_id = pci_requester_id(dev);
362     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
363                          attrs, NULL);
364 }
365 
366 static void pci_reset_regions(PCIDevice *dev)
367 {
368     int r;
369     if (pci_is_vf(dev)) {
370         return;
371     }
372 
373     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
374         PCIIORegion *region = &dev->io_regions[r];
375         if (!region->size) {
376             continue;
377         }
378 
379         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
380             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
381             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
382         } else {
383             pci_set_long(dev->config + pci_bar(dev, r), region->type);
384         }
385     }
386 }
387 
388 static void pci_do_device_reset(PCIDevice *dev)
389 {
390     pci_device_deassert_intx(dev);
391     assert(dev->irq_state == 0);
392 
393     /* Clear all writable bits */
394     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
395                                  pci_get_word(dev->wmask + PCI_COMMAND) |
396                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
397     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
398                                  pci_get_word(dev->wmask + PCI_STATUS) |
399                                  pci_get_word(dev->w1cmask + PCI_STATUS));
400     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
401     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
402                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
403                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
404     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
405     pci_reset_regions(dev);
406     pci_update_mappings(dev);
407 
408     msi_reset(dev);
409     msix_reset(dev);
410 }
411 
412 /*
413  * This function is called on #RST and FLR.
414  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
415  */
416 void pci_device_reset(PCIDevice *dev)
417 {
418     device_cold_reset(&dev->qdev);
419     pci_do_device_reset(dev);
420 }
421 
422 /*
423  * Trigger pci bus reset under a given bus.
424  * Called via bus_cold_reset on RST# assert, after the devices
425  * have been reset device_cold_reset-ed already.
426  */
427 static void pcibus_reset(BusState *qbus)
428 {
429     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
430     int i;
431 
432     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
433         if (bus->devices[i]) {
434             pci_do_device_reset(bus->devices[i]);
435         }
436     }
437 
438     for (i = 0; i < bus->nirq; i++) {
439         assert(bus->irq_count[i] == 0);
440     }
441 }
442 
443 static void pci_host_bus_register(DeviceState *host)
444 {
445     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
446 
447     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
448 }
449 
450 static void pci_host_bus_unregister(DeviceState *host)
451 {
452     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
453 
454     QLIST_REMOVE(host_bridge, next);
455 }
456 
457 PCIBus *pci_device_root_bus(const PCIDevice *d)
458 {
459     PCIBus *bus = pci_get_bus(d);
460 
461     while (!pci_bus_is_root(bus)) {
462         d = bus->parent_dev;
463         assert(d != NULL);
464 
465         bus = pci_get_bus(d);
466     }
467 
468     return bus;
469 }
470 
471 const char *pci_root_bus_path(PCIDevice *dev)
472 {
473     PCIBus *rootbus = pci_device_root_bus(dev);
474     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
475     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
476 
477     assert(host_bridge->bus == rootbus);
478 
479     if (hc->root_bus_path) {
480         return (*hc->root_bus_path)(host_bridge, rootbus);
481     }
482 
483     return rootbus->qbus.name;
484 }
485 
486 bool pci_bus_bypass_iommu(PCIBus *bus)
487 {
488     PCIBus *rootbus = bus;
489     PCIHostState *host_bridge;
490 
491     if (!pci_bus_is_root(bus)) {
492         rootbus = pci_device_root_bus(bus->parent_dev);
493     }
494 
495     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
496 
497     assert(host_bridge->bus == rootbus);
498 
499     return host_bridge->bypass_iommu;
500 }
501 
502 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
503                                        MemoryRegion *address_space_mem,
504                                        MemoryRegion *address_space_io,
505                                        uint8_t devfn_min)
506 {
507     assert(PCI_FUNC(devfn_min) == 0);
508     bus->devfn_min = devfn_min;
509     bus->slot_reserved_mask = 0x0;
510     bus->address_space_mem = address_space_mem;
511     bus->address_space_io = address_space_io;
512     bus->flags |= PCI_BUS_IS_ROOT;
513 
514     /* host bridge */
515     QLIST_INIT(&bus->child);
516 
517     pci_host_bus_register(parent);
518 }
519 
520 static void pci_bus_uninit(PCIBus *bus)
521 {
522     pci_host_bus_unregister(BUS(bus)->parent);
523 }
524 
525 bool pci_bus_is_express(const PCIBus *bus)
526 {
527     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
528 }
529 
530 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
531                        const char *name,
532                        MemoryRegion *address_space_mem,
533                        MemoryRegion *address_space_io,
534                        uint8_t devfn_min, const char *typename)
535 {
536     qbus_init(bus, bus_size, typename, parent, name);
537     pci_root_bus_internal_init(bus, parent, address_space_mem,
538                                address_space_io, devfn_min);
539 }
540 
541 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
542                          MemoryRegion *address_space_mem,
543                          MemoryRegion *address_space_io,
544                          uint8_t devfn_min, const char *typename)
545 {
546     PCIBus *bus;
547 
548     bus = PCI_BUS(qbus_new(typename, parent, name));
549     pci_root_bus_internal_init(bus, parent, address_space_mem,
550                                address_space_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 *address_space_mem,
590                               MemoryRegion *address_space_io,
591                               uint8_t devfn_min, int nirq,
592                               const char *typename)
593 {
594     PCIBus *bus;
595 
596     bus = pci_root_bus_new(parent, name, address_space_mem,
597                            address_space_io, devfn_min, typename);
598     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
599     pci_bus_map_irqs(bus, map_irq);
600     return bus;
601 }
602 
603 void pci_unregister_root_bus(PCIBus *bus)
604 {
605     pci_bus_irqs_cleanup(bus);
606     pci_root_bus_cleanup(bus);
607 }
608 
609 int pci_bus_num(PCIBus *s)
610 {
611     return PCI_BUS_GET_CLASS(s)->bus_num(s);
612 }
613 
614 /* Returns the min and max bus numbers of a PCI bus hierarchy */
615 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
616 {
617     int i;
618     *min_bus = *max_bus = pci_bus_num(bus);
619 
620     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
621         PCIDevice *dev = bus->devices[i];
622 
623         if (dev && IS_PCI_BRIDGE(dev)) {
624             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
625             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
626         }
627     }
628 }
629 
630 int pci_bus_numa_node(PCIBus *bus)
631 {
632     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
633 }
634 
635 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
636                                  const VMStateField *field)
637 {
638     PCIDevice *s = container_of(pv, PCIDevice, config);
639     uint8_t *config;
640     int i;
641 
642     assert(size == pci_config_size(s));
643     config = g_malloc(size);
644 
