xref: /openbmc/qemu/hw/pci/pci.c (revision 139610ae)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu/datadir.h"
27 #include "qemu/units.h"
28 #include "hw/irq.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bridge.h"
31 #include "hw/pci/pci_bus.h"
32 #include "hw/pci/pci_host.h"
33 #include "hw/qdev-properties.h"
34 #include "hw/qdev-properties-system.h"
35 #include "migration/qemu-file-types.h"
36 #include "migration/vmstate.h"
37 #include "net/net.h"
38 #include "sysemu/numa.h"
39 #include "sysemu/runstate.h"
40 #include "sysemu/sysemu.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "qapi/error.h"
50 #include "qemu/cutils.h"
51 #include "pci-internal.h"
52 
53 #include "hw/xen/xen.h"
54 #include "hw/i386/kvm/xen_evtchn.h"
55 
56 //#define DEBUG_PCI
57 #ifdef DEBUG_PCI
58 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
59 #else
60 # define PCI_DPRINTF(format, ...)       do { } while (0)
61 #endif
62 
63 bool pci_available = true;
64 
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset_hold(Object *obj, ResetType type);
68 static bool pcie_has_upstream_port(PCIDevice *dev);
69 
70 static Property pci_props[] = {
71     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
72     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
73     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, -1),
74     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
75     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
76                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
77     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
78                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
79     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
80                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
81     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
82                        failover_pair_id),
83     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
84     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
85                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
86     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
87                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
88     DEFINE_PROP_END_OF_LIST()
89 };
90 
91 static const VMStateDescription vmstate_pcibus = {
92     .name = "PCIBUS",
93     .version_id = 1,
94     .minimum_version_id = 1,
95     .fields = (const VMStateField[]) {
96         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
97         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
98                              nirq, 0, vmstate_info_int32,
99                              int32_t),
100         VMSTATE_END_OF_LIST()
101     }
102 };
103 
104 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
105 {
106     return a - b;
107 }
108 
109 static GSequence *pci_acpi_index_list(void)
110 {
111     static GSequence *used_acpi_index_list;
112 
113     if (!used_acpi_index_list) {
114         used_acpi_index_list = g_sequence_new(NULL);
115     }
116     return used_acpi_index_list;
117 }
118 
119 static void pci_init_bus_master(PCIDevice *pci_dev)
120 {
121     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
122 
123     memory_region_init_alias(&pci_dev->bus_master_enable_region,
124                              OBJECT(pci_dev), "bus master",
125                              dma_as->root, 0, memory_region_size(dma_as->root));
126     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
127     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
128                                 &pci_dev->bus_master_enable_region);
129 }
130 
131 static void pcibus_machine_done(Notifier *notifier, void *data)
132 {
133     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
134     int i;
135 
136     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
137         if (bus->devices[i]) {
138             pci_init_bus_master(bus->devices[i]);
139         }
140     }
141 }
142 
143 static void pci_bus_realize(BusState *qbus, Error **errp)
144 {
145     PCIBus *bus = PCI_BUS(qbus);
146 
147     bus->machine_done.notify = pcibus_machine_done;
148     qemu_add_machine_init_done_notifier(&bus->machine_done);
149 
150     vmstate_register_any(NULL, &vmstate_pcibus, bus);
151 }
152 
153 static void pcie_bus_realize(BusState *qbus, Error **errp)
154 {
155     PCIBus *bus = PCI_BUS(qbus);
156     Error *local_err = NULL;
157 
158     pci_bus_realize(qbus, &local_err);
159     if (local_err) {
160         error_propagate(errp, local_err);
161         return;
162     }
163 
164     /*
165      * A PCI-E bus can support extended config space if it's the root
166      * bus, or if the bus/bridge above it does as well
167      */
168     if (pci_bus_is_root(bus)) {
169         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
170     } else {
171         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
172 
173         if (pci_bus_allows_extended_config_space(parent_bus)) {
174             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
175         }
176     }
177 }
178 
179 static void pci_bus_unrealize(BusState *qbus)
180 {
181     PCIBus *bus = PCI_BUS(qbus);
182 
183     qemu_remove_machine_init_done_notifier(&bus->machine_done);
184 
185     vmstate_unregister(NULL, &vmstate_pcibus, bus);
186 }
187 
188 static int pcibus_num(PCIBus *bus)
189 {
190     if (pci_bus_is_root(bus)) {
191         return 0; /* pci host bridge */
192     }
193     return bus->parent_dev->config[PCI_SECONDARY_BUS];
194 }
195 
196 static uint16_t pcibus_numa_node(PCIBus *bus)
197 {
198     return NUMA_NODE_UNASSIGNED;
199 }
200 
201 static void pci_bus_class_init(ObjectClass *klass, void *data)
202 {
203     BusClass *k = BUS_CLASS(klass);
204     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
205     ResettableClass *rc = RESETTABLE_CLASS(klass);
206 
207     k->print_dev = pcibus_dev_print;
208     k->get_dev_path = pcibus_get_dev_path;
209     k->get_fw_dev_path = pcibus_get_fw_dev_path;
210     k->realize = pci_bus_realize;
211     k->unrealize = pci_bus_unrealize;
212 
213     rc->phases.hold = pcibus_reset_hold;
214 
215     pbc->bus_num = pcibus_num;
216     pbc->numa_node = pcibus_numa_node;
217 }
218 
219 static const TypeInfo pci_bus_info = {
220     .name = TYPE_PCI_BUS,
221     .parent = TYPE_BUS,
222     .instance_size = sizeof(PCIBus),
223     .class_size = sizeof(PCIBusClass),
224     .class_init = pci_bus_class_init,
225 };
226 
227 static const TypeInfo cxl_interface_info = {
228     .name          = INTERFACE_CXL_DEVICE,
229     .parent        = TYPE_INTERFACE,
230 };
231 
232 static const TypeInfo pcie_interface_info = {
233     .name          = INTERFACE_PCIE_DEVICE,
234     .parent        = TYPE_INTERFACE,
235 };
236 
237 static const TypeInfo conventional_pci_interface_info = {
238     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
239     .parent        = TYPE_INTERFACE,
240 };
241 
242 static void pcie_bus_class_init(ObjectClass *klass, void *data)
243 {
244     BusClass *k = BUS_CLASS(klass);
245 
246     k->realize = pcie_bus_realize;
247 }
248 
249 static const TypeInfo pcie_bus_info = {
250     .name = TYPE_PCIE_BUS,
251     .parent = TYPE_PCI_BUS,
252     .class_init = pcie_bus_class_init,
253 };
254 
255 static const TypeInfo cxl_bus_info = {
256     .name       = TYPE_CXL_BUS,
257     .parent     = TYPE_PCIE_BUS,
258     .class_init = pcie_bus_class_init,
259 };
260 
261 static void pci_update_mappings(PCIDevice *d);
262 static void pci_irq_handler(void *opaque, int irq_num, int level);
263 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
264 static void pci_del_option_rom(PCIDevice *pdev);
265 
266 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
267 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
268 
269 PCIHostStateList pci_host_bridges;
270 
271 int pci_bar(PCIDevice *d, int reg)
272 {
273     uint8_t type;
274 
275     /* PCIe virtual functions do not have their own BARs */
276     assert(!pci_is_vf(d));
277 
278     if (reg != PCI_ROM_SLOT)
279         return PCI_BASE_ADDRESS_0 + reg * 4;
280 
281     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
282     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
283 }
284 
285 static inline int pci_irq_state(PCIDevice *d, int irq_num)
286 {
287         return (d->irq_state >> irq_num) & 0x1;
288 }
289 
290 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
291 {
292         d->irq_state &= ~(0x1 << irq_num);
293         d->irq_state |= level << irq_num;
294 }
295 
296 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
297 {
298     assert(irq_num >= 0);
299     assert(irq_num < bus->nirq);
300     bus->irq_count[irq_num] += change;
301     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
302 }
303 
304 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
305 {
306     PCIBus *bus;
307     for (;;) {
308         int dev_irq = irq_num;
309         bus = pci_get_bus(pci_dev);
310         assert(bus->map_irq);
311         irq_num = bus->map_irq(pci_dev, irq_num);
312         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
313                             pci_bus_is_root(bus) ? "root-complex"
314                                     : DEVICE(bus->parent_dev)->canonical_path);
315         if (bus->set_irq)
316             break;
317         pci_dev = bus->parent_dev;
318     }
319     pci_bus_change_irq_level(bus, irq_num, change);
320 }
321 
322 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
323 {
324     assert(irq_num >= 0);
325     assert(irq_num < bus->nirq);
326     return !!bus->irq_count[irq_num];
327 }
328 
329 /* Update interrupt status bit in config space on interrupt
330  * state change. */
331 static void pci_update_irq_status(PCIDevice *dev)
332 {
333     if (dev->irq_state) {
334         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
335     } else {
336         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
337     }
338 }
339 
340 void pci_device_deassert_intx(PCIDevice *dev)
341 {
342     int i;
343     for (i = 0; i < PCI_NUM_PINS; ++i) {
344         pci_irq_handler(dev, i, 0);
345     }
346 }
347 
348 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
349 {
350     MemTxAttrs attrs = {};
351 
352     /*
353      * Xen uses the high bits of the address to contain some of the bits
354      * of the PIRQ#. Therefore we can't just send the write cycle and
355      * trust that it's caught by the APIC at 0xfee00000 because the
356      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
357      * So we intercept the delivery here instead of in kvm_send_msi().
358      */
359     if (xen_mode == XEN_EMULATE &&
360         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
361         return;
362     }
363     attrs.requester_id = pci_requester_id(dev);
364     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
365                          attrs, NULL);
366 }
367 
368 static void pci_reset_regions(PCIDevice *dev)
369 {
370     int r;
371     if (pci_is_vf(dev)) {
372         return;
373     }
374 
375     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
376         PCIIORegion *region = &dev->io_regions[r];
377         if (!region->size) {
378             continue;
379         }
380 
381         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
382             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
383             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
384         } else {
385             pci_set_long(dev->config + pci_bar(dev, r), region->type);
386         }
387     }
388 }
389 
390 static void pci_do_device_reset(PCIDevice *dev)
391 {
392     pci_device_deassert_intx(dev);
393     assert(dev->irq_state == 0);
394 
395     /* Clear all writable bits */
396     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
397                                  pci_get_word(dev->wmask + PCI_COMMAND) |
398                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
399     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
400                                  pci_get_word(dev->wmask + PCI_STATUS) |
401                                  pci_get_word(dev->w1cmask + PCI_STATUS));
402     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
403     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
404                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
405                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
406     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
407     pci_reset_regions(dev);
408     pci_update_mappings(dev);
409 
410     msi_reset(dev);
411     msix_reset(dev);
412     pcie_sriov_pf_reset(dev);
413 }
414 
415 /*
416  * This function is called on #RST and FLR.
