xref: /openbmc/qemu/hw/pci/pci.c (revision 700784bf)
1 /*
2  * QEMU PCI bus manager
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to deal
8  * in the Software without restriction, including without limitation the rights
9  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
10  * copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
22  * THE SOFTWARE.
23  */
24 
25 #include "qemu/osdep.h"
26 #include "qemu/datadir.h"
27 #include "qemu/units.h"
28 #include "hw/irq.h"
29 #include "hw/pci/pci.h"
30 #include "hw/pci/pci_bridge.h"
31 #include "hw/pci/pci_bus.h"
32 #include "hw/pci/pci_host.h"
33 #include "hw/qdev-properties.h"
34 #include "hw/qdev-properties-system.h"
35 #include "migration/qemu-file-types.h"
36 #include "migration/vmstate.h"
37 #include "net/net.h"
38 #include "sysemu/numa.h"
39 #include "sysemu/runstate.h"
40 #include "sysemu/sysemu.h"
41 #include "hw/loader.h"
42 #include "qemu/error-report.h"
43 #include "qemu/range.h"
44 #include "trace.h"
45 #include "hw/pci/msi.h"
46 #include "hw/pci/msix.h"
47 #include "hw/hotplug.h"
48 #include "hw/boards.h"
49 #include "qapi/error.h"
50 #include "qemu/cutils.h"
51 #include "pci-internal.h"
52 
53 #include "hw/xen/xen.h"
54 #include "hw/i386/kvm/xen_evtchn.h"
55 
56 //#define DEBUG_PCI
57 #ifdef DEBUG_PCI
58 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
59 #else
60 # define PCI_DPRINTF(format, ...)       do { } while (0)
61 #endif
62 
63 bool pci_available = true;
64 
65 static char *pcibus_get_dev_path(DeviceState *dev);
66 static char *pcibus_get_fw_dev_path(DeviceState *dev);
67 static void pcibus_reset_hold(Object *obj, ResetType type);
68 static bool pcie_has_upstream_port(PCIDevice *dev);
69 
70 static void prop_pci_busnr_get(Object *obj, Visitor *v, const char *name,
71                              void *opaque, Error **errp)
72 {
73     uint8_t busnr = pci_dev_bus_num(PCI_DEVICE(obj));
74 
75     visit_type_uint8(v, name, &busnr, errp);
76 }
77 
78 static const PropertyInfo prop_pci_busnr = {
79     .name = "busnr",
80     .get = prop_pci_busnr_get,
81 };
82 
83 static Property pci_props[] = {
84     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
85     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
86     DEFINE_PROP_UINT32("romsize", PCIDevice, romsize, UINT32_MAX),
87     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
88     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
89                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
90     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
91                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
92     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
93                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
94     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
95                        failover_pair_id),
96     DEFINE_PROP_UINT32("acpi-index",  PCIDevice, acpi_index, 0),
97     DEFINE_PROP_BIT("x-pcie-err-unc-mask", PCIDevice, cap_present,
98                     QEMU_PCIE_ERR_UNC_MASK_BITNR, true),
99     DEFINE_PROP_BIT("x-pcie-ari-nextfn-1", PCIDevice, cap_present,
100                     QEMU_PCIE_ARI_NEXTFN_1_BITNR, false),
101     DEFINE_PROP_SIZE32("x-max-bounce-buffer-size", PCIDevice,
102                      max_bounce_buffer_size, DEFAULT_MAX_BOUNCE_BUFFER_SIZE),
103     DEFINE_PROP_BIT("x-pcie-ext-tag", PCIDevice, cap_present,
104                     QEMU_PCIE_EXT_TAG_BITNR, true),
105     { .name = "busnr", .info = &prop_pci_busnr },
106     DEFINE_PROP_END_OF_LIST()
107 };
108 
109 static const VMStateDescription vmstate_pcibus = {
110     .name = "PCIBUS",
111     .version_id = 1,
112     .minimum_version_id = 1,
113     .fields = (const VMStateField[]) {
114         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
115         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
116                              nirq, 0, vmstate_info_int32,
117                              int32_t),
118         VMSTATE_END_OF_LIST()
119     }
120 };
121 
122 static gint g_cmp_uint32(gconstpointer a, gconstpointer b, gpointer user_data)
123 {
124     return a - b;
125 }
126 
127 static GSequence *pci_acpi_index_list(void)
128 {
129     static GSequence *used_acpi_index_list;
130 
131     if (!used_acpi_index_list) {
132         used_acpi_index_list = g_sequence_new(NULL);
133     }
134     return used_acpi_index_list;
135 }
136 
137 static void pci_init_bus_master(PCIDevice *pci_dev)
138 {
139     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
140 
141     memory_region_init_alias(&pci_dev->bus_master_enable_region,
142                              OBJECT(pci_dev), "bus master",