645     qemu_get_buffer(f, config, size);
646     for (i = 0; i < size; ++i) {
647         if ((config[i] ^ s->config[i]) &
648             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
649             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
650                          "cmask: %x wmask: %x w1cmask:%x", __func__,
651                          i, config[i], s->config[i],
652                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
653             g_free(config);
654             return -EINVAL;
655         }
656     }
657     memcpy(s->config, config, size);
658 
659     pci_update_mappings(s);
660     if (IS_PCI_BRIDGE(s)) {
661         pci_bridge_update_mappings(PCI_BRIDGE(s));
662     }
663 
664     memory_region_set_enabled(&s->bus_master_enable_region,
665                               pci_get_word(s->config + PCI_COMMAND)
666                               & PCI_COMMAND_MASTER);
667 
668     g_free(config);
669     return 0;
670 }
671 
672 /* just put buffer */
673 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
674                                  const VMStateField *field, JSONWriter *vmdesc)
675 {
676     const uint8_t **v = pv;
677     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
678     qemu_put_buffer(f, *v, size);
679 
680     return 0;
681 }
682 
683 static VMStateInfo vmstate_info_pci_config = {
684     .name = "pci config",
685     .get  = get_pci_config_device,
686     .put  = put_pci_config_device,
687 };
688 
689 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
690                              const VMStateField *field)
691 {
692     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
693     uint32_t irq_state[PCI_NUM_PINS];
694     int i;
695     for (i = 0; i < PCI_NUM_PINS; ++i) {
696         irq_state[i] = qemu_get_be32(f);
697         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
698             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
699                     irq_state[i]);
700             return -EINVAL;
701         }
702     }
703 
704     for (i = 0; i < PCI_NUM_PINS; ++i) {
705         pci_set_irq_state(s, i, irq_state[i]);
706     }
707 
708     return 0;
709 }
710 
711 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
712                              const VMStateField *field, JSONWriter *vmdesc)
713 {
714     int i;
715     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
716 
717     for (i = 0; i < PCI_NUM_PINS; ++i) {
718         qemu_put_be32(f, pci_irq_state(s, i));
719     }
720 
721     return 0;
722 }
723 
724 static VMStateInfo vmstate_info_pci_irq_state = {
725     .name = "pci irq state",
726     .get  = get_pci_irq_state,
727     .put  = put_pci_irq_state,
728 };
729 
730 static bool migrate_is_pcie(void *opaque, int version_id)
731 {
732     return pci_is_express((PCIDevice *)opaque);
733 }
734 
735 static bool migrate_is_not_pcie(void *opaque, int version_id)
736 {
737     return !pci_is_express((PCIDevice *)opaque);
738 }
739 
740 const VMStateDescription vmstate_pci_device = {
741     .name = "PCIDevice",
742     .version_id = 2,
743     .minimum_version_id = 1,
744     .fields = (VMStateField[]) {
745         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
746         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
747                                    migrate_is_not_pcie,
748                                    0, vmstate_info_pci_config,
749                                    PCI_CONFIG_SPACE_SIZE),
750         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
751                                    migrate_is_pcie,
752                                    0, vmstate_info_pci_config,
753                                    PCIE_CONFIG_SPACE_SIZE),
754         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
755                                    vmstate_info_pci_irq_state,
756                                    PCI_NUM_PINS * sizeof(int32_t)),
757         VMSTATE_END_OF_LIST()
758     }
759 };
760 
761 
762 void pci_device_save(PCIDevice *s, QEMUFile *f)
763 {
764     /* Clear interrupt status bit: it is implicit
765      * in irq_state which we are saving.
766      * This makes us compatible with old devices
767      * which never set or clear this bit. */
768     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
769     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
770     /* Restore the interrupt status bit. */
771     pci_update_irq_status(s);
772 }
773 
774 int pci_device_load(PCIDevice *s, QEMUFile *f)
775 {
776     int ret;
777     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
778     /* Restore the interrupt status bit. */
779     pci_update_irq_status(s);
780     return ret;
781 }
782 
783 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
784 {
785     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
786                  pci_default_sub_vendor_id);
787     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
788                  pci_default_sub_device_id);
789 }
790 
791 /*
792  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
793  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
794  */
795 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
796                              unsigned int *slotp, unsigned int *funcp)
797 {
798     const char *p;
799     char *e;
800     unsigned long val;
801     unsigned long dom = 0, bus = 0;
802     unsigned int slot = 0;
803     unsigned int func = 0;
804 
805     p = addr;
806     val = strtoul(p, &e, 16);
807     if (e == p)
808         return -1;
809     if (*e == ':') {
810         bus = val;
811         p = e + 1;
812         val = strtoul(p, &e, 16);
813         if (e == p)
814             return -1;
815         if (*e == ':') {
816             dom = bus;
817             bus = val;
818             p = e + 1;
819             val = strtoul(p, &e, 16);
820             if (e == p)
821                 return -1;
822         }
823     }
824 
825     slot = val;
826 
827     if (funcp != NULL) {
828         if (*e != '.')
829             return -1;
830 
831         p = e + 1;
832         val = strtoul(p, &e, 16);
833         if (e == p)
834             return -1;
835 
836         func = val;
837     }
838 
839     /* if funcp == NULL func is 0 */
840     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
841         return -1;
842 
843     if (*e)
844         return -1;
845 
846     *domp = dom;
847     *busp = bus;
848     *slotp = slot;
849     if (funcp != NULL)
850         *funcp = func;
851     return 0;
852 }
853 
854 static void pci_init_cmask(PCIDevice *dev)
855 {
856     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
857     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
858     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
859     dev->cmask[PCI_REVISION_ID] = 0xff;
860     dev->cmask[PCI_CLASS_PROG] = 0xff;
861     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
862     dev->cmask[PCI_HEADER_TYPE] = 0xff;
863     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
864 }
865 
866 static void pci_init_wmask(PCIDevice *dev)
867 {
868     int config_size = pci_config_size(dev);
869 
870     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
871     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
872     pci_set_word(dev->wmask + PCI_COMMAND,
873                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
874                  PCI_COMMAND_INTX_DISABLE);
875     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
876 
877     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
878            config_size - PCI_CONFIG_HEADER_SIZE);
879 }
880 
881 static void pci_init_w1cmask(PCIDevice *dev)
882 {
883     /*
884      * Note: It's okay to set w1cmask even for readonly bits as
885      * long as their value is hardwired to 0.