417  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
418  */
419 void pci_device_reset(PCIDevice *dev)
420 {
421     device_cold_reset(&dev->qdev);
422     pci_do_device_reset(dev);
423 }
424 
425 /*
426  * Trigger pci bus reset under a given bus.
427  * Called via bus_cold_reset on RST# assert, after the devices
428  * have been reset device_cold_reset-ed already.
429  */
430 static void pcibus_reset_hold(Object *obj, ResetType type)
431 {
432     PCIBus *bus = PCI_BUS(obj);
433     int i;
434 
435     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
436         if (bus->devices[i]) {
437             pci_do_device_reset(bus->devices[i]);
438         }
439     }
440 
441     for (i = 0; i < bus->nirq; i++) {
442         assert(bus->irq_count[i] == 0);
443     }
444 }
445 
446 static void pci_host_bus_register(DeviceState *host)
447 {
448     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
449 
450     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
451 }
452 
453 static void pci_host_bus_unregister(DeviceState *host)
454 {
455     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
456 
457     QLIST_REMOVE(host_bridge, next);
458 }
459 
460 PCIBus *pci_device_root_bus(const PCIDevice *d)
461 {
462     PCIBus *bus = pci_get_bus(d);
463 
464     while (!pci_bus_is_root(bus)) {
465         d = bus->parent_dev;
466         assert(d != NULL);
467 
468         bus = pci_get_bus(d);
469     }
470 
471     return bus;
472 }
473 
474 const char *pci_root_bus_path(PCIDevice *dev)
475 {
476     PCIBus *rootbus = pci_device_root_bus(dev);
477     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
478     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
479 
480     assert(host_bridge->bus == rootbus);
481 
482     if (hc->root_bus_path) {
483         return (*hc->root_bus_path)(host_bridge, rootbus);
484     }
485 
486     return rootbus->qbus.name;
487 }
488 
489 bool pci_bus_bypass_iommu(PCIBus *bus)
490 {
491     PCIBus *rootbus = bus;
492     PCIHostState *host_bridge;
493 
494     if (!pci_bus_is_root(bus)) {
495         rootbus = pci_device_root_bus(bus->parent_dev);
496     }
497 
498     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
499 
500     assert(host_bridge->bus == rootbus);
501 
502     return host_bridge->bypass_iommu;
503 }
504 
505 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
506                                        MemoryRegion *mem, MemoryRegion *io,
507                                        uint8_t devfn_min)
508 {
509     assert(PCI_FUNC(devfn_min) == 0);
510     bus->devfn_min = devfn_min;
511     bus->slot_reserved_mask = 0x0;
512     bus->address_space_mem = mem;
513     bus->address_space_io = io;
514     bus->flags |= PCI_BUS_IS_ROOT;
515 
516     /* host bridge */
517     QLIST_INIT(&bus->child);
518 
519     pci_host_bus_register(parent);
520 }
521 
522 static void pci_bus_uninit(PCIBus *bus)
523 {
524     pci_host_bus_unregister(BUS(bus)->parent);
525 }
526 
527 bool pci_bus_is_express(const PCIBus *bus)
528 {
529     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
530 }
531 
532 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
533                        const char *name,
534                        MemoryRegion *mem, MemoryRegion *io,
535                        uint8_t devfn_min, const char *typename)
536 {
537     qbus_init(bus, bus_size, typename, parent, name);
538     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
539 }
540 
541 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
542                          MemoryRegion *mem, MemoryRegion *io,
543                          uint8_t devfn_min, const char *typename)
544 {
545     PCIBus *bus;
546 
547     bus = PCI_BUS(qbus_new(typename, parent, name));
548     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
549     return bus;
550 }
551 
552 void pci_root_bus_cleanup(PCIBus *bus)
553 {
554     pci_bus_uninit(bus);
555     /* the caller of the unplug hotplug handler will delete this device */
556     qbus_unrealize(BUS(bus));
557 }
558 
559 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
560                   void *irq_opaque, int nirq)
561 {
562     bus->set_irq = set_irq;
563     bus->irq_opaque = irq_opaque;
564     bus->nirq = nirq;
565     g_free(bus->irq_count);
566     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
567 }
568 
569 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
570 {
571     bus->map_irq = map_irq;
572 }
573 
574 void pci_bus_irqs_cleanup(PCIBus *bus)
575 {
576     bus->set_irq = NULL;
577     bus->map_irq = NULL;
578     bus->irq_opaque = NULL;
579     bus->nirq = 0;
580     g_free(bus->irq_count);
581     bus->irq_count = NULL;
582 }
583 
584 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
585                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
586                               void *irq_opaque,
587                               MemoryRegion *mem, MemoryRegion *io,
588                               uint8_t devfn_min, int nirq,
589                               const char *typename)
590 {
591     PCIBus *bus;
592 
593     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
594     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
595     pci_bus_map_irqs(bus, map_irq);
596     return bus;
597 }
598 
599 void pci_unregister_root_bus(PCIBus *bus)
600 {
601     pci_bus_irqs_cleanup(bus);
602     pci_root_bus_cleanup(bus);
603 }
604 
605 int pci_bus_num(PCIBus *s)
606 {
607     return PCI_BUS_GET_CLASS(s)->bus_num(s);
608 }
609 
610 /* Returns the min and max bus numbers of a PCI bus hierarchy */
611 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
612 {
613     int i;
614     *min_bus = *max_bus = pci_bus_num(bus);
615 
616     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
617         PCIDevice *dev = bus->devices[i];
618 
619         if (dev && IS_PCI_BRIDGE(dev)) {
620             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
621             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
622         }
623     }
624 }
625 
626 int pci_bus_numa_node(PCIBus *bus)
627 {
628     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
629 }
630 
631 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
632                                  const VMStateField *field)
633 {
634     PCIDevice *s = container_of(pv, PCIDevice, config);
635     uint8_t *config;
636     int i;
637 
638     assert(size == pci_config_size(s));
639     config = g_malloc(size);
640 
641     qemu_get_buffer(f, config, size);
642     for (i = 0; i < size; ++i) {
643         if ((config[i] ^ s->config[i]) &
644             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
645             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
646                          "cmask: %x wmask: %x w1cmask:%x", __func__,
647                          i, config[i], s->config[i],
648                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
649             g_free(config);
650             return -EINVAL;
651         }
652     }
653     memcpy(s->config, config, size);
654 
655     pci_update_mappings(s);
656     if (IS_PCI_BRIDGE(s)) {
657         pci_bridge_update_mappings(PCI_BRIDGE(s));
658     }
659 
660     memory_region_set_enabled(&s->bus_master_enable_region,
661                               pci_get_word(s->config + PCI_COMMAND)
662                               & PCI_COMMAND_MASTER);
663 
664     g_free(config);
665     return 0;
666 }
667 
668 /* just put buffer */
669 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
670                                  const VMStateField *field, JSONWriter *vmdesc)
671 {
672     const uint8_t **v = pv;
673     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
674     qemu_put_buffer(f, *v, size);
675 
676     return 0;
677 }
678 
679 static const VMStateInfo vmstate_info_pci_config = {
680     .name = "pci config",
681     .get  = get_pci_config_device,
682     .put  = put_pci_config_device,
683 };
684 
685 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
686                              const VMStateField *field)
687 {
688     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
689     uint32_t irq_state[PCI_NUM_PINS];
690     int i;
691     for (i = 0; i < PCI_NUM_PINS; ++i) {
692         irq_state[i] = qemu_get_be32(f);
693         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
694             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
695                     irq_state[i]);
696             return -EINVAL;
697         }
698     }
699 
700     for (i = 0; i < PCI_NUM_PINS; ++i) {
701         pci_set_irq_state(s, i, irq_state[i]);
702     }
703 
704     return 0;
705 }
706 
707 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
708                              const VMStateField *field, JSONWriter *vmdesc)
709 {
710     int i;
711     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
712 
713     for (i = 0; i < PCI_NUM_PINS; ++i) {
714         qemu_put_be32(f, pci_irq_state(s, i));
715     }
716 
717     return 0;
718 }
719 
720 static const VMStateInfo vmstate_info_pci_irq_state = {
721     .name = "pci irq state",
722     .get  = get_pci_irq_state,
723     .put  = put_pci_irq_state,
724 };
725 
726 static bool migrate_is_pcie(void *opaque, int version_id)
727 {
728     return pci_is_express((PCIDevice *)opaque);
729 }
730 
731 static bool migrate_is_not_pcie(void *opaque, int version_id)
732 {
733     return !pci_is_express((PCIDevice *)opaque);
734 }
735 
736 const VMStateDescription vmstate_pci_device = {
737     .name = "PCIDevice",
738     .version_id = 2,
739     .minimum_version_id = 1,
740     .fields = (const VMStateField[]) {
741         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
742         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
743                                    migrate_is_not_pcie,
744                                    0, vmstate_info_pci_config,
745                                    PCI_CONFIG_SPACE_SIZE),
746         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
747                                    migrate_is_pcie,
748                                    0, vmstate_info_pci_config,
749                                    PCIE_CONFIG_SPACE_SIZE),
750         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
751                                    vmstate_info_pci_irq_state,
752                                    PCI_NUM_PINS * sizeof(int32_t)),
753         VMSTATE_END_OF_LIST()
754     }
755 };
756 
757 
758 void pci_device_save(PCIDevice *s, QEMUFile *f)
759 {
760     /* Clear interrupt status bit: it is implicit
761      * in irq_state which we are saving.
762      * This makes us compatible with old devices
763      * which never set or clear this bit. */
764     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
765     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
766     /* Restore the interrupt status bit. */
767     pci_update_irq_status(s);
768 }
769 
770 int pci_device_load(PCIDevice *s, QEMUFile *f)
771 {
772     int ret;
773     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
774     /* Restore the interrupt status bit. */
775     pci_update_irq_status(s);
776     return ret;
777 }
778 
779 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
780 {
781     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
782                  pci_default_sub_vendor_id);
783     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
784                  pci_default_sub_device_id);
785 }
786 
787 /*
788  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
789  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
790  */
791 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
792                              unsigned int *slotp, unsigned int *funcp)
793 {
794     const char *p;
795     char *e;
796     unsigned long val;
797     unsigned long dom = 0, bus = 0;
798     unsigned int slot = 0;
799     unsigned int func = 0;
800 
801     p = addr;
802     val = strtoul(p, &e, 16);
803     if (e == p)
804         return -1;
805     if (*e == ':') {
806         bus = val;
807         p = e + 1;
808         val = strtoul(p, &e, 16);
809         if (e == p)
810             return -1;
811         if (*e == ':') {
812             dom = bus;
813             bus = val;
814             p = e + 1;
815             val = strtoul(p, &e, 16);
816             if (e == p)
817                 return -1;
818         }
819     }
820 
821     slot = val;
822 
823     if (funcp != NULL) {
824         if (*e != '.')