143                              dma_as->root, 0, memory_region_size(dma_as->root));
144     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
145     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
146                                 &pci_dev->bus_master_enable_region);
147 }
148 
149 static void pcibus_machine_done(Notifier *notifier, void *data)
150 {
151     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
152     int i;
153 
154     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
155         if (bus->devices[i]) {
156             pci_init_bus_master(bus->devices[i]);
157         }
158     }
159 }
160 
161 static void pci_bus_realize(BusState *qbus, Error **errp)
162 {
163     PCIBus *bus = PCI_BUS(qbus);
164 
165     bus->machine_done.notify = pcibus_machine_done;
166     qemu_add_machine_init_done_notifier(&bus->machine_done);
167 
168     vmstate_register_any(NULL, &vmstate_pcibus, bus);
169 }
170 
171 static void pcie_bus_realize(BusState *qbus, Error **errp)
172 {
173     PCIBus *bus = PCI_BUS(qbus);
174     Error *local_err = NULL;
175 
176     pci_bus_realize(qbus, &local_err);
177     if (local_err) {
178         error_propagate(errp, local_err);
179         return;
180     }
181 
182     /*
183      * A PCI-E bus can support extended config space if it's the root
184      * bus, or if the bus/bridge above it does as well
185      */
186     if (pci_bus_is_root(bus)) {
187         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
188     } else {
189         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
190 
191         if (pci_bus_allows_extended_config_space(parent_bus)) {
192             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
193         }
194     }
195 }
196 
197 static void pci_bus_unrealize(BusState *qbus)
198 {
199     PCIBus *bus = PCI_BUS(qbus);
200 
201     qemu_remove_machine_init_done_notifier(&bus->machine_done);
202 
203     vmstate_unregister(NULL, &vmstate_pcibus, bus);
204 }
205 
206 static int pcibus_num(PCIBus *bus)
207 {
208     if (pci_bus_is_root(bus)) {
209         return 0; /* pci host bridge */
210     }
211     return bus->parent_dev->config[PCI_SECONDARY_BUS];
212 }
213 
214 static uint16_t pcibus_numa_node(PCIBus *bus)
215 {
216     return NUMA_NODE_UNASSIGNED;
217 }
218 
219 static void pci_bus_class_init(ObjectClass *klass, void *data)
220 {
221     BusClass *k = BUS_CLASS(klass);
222     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
223     ResettableClass *rc = RESETTABLE_CLASS(klass);
224 
225     k->print_dev = pcibus_dev_print;
226     k->get_dev_path = pcibus_get_dev_path;
227     k->get_fw_dev_path = pcibus_get_fw_dev_path;
228     k->realize = pci_bus_realize;
229     k->unrealize = pci_bus_unrealize;
230 
231     rc->phases.hold = pcibus_reset_hold;
232 
233     pbc->bus_num = pcibus_num;
234     pbc->numa_node = pcibus_numa_node;
235 }
236 
237 static const TypeInfo pci_bus_info = {
238     .name = TYPE_PCI_BUS,
239     .parent = TYPE_BUS,
240     .instance_size = sizeof(PCIBus),
241     .class_size = sizeof(PCIBusClass),
242     .class_init = pci_bus_class_init,
243 };
244 
245 static const TypeInfo cxl_interface_info = {
246     .name          = INTERFACE_CXL_DEVICE,
247     .parent        = TYPE_INTERFACE,
248 };
249 
250 static const TypeInfo pcie_interface_info = {
251     .name          = INTERFACE_PCIE_DEVICE,
252     .parent        = TYPE_INTERFACE,
253 };
254 
255 static const TypeInfo conventional_pci_interface_info = {
256     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
257     .parent        = TYPE_INTERFACE,
258 };
259 
260 static void pcie_bus_class_init(ObjectClass *klass, void *data)
261 {
262     BusClass *k = BUS_CLASS(klass);
263 
264     k->realize = pcie_bus_realize;
265 }
266 
267 static const TypeInfo pcie_bus_info = {
268     .name = TYPE_PCIE_BUS,
269     .parent = TYPE_PCI_BUS,
270     .class_init = pcie_bus_class_init,
271 };
272 
273 static const TypeInfo cxl_bus_info = {
274     .name       = TYPE_CXL_BUS,
275     .parent     = TYPE_PCIE_BUS,
276     .class_init = pcie_bus_class_init,
277 };
278 
279 static void pci_update_mappings(PCIDevice *d);
280 static void pci_irq_handler(void *opaque, int irq_num, int level);
281 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
282 static void pci_del_option_rom(PCIDevice *pdev);
283 
284 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
285 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
286 
287 PCIHostStateList pci_host_bridges;
288 
289 int pci_bar(PCIDevice *d, int reg)
290 {
291     uint8_t type;
292 
293     /* PCIe virtual functions do not have their own BARs */
294     assert(!pci_is_vf(d));
295 
296     if (reg != PCI_ROM_SLOT)
297         return PCI_BASE_ADDRESS_0 + reg * 4;