886      */
887     pci_set_word(dev->w1cmask + PCI_STATUS,
888                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
889                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
890                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
891 }
892 
893 static void pci_init_mask_bridge(PCIDevice *d)
894 {
895     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
896        PCI_SEC_LATENCY_TIMER */
897     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
898 
899     /* base and limit */
900     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
901     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
902     pci_set_word(d->wmask + PCI_MEMORY_BASE,
903                  PCI_MEMORY_RANGE_MASK & 0xffff);
904     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
905                  PCI_MEMORY_RANGE_MASK & 0xffff);
906     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
907                  PCI_PREF_RANGE_MASK & 0xffff);
908     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
909                  PCI_PREF_RANGE_MASK & 0xffff);
910 
911     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
912     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
913 
914     /* Supported memory and i/o types */
915     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
916     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
917     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
918                                PCI_PREF_RANGE_TYPE_64);
919     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
920                                PCI_PREF_RANGE_TYPE_64);
921 
922     /*
923      * TODO: Bridges default to 10-bit VGA decoding but we currently only
924      * implement 16-bit decoding (no alias support).
925      */
926     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
927                  PCI_BRIDGE_CTL_PARITY |
928                  PCI_BRIDGE_CTL_SERR |
929                  PCI_BRIDGE_CTL_ISA |
930                  PCI_BRIDGE_CTL_VGA |
931                  PCI_BRIDGE_CTL_VGA_16BIT |
932                  PCI_BRIDGE_CTL_MASTER_ABORT |
933                  PCI_BRIDGE_CTL_BUS_RESET |
934                  PCI_BRIDGE_CTL_FAST_BACK |
935                  PCI_BRIDGE_CTL_DISCARD |
936                  PCI_BRIDGE_CTL_SEC_DISCARD |
937                  PCI_BRIDGE_CTL_DISCARD_SERR);
938     /* Below does not do anything as we never set this bit, put here for
939      * completeness. */
940     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
941                  PCI_BRIDGE_CTL_DISCARD_STATUS);
942     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
943     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
944     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
945                                PCI_PREF_RANGE_TYPE_MASK);
946     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
947                                PCI_PREF_RANGE_TYPE_MASK);
948 }
949 
950 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
951 {
952     uint8_t slot = PCI_SLOT(dev->devfn);
953     uint8_t func;
954 
955     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
956         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
957     }
958 
959     /*
960      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
961      * device, as it may just be a VF that ended up with function 0 in
962      * the legacy PCI interpretation. Avoid failing in such cases:
963      */
964     if (pci_is_vf(dev) &&
965         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
966         return;
967     }
968 
969     /*
970      * multifunction bit is interpreted in two ways as follows.
971      *   - all functions must set the bit to 1.
972      *     Example: Intel X53
973      *   - function 0 must set the bit, but the rest function (> 0)
974      *     is allowed to leave the bit to 0.
975      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
976      *
977      * So OS (at least Linux) checks the bit of only function 0,
978      * and doesn't see the bit of function > 0.
979      *
980      * The below check allows both interpretation.
981      */
982     if (PCI_FUNC(dev->devfn)) {
983         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
984         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
985             /* function 0 should set multifunction bit */
986             error_setg(errp, "PCI: single function device can't be populated "
987                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
988             return;
989         }
990         return;
991     }
992 
993     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
994         return;
995     }
996     /* function 0 indicates single function, so function > 0 must be NULL */
997     for (func = 1; func < PCI_FUNC_MAX; ++func) {
998         if (bus->devices[PCI_DEVFN(slot, func)]) {
999             error_setg(errp, "PCI: %x.0 indicates single function, "
1000                        "but %x.%x is already populated.",
1001                        slot, slot, func);
1002             return;
1003         }
1004     }
1005 }
1006 
1007 static void pci_config_alloc(PCIDevice *pci_dev)
1008 {
1009     int config_size = pci_config_size(pci_dev);
1010 
1011     pci_dev->config = g_malloc0(config_size);
1012     pci_dev->cmask = g_malloc0(config_size);
1013     pci_dev->wmask = g_malloc0(config_size);
1014     pci_dev->w1cmask = g_malloc0(config_size);
1015     pci_dev->used = g_malloc0(config_size);
1016 }
1017 
1018 static void pci_config_free(PCIDevice *pci_dev)
1019 {
1020     g_free(pci_dev->config);
1021     g_free(pci_dev->cmask);
1022     g_free(pci_dev->wmask);
1023     g_free(pci_dev->w1cmask);
1024     g_free(pci_dev->used);
1025 }
1026 
1027 static void do_pci_unregister_device(PCIDevice *pci_dev)
1028 {
1029     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1030     pci_config_free(pci_dev);
1031 
1032     if (xen_mode == XEN_EMULATE) {
1033         xen_evtchn_remove_pci_device(pci_dev);
1034     }
1035     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1036         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1037                                     &pci_dev->bus_master_enable_region);
1038     }
1039     address_space_destroy(&pci_dev->bus_master_as);
1040 }
1041 
1042 /* Extract PCIReqIDCache into BDF format */
1043 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1044 {
1045     uint8_t bus_n;
1046     uint16_t result;
1047 
1048     switch (cache->type) {
1049     case PCI_REQ_ID_BDF:
1050         result = pci_get_bdf(cache->dev);
1051         break;
1052     case PCI_REQ_ID_SECONDARY_BUS:
1053         bus_n = pci_dev_bus_num(cache->dev);
1054         result = PCI_BUILD_BDF(bus_n, 0);
1055         break;
1056     default:
1057         error_report("Invalid PCI requester ID cache type: %d",
1058                      cache->type);
1059         exit(1);
1060         break;
1061     }
1062 
1063     return result;
1064 }
1065 
1066 /* Parse bridges up to the root complex and return requester ID
1067  * cache for specific device.  For full PCIe topology, the cache
1068  * result would be exactly the same as getting BDF of the device.
1069  * However, several tricks are required when system mixed up with
1070  * legacy PCI devices and PCIe-to-PCI bridges.
1071  *
1072  * Here we cache the proxy device (and type) not requester ID since
1073  * bus number might change from time to time.