825             return -1;
826 
827         p = e + 1;
828         val = strtoul(p, &e, 16);
829         if (e == p)
830             return -1;
831 
832         func = val;
833     }
834 
835     /* if funcp == NULL func is 0 */
836     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
837         return -1;
838 
839     if (*e)
840         return -1;
841 
842     *domp = dom;
843     *busp = bus;
844     *slotp = slot;
845     if (funcp != NULL)
846         *funcp = func;
847     return 0;
848 }
849 
850 static void pci_init_cmask(PCIDevice *dev)
851 {
852     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
853     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
854     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
855     dev->cmask[PCI_REVISION_ID] = 0xff;
856     dev->cmask[PCI_CLASS_PROG] = 0xff;
857     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
858     dev->cmask[PCI_HEADER_TYPE] = 0xff;
859     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
860 }
861 
862 static void pci_init_wmask(PCIDevice *dev)
863 {
864     int config_size = pci_config_size(dev);
865 
866     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
867     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
868     pci_set_word(dev->wmask + PCI_COMMAND,
869                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
870                  PCI_COMMAND_INTX_DISABLE);
871     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
872 
873     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
874            config_size - PCI_CONFIG_HEADER_SIZE);
875 }
876 
877 static void pci_init_w1cmask(PCIDevice *dev)
878 {
879     /*
880      * Note: It's okay to set w1cmask even for readonly bits as
881      * long as their value is hardwired to 0.
882      */
883     pci_set_word(dev->w1cmask + PCI_STATUS,
884                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
885                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
886                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
887 }
888 
889 static void pci_init_mask_bridge(PCIDevice *d)
890 {
891     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
892        PCI_SEC_LATENCY_TIMER */
893     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
894 
895     /* base and limit */
896     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
897     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
898     pci_set_word(d->wmask + PCI_MEMORY_BASE,
899                  PCI_MEMORY_RANGE_MASK & 0xffff);
900     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
901                  PCI_MEMORY_RANGE_MASK & 0xffff);
902     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
903                  PCI_PREF_RANGE_MASK & 0xffff);
904     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
905                  PCI_PREF_RANGE_MASK & 0xffff);
906 
907     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
908     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
909 
910     /* Supported memory and i/o types */
911     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
912     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
913     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
914                                PCI_PREF_RANGE_TYPE_64);
915     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
916                                PCI_PREF_RANGE_TYPE_64);
917 
918     /*
919      * TODO: Bridges default to 10-bit VGA decoding but we currently only
920      * implement 16-bit decoding (no alias support).
921      */
922     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
923                  PCI_BRIDGE_CTL_PARITY |
924                  PCI_BRIDGE_CTL_SERR |
925                  PCI_BRIDGE_CTL_ISA |
926                  PCI_BRIDGE_CTL_VGA |
927                  PCI_BRIDGE_CTL_VGA_16BIT |
928                  PCI_BRIDGE_CTL_MASTER_ABORT |
929                  PCI_BRIDGE_CTL_BUS_RESET |
930                  PCI_BRIDGE_CTL_FAST_BACK |
931                  PCI_BRIDGE_CTL_DISCARD |
932                  PCI_BRIDGE_CTL_SEC_DISCARD |
933                  PCI_BRIDGE_CTL_DISCARD_SERR);
934     /* Below does not do anything as we never set this bit, put here for
935      * completeness. */
936     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
937                  PCI_BRIDGE_CTL_DISCARD_STATUS);
938     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
939     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
940     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
941                                PCI_PREF_RANGE_TYPE_MASK);
942     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
943                                PCI_PREF_RANGE_TYPE_MASK);
944 }
945 
946 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
947 {
948     uint8_t slot = PCI_SLOT(dev->devfn);
949     uint8_t func;
950 
951     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
952         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
953     }
954 
955     /*
956      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
957      * device, as it may just be a VF that ended up with function 0 in
958      * the legacy PCI interpretation. Avoid failing in such cases:
959      */
960     if (pci_is_vf(dev) &&
961         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
962         return;
963     }
964 
965     /*
966      * multifunction bit is interpreted in two ways as follows.
967      *   - all functions must set the bit to 1.
968      *     Example: Intel X53
969      *   - function 0 must set the bit, but the rest function (> 0)
970      *     is allowed to leave the bit to 0.
971      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
972      *
973      * So OS (at least Linux) checks the bit of only function 0,
974      * and doesn't see the bit of function > 0.
975      *
976      * The below check allows both interpretation.
977      */
978     if (PCI_FUNC(dev->devfn)) {
979         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
980         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
981             /* function 0 should set multifunction bit */
982             error_setg(errp, "PCI: single function device can't be populated "
983                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
984             return;
985         }
986         return;
987     }
988 
989     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
990         return;
991     }
992     /* function 0 indicates single function, so function > 0 must be NULL */
993     for (func = 1; func < PCI_FUNC_MAX; ++func) {
994         if (bus->devices[PCI_DEVFN(slot, func)]) {
995             error_setg(errp, "PCI: %x.0 indicates single function, "
996                        "but %x.%x is already populated.",
997                        slot, slot, func);
998             return;
999         }
1000     }
1001 }
1002 
1003 static void pci_config_alloc(PCIDevice *pci_dev)
1004 {
1005     int config_size = pci_config_size(pci_dev);
1006 
1007     pci_dev->config = g_malloc0(config_size);
1008     pci_dev->cmask = g_malloc0(config_size);
1009     pci_dev->wmask = g_malloc0(config_size);
1010     pci_dev->w1cmask = g_malloc0(config_size);
1011     pci_dev->used = g_malloc0(config_size);
1012 }
1013 
1014 static void pci_config_free(PCIDevice *pci_dev)
1015 {
1016     g_free(pci_dev->config);
1017     g_free(pci_dev->cmask);
1018     g_free(pci_dev->wmask);
1019     g_free(pci_dev->w1cmask);
1020     g_free(pci_dev->used);
1021 }
1022 
1023 static void do_pci_unregister_device(PCIDevice *pci_dev)
1024 {
1025     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1026     pci_config_free(pci_dev);
1027 
1028     if (xen_mode == XEN_EMULATE) {
1029         xen_evtchn_remove_pci_device(pci_dev);
1030     }
1031     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1032         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1033                                     &pci_dev->bus_master_enable_region);
1034     }
1035     address_space_destroy(&pci_dev->bus_master_as);
1036 }
1037 
1038 /* Extract PCIReqIDCache into BDF format */
1039 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1040 {
1041     uint8_t bus_n;
1042     uint16_t result;
1043 
1044     switch (cache->type) {
1045     case PCI_REQ_ID_BDF:
1046         result = pci_get_bdf(cache->dev);
1047         break;
1048     case PCI_REQ_ID_SECONDARY_BUS:
1049         bus_n = pci_dev_bus_num(cache->dev);
1050         result = PCI_BUILD_BDF(bus_n, 0);
1051         break;
1052     default:
1053         error_report("Invalid PCI requester ID cache type: %d",
1054                      cache->type);
1055         exit(1);
1056         break;
1057     }
1058 
1059     return result;
1060 }
1061 
1062 /* Parse bridges up to the root complex and return requester ID
1063  * cache for specific device.  For full PCIe topology, the cache
1064  * result would be exactly the same as getting BDF of the device.
1065  * However, several tricks are required when system mixed up with
1066  * legacy PCI devices and PCIe-to-PCI bridges.
1067  *
1068  * Here we cache the proxy device (and type) not requester ID since
1069  * bus number might change from time to time.
1070  */
1071 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1072 {
1073     PCIDevice *parent;
1074     PCIReqIDCache cache = {
1075         .dev = dev,
1076         .type = PCI_REQ_ID_BDF,
1077     };
1078 
1079     while (!pci_bus_is_root(pci_get_bus(dev))) {
1080         /* We are under PCI/PCIe bridges */
1081         parent = pci_get_bus(dev)->parent_dev;
1082         if (pci_is_express(parent)) {
1083             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1084                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1085                  * override the requester ID using secondary bus
1086                  * number of parent bridge with zeroed devfn
1087                  * (pcie-to-pci bridge spec chap 2.3). */
1088                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1089                 cache.dev = dev;
1090             }
1091         } else {
1092             /* Legacy PCI, override requester ID with the bridge's
1093              * BDF upstream.  When the root complex connects to
1094              * legacy PCI devices (including buses), it can only
1095              * obtain requester ID info from directly attached
1096              * devices.  If devices are attached under bridges, only
1097              * the requester ID of the bridge that is directly
1098              * attached to the root complex can be recognized. */
1099             cache.type = PCI_REQ_ID_BDF;
1100             cache.dev = parent;
1101         }
1102         dev = parent;
1103     }
1104 
1105     return cache;
1106 }
1107 
1108 uint16_t pci_requester_id(PCIDevice *dev)
1109 {
1110     return pci_req_id_cache_extract(&dev->requester_id_cache);
1111 }
1112 
1113 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1114 {
1115     return !(bus->devices[devfn]);
1116 }
1117 
1118 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1119 {
1120     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1121 }
1122 
1123 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1124 {
1125     return bus->slot_reserved_mask;
1126 }
1127 
1128 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1129 {
1130     bus->slot_reserved_mask |= mask;
1131 }
1132 
1133 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1134 {
1135     bus->slot_reserved_mask &= ~mask;
1136 }
1137 
1138 /* -1 for devfn means auto assign */
1139 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1140                                          const char *name, int devfn,
1141                                          Error **errp)
1142 {
1143     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1144     PCIConfigReadFunc *config_read = pc->config_read;
1145     PCIConfigWriteFunc *config_write = pc->config_write;
1146     Error *local_err = NULL;
1147     DeviceState *dev = DEVICE(pci_dev);
1148     PCIBus *bus = pci_get_bus(pci_dev);
1149     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1150 
1151     /* Only pci bridges can be attached to extra PCI root buses */
1152     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1153         error_setg(errp,
1154                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1155                     bus->parent_dev->name);
1156         return NULL;
1157     }
1158 
1159     if (devfn < 0) {
1160         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1161             devfn += PCI_FUNC_MAX) {
1162             if (pci_bus_devfn_available(bus, devfn) &&
1163                    !pci_bus_devfn_reserved(bus, devfn)) {
1164                 goto found;
1165             }
1166         }
1167         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1168                    "or reserved", name);
1169         return NULL;
1170     found: ;
1171     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1172         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1173                    " reserved",
1174                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1175         return NULL;
1176     } else if (!pci_bus_devfn_available(bus, devfn)) {
1177         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1178                    " in use by %s,id=%s",
1179                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1180                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1181         return NULL;
1182     } /*
1183        * Populating function 0 triggers a scan from the guest that
1184        * exposes other non-zero functions. Hence we need to ensure that
1185        * function 0 wasn't added yet.