298 
299     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
300     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
301 }
302 
303 static inline int pci_irq_state(PCIDevice *d, int irq_num)
304 {
305         return (d->irq_state >> irq_num) & 0x1;
306 }
307 
308 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
309 {
310         d->irq_state &= ~(0x1 << irq_num);
311         d->irq_state |= level << irq_num;
312 }
313 
314 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
315 {
316     assert(irq_num >= 0);
317     assert(irq_num < bus->nirq);
318     bus->irq_count[irq_num] += change;
319     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
320 }
321 
322 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
323 {
324     PCIBus *bus;
325     for (;;) {
326         int dev_irq = irq_num;
327         bus = pci_get_bus(pci_dev);
328         assert(bus->map_irq);
329         irq_num = bus->map_irq(pci_dev, irq_num);
330         trace_pci_route_irq(dev_irq, DEVICE(pci_dev)->canonical_path, irq_num,
331                             pci_bus_is_root(bus) ? "root-complex"
332                                     : DEVICE(bus->parent_dev)->canonical_path);
333         if (bus->set_irq)
334             break;
335         pci_dev = bus->parent_dev;
336     }
337     pci_bus_change_irq_level(bus, irq_num, change);
338 }
339 
340 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
341 {
342     assert(irq_num >= 0);
343     assert(irq_num < bus->nirq);
344     return !!bus->irq_count[irq_num];
345 }
346 
347 /* Update interrupt status bit in config space on interrupt
348  * state change. */
349 static void pci_update_irq_status(PCIDevice *dev)
350 {
351     if (dev->irq_state) {
352         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
353     } else {
354         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
355     }
356 }
357 
358 void pci_device_deassert_intx(PCIDevice *dev)
359 {
360     int i;
361     for (i = 0; i < PCI_NUM_PINS; ++i) {
362         pci_irq_handler(dev, i, 0);
363     }
364 }
365 
366 static void pci_msi_trigger(PCIDevice *dev, MSIMessage msg)
367 {
368     MemTxAttrs attrs = {};
369 
370     /*
371      * Xen uses the high bits of the address to contain some of the bits
372      * of the PIRQ#. Therefore we can't just send the write cycle and
373      * trust that it's caught by the APIC at 0xfee00000 because the
374      * target of the write might be e.g. 0x0x1000fee46000 for PIRQ#4166.
375      * So we intercept the delivery here instead of in kvm_send_msi().
376      */
377     if (xen_mode == XEN_EMULATE &&
378         xen_evtchn_deliver_pirq_msi(msg.address, msg.data)) {
379         return;
380     }
381     attrs.requester_id = pci_requester_id(dev);
382     address_space_stl_le(&dev->bus_master_as, msg.address, msg.data,
383                          attrs, NULL);
384 }
385 
386 static void pci_reset_regions(PCIDevice *dev)
387 {
388     int r;
389     if (pci_is_vf(dev)) {
390         return;
391     }
392 
393     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
394         PCIIORegion *region = &dev->io_regions[r];
395         if (!region->size) {
396             continue;
397         }
398 
399         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
400             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
401             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
402         } else {
403             pci_set_long(dev->config + pci_bar(dev, r), region->type);
404         }
405     }
406 }
407 
408 static void pci_do_device_reset(PCIDevice *dev)
409 {
410     pci_device_deassert_intx(dev);
411     assert(dev->irq_state == 0);
412 
413     /* Clear all writable bits */
414     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
415                                  pci_get_word(dev->wmask + PCI_COMMAND) |
416                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
417     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
418                                  pci_get_word(dev->wmask + PCI_STATUS) |
419                                  pci_get_word(dev->w1cmask + PCI_STATUS));
420     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
421     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
422                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
423                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
424     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
425     pci_reset_regions(dev);
426     pci_update_mappings(dev);
427 
428     msi_reset(dev);
429     msix_reset(dev);
430     pcie_sriov_pf_reset(dev);
431 }
432 
433 /*
434  * This function is called on #RST and FLR.