1074  */
1075 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1076 {
1077     PCIDevice *parent;
1078     PCIReqIDCache cache = {
1079         .dev = dev,
1080         .type = PCI_REQ_ID_BDF,
1081     };
1082 
1083     while (!pci_bus_is_root(pci_get_bus(dev))) {
1084         /* We are under PCI/PCIe bridges */
1085         parent = pci_get_bus(dev)->parent_dev;
1086         if (pci_is_express(parent)) {
1087             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1088                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1089                  * override the requester ID using secondary bus
1090                  * number of parent bridge with zeroed devfn
1091                  * (pcie-to-pci bridge spec chap 2.3). */
1092                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1093                 cache.dev = dev;
1094             }
1095         } else {
1096             /* Legacy PCI, override requester ID with the bridge's
1097              * BDF upstream.  When the root complex connects to
1098              * legacy PCI devices (including buses), it can only
1099              * obtain requester ID info from directly attached
1100              * devices.  If devices are attached under bridges, only
1101              * the requester ID of the bridge that is directly
1102              * attached to the root complex can be recognized. */
1103             cache.type = PCI_REQ_ID_BDF;
1104             cache.dev = parent;
1105         }
1106         dev = parent;
1107     }
1108 
1109     return cache;
1110 }
1111 
1112 uint16_t pci_requester_id(PCIDevice *dev)
1113 {
1114     return pci_req_id_cache_extract(&dev->requester_id_cache);
1115 }
1116 
1117 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1118 {
1119     return !(bus->devices[devfn]);
1120 }
1121 
1122 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1123 {
1124     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1125 }
1126 
1127 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1128 {
1129     return bus->slot_reserved_mask;
1130 }
1131 
1132 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1133 {
1134     bus->slot_reserved_mask |= mask;
1135 }
1136 
1137 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1138 {
1139     bus->slot_reserved_mask &= ~mask;
1140 }
1141 
1142 /* -1 for devfn means auto assign */
1143 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1144                                          const char *name, int devfn,
1145                                          Error **errp)
1146 {
1147     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1148     PCIConfigReadFunc *config_read = pc->config_read;
1149     PCIConfigWriteFunc *config_write = pc->config_write;
1150     Error *local_err = NULL;
1151     DeviceState *dev = DEVICE(pci_dev);
1152     PCIBus *bus = pci_get_bus(pci_dev);
1153     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1154 
1155     /* Only pci bridges can be attached to extra PCI root buses */
1156     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1157         error_setg(errp,
1158                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1159                     bus->parent_dev->name);
1160         return NULL;
1161     }
1162 
1163     if (devfn < 0) {
1164         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1165             devfn += PCI_FUNC_MAX) {
1166             if (pci_bus_devfn_available(bus, devfn) &&
1167                    !pci_bus_devfn_reserved(bus, devfn)) {
1168                 goto found;
1169             }
1170         }
1171         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1172                    "or reserved", name);
1173         return NULL;
1174     found: ;
1175     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1176         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1177                    " reserved",
1178                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1179         return NULL;
1180     } else if (!pci_bus_devfn_available(bus, devfn)) {
1181         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1182                    " in use by %s,id=%s",
1183                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1184                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1185         return NULL;
1186     } /*
1187        * Populating function 0 triggers a scan from the guest that
1188        * exposes other non-zero functions. Hence we need to ensure that
1189        * function 0 wasn't added yet.
1190        */
1191     else if (dev->hotplugged &&
1192              !pci_is_vf(pci_dev) &&
1193              pci_get_function_0(pci_dev)) {
1194         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1195                    " new func %s cannot be exposed to guest.",
1196                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1197                    pci_get_function_0(pci_dev)->name,
1198                    name);
1199 
1200        return NULL;
1201     }
1202 
1203     pci_dev->devfn = devfn;
1204     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1205     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1206 
1207     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1208                        "bus master container", UINT64_MAX);
1209     address_space_init(&pci_dev->bus_master_as,
1210                        &pci_dev->bus_master_container_region, pci_dev->name);
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 /* Initialize a PCI NIC.  */
1864 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1865                                const char *default_model,
1866                                const char *default_devaddr)
1867 {
1868     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1869     GPtrArray *pci_nic_models;
1870     PCIBus *bus;
1871     PCIDevice *pci_dev;
1872     DeviceState *dev;
1873     int devfn;
1874     int i;
1875     int dom, busnr;
1876     unsigned slot;
1877 
1878     if (nd->model && !strcmp(nd->model, "virtio")) {
1879         g_free(nd->model);
1880         nd->model = g_strdup("virtio-net-pci");
1881     }
1882 
1883     pci_nic_models = qemu_get_nic_models(TYPE_PCI_DEVICE);
1884 
1885     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1886         exit(0);
1887     }
1888 
1889     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1890                             default_model);
1891     if (i < 0) {
1892         exit(1);
1893     }
1894 
1895     if (!rootbus) {
1896         error_report("No primary PCI bus");
1897         exit(1);
1898     }
1899 
1900     assert(!rootbus->parent_dev);
1901 
1902     if (!devaddr) {
1903         devfn = -1;
1904         busnr = 0;
1905     } else {
1906         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1907             error_report("Invalid PCI device address %s for device %s",
1908                          devaddr, nd->model);
1909             exit(1);
1910         }
1911 
1912         if (dom != 0) {
1913             error_report("No support for non-zero PCI domains");
1914             exit(1);
1915         }
1916 
1917         devfn = PCI_DEVFN(slot, 0);
1918     }
1919 