1186        */
1187     else if (dev->hotplugged &&
1188              !pci_is_vf(pci_dev) &&
1189              pci_get_function_0(pci_dev)) {
1190         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1191                    " new func %s cannot be exposed to guest.",
1192                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1193                    pci_get_function_0(pci_dev)->name,
1194                    name);
1195 
1196        return NULL;
1197     }
1198 
1199     pci_dev->devfn = devfn;
1200     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1201     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1202 
1203     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1204                        "bus master container", UINT64_MAX);
1205     address_space_init(&pci_dev->bus_master_as,
1206                        &pci_dev->bus_master_container_region, pci_dev->name);
1207 
1208     if (phase_check(PHASE_MACHINE_READY)) {
1209         pci_init_bus_master(pci_dev);
1210     }
1211     pci_dev->irq_state = 0;
1212     pci_config_alloc(pci_dev);
1213 
1214     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1215     pci_config_set_device_id(pci_dev->config, pc->device_id);
1216     pci_config_set_revision(pci_dev->config, pc->revision);
1217     pci_config_set_class(pci_dev->config, pc->class_id);
1218 
1219     if (!is_bridge) {
1220         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1221             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1222                          pc->subsystem_vendor_id);
1223             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1224                          pc->subsystem_id);
1225         } else {
1226             pci_set_default_subsystem_id(pci_dev);
1227         }
1228     } else {
1229         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1230         assert(!pc->subsystem_vendor_id);
1231         assert(!pc->subsystem_id);
1232     }
1233     pci_init_cmask(pci_dev);
1234     pci_init_wmask(pci_dev);
1235     pci_init_w1cmask(pci_dev);
1236     if (is_bridge) {
1237         pci_init_mask_bridge(pci_dev);
1238     }
1239     pci_init_multifunction(bus, pci_dev, &local_err);
1240     if (local_err) {
1241         error_propagate(errp, local_err);
1242         do_pci_unregister_device(pci_dev);
1243         return NULL;
1244     }
1245 
1246     if (!config_read)
1247         config_read = pci_default_read_config;
1248     if (!config_write)
1249         config_write = pci_default_write_config;
1250     pci_dev->config_read = config_read;
1251     pci_dev->config_write = config_write;
1252     bus->devices[devfn] = pci_dev;
1253     pci_dev->version_id = 2; /* Current pci device vmstate version */
1254     return pci_dev;
1255 }
1256 
1257 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1258 {
1259     PCIIORegion *r;
1260     int i;
1261 
1262     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1263         r = &pci_dev->io_regions[i];
1264         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1265             continue;
1266         memory_region_del_subregion(r->address_space, r->memory);
1267     }
1268 
1269     pci_unregister_vga(pci_dev);
1270 }
1271 
1272 static void pci_qdev_unrealize(DeviceState *dev)
1273 {
1274     PCIDevice *pci_dev = PCI_DEVICE(dev);
1275     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1276 
1277     pci_unregister_io_regions(pci_dev);
1278     pci_del_option_rom(pci_dev);
1279 
1280     if (pc->exit) {
1281         pc->exit(pci_dev);
1282     }
1283 
1284     pci_device_deassert_intx(pci_dev);
1285     do_pci_unregister_device(pci_dev);
1286 
1287     pci_dev->msi_trigger = NULL;
1288 
1289     /*
1290      * clean up acpi-index so it could reused by another device
1291      */
1292     if (pci_dev->acpi_index) {
1293         GSequence *used_indexes = pci_acpi_index_list();
1294 
1295         g_sequence_remove(g_sequence_lookup(used_indexes,
1296                           GINT_TO_POINTER(pci_dev->acpi_index),
1297                           g_cmp_uint32, NULL));
1298     }
1299 }
1300 
1301 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1302                       uint8_t type, MemoryRegion *memory)
1303 {
1304     PCIIORegion *r;
1305     uint32_t addr; /* offset in pci config space */
1306     uint64_t wmask;
1307     pcibus_t size = memory_region_size(memory);
1308     uint8_t hdr_type;
1309 
1310     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1311     assert(region_num >= 0);
1312     assert(region_num < PCI_NUM_REGIONS);
1313     assert(is_power_of_2(size));
1314 
1315     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1316     hdr_type =
1317         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1318     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1319 
1320     r = &pci_dev->io_regions[region_num];
1321     r->addr = PCI_BAR_UNMAPPED;
1322     r->size = size;
1323     r->type = type;
1324     r->memory = memory;
1325     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1326                         ? pci_get_bus(pci_dev)->address_space_io
1327                         : pci_get_bus(pci_dev)->address_space_mem;
1328 
1329     wmask = ~(size - 1);
1330     if (region_num == PCI_ROM_SLOT) {
1331         /* ROM enable bit is writable */
1332         wmask |= PCI_ROM_ADDRESS_ENABLE;
1333     }
1334 
1335     addr = pci_bar(pci_dev, region_num);
1336     pci_set_long(pci_dev->config + addr, type);
1337 
1338     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1339         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1340         pci_set_quad(pci_dev->wmask + addr, wmask);
1341         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1342     } else {
1343         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1344         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1345     }
1346 }
1347 
1348 static void pci_update_vga(PCIDevice *pci_dev)
1349 {
1350     uint16_t cmd;
1351 
1352     if (!pci_dev->has_vga) {
1353         return;
1354     }
1355 
1356     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1357 
1358     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1359                               cmd & PCI_COMMAND_MEMORY);
1360     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1361                               cmd & PCI_COMMAND_IO);
1362     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1363                               cmd & PCI_COMMAND_IO);
1364 }
1365 
1366 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1367                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1368 {
1369     PCIBus *bus = pci_get_bus(pci_dev);
1370 
1371     assert(!pci_dev->has_vga);
1372 
1373     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1374     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1375     memory_region_add_subregion_overlap(bus->address_space_mem,
1376                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1377 
1378     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1379     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1380     memory_region_add_subregion_overlap(bus->address_space_io,
1381                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1382 
1383     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1384     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1385     memory_region_add_subregion_overlap(bus->address_space_io,
1386                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1387     pci_dev->has_vga = true;
1388 
1389     pci_update_vga(pci_dev);
1390 }
1391 
1392 void pci_unregister_vga(PCIDevice *pci_dev)
1393 {
1394     PCIBus *bus = pci_get_bus(pci_dev);
1395 
1396     if (!pci_dev->has_vga) {
1397         return;
1398     }
1399 
1400     memory_region_del_subregion(bus->address_space_mem,
1401                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1402     memory_region_del_subregion(bus->address_space_io,
1403                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1404     memory_region_del_subregion(bus->address_space_io,
1405                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1406     pci_dev->has_vga = false;
1407 }
1408 
1409 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1410 {
1411     return pci_dev->io_regions[region_num].addr;
1412 }
1413 
1414 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1415                                         uint8_t type, pcibus_t size)
1416 {
1417     pcibus_t new_addr;
1418     if (!pci_is_vf(d)) {
1419         int bar = pci_bar(d, reg);
1420         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1421             new_addr = pci_get_quad(d->config + bar);
1422         } else {
1423             new_addr = pci_get_long(d->config + bar);
1424         }
1425     } else {
1426         PCIDevice *pf = d->exp.sriov_vf.pf;
1427         uint16_t sriov_cap = pf->exp.sriov_cap;
1428         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1429         uint16_t vf_offset =
1430             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1431         uint16_t vf_stride =
1432             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1433         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1434 
1435         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1436             new_addr = pci_get_quad(pf->config + bar);
1437         } else {
1438             new_addr = pci_get_long(pf->config + bar);
1439         }
1440         new_addr += vf_num * size;
1441     }
1442     /* The ROM slot has a specific enable bit, keep it intact */
1443     if (reg != PCI_ROM_SLOT) {
1444         new_addr &= ~(size - 1);
1445     }
1446     return new_addr;
1447 }
1448 
1449 pcibus_t pci_bar_address(PCIDevice *d,
1450                          int reg, uint8_t type, pcibus_t size)
1451 {
1452     pcibus_t new_addr, last_addr;
1453     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1454     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1455     bool allow_0_address = mc->pci_allow_0_address;
1456 
1457     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1458         if (!(cmd & PCI_COMMAND_IO)) {
1459             return PCI_BAR_UNMAPPED;
1460         }
1461         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1462         last_addr = new_addr + size - 1;
1463         /* Check if 32 bit BAR wraps around explicitly.
1464          * TODO: make priorities correct and remove this work around.
1465          */
1466         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1467             (!allow_0_address && new_addr == 0)) {
1468             return PCI_BAR_UNMAPPED;
1469         }
1470         return new_addr;
1471     }
1472 
1473     if (!(cmd & PCI_COMMAND_MEMORY)) {
1474         return PCI_BAR_UNMAPPED;
1475     }
1476     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1477     /* the ROM slot has a specific enable bit */
1478     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1479         return PCI_BAR_UNMAPPED;
1480     }
1481     new_addr &= ~(size - 1);
1482     last_addr = new_addr + size - 1;
1483     /* NOTE: we do not support wrapping */
1484     /* XXX: as we cannot support really dynamic
1485        mappings, we handle specific values as invalid
1486        mappings. */
1487     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1488         (!allow_0_address && new_addr == 0)) {
1489         return PCI_BAR_UNMAPPED;
1490     }
1491 
1492     /* Now pcibus_t is 64bit.
1493      * Check if 32 bit BAR wraps around explicitly.
1494      * Without this, PC ide doesn't work well.
1495      * TODO: remove this work around.
1496      */
1497     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1498         return PCI_BAR_UNMAPPED;
1499     }
1500 
1501     /*
1502      * OS is allowed to set BAR beyond its addressable
1503      * bits. For example, 32 bit OS can set 64bit bar
1504      * to >4G. Check it. TODO: we might need to support
1505      * it in the future for e.g. PAE.