435  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
436  */
437 void pci_device_reset(PCIDevice *dev)
438 {
439     device_cold_reset(&dev->qdev);
440     pci_do_device_reset(dev);
441 }
442 
443 /*
444  * Trigger pci bus reset under a given bus.
445  * Called via bus_cold_reset on RST# assert, after the devices
446  * have been reset device_cold_reset-ed already.
447  */
448 static void pcibus_reset_hold(Object *obj, ResetType type)
449 {
450     PCIBus *bus = PCI_BUS(obj);
451     int i;
452 
453     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
454         if (bus->devices[i]) {
455             pci_do_device_reset(bus->devices[i]);
456         }
457     }
458 
459     for (i = 0; i < bus->nirq; i++) {
460         assert(bus->irq_count[i] == 0);
461     }
462 }
463 
464 static void pci_host_bus_register(DeviceState *host)
465 {
466     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
467 
468     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
469 }
470 
471 static void pci_host_bus_unregister(DeviceState *host)
472 {
473     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
474 
475     QLIST_REMOVE(host_bridge, next);
476 }
477 
478 PCIBus *pci_device_root_bus(const PCIDevice *d)
479 {
480     PCIBus *bus = pci_get_bus(d);
481 
482     while (!pci_bus_is_root(bus)) {
483         d = bus->parent_dev;
484         assert(d != NULL);
485 
486         bus = pci_get_bus(d);
487     }
488 
489     return bus;
490 }
491 
492 const char *pci_root_bus_path(PCIDevice *dev)
493 {
494     PCIBus *rootbus = pci_device_root_bus(dev);
495     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
496     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
497 
498     assert(host_bridge->bus == rootbus);
499 
500     if (hc->root_bus_path) {
501         return (*hc->root_bus_path)(host_bridge, rootbus);
502     }
503 
504     return rootbus->qbus.name;
505 }
506 
507 bool pci_bus_bypass_iommu(PCIBus *bus)
508 {
509     PCIBus *rootbus = bus;
510     PCIHostState *host_bridge;
511 
512     if (!pci_bus_is_root(bus)) {
513         rootbus = pci_device_root_bus(bus->parent_dev);
514     }
515 
516     host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
517 
518     assert(host_bridge->bus == rootbus);
519 
520     return host_bridge->bypass_iommu;
521 }
522 
523 static void pci_root_bus_internal_init(PCIBus *bus, DeviceState *parent,
524                                        MemoryRegion *mem, MemoryRegion *io,
525                                        uint8_t devfn_min)
526 {
527     assert(PCI_FUNC(devfn_min) == 0);
528     bus->devfn_min = devfn_min;
529     bus->slot_reserved_mask = 0x0;
530     bus->address_space_mem = mem;
531     bus->address_space_io = io;
532     bus->flags |= PCI_BUS_IS_ROOT;
533 
534     /* host bridge */
535     QLIST_INIT(&bus->child);
536 
537     pci_host_bus_register(parent);
538 }
539 
540 static void pci_bus_uninit(PCIBus *bus)
541 {
542     pci_host_bus_unregister(BUS(bus)->parent);
543 }
544 
545 bool pci_bus_is_express(const PCIBus *bus)
546 {
547     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
548 }
549 
550 void pci_root_bus_init(PCIBus *bus, size_t bus_size, DeviceState *parent,
551                        const char *name,
552                        MemoryRegion *mem, MemoryRegion *io,
553                        uint8_t devfn_min, const char *typename)
554 {
555     qbus_init(bus, bus_size, typename, parent, name);
556     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
557 }
558 
559 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
560                          MemoryRegion *mem, MemoryRegion *io,
561                          uint8_t devfn_min, const char *typename)
562 {
563     PCIBus *bus;
564 
565     bus = PCI_BUS(qbus_new(typename, parent, name));
566     pci_root_bus_internal_init(bus, parent, mem, io, devfn_min);
567     return bus;
568 }
569 
570 void pci_root_bus_cleanup(PCIBus *bus)
571 {
572     pci_bus_uninit(bus);
573     /* the caller of the unplug hotplug handler will delete this device */
574     qbus_unrealize(BUS(bus));
575 }
576 
577 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq,
578                   void *irq_opaque, int nirq)
579 {
580     bus->set_irq = set_irq;
581     bus->irq_opaque = irq_opaque;
582     bus->nirq = nirq;
583     g_free(bus->irq_count);
584     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
585 }
586 
587 void pci_bus_map_irqs(PCIBus *bus, pci_map_irq_fn map_irq)
588 {
589     bus->map_irq = map_irq;
590 }
591 
592 void pci_bus_irqs_cleanup(PCIBus *bus)
593 {
594     bus->set_irq = NULL;
595     bus->map_irq = NULL;
596     bus->irq_opaque = NULL;
597     bus->nirq = 0;