1920     bus = pci_find_bus_nr(rootbus, busnr);
1921     if (!bus) {
1922         error_report("Invalid PCI device address %s for device %s",
1923                      devaddr, nd->model);
1924         exit(1);
1925     }
1926 
1927     pci_dev = pci_new(devfn, nd->model);
1928     dev = &pci_dev->qdev;
1929     qdev_set_nic_properties(dev, nd);
1930     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1931     g_ptr_array_free(pci_nic_models, true);
1932     return pci_dev;
1933 }
1934 
1935 PCIDevice *pci_vga_init(PCIBus *bus)
1936 {
1937     vga_interface_created = true;
1938     switch (vga_interface_type) {
1939     case VGA_CIRRUS:
1940         return pci_create_simple(bus, -1, "cirrus-vga");
1941     case VGA_QXL:
1942         return pci_create_simple(bus, -1, "qxl-vga");
1943     case VGA_STD:
1944         return pci_create_simple(bus, -1, "VGA");
1945     case VGA_VMWARE:
1946         return pci_create_simple(bus, -1, "vmware-svga");
1947     case VGA_VIRTIO:
1948         return pci_create_simple(bus, -1, "virtio-vga");
1949     case VGA_NONE:
1950     default: /* Other non-PCI types. Checking for unsupported types is already
1951                 done in vl.c. */
1952         return NULL;
1953     }
1954 }
1955 
1956 /* Whether a given bus number is in range of the secondary
1957  * bus of the given bridge device. */
1958 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1959 {
1960     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1961              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1962         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1963         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1964 }
1965 
1966 /* Whether a given bus number is in a range of a root bus */
1967 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1968 {
1969     int i;
1970 
1971     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1972         PCIDevice *dev = bus->devices[i];
1973 
1974         if (dev && IS_PCI_BRIDGE(dev)) {
1975             if (pci_secondary_bus_in_range(dev, bus_num)) {
1976                 return true;
1977             }
1978         }
1979     }
1980 
1981     return false;
1982 }
1983 
1984 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1985 {
1986     PCIBus *sec;
1987 
1988     if (!bus) {
1989         return NULL;
1990     }
1991 
1992     if (pci_bus_num(bus) == bus_num) {
1993         return bus;
1994     }
1995 
1996     /* Consider all bus numbers in range for the host pci bridge. */
1997     if (!pci_bus_is_root(bus) &&
1998         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1999         return NULL;
2000     }
2001 
2002     /* try child bus */
2003     for (; bus; bus = sec) {
2004         QLIST_FOREACH(sec, &bus->child, sibling) {
2005             if (pci_bus_num(sec) == bus_num) {
2006                 return sec;
2007             }
2008             /* PXB buses assumed to be children of bus 0 */
2009             if (pci_bus_is_root(sec)) {
2010                 if (pci_root_bus_in_range(sec, bus_num)) {
2011                     break;
2012                 }
2013             } else {
2014                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2015                     break;
2016                 }
2017             }
2018         }
2019     }
2020 
2021     return NULL;
2022 }
2023 
2024 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
2025                                   pci_bus_fn end, void *parent_state)
2026 {
2027     PCIBus *sec;
2028     void *state;
2029 
2030     if (!bus) {
2031         return;
2032     }
2033 
2034     if (begin) {
2035         state = begin(bus, parent_state);
2036     } else {
2037         state = parent_state;
2038     }
2039 
2040     QLIST_FOREACH(sec, &bus->child, sibling) {
2041         pci_for_each_bus_depth_first(sec, begin, end, state);
2042     }
2043 
2044     if (end) {
2045         end(bus, state);
2046     }
2047 }
2048 
2049 
2050 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2051 {
2052     bus = pci_find_bus_nr(bus, bus_num);
2053 
2054     if (!bus)
2055         return NULL;
2056 
2057     return bus->devices[devfn];
2058 }
2059 
2060 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2061 
2062 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2063 {
2064     PCIDevice *pci_dev = (PCIDevice *)qdev;
2065     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2066     ObjectClass *klass = OBJECT_CLASS(pc);
2067     Error *local_err = NULL;
2068     bool is_default_rom;
2069     uint16_t class_id;
2070 
2071     /*
2072      * capped by systemd (see: udev-builtin-net_id.c)
2073      * as it's the only known user honor it to avoid users
2074      * misconfigure QEMU and then wonder why acpi-index doesn't work
2075      */
2076     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2077         error_setg(errp, "acpi-index should be less or equal to %u",
2078                    ONBOARD_INDEX_MAX);
2079         return;
2080     }
2081 
2082     /*
2083      * make sure that acpi-index is unique across all present PCI devices
2084      */
2085     if (pci_dev->acpi_index) {
2086         GSequence *used_indexes = pci_acpi_index_list();
2087 
2088         if (g_sequence_lookup(used_indexes,
2089                               GINT_TO_POINTER(pci_dev->acpi_index),
2090                               g_cmp_uint32, NULL)) {
2091             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2092                        " already exist", pci_dev->acpi_index);
2093             return;
2094         }
2095         g_sequence_insert_sorted(used_indexes,
2096                                  GINT_TO_POINTER(pci_dev->acpi_index),
2097                                  g_cmp_uint32, NULL);
2098     }
2099 
2100     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2101         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2102         return;
2103     }
2104 
2105     /* initialize cap_present for pci_is_express() and pci_config_size(),
2106      * Note that hybrid PCIs are not set automatically and need to manage
2107      * QEMU_PCI_CAP_EXPRESS manually */
2108     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2109        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2110         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2111     }
2112 
2113     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2114         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2115     }
2116 
2117     pci_dev = do_pci_register_device(pci_dev,
2118                                      object_get_typename(OBJECT(qdev)),
2119                                      pci_dev->devfn, errp);
2120     if (pci_dev == NULL)
2121         return;
2122 
2123     if (pc->realize) {
2124         pc->realize(pci_dev, &local_err);
2125         if (local_err) {
2126             error_propagate(errp, local_err);
2127             do_pci_unregister_device(pci_dev);
2128             return;
2129         }
2130     }
2131 
2132     /*
2133      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2134      * associate only Device 0 with the device attached to the bus
2135      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2136      * sec 7.3.1).