1506      */
1507     if (last_addr >= HWADDR_MAX) {
1508         return PCI_BAR_UNMAPPED;
1509     }
1510 
1511     return new_addr;
1512 }
1513 
1514 static void pci_update_mappings(PCIDevice *d)
1515 {
1516     PCIIORegion *r;
1517     int i;
1518     pcibus_t new_addr;
1519 
1520     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1521         r = &d->io_regions[i];
1522 
1523         /* this region isn't registered */
1524         if (!r->size)
1525             continue;
1526 
1527         new_addr = pci_bar_address(d, i, r->type, r->size);
1528         if (!d->enabled) {
1529             new_addr = PCI_BAR_UNMAPPED;
1530         }
1531 
1532         /* This bar isn't changed */
1533         if (new_addr == r->addr)
1534             continue;
1535 
1536         /* now do the real mapping */
1537         if (r->addr != PCI_BAR_UNMAPPED) {
1538             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1539                                           PCI_SLOT(d->devfn),
1540                                           PCI_FUNC(d->devfn),
1541                                           i, r->addr, r->size);
1542             memory_region_del_subregion(r->address_space, r->memory);
1543         }
1544         r->addr = new_addr;
1545         if (r->addr != PCI_BAR_UNMAPPED) {
1546             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1547                                           PCI_SLOT(d->devfn),
1548                                           PCI_FUNC(d->devfn),
1549                                           i, r->addr, r->size);
1550             memory_region_add_subregion_overlap(r->address_space,
1551                                                 r->addr, r->memory, 1);
1552         }
1553     }
1554 
1555     pci_update_vga(d);
1556 }
1557 
1558 static inline int pci_irq_disabled(PCIDevice *d)
1559 {
1560     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1561 }
1562 
1563 /* Called after interrupt disabled field update in config space,
1564  * assert/deassert interrupts if necessary.
1565  * Gets original interrupt disable bit value (before update). */
1566 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1567 {
1568     int i, disabled = pci_irq_disabled(d);
1569     if (disabled == was_irq_disabled)
1570         return;
1571     for (i = 0; i < PCI_NUM_PINS; ++i) {
1572         int state = pci_irq_state(d, i);
1573         pci_change_irq_level(d, i, disabled ? -state : state);
1574     }
1575 }
1576 
1577 uint32_t pci_default_read_config(PCIDevice *d,
1578                                  uint32_t address, int len)
1579 {
1580     uint32_t val = 0;
1581 
1582     assert(address + len <= pci_config_size(d));
1583 
1584     if (pci_is_express_downstream_port(d) &&
1585         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1586         pcie_sync_bridge_lnk(d);
1587     }
1588     memcpy(&val, d->config + address, len);
1589     return le32_to_cpu(val);
1590 }
1591 
1592 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1593 {
1594     int i, was_irq_disabled = pci_irq_disabled(d);
1595     uint32_t val = val_in;
1596 
1597     assert(addr + l <= pci_config_size(d));
1598 
1599     for (i = 0; i < l; val >>= 8, ++i) {
1600         uint8_t wmask = d->wmask[addr + i];
1601         uint8_t w1cmask = d->w1cmask[addr + i];
1602         assert(!(wmask & w1cmask));
1603         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1604         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1605     }
1606     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1607         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1608         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1609         range_covers_byte(addr, l, PCI_COMMAND))
1610         pci_update_mappings(d);
1611 
1612     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1613         pci_update_irq_disabled(d, was_irq_disabled);
1614         memory_region_set_enabled(&d->bus_master_enable_region,
1615                                   (pci_get_word(d->config + PCI_COMMAND)
1616                                    & PCI_COMMAND_MASTER) && d->enabled);
1617     }
1618 
1619     msi_write_config(d, addr, val_in, l);
1620     msix_write_config(d, addr, val_in, l);
1621     pcie_sriov_config_write(d, addr, val_in, l);
1622 }
1623 
1624 /***********************************************************/
1625 /* generic PCI irq support */
1626 
1627 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1628 static void pci_irq_handler(void *opaque, int irq_num, int level)
1629 {
1630     PCIDevice *pci_dev = opaque;
1631     int change;
1632 
1633     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1634     assert(level == 0 || level == 1);
1635     change = level - pci_irq_state(pci_dev, irq_num);
1636     if (!change)
1637         return;
1638 
1639     pci_set_irq_state(pci_dev, irq_num, level);
1640     pci_update_irq_status(pci_dev);
1641     if (pci_irq_disabled(pci_dev))
1642         return;
1643     pci_change_irq_level(pci_dev, irq_num, change);
1644 }
1645 
1646 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1647 {
1648     int intx = pci_intx(pci_dev);
1649     assert(0 <= intx && intx < PCI_NUM_PINS);
1650 
1651     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1652 }
1653 
1654 void pci_set_irq(PCIDevice *pci_dev, int level)
1655 {
1656     int intx = pci_intx(pci_dev);
1657     pci_irq_handler(pci_dev, intx, level);
1658 }
1659 
1660 /* Special hooks used by device assignment */
1661 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1662 {
1663     assert(pci_bus_is_root(bus));
1664     bus->route_intx_to_irq = route_intx_to_irq;
1665 }
1666 
1667 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1668 {
1669     PCIBus *bus;
1670 
1671     do {
1672         int dev_irq = pin;
1673         bus = pci_get_bus(dev);
1674         pin = bus->map_irq(dev, pin);
1675         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1676                             pci_bus_is_root(bus) ? "root-complex"
1677                                     : DEVICE(bus->parent_dev)->canonical_path);
1678         dev = bus->parent_dev;
1679     } while (dev);
1680 
1681     if (!bus->route_intx_to_irq) {
1682         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1683                      object_get_typename(OBJECT(bus->qbus.parent)));
1684         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1685     }
1686 
1687     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1688 }
1689 
1690 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1691 {
1692     return old->mode != new->mode || old->irq != new->irq;
1693 }
1694 
1695 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1696 {
1697     PCIDevice *dev;
1698     PCIBus *sec;
1699     int i;
1700 
1701     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1702         dev = bus->devices[i];
1703         if (dev && dev->intx_routing_notifier) {
1704             dev->intx_routing_notifier(dev);
1705         }
1706     }
1707 
1708     QLIST_FOREACH(sec, &bus->child, sibling) {
1709         pci_bus_fire_intx_routing_notifier(sec);
1710     }
1711 }
1712 
1713 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1714                                           PCIINTxRoutingNotifier notifier)
1715 {
1716     dev->intx_routing_notifier = notifier;
1717 }
1718 
1719 /*
1720  * PCI-to-PCI bridge specification
1721  * 9.1: Interrupt routing. Table 9-1
1722  *
1723  * the PCI Express Base Specification, Revision 2.1
1724  * 2.2.8.1: INTx interrupt signaling - Rules
1725  *          the Implementation Note
1726  *          Table 2-20
1727  */
1728 /*
1729  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1730  * 0-origin unlike PCI interrupt pin register.
1731  */
1732 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1733 {
1734     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1735 }
1736 
1737 /***********************************************************/
1738 /* monitor info on PCI */
1739 
1740 static const pci_class_desc pci_class_descriptions[] =
1741 {
1742     { 0x0001, "VGA controller", "display"},
1743     { 0x0100, "SCSI controller", "scsi"},
1744     { 0x0101, "IDE controller", "ide"},
1745     { 0x0102, "Floppy controller", "fdc"},
1746     { 0x0103, "IPI controller", "ipi"},
1747     { 0x0104, "RAID controller", "raid"},
1748     { 0x0106, "SATA controller"},
1749     { 0x0107, "SAS controller"},
1750     { 0x0180, "Storage controller"},
1751     { 0x0200, "Ethernet controller", "ethernet"},
1752     { 0x0201, "Token Ring controller", "token-ring"},
1753     { 0x0202, "FDDI controller", "fddi"},
1754     { 0x0203, "ATM controller", "atm"},
1755     { 0x0280, "Network controller"},
1756     { 0x0300, "VGA controller", "display", 0x00ff},
1757     { 0x0301, "XGA controller"},
1758     { 0x0302, "3D controller"},
1759     { 0x0380, "Display controller"},
1760     { 0x0400, "Video controller", "video"},
1761     { 0x0401, "Audio controller", "sound"},
1762     { 0x0402, "Phone"},
1763     { 0x0403, "Audio controller", "sound"},
1764     { 0x0480, "Multimedia controller"},
1765     { 0x0500, "RAM controller", "memory"},
1766     { 0x0501, "Flash controller", "flash"},
1767     { 0x0580, "Memory controller"},
1768     { 0x0600, "Host bridge", "host"},
1769     { 0x0601, "ISA bridge", "isa"},
1770     { 0x0602, "EISA bridge", "eisa"},
1771     { 0x0603, "MC bridge", "mca"},
1772     { 0x0604, "PCI bridge", "pci-bridge"},
1773     { 0x0605, "PCMCIA bridge", "pcmcia"},
1774     { 0x0606, "NUBUS bridge", "nubus"},
1775     { 0x0607, "CARDBUS bridge", "cardbus"},
1776     { 0x0608, "RACEWAY bridge"},
1777     { 0x0680, "Bridge"},
1778     { 0x0700, "Serial port", "serial"},
1779     { 0x0701, "Parallel port", "parallel"},
1780     { 0x0800, "Interrupt controller", "interrupt-controller"},
1781     { 0x0801, "DMA controller", "dma-controller"},
1782     { 0x0802, "Timer", "timer"},
1783     { 0x0803, "RTC", "rtc"},
1784     { 0x0900, "Keyboard", "keyboard"},
1785     { 0x0901, "Pen", "pen"},
1786     { 0x0902, "Mouse", "mouse"},
1787     { 0x0A00, "Dock station", "dock", 0x00ff},
1788     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1789     { 0x0c00, "Firewire controller", "firewire"},