598     g_free(bus->irq_count);
599     bus->irq_count = NULL;
600 }
601 
602 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
603                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
604                               void *irq_opaque,
605                               MemoryRegion *mem, MemoryRegion *io,
606                               uint8_t devfn_min, int nirq,
607                               const char *typename)
608 {
609     PCIBus *bus;
610 
611     bus = pci_root_bus_new(parent, name, mem, io, devfn_min, typename);
612     pci_bus_irqs(bus, set_irq, irq_opaque, nirq);
613     pci_bus_map_irqs(bus, map_irq);
614     return bus;
615 }
616 
617 void pci_unregister_root_bus(PCIBus *bus)
618 {
619     pci_bus_irqs_cleanup(bus);
620     pci_root_bus_cleanup(bus);
621 }
622 
623 int pci_bus_num(PCIBus *s)
624 {
625     return PCI_BUS_GET_CLASS(s)->bus_num(s);
626 }
627 
628 /* Returns the min and max bus numbers of a PCI bus hierarchy */
629 void pci_bus_range(PCIBus *bus, int *min_bus, int *max_bus)
630 {
631     int i;
632     *min_bus = *max_bus = pci_bus_num(bus);
633 
634     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
635         PCIDevice *dev = bus->devices[i];
636 
637         if (dev && IS_PCI_BRIDGE(dev)) {
638             *min_bus = MIN(*min_bus, dev->config[PCI_SECONDARY_BUS]);
639             *max_bus = MAX(*max_bus, dev->config[PCI_SUBORDINATE_BUS]);
640         }
641     }
642 }
643 
644 int pci_bus_numa_node(PCIBus *bus)
645 {
646     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
647 }
648 
649 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
650                                  const VMStateField *field)
651 {
652     PCIDevice *s = container_of(pv, PCIDevice, config);
653     uint8_t *config;
654     int i;
655 
656     assert(size == pci_config_size(s));
657     config = g_malloc(size);
658 
659     qemu_get_buffer(f, config, size);
660     for (i = 0; i < size; ++i) {
661         if ((config[i] ^ s->config[i]) &
662             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
663             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
664                          "cmask: %x wmask: %x w1cmask:%x", __func__,
665                          i, config[i], s->config[i],
666                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
667             g_free(config);
668             return -EINVAL;
669         }
670     }
671     memcpy(s->config, config, size);
672 
673     pci_update_mappings(s);
674     if (IS_PCI_BRIDGE(s)) {
675         pci_bridge_update_mappings(PCI_BRIDGE(s));
676     }
677 
678     memory_region_set_enabled(&s->bus_master_enable_region,
679                               pci_get_word(s->config + PCI_COMMAND)
680                               & PCI_COMMAND_MASTER);
681 
682     g_free(config);
683     return 0;
684 }
685 
686 /* just put buffer */
687 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
688                                  const VMStateField *field, JSONWriter *vmdesc)
689 {
690     const uint8_t **v = pv;
691     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
692     qemu_put_buffer(f, *v, size);
693 
694     return 0;
695 }
696 
697 static const VMStateInfo vmstate_info_pci_config = {
698     .name = "pci config",
699     .get  = get_pci_config_device,
700     .put  = put_pci_config_device,
701 };
702 
703 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
704                              const VMStateField *field)
705 {
706     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
707     uint32_t irq_state[PCI_NUM_PINS];
708     int i;
709     for (i = 0; i < PCI_NUM_PINS; ++i) {
710         irq_state[i] = qemu_get_be32(f);
711         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
712             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
713                     irq_state[i]);
714             return -EINVAL;
715         }
716     }
717 
718     for (i = 0; i < PCI_NUM_PINS; ++i) {
719         pci_set_irq_state(s, i, irq_state[i]);
720     }
721 
722     return 0;
723 }
724 
725 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
726                              const VMStateField *field, JSONWriter *vmdesc)
727 {
728     int i;
729     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
730 
731     for (i = 0; i < PCI_NUM_PINS; ++i) {
732         qemu_put_be32(f, pci_irq_state(s, i));
733     }
734 
735     return 0;
736 }
737 
738 static const VMStateInfo vmstate_info_pci_irq_state = {
739     .name = "pci irq state",
740     .get  = get_pci_irq_state,
741     .put  = put_pci_irq_state,
742 };
743 
744 static bool migrate_is_pcie(void *opaque, int version_id)