2137      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2138      * 5-bit Device Number and 3-bit Function Number fields in its associated
2139      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2140      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2141      */
2142     if (pci_is_express(pci_dev) &&
2143         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2144         pcie_has_upstream_port(pci_dev) &&
2145         PCI_SLOT(pci_dev->devfn)) {
2146         warn_report("PCI: slot %d is not valid for %s,"
2147                     " parent device only allows plugging into slot 0.",
2148                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2149     }
2150 
2151     if (pci_dev->failover_pair_id) {
2152         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2153             error_setg(errp, "failover primary device must be on "
2154                              "PCIExpress bus");
2155             pci_qdev_unrealize(DEVICE(pci_dev));
2156             return;
2157         }
2158         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2159         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2160             error_setg(errp, "failover primary device is not an "
2161                              "Ethernet device");
2162             pci_qdev_unrealize(DEVICE(pci_dev));
2163             return;
2164         }
2165         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2166             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2167             error_setg(errp, "failover: primary device must be in its own "
2168                               "PCI slot");
2169             pci_qdev_unrealize(DEVICE(pci_dev));
2170             return;
2171         }
2172         qdev->allow_unplug_during_migration = true;
2173     }
2174 
2175     /* rom loading */
2176     is_default_rom = false;
2177     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2178         pci_dev->romfile = g_strdup(pc->romfile);
2179         is_default_rom = true;
2180     }
2181 
2182     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2183     if (local_err) {
2184         error_propagate(errp, local_err);
2185         pci_qdev_unrealize(DEVICE(pci_dev));
2186         return;
2187     }
2188 
2189     pci_set_power(pci_dev, true);
2190 
2191     pci_dev->msi_trigger = pci_msi_trigger;
2192 }
2193 
2194 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2195                                    const char *name)
2196 {
2197     DeviceState *dev;
2198 
2199     dev = qdev_new(name);
2200     qdev_prop_set_int32(dev, "addr", devfn);
2201     qdev_prop_set_bit(dev, "multifunction", multifunction);
2202     return PCI_DEVICE(dev);
2203 }
2204 
2205 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2206 {
2207     return pci_new_internal(devfn, true, name);
2208 }
2209 
2210 PCIDevice *pci_new(int devfn, const char *name)
2211 {
2212     return pci_new_internal(devfn, false, name);
2213 }
2214 
2215 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2216 {
2217     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2218 }
2219 
2220 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2221                                            const char *name)
2222 {
2223     PCIDevice *dev = pci_new_multifunction(devfn, name);
2224     pci_realize_and_unref(dev, bus, &error_fatal);
2225     return dev;
2226 }
2227 
2228 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2229 {
2230     PCIDevice *dev = pci_new(devfn, name);
2231     pci_realize_and_unref(dev, bus, &error_fatal);
2232     return dev;
2233 }
2234 
2235 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2236 {
2237     int offset = PCI_CONFIG_HEADER_SIZE;
2238     int i;
2239     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2240         if (pdev->used[i])
2241             offset = i + 1;
2242         else if (i - offset + 1 == size)
2243             return offset;
2244     }
2245     return 0;
2246 }
2247 
2248 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2249                                         uint8_t *prev_p)
2250 {
2251     uint8_t next, prev;
2252 
2253     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2254         return 0;
2255 
2256     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2257          prev = next + PCI_CAP_LIST_NEXT)
2258         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2259             break;
2260 
2261     if (prev_p)
2262         *prev_p = prev;
2263     return next;
2264 }
2265 
2266 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2267 {
2268     uint8_t next, prev, found = 0;
2269 
2270     if (!(pdev->used[offset])) {
2271         return 0;
2272     }
2273 
2274     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2275 
2276     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2277          prev = next + PCI_CAP_LIST_NEXT) {
2278         if (next <= offset && next > found) {
2279             found = next;
2280         }
2281     }
2282     return found;
2283 }
2284 
2285 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2286    This is needed for an option rom which is used for more than one device. */
2287 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2288 {
2289     uint16_t vendor_id;
2290     uint16_t device_id;
2291     uint16_t rom_vendor_id;
2292     uint16_t rom_device_id;
2293     uint16_t rom_magic;
2294     uint16_t pcir_offset;
2295     uint8_t checksum;
2296 
2297     /* Words in rom data are little endian (like in PCI configuration),
2298        so they can be read / written with pci_get_word / pci_set_word. */
2299 
2300     /* Only a valid rom will be patched. */
2301     rom_magic = pci_get_word(ptr);
2302     if (rom_magic != 0xaa55) {
2303         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2304         return;
2305     }
2306     pcir_offset = pci_get_word(ptr + 0x18);
2307     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2308         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2309         return;
2310     }
2311 
2312     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2313     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2314     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2315     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2316 
2317     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2318                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2319 
2320     checksum = ptr[6];
2321 
2322     if (vendor_id != rom_vendor_id) {
2323         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2324         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2325         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2326         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2327         ptr[6] = checksum;
2328         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2329     }
2330 
2331     if (device_id != rom_device_id) {
2332         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2333         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2334         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2335         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2336         ptr[6] = checksum;
2337         pci_set_word(ptr + pcir_offset + 6, device_id);
2338     }
2339 }
2340 
2341 /* Add an option rom for the device */
2342 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2343                                Error **errp)
2344 {
2345     int64_t size = 0;
2346     g_autofree char *path = NULL;
2347     char name[32];
2348     const VMStateDescription *vmsd;
2349 
2350     /*
2351      * In case of incoming migration ROM will come with migration stream, no
2352      * reason to load the file.  Neither we want to fail if local ROM file
2353      * mismatches with specified romsize.
2354      */
2355     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2356 
2357     if (!pdev->romfile || !strlen(pdev->romfile)) {
2358         return;
2359     }
2360 
2361     if (!pdev->rom_bar) {
2362         /*
2363          * Load rom via fw_cfg instead of creating a rom bar,
2364          * for 0.11 compatibility.
2365          */
2366         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2367 
2368         /*
2369          * Hot-plugged devices can't use the option ROM
2370          * if the rom bar is disabled.
2371          */
2372         if (DEVICE(pdev)->hotplugged) {
2373             error_setg(errp, "Hot-plugged device without ROM bar"
2374                        " can't have an option ROM");
2375             return;
2376         }
2377 
2378         if (class == 0x0300) {
2379             rom_add_vga(pdev->romfile);
2380         } else {
2381             rom_add_option(pdev->romfile, -1);
2382         }
2383         return;
2384     }
2385 
2386     if (load_file || pdev->romsize == -1) {
2387         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2388         if (path == NULL) {
2389             path = g_strdup(pdev->romfile);
2390         }
2391 
2392         size = get_image_size(path);
2393         if (size < 0) {
2394             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2395             return;
2396         } else if (size == 0) {
2397             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2398             return;
2399         } else if (size > 2 * GiB) {
2400             error_setg(errp,
2401                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2402                        pdev->romfile);
2403             return;
2404         }
2405         if (pdev->romsize != -1) {
2406             if (size > pdev->romsize) {
2407                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2408                            "is too large for ROM size %u",
2409                            pdev->romfile, (uint32_t)size, pdev->romsize);
2410                 return;
2411             }
2412         } else {
2413             pdev->romsize = pow2ceil(size);
2414         }
2415     }
2416 
2417     vmsd = qdev_get_vmsd(DEVICE(pdev));
2418     snprintf(name, sizeof(name), "%s.rom",
2419              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2420 
2421     pdev->has_rom = true;
2422     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2423                            &error_fatal);
2424 
2425     if (load_file) {
2426         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2427 
2428         if (load_image_size(path, ptr, size) < 0) {
2429             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2430             return;
2431         }
2432 
2433         if (is_default_rom) {
2434             /* Only the default rom images will be patched (if needed). */
2435             pci_patch_ids(pdev, ptr, size);
2436         }
2437     }
2438 
2439     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2440 }
2441 
2442 static void pci_del_option_rom(PCIDevice *pdev)
2443 {
2444     if (!pdev->has_rom)
2445         return;
2446 
2447     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2448     pdev->has_rom = false;
2449 }
2450 
2451 /*
2452  * On success, pci_add_capability() returns a positive value
2453  * that the offset of the pci capability.