1790     { 0x0c01, "Access bus controller", "access-bus"},
1791     { 0x0c02, "SSA controller", "ssa"},
1792     { 0x0c03, "USB controller", "usb"},
1793     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1794     { 0x0c05, "SMBus"},
1795     { 0, NULL}
1796 };
1797 
1798 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1799                                            pci_bus_dev_fn fn,
1800                                            void *opaque)
1801 {
1802     PCIDevice *d;
1803     int devfn;
1804 
1805     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1806         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1807         if (d) {
1808             fn(bus, d, opaque);
1809         }
1810     }
1811 }
1812 
1813 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1814                                  pci_bus_dev_fn fn, void *opaque)
1815 {
1816     bus = pci_find_bus_nr(bus, bus_num);
1817 
1818     if (bus) {
1819         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1820     }
1821 }
1822 
1823 void pci_for_each_device_under_bus(PCIBus *bus,
1824                                    pci_bus_dev_fn fn, void *opaque)
1825 {
1826     PCIDevice *d;
1827     int devfn;
1828 
1829     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1830         d = bus->devices[devfn];
1831         if (d) {
1832             fn(bus, d, opaque);
1833         }
1834     }
1835 }
1836 
1837 void pci_for_each_device(PCIBus *bus, int bus_num,
1838                          pci_bus_dev_fn fn, void *opaque)
1839 {
1840     bus = pci_find_bus_nr(bus, bus_num);
1841 
1842     if (bus) {
1843         pci_for_each_device_under_bus(bus, fn, opaque);
1844     }
1845 }
1846 
1847 const pci_class_desc *get_class_desc(int class)
1848 {
1849     const pci_class_desc *desc;
1850 
1851     desc = pci_class_descriptions;
1852     while (desc->desc && class != desc->class) {
1853         desc++;
1854     }
1855 
1856     return desc;
1857 }
1858 
1859 void pci_init_nic_devices(PCIBus *bus, const char *default_model)
1860 {
1861     qemu_create_nic_bus_devices(&bus->qbus, TYPE_PCI_DEVICE, default_model,
1862                                 "virtio", "virtio-net-pci");
1863 }
1864 
1865 bool pci_init_nic_in_slot(PCIBus *rootbus, const char *model,
1866                           const char *alias, const char *devaddr)
1867 {
1868     NICInfo *nd = qemu_find_nic_info(model, true, alias);
1869     int dom, busnr, devfn;
1870     PCIDevice *pci_dev;
1871     unsigned slot;
1872     PCIBus *bus;
1873 
1874     if (!nd) {
1875         return false;
1876     }
1877 
1878     if (!devaddr || pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1879         error_report("Invalid PCI device address %s for device %s",
1880                      devaddr, model);
1881         exit(1);
1882     }
1883 
1884     if (dom != 0) {
1885         error_report("No support for non-zero PCI domains");
1886         exit(1);
1887     }
1888 
1889     devfn = PCI_DEVFN(slot, 0);
1890 
1891     bus = pci_find_bus_nr(rootbus, busnr);
1892     if (!bus) {
1893         error_report("Invalid PCI device address %s for device %s",
1894                      devaddr, model);
1895         exit(1);
1896     }
1897 
1898     pci_dev = pci_new(devfn, model);
1899     qdev_set_nic_properties(&pci_dev->qdev, nd);
1900     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1901     return true;
1902 }
1903 
1904 PCIDevice *pci_vga_init(PCIBus *bus)
1905 {
1906     vga_interface_created = true;
1907     switch (vga_interface_type) {
1908     case VGA_CIRRUS:
1909         return pci_create_simple(bus, -1, "cirrus-vga");
1910     case VGA_QXL:
1911         return pci_create_simple(bus, -1, "qxl-vga");
1912     case VGA_STD:
1913         return pci_create_simple(bus, -1, "VGA");
1914     case VGA_VMWARE:
1915         return pci_create_simple(bus, -1, "vmware-svga");
1916     case VGA_VIRTIO:
1917         return pci_create_simple(bus, -1, "virtio-vga");
1918     case VGA_NONE:
1919     default: /* Other non-PCI types. Checking for unsupported types is already
1920                 done in vl.c. */
1921         return NULL;
1922     }
1923 }
1924 
1925 /* Whether a given bus number is in range of the secondary
1926  * bus of the given bridge device. */
1927 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1928 {
1929     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1930              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1931         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1932         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1933 }
1934 
1935 /* Whether a given bus number is in a range of a root bus */
1936 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1937 {
1938     int i;
1939 
1940     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1941         PCIDevice *dev = bus->devices[i];
1942 
1943         if (dev && IS_PCI_BRIDGE(dev)) {
1944             if (pci_secondary_bus_in_range(dev, bus_num)) {
1945                 return true;
1946             }
1947         }
1948     }
1949 
1950     return false;
1951 }
1952 
1953 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1954 {
1955     PCIBus *sec;
1956 
1957     if (!bus) {
1958         return NULL;
1959     }
1960 
1961     if (pci_bus_num(bus) == bus_num) {
1962         return bus;
1963     }
1964 
1965     /* Consider all bus numbers in range for the host pci bridge. */
1966     if (!pci_bus_is_root(bus) &&
1967         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1968         return NULL;
1969     }
1970 
1971     /* try child bus */
1972     for (; bus; bus = sec) {
1973         QLIST_FOREACH(sec, &bus->child, sibling) {
1974             if (pci_bus_num(sec) == bus_num) {
1975                 return sec;
1976             }
1977             /* PXB buses assumed to be children of bus 0 */
1978             if (pci_bus_is_root(sec)) {
1979                 if (pci_root_bus_in_range(sec, bus_num)) {
1980                     break;
1981                 }
1982             } else {
1983                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1984                     break;
1985                 }
1986             }
1987         }
1988     }
1989 
1990     return NULL;
1991 }
1992 
1993 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
1994                                   pci_bus_fn end, void *parent_state)
1995 {
1996     PCIBus *sec;
1997     void *state;
1998 
1999     if (!bus) {
2000         return;
2001     }
2002 
2003     if (begin) {
2004         state = begin(bus, parent_state);
2005     } else {
2006         state = parent_state;
2007     }
2008 
2009     QLIST_FOREACH(sec, &bus->child, sibling) {
2010         pci_for_each_bus_depth_first(sec, begin, end, state);
2011     }
2012 
2013     if (end) {
2014         end(bus, state);
2015     }
2016 }
2017 
2018 
2019 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2020 {
2021     bus = pci_find_bus_nr(bus, bus_num);
2022 
2023     if (!bus)
2024         return NULL;
2025 
2026     return bus->devices[devfn];
2027 }
2028 
2029 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2030 
2031 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2032 {
2033     PCIDevice *pci_dev = (PCIDevice *)qdev;
2034     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2035     ObjectClass *klass = OBJECT_CLASS(pc);
2036     Error *local_err = NULL;
2037     bool is_default_rom;
2038     uint16_t class_id;
2039 
2040     /*
2041      * capped by systemd (see: udev-builtin-net_id.c)
2042      * as it's the only known user honor it to avoid users
2043      * misconfigure QEMU and then wonder why acpi-index doesn't work
2044      */
2045     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2046         error_setg(errp, "acpi-index should be less or equal to %u",
2047                    ONBOARD_INDEX_MAX);
2048         return;
2049     }
2050 
2051     /*
2052      * make sure that acpi-index is unique across all present PCI devices
2053      */
2054     if (pci_dev->acpi_index) {
2055         GSequence *used_indexes = pci_acpi_index_list();
2056 
2057         if (g_sequence_lookup(used_indexes,
2058                               GINT_TO_POINTER(pci_dev->acpi_index),
2059                               g_cmp_uint32, NULL)) {
2060             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2061                        " already exist", pci_dev->acpi_index);
2062             return;
2063         }
2064         g_sequence_insert_sorted(used_indexes,
2065                                  GINT_TO_POINTER(pci_dev->acpi_index),
2066                                  g_cmp_uint32, NULL);
2067     }
2068 
2069     if (pci_dev->romsize != -1 && !is_power_of_2(pci_dev->romsize)) {
2070         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2071         return;
2072     }
2073 
2074     /* initialize cap_present for pci_is_express() and pci_config_size(),
2075      * Note that hybrid PCIs are not set automatically and need to manage
2076      * QEMU_PCI_CAP_EXPRESS manually */
2077     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2078        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2079         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2080     }
2081 
2082     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2083         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2084     }
2085 
2086     pci_dev = do_pci_register_device(pci_dev,
2087                                      object_get_typename(OBJECT(qdev)),
2088                                      pci_dev->devfn, errp);
2089     if (pci_dev == NULL)
2090         return;
2091 
2092     if (pc->realize) {
2093         pc->realize(pci_dev, &local_err);
2094         if (local_err) {
2095             error_propagate(errp, local_err);
2096             do_pci_unregister_device(pci_dev);
2097             return;
2098         }
2099     }
2100 
2101     /*
2102      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2103      * associate only Device 0 with the device attached to the bus
2104      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2105      * sec 7.3.1).
2106      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2107      * 5-bit Device Number and 3-bit Function Number fields in its associated
2108      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2109      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2110      */
2111     if (pci_is_express(pci_dev) &&
2112         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2113         pcie_has_upstream_port(pci_dev) &&
2114         PCI_SLOT(pci_dev->devfn)) {
2115         warn_report("PCI: slot %d is not valid for %s,"
2116                     " parent device only allows plugging into slot 0.",
2117                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2118     }
2119 
2120     if (pci_dev->failover_pair_id) {
2121         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2122             error_setg(errp, "failover primary device must be on "
2123                              "PCIExpress bus");
2124             pci_qdev_unrealize(DEVICE(pci_dev));
2125             return;
2126         }
2127         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2128         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2129             error_setg(errp, "failover primary device is not an "
2130                              "Ethernet device");
2131             pci_qdev_unrealize(DEVICE(pci_dev));
2132             return;
2133         }
2134         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2135             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2136             error_setg(errp, "failover: primary device must be in its own "
2137                               "PCI slot");
2138             pci_qdev_unrealize(DEVICE(pci_dev));
2139             return;
2140         }
2141         qdev->allow_unplug_during_migration = true;
2142     }
2143 
2144     /* rom loading */
2145     is_default_rom = false;
2146     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2147         pci_dev->romfile = g_strdup(pc->romfile);
2148         is_default_rom = true;
2149     }
2150 
2151     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2152     if (local_err) {
2153         error_propagate(errp, local_err);
2154         pci_qdev_unrealize(DEVICE(pci_dev));
2155         return;
2156     }
2157 
2158     pci_set_power(pci_dev, true);
2159 
2160     pci_dev->msi_trigger = pci_msi_trigger;
2161 }
2162 
2163 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2164                                    const char *name)
2165 {
2166     DeviceState *dev;
2167 
2168     dev = qdev_new(name);
2169     qdev_prop_set_int32(dev, "addr", devfn);
2170     qdev_prop_set_bit(dev, "multifunction", multifunction);
2171     return PCI_DEVICE(dev);
2172 }
2173 
2174 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2175 {
2176     return pci_new_internal(devfn, true, name);
2177 }
2178 
2179 PCIDevice *pci_new(int devfn, const char *name)
2180 {
2181     return pci_new_internal(devfn, false, name);
2182 }
2183 
2184 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2185 {
2186     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2187 }
2188 
2189 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2190                                            const char *name)
2191 {
2192     PCIDevice *dev = pci_new_multifunction(devfn, name);
2193     pci_realize_and_unref(dev, bus, &error_fatal);
2194     return dev;
2195 }
2196 
2197 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2198 {
2199     PCIDevice *dev = pci_new(devfn, name);
2200     pci_realize_and_unref(dev, bus, &error_fatal);
2201     return dev;
2202 }
2203 
2204 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2205 {
2206     int offset = PCI_CONFIG_HEADER_SIZE;
2207     int i;
2208     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2209         if (pdev->used[i])
2210             offset = i + 1;
2211         else if (i - offset + 1 == size)
2212             return offset;
2213     }
2214     return 0;
2215 }
2216 
2217 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2218                                         uint8_t *prev_p)
2219 {
2220     uint8_t next, prev;
2221 
2222     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2223         return 0;
2224 
2225     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2226          prev = next + PCI_CAP_LIST_NEXT)
2227         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2228             break;
2229 
2230     if (prev_p)
2231         *prev_p = prev;
2232     return next;
2233 }
2234 
2235 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2236 {
2237     uint8_t next, prev, found = 0;
2238 
2239     if (!(pdev->used[offset])) {
2240         return 0;
2241     }
2242 
2243     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2244 
2245     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2246          prev = next + PCI_CAP_LIST_NEXT) {
2247         if (next <= offset && next > found) {
2248             found = next;
2249         }
2250     }
2251     return found;
2252 }
2253 
2254 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2255    This is needed for an option rom which is used for more than one device. */
2256 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2257 {
2258     uint16_t vendor_id;
2259     uint16_t device_id;
2260     uint16_t rom_vendor_id;
2261     uint16_t rom_device_id;
2262     uint16_t rom_magic;
2263     uint16_t pcir_offset;
2264     uint8_t checksum;
2265 
2266     /* Words in rom data are little endian (like in PCI configuration),
2267        so they can be read / written with pci_get_word / pci_set_word. */
2268 
2269     /* Only a valid rom will be patched. */
2270     rom_magic = pci_get_word(ptr);
2271     if (rom_magic != 0xaa55) {
2272         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2273         return;
2274     }
2275     pcir_offset = pci_get_word(ptr + 0x18);
2276     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2277         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2278         return;
2279     }
2280 
2281     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2282     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2283     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2284     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2285 
2286     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2287                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2288 
2289     checksum = ptr[6];
2290 
2291     if (vendor_id != rom_vendor_id) {
2292         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2293         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2294         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2295         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2296         ptr[6] = checksum;
2297         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2298     }
2299 
2300     if (device_id != rom_device_id) {
2301         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2302         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2303         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2304         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2305         ptr[6] = checksum;
2306         pci_set_word(ptr + pcir_offset + 6, device_id);
2307     }
2308 }
2309 
2310 /* Add an option rom for the device */
2311 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2312                                Error **errp)
2313 {
2314     int64_t size = 0;
2315     g_autofree char *path = NULL;
2316     char name[32];
2317     const VMStateDescription *vmsd;
2318 
2319     /*
2320      * In case of incoming migration ROM will come with migration stream, no
2321      * reason to load the file.  Neither we want to fail if local ROM file
2322      * mismatches with specified romsize.
2323      */
2324     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2325 
2326     if (!pdev->romfile || !strlen(pdev->romfile)) {
2327         return;
2328     }
2329 
2330     if (!pdev->rom_bar) {
2331         /*
2332          * Load rom via fw_cfg instead of creating a rom bar,
2333          * for 0.11 compatibility.
2334          */
2335         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2336 
2337         /*
2338          * Hot-plugged devices can't use the option ROM
2339          * if the rom bar is disabled.
2340          */
2341         if (DEVICE(pdev)->hotplugged) {
2342             error_setg(errp, "Hot-plugged device without ROM bar"
2343                        " can't have an option ROM");
2344             return;
2345         }
2346 
2347         if (class == 0x0300) {
2348             rom_add_vga(pdev->romfile);
2349         } else {
2350             rom_add_option(pdev->romfile, -1);
2351         }
2352         return;
2353     }
2354 
2355     if (load_file || pdev->romsize == -1) {
2356         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2357         if (path == NULL) {
2358             path = g_strdup(pdev->romfile);
2359         }
2360 
2361         size = get_image_size(path);
2362         if (size < 0) {
2363             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2364             return;
2365         } else if (size == 0) {
2366             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2367             return;
2368         } else if (size > 2 * GiB) {
2369             error_setg(errp,
2370                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2371                        pdev->romfile);
2372             return;
2373         }
2374         if (pdev->romsize != -1) {
2375             if (size > pdev->romsize) {
2376                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2377                            "is too large for ROM size %u",
2378                            pdev->romfile, (uint32_t)size, pdev->romsize);
2379                 return;
2380             }
2381         } else {
2382             pdev->romsize = pow2ceil(size);
2383         }
2384     }
2385 
2386     vmsd = qdev_get_vmsd(DEVICE(pdev));
2387     snprintf(name, sizeof(name), "%s.rom",
2388              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2389 
2390     pdev->has_rom = true;
2391     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2392                            &error_fatal);
2393 
2394     if (load_file) {
2395         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2396 
2397         if (load_image_size(path, ptr, size) < 0) {
2398             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2399             return;
2400         }
2401 
2402         if (is_default_rom) {
2403             /* Only the default rom images will be patched (if needed). */
2404             pci_patch_ids(pdev, ptr, size);
2405         }
2406     }
2407 
2408     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2409 }
2410 
2411 static void pci_del_option_rom(PCIDevice *pdev)
2412 {
2413     if (!pdev->has_rom)
2414         return;
2415 
2416     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2417     pdev->has_rom = false;
2418 }
2419 
2420 /*
2421  * On success, pci_add_capability() returns a positive value
2422  * that the offset of the pci capability.
2423  * On failure, it sets an error and returns a negative error
2424  * code.
2425  */
2426 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2427                        uint8_t offset, uint8_t size,
2428                        Error **errp)
2429 {
2430     uint8_t *config;
2431     int i, overlapping_cap;
2432 
2433     if (!offset) {
2434         offset = pci_find_space(pdev, size);
2435         /* out of PCI config space is programming error */
2436         assert(offset);
2437     } else {
2438         /* Verify that capabilities don't overlap.  Note: device assignment
2439          * depends on this check to verify that the device is not broken.
2440          * Should never trigger for emulated devices, but it's helpful
2441          * for debugging these. */
2442         for (i = offset; i < offset + size; i++) {
2443             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2444             if (overlapping_cap) {
2445                 error_setg(errp, "%s:%02x:%02x.%x "
2446                            "Attempt to add PCI capability %x at offset "
2447                            "%x overlaps existing capability %x at offset %x",
2448                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2449                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2450                            cap_id, offset, overlapping_cap, i);
2451                 return -EINVAL;
2452             }
2453         }
2454     }
2455 
2456     config = pdev->config + offset;
2457     config[PCI_CAP_LIST_ID] = cap_id;
2458     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2459     pdev->config[PCI_CAPABILITY_LIST] = offset;
2460     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2461     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2462     /* Make capability read-only by default */
2463     memset(pdev->wmask + offset, 0, size);
2464     /* Check capability by default */
2465     memset(pdev->cmask + offset, 0xFF, size);
2466     return offset;
2467 }
2468 
2469 /* Unlink capability from the pci config space. */
2470 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2471 {
2472     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2473     if (!offset)
2474         return;
2475     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2476     /* Make capability writable again */
2477     memset(pdev->wmask + offset, 0xff, size);
2478     memset(pdev->w1cmask + offset, 0, size);
2479     /* Clear cmask as device-specific registers can't be checked */
2480     memset(pdev->cmask + offset, 0, size);
2481     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2482 
2483     if (!pdev->config[PCI_CAPABILITY_LIST])
2484         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2485 }
2486 
2487 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2488 {
2489     return pci_find_capability_list(pdev, cap_id, NULL);
2490 }
2491 
2492 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2493 {
2494     PCIDevice *d = (PCIDevice *)dev;
2495     const char *name = NULL;
2496     const pci_class_desc *desc =  pci_class_descriptions;
2497     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2498 
2499     while (desc->desc &&
2500           (class & ~desc->fw_ign_bits) !=
2501           (desc->class & ~desc->fw_ign_bits)) {
2502         desc++;
2503     }
2504 
2505     if (desc->desc) {
2506         name = desc->fw_name;
2507     }
2508 
2509     if (name) {
2510         pstrcpy(buf, len, name);
2511     } else {
2512         snprintf(buf, len, "pci%04x,%04x",
2513                  pci_get_word(d->config + PCI_VENDOR_ID),
2514                  pci_get_word(d->config + PCI_DEVICE_ID));
2515     }
2516 
2517     return buf;
2518 }
2519 
2520 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2521 {
2522     PCIDevice *d = (PCIDevice *)dev;
2523     char name[33];
2524     int has_func = !!PCI_FUNC(d->devfn);
2525 
2526     return g_strdup_printf("%s@%x%s%.*x",
2527                            pci_dev_fw_name(dev, name, sizeof(name)),
2528                            PCI_SLOT(d->devfn),
2529                            has_func ? "," : "",
2530                            has_func,
2531                            PCI_FUNC(d->devfn));
2532 }
2533 
2534 static char *pcibus_get_dev_path(DeviceState *dev)
2535 {
2536     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2537     PCIDevice *t;
2538     int slot_depth;
2539     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2540      * 00 is added here to make this format compatible with
2541      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2542      * Slot.Function list specifies the slot and function numbers for all
2543      * devices on the path from root to the specific device. */
2544     const char *root_bus_path;
2545     int root_bus_len;
2546     char slot[] = ":SS.F";
2547     int slot_len = sizeof slot - 1 /* For '\0' */;
2548     int path_len;
2549     char *path, *p;
2550     int s;
2551 
2552     root_bus_path = pci_root_bus_path(d);
2553     root_bus_len = strlen(root_bus_path);
2554 
2555     /* Calculate # of slots on path between device and root. */;
2556     slot_depth = 0;
2557     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2558         ++slot_depth;
2559     }
2560 
2561     path_len = root_bus_len + slot_len * slot_depth;
2562 
2563     /* Allocate memory, fill in the terminating null byte. */
2564     path = g_malloc(path_len + 1 /* For '\0' */);
2565     path[path_len] = '\0';
2566 
2567     memcpy(path, root_bus_path, root_bus_len);
2568 
2569     /* Fill in slot numbers. We walk up from device to root, so need to print
2570      * them in the reverse order, last to first. */
2571     p = path + path_len;
2572     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2573         p -= slot_len;
2574         s = snprintf(slot, sizeof slot, ":%02x.%x",
2575                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2576         assert(s == slot_len);
2577         memcpy(p, slot, slot_len);
2578     }
2579 
2580     return path;
2581 }
2582 
2583 static int pci_qdev_find_recursive(PCIBus *bus,
2584                                    const char *id, PCIDevice **pdev)
2585 {
2586     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2587     if (!qdev) {
2588         return -ENODEV;
2589     }
2590 
2591     /* roughly check if given qdev is pci device */
2592     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2593         *pdev = PCI_DEVICE(qdev);
2594         return 0;
2595     }
2596     return -EINVAL;
2597 }
2598 
2599 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2600 {
2601     PCIHostState *host_bridge;
2602     int rc = -ENODEV;
2603 
2604     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2605         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2606         if (!tmp) {
2607             rc = 0;
2608             break;
2609         }
2610         if (tmp != -ENODEV) {
2611             rc = tmp;
2612         }
2613     }
2614 
2615     return rc;
2616 }
2617 
2618 MemoryRegion *pci_address_space(PCIDevice *dev)
2619 {
2620     return pci_get_bus(dev)->address_space_mem;
2621 }
2622 
2623 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2624 {
2625     return pci_get_bus(dev)->address_space_io;
2626 }
2627 
2628 static void pci_device_class_init(ObjectClass *klass, void *data)
2629 {
2630     DeviceClass *k = DEVICE_CLASS(klass);
2631 
2632     k->realize = pci_qdev_realize;
2633     k->unrealize = pci_qdev_unrealize;
2634     k->bus_type = TYPE_PCI_BUS;
2635     device_class_set_props(k, pci_props);
2636 }
2637 
2638 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2639 {
2640     if (!object_class_is_abstract(klass)) {
2641         ObjectClass *conventional =
2642             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2643         ObjectClass *pcie =
2644             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2645         ObjectClass *cxl =
2646             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2647         assert(conventional || pcie || cxl);
2648     }
2649 }
2650 
2651 /*
2652  * Get IOMMU root bus, aliased bus and devfn of a PCI device
2653  *
2654  * IOMMU root bus is needed by all call sites to call into iommu_ops.