2454  * On failure, it sets an error and returns a negative error
2455  * code.
2456  */
2457 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2458                        uint8_t offset, uint8_t size,
2459                        Error **errp)
2460 {
2461     uint8_t *config;
2462     int i, overlapping_cap;
2463 
2464     if (!offset) {
2465         offset = pci_find_space(pdev, size);
2466         /* out of PCI config space is programming error */
2467         assert(offset);
2468     } else {
2469         /* Verify that capabilities don't overlap.  Note: device assignment
2470          * depends on this check to verify that the device is not broken.
2471          * Should never trigger for emulated devices, but it's helpful
2472          * for debugging these. */
2473         for (i = offset; i < offset + size; i++) {
2474             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2475             if (overlapping_cap) {
2476                 error_setg(errp, "%s:%02x:%02x.%x "
2477                            "Attempt to add PCI capability %x at offset "
2478                            "%x overlaps existing capability %x at offset %x",
2479                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2480                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2481                            cap_id, offset, overlapping_cap, i);
2482                 return -EINVAL;
2483             }
2484         }
2485     }
2486 
2487     config = pdev->config + offset;
2488     config[PCI_CAP_LIST_ID] = cap_id;
2489     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2490     pdev->config[PCI_CAPABILITY_LIST] = offset;
2491     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2492     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2493     /* Make capability read-only by default */
2494     memset(pdev->wmask + offset, 0, size);
2495     /* Check capability by default */
2496     memset(pdev->cmask + offset, 0xFF, size);
2497     return offset;
2498 }
2499 
2500 /* Unlink capability from the pci config space. */
2501 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2502 {
2503     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2504     if (!offset)
2505         return;
2506     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2507     /* Make capability writable again */
2508     memset(pdev->wmask + offset, 0xff, size);
2509     memset(pdev->w1cmask + offset, 0, size);
2510     /* Clear cmask as device-specific registers can't be checked */
2511     memset(pdev->cmask + offset, 0, size);
2512     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2513 
2514     if (!pdev->config[PCI_CAPABILITY_LIST])
2515         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2516 }
2517 
2518 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2519 {
2520     return pci_find_capability_list(pdev, cap_id, NULL);
2521 }
2522 
2523 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2524 {
2525     PCIDevice *d = (PCIDevice *)dev;
2526     const char *name = NULL;
2527     const pci_class_desc *desc =  pci_class_descriptions;
2528     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2529 
2530     while (desc->desc &&
2531           (class & ~desc->fw_ign_bits) !=
2532           (desc->class & ~desc->fw_ign_bits)) {
2533         desc++;
2534     }
2535 
2536     if (desc->desc) {
2537         name = desc->fw_name;
2538     }
2539 
2540     if (name) {
2541         pstrcpy(buf, len, name);
2542     } else {
2543         snprintf(buf, len, "pci%04x,%04x",
2544                  pci_get_word(d->config + PCI_VENDOR_ID),
2545                  pci_get_word(d->config + PCI_DEVICE_ID));
2546     }
2547 
2548     return buf;
2549 }
2550 
2551 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2552 {
2553     PCIDevice *d = (PCIDevice *)dev;
2554     char name[33];
2555     int has_func = !!PCI_FUNC(d->devfn);
2556 
2557     return g_strdup_printf("%s@%x%s%.*x",
2558                            pci_dev_fw_name(dev, name, sizeof(name)),
2559                            PCI_SLOT(d->devfn),
2560                            has_func ? "," : "",
2561                            has_func,
2562                            PCI_FUNC(d->devfn));
2563 }
2564 
2565 static char *pcibus_get_dev_path(DeviceState *dev)
2566 {
2567     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2568     PCIDevice *t;
2569     int slot_depth;
2570     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2571      * 00 is added here to make this format compatible with
2572      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2573      * Slot.Function list specifies the slot and function numbers for all
2574      * devices on the path from root to the specific device. */
2575     const char *root_bus_path;
2576     int root_bus_len;
2577     char slot[] = ":SS.F";
2578     int slot_len = sizeof slot - 1 /* For '\0' */;
2579     int path_len;
2580     char *path, *p;
2581     int s;
2582 
2583     root_bus_path = pci_root_bus_path(d);
2584     root_bus_len = strlen(root_bus_path);
2585 
2586     /* Calculate # of slots on path between device and root. */;
2587     slot_depth = 0;
2588     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2589         ++slot_depth;
2590     }
2591 
2592     path_len = root_bus_len + slot_len * slot_depth;
2593 
2594     /* Allocate memory, fill in the terminating null byte. */
2595     path = g_malloc(path_len + 1 /* For '\0' */);
2596     path[path_len] = '\0';
2597 
2598     memcpy(path, root_bus_path, root_bus_len);
2599 
2600     /* Fill in slot numbers. We walk up from device to root, so need to print
2601      * them in the reverse order, last to first. */
2602     p = path + path_len;
2603     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2604         p -= slot_len;
2605         s = snprintf(slot, sizeof slot, ":%02x.%x",
2606                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2607         assert(s == slot_len);
2608         memcpy(p, slot, slot_len);
2609     }
2610 
2611     return path;
2612 }
2613 
2614 static int pci_qdev_find_recursive(PCIBus *bus,
2615                                    const char *id, PCIDevice **pdev)
2616 {
2617     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2618     if (!qdev) {
2619         return -ENODEV;
2620     }
2621 
2622     /* roughly check if given qdev is pci device */
2623     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2624         *pdev = PCI_DEVICE(qdev);
2625         return 0;
2626     }
2627     return -EINVAL;
2628 }
2629 
2630 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2631 {
2632     PCIHostState *host_bridge;
2633     int rc = -ENODEV;
2634 
2635     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2636         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2637         if (!tmp) {
2638             rc = 0;
2639             break;
2640         }
2641         if (tmp != -ENODEV) {
2642             rc = tmp;
2643         }
2644     }
2645 
2646     return rc;
2647 }
2648 
2649 MemoryRegion *pci_address_space(PCIDevice *dev)
2650 {
2651     return pci_get_bus(dev)->address_space_mem;
2652 }
2653 
2654 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2655 {
2656     return pci_get_bus(dev)->address_space_io;
2657 }
2658 
2659 static void pci_device_class_init(ObjectClass *klass, void *data)
2660 {
2661     DeviceClass *k = DEVICE_CLASS(klass);
2662 
2663     k->realize = pci_qdev_realize;
2664     k->unrealize = pci_qdev_unrealize;
2665     k->bus_type = TYPE_PCI_BUS;
2666     device_class_set_props(k, pci_props);
2667 }
2668 
2669 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2670 {
2671     if (!object_class_is_abstract(klass)) {
2672         ObjectClass *conventional =
2673             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2674         ObjectClass *pcie =
2675             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2676         ObjectClass *cxl =
2677             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2678         assert(conventional || pcie || cxl);
2679     }
2680 }
2681 
2682 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2683 {
2684     PCIBus *bus = pci_get_bus(dev);
2685     PCIBus *iommu_bus = bus;
2686     uint8_t devfn = dev->devfn;
2687 
2688     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2689         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2690 
2691         /*
2692          * The requester ID of the provided device may be aliased, as seen from
2693          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2694          * requester ID to provide a unique AddressSpace for devices, but
2695          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2696          * to-PCI bridge creates and accepts transactions on behalf of down-
2697          * stream devices.  When doing so, all downstream devices are masked
2698          * (aliased) behind a single requester ID.  The requester ID used
2699          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2700          * bridges, with a PCIe capability indicating such, follow the
2701          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2702          * where the bridge uses the seconary bus as the bridge portion of the
2703          * requester ID and devfn of 00.0.  For other bridges, typically those
2704          * found on the root complex such as the dmi-to-pci-bridge, we follow
2705          * the convention of typical bare-metal hardware, which uses the
2706          * requester ID of the bridge itself.  There are device specific
2707          * exceptions to these rules, but these are the defaults that the
2708          * Linux kernel uses when determining DMA aliases itself and believed
2709          * to be true for the bare metal equivalents of the devices emulated
2710          * in QEMU.
2711          */
2712         if (!pci_bus_is_express(iommu_bus)) {
2713             PCIDevice *parent = iommu_bus->parent_dev;
2714 
2715             if (pci_is_express(parent) &&
2716                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2717                 devfn = PCI_DEVFN(0, 0);
2718                 bus = iommu_bus;
2719             } else {
2720                 devfn = parent->devfn;
2721                 bus = parent_bus;
2722             }
2723         }
2724 
2725         iommu_bus = parent_bus;
2726     }
2727     if (!pci_bus_bypass_iommu(bus) && iommu_bus && iommu_bus->iommu_fn) {
2728         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2729     }
2730     return &address_space_memory;
2731 }
2732 
2733 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2734 {
2735     bus->iommu_fn = fn;
2736     bus->iommu_opaque = opaque;
2737 }
2738 
2739 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2740 {
2741     Range *range = opaque;
2742     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2743     int i;
2744 
2745     if (!(cmd & PCI_COMMAND_MEMORY)) {
2746         return;
2747     }
2748 
2749     if (IS_PCI_BRIDGE(dev)) {
2750         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2751         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2752 
2753         base = MAX(base, 0x1ULL << 32);
2754 
2755         if (limit >= base) {
2756             Range pref_range;
2757             range_set_bounds(&pref_range, base, limit);
2758             range_extend(range, &pref_range);
2759         }
2760     }
2761     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2762         PCIIORegion *r = &dev->io_regions[i];
2763         pcibus_t lob, upb;
2764         Range region_range;
2765 
2766         if (!r->size ||
2767             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2768             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2769             continue;
2770         }
2771 
2772         lob = pci_bar_address(dev, i, r->type, r->size);
2773         upb = lob + r->size - 1;
2774         if (lob == PCI_BAR_UNMAPPED) {
2775             continue;
2776         }
2777 
2778         lob = MAX(lob, 0x1ULL << 32);
2779 
2780         if (upb >= lob) {
2781             range_set_bounds(&region_range, lob, upb);
2782             range_extend(range, &region_range);
2783         }
2784     }
2785 }
2786 
2787 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2788 {
2789     range_make_empty(range);
2790     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2791 }
2792 
2793 static bool pcie_has_upstream_port(PCIDevice *dev)
2794 {
2795     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2796 
2797     /* Device associated with an upstream port.
2798      * As there are several types of these, it's easier to check the
2799      * parent device: upstream ports are always connected to
2800      * root or downstream ports.
2801      */
2802     return parent_dev &&
2803         pci_is_express(parent_dev) &&
2804         parent_dev->exp.exp_cap &&
2805         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2806          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2807 }
2808 
2809 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2810 {
2811     PCIBus *bus = pci_get_bus(pci_dev);
2812 
2813     if(pcie_has_upstream_port(pci_dev)) {
2814         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2815         return bus->devices[0];
2816     } else {
2817         /* Other bus types might support multiple devices at slots 0-31 */
2818         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2819     }
2820 }
2821 
2822 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2823 {
2824     MSIMessage msg;
2825     if (msix_enabled(dev)) {
2826         msg = msix_get_message(dev, vector);
2827     } else if (msi_enabled(dev)) {
2828         msg = msi_get_message(dev, vector);
2829     } else {
2830         /* Should never happen */
2831         error_report("%s: unknown interrupt type", __func__);
2832         abort();
2833     }
2834     return msg;
2835 }
2836 
2837 void pci_set_power(PCIDevice *d, bool state)
2838 {
2839     if (d->has_power == state) {
2840         return;
2841     }
2842 
2843     d->has_power = state;
2844     pci_update_mappings(d);
2845     memory_region_set_enabled(&d->bus_master_enable_region,
2846                               (pci_get_word(d->config + PCI_COMMAND)
2847                                & PCI_COMMAND_MASTER) && d->has_power);
2848     if (!d->has_power) {
2849         pci_device_reset(d);
2850     }
2851 }
2852 
2853 static const TypeInfo pci_device_type_info = {
2854     .name = TYPE_PCI_DEVICE,
2855     .parent = TYPE_DEVICE,
2856     .instance_size = sizeof(PCIDevice),
2857     .abstract = true,
2858     .class_size = sizeof(PCIDeviceClass),
2859     .class_init = pci_device_class_init,
2860     .class_base_init = pci_device_class_base_init,
2861 };
2862 
2863 static void pci_register_types(void)
2864 {
2865     type_register_static(&pci_bus_info);
2866     type_register_static(&pcie_bus_info);
2867     type_register_static(&cxl_bus_info);
2868     type_register_static(&conventional_pci_interface_info);
2869     type_register_static(&cxl_interface_info);
2870     type_register_static(&pcie_interface_info);
2871     type_register_static(&pci_device_type_info);
2872 }
2873 
2874 type_init(pci_register_types)
2875