2655  * For call sites which don't need aliased BDF, passing NULL to
2656  * aliased_[bus|devfn] is allowed.
2657  *
2658  * @piommu_bus: return root #PCIBus backed by an IOMMU for the PCI device.
2659  *
2660  * @aliased_bus: return aliased #PCIBus of the PCI device, optional.
2661  *
2662  * @aliased_devfn: return aliased devfn of the PCI device, optional.
2663  */
2664 static void pci_device_get_iommu_bus_devfn(PCIDevice *dev,
2665                                            PCIBus **piommu_bus,
2666                                            PCIBus **aliased_bus,
2667                                            int *aliased_devfn)
2668 {
2669     PCIBus *bus = pci_get_bus(dev);
2670     PCIBus *iommu_bus = bus;
2671     int devfn = dev->devfn;
2672 
2673     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2674         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2675 
2676         /*
2677          * The requester ID of the provided device may be aliased, as seen from
2678          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2679          * requester ID to provide a unique AddressSpace for devices, but
2680          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2681          * to-PCI bridge creates and accepts transactions on behalf of down-
2682          * stream devices.  When doing so, all downstream devices are masked
2683          * (aliased) behind a single requester ID.  The requester ID used
2684          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2685          * bridges, with a PCIe capability indicating such, follow the
2686          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2687          * where the bridge uses the seconary bus as the bridge portion of the
2688          * requester ID and devfn of 00.0.  For other bridges, typically those
2689          * found on the root complex such as the dmi-to-pci-bridge, we follow
2690          * the convention of typical bare-metal hardware, which uses the
2691          * requester ID of the bridge itself.  There are device specific
2692          * exceptions to these rules, but these are the defaults that the
2693          * Linux kernel uses when determining DMA aliases itself and believed
2694          * to be true for the bare metal equivalents of the devices emulated
2695          * in QEMU.
2696          */
2697         if (!pci_bus_is_express(iommu_bus)) {
2698             PCIDevice *parent = iommu_bus->parent_dev;
2699 
2700             if (pci_is_express(parent) &&
2701                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2702                 devfn = PCI_DEVFN(0, 0);
2703                 bus = iommu_bus;
2704             } else {
2705                 devfn = parent->devfn;
2706                 bus = parent_bus;
2707             }
2708         }
2709 
2710         iommu_bus = parent_bus;
2711     }
2712 
2713     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
2714     assert(iommu_bus);
2715 
2716     if (pci_bus_bypass_iommu(bus) || !iommu_bus->iommu_ops) {
2717         iommu_bus = NULL;
2718     }
2719 
2720     *piommu_bus = iommu_bus;
2721 
2722     if (aliased_bus) {
2723         *aliased_bus = bus;
2724     }
2725 
2726     if (aliased_devfn) {
2727         *aliased_devfn = devfn;
2728     }
2729 }
2730 
2731 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2732 {
2733     PCIBus *bus;
2734     PCIBus *iommu_bus;
2735     int devfn;
2736 
2737     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn);
2738     if (iommu_bus) {
2739         return iommu_bus->iommu_ops->get_address_space(bus,
2740                                  iommu_bus->iommu_opaque, devfn);
2741     }
2742     return &address_space_memory;
2743 }
2744 
2745 bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod,
2746                                  Error **errp)
2747 {
2748     PCIBus *iommu_bus, *aliased_bus;
2749     int aliased_devfn;
2750 
2751     /* set_iommu_device requires device's direct BDF instead of aliased BDF */
2752     pci_device_get_iommu_bus_devfn(dev, &iommu_bus,
2753                                    &aliased_bus, &aliased_devfn);
2754     if (iommu_bus && iommu_bus->iommu_ops->set_iommu_device) {
2755         hiod->aliased_bus = aliased_bus;
2756         hiod->aliased_devfn = aliased_devfn;
2757         return iommu_bus->iommu_ops->set_iommu_device(pci_get_bus(dev),
2758                                                       iommu_bus->iommu_opaque,
2759                                                       dev->devfn, hiod, errp);
2760     }
2761     return true;
2762 }
2763 
2764 void pci_device_unset_iommu_device(PCIDevice *dev)
2765 {
2766     PCIBus *iommu_bus;
2767 
2768     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, NULL, NULL);
2769     if (iommu_bus && iommu_bus->iommu_ops->unset_iommu_device) {
2770         return iommu_bus->iommu_ops->unset_iommu_device(pci_get_bus(dev),
2771                                                         iommu_bus->iommu_opaque,
2772                                                         dev->devfn);
2773     }
2774 }
2775 
2776 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2777 {
2778     /*
2779      * If called, pci_setup_iommu() should provide a minimum set of
2780      * useful callbacks for the bus.
2781      */
2782     assert(ops);
2783     assert(ops->get_address_space);
2784 
2785     bus->iommu_ops = ops;
2786     bus->iommu_opaque = opaque;
2787 }
2788 
2789 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2790 {
2791     Range *range = opaque;
2792     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2793     int i;
2794 
2795     if (!(cmd & PCI_COMMAND_MEMORY)) {
2796         return;
2797     }
2798 
2799     if (IS_PCI_BRIDGE(dev)) {
2800         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2801         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2802 
2803         base = MAX(base, 0x1ULL << 32);
2804 
2805         if (limit >= base) {
2806             Range pref_range;
2807             range_set_bounds(&pref_range, base, limit);
2808             range_extend(range, &pref_range);
2809         }
2810     }
2811     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2812         PCIIORegion *r = &dev->io_regions[i];
2813         pcibus_t lob, upb;
2814         Range region_range;
2815 
2816         if (!r->size ||
2817             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2818             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2819             continue;
2820         }
2821 
2822         lob = pci_bar_address(dev, i, r->type, r->size);
2823         upb = lob + r->size - 1;
2824         if (lob == PCI_BAR_UNMAPPED) {
2825             continue;
2826         }
2827 
2828         lob = MAX(lob, 0x1ULL << 32);
2829 
2830         if (upb >= lob) {
2831             range_set_bounds(&region_range, lob, upb);
2832             range_extend(range, &region_range);
2833         }
2834     }
2835 }
2836 
2837 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2838 {
2839     range_make_empty(range);
2840     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2841 }
2842 
2843 static bool pcie_has_upstream_port(PCIDevice *dev)
2844 {
2845     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2846 
2847     /* Device associated with an upstream port.
2848      * As there are several types of these, it's easier to check the
2849      * parent device: upstream ports are always connected to
2850      * root or downstream ports.
2851      */
2852     return parent_dev &&
2853         pci_is_express(parent_dev) &&
2854         parent_dev->exp.exp_cap &&
2855         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2856          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2857 }
2858 
2859 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2860 {
2861     PCIBus *bus = pci_get_bus(pci_dev);
2862 
2863     if(pcie_has_upstream_port(pci_dev)) {
2864         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2865         return bus->devices[0];
2866     } else {
2867         /* Other bus types might support multiple devices at slots 0-31 */
2868         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2869     }
2870 }
2871 
2872 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2873 {
2874     MSIMessage msg;
2875     if (msix_enabled(dev)) {
2876         msg = msix_get_message(dev, vector);
2877     } else if (msi_enabled(dev)) {
2878         msg = msi_get_message(dev, vector);
2879     } else {
2880         /* Should never happen */
2881         error_report("%s: unknown interrupt type", __func__);
2882         abort();
2883     }
2884     return msg;
2885 }
2886 
2887 void pci_set_enabled(PCIDevice *d, bool state)
2888 {
2889     if (d->enabled == state) {
2890         return;
2891     }
2892 
2893     d->enabled = state;
2894     pci_update_mappings(d);
2895     memory_region_set_enabled(&d->bus_master_enable_region,
2896                               (pci_get_word(d->config + PCI_COMMAND)
2897                                & PCI_COMMAND_MASTER) && d->enabled);
2898     if (d->qdev.realized) {
2899         pci_device_reset(d);
2900     }
2901 }
2902 
2903 static const TypeInfo pci_device_type_info = {
2904     .name = TYPE_PCI_DEVICE,
2905     .parent = TYPE_DEVICE,
2906     .instance_size = sizeof(PCIDevice),
2907     .abstract = true,
2908     .class_size = sizeof(PCIDeviceClass),
2909     .class_init = pci_device_class_init,
2910     .class_base_init = pci_device_class_base_init,
2911 };
2912 
2913 static void pci_register_types(void)
2914 {
2915     type_register_static(&pci_bus_info);
2916     type_register_static(&pcie_bus_info);
2917     type_register_static(&cxl_bus_info);
2918     type_register_static(&conventional_pci_interface_info);
2919     type_register_static(&cxl_interface_info);
2920     type_register_static(&pcie_interface_info);
2921     type_register_static(&pci_device_type_info);
2922 }
2923 
2924 type_init(pci_register_types)
2925