745 {
746     return pci_is_express((PCIDevice *)opaque);
747 }
748 
749 static bool migrate_is_not_pcie(void *opaque, int version_id)
750 {
751     return !pci_is_express((PCIDevice *)opaque);
752 }
753 
754 const VMStateDescription vmstate_pci_device = {
755     .name = "PCIDevice",
756     .version_id = 2,
757     .minimum_version_id = 1,
758     .fields = (const VMStateField[]) {
759         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
760         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
761                                    migrate_is_not_pcie,
762                                    0, vmstate_info_pci_config,
763                                    PCI_CONFIG_SPACE_SIZE),
764         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
765                                    migrate_is_pcie,
766                                    0, vmstate_info_pci_config,
767                                    PCIE_CONFIG_SPACE_SIZE),
768         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
769                                    vmstate_info_pci_irq_state,
770                                    PCI_NUM_PINS * sizeof(int32_t)),
771         VMSTATE_END_OF_LIST()
772     }
773 };
774 
775 
776 void pci_device_save(PCIDevice *s, QEMUFile *f)
777 {
778     /* Clear interrupt status bit: it is implicit
779      * in irq_state which we are saving.
780      * This makes us compatible with old devices
781      * which never set or clear this bit. */
782     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
783     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
784     /* Restore the interrupt status bit. */
785     pci_update_irq_status(s);
786 }
787 
788 int pci_device_load(PCIDevice *s, QEMUFile *f)
789 {
790     int ret;
791     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
792     /* Restore the interrupt status bit. */
793     pci_update_irq_status(s);
794     return ret;
795 }
796 
797 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
798 {
799     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
800                  pci_default_sub_vendor_id);
801     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
802                  pci_default_sub_device_id);
803 }
804 
805 /*
806  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
807  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
808  */
809 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
810                              unsigned int *slotp, unsigned int *funcp)
811 {
812     const char *p;
813     char *e;
814     unsigned long val;
815     unsigned long dom = 0, bus = 0;
816     unsigned int slot = 0;
817     unsigned int func = 0;
818 
819     p = addr;
820     val = strtoul(p, &e, 16);
821     if (e == p)
822         return -1;
823     if (*e == ':') {
824         bus = val;
825         p = e + 1;
826         val = strtoul(p, &e, 16);
827         if (e == p)
828             return -1;
829         if (*e == ':') {
830             dom = bus;
831             bus = val;
832             p = e + 1;
833             val = strtoul(p, &e, 16);
834             if (e == p)
835                 return -1;
836         }
837     }
838 
839     slot = val;
840 
841     if (funcp != NULL) {
842         if (*e != '.')
843             return -1;
844 
845         p = e + 1;
846         val = strtoul(p, &e, 16);
847         if (e == p)
848             return -1;
849 
850         func = val;
851     }
852 
853     /* if funcp == NULL func is 0 */
854     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
855         return -1;
856 
857     if (*e)
858         return -1;
859 
860     *domp = dom;
861     *busp = bus;
862     *slotp = slot;
863     if (funcp != NULL)
864         *funcp = func;
865     return 0;
866 }
867 
868 static void pci_init_cmask(PCIDevice *dev)
869 {
870     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
871     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
872     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
873     dev->cmask[PCI_REVISION_ID] = 0xff;
874     dev->cmask[PCI_CLASS_PROG] = 0xff;
875     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
876     dev->cmask[PCI_HEADER_TYPE] = 0xff;
877     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
878 }
879 
880 static void pci_init_wmask(PCIDevice *dev)
881 {
882     int config_size = pci_config_size(dev);
883 
884     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
885     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
886     pci_set_word(dev->wmask + PCI_COMMAND,
887                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
888                  PCI_COMMAND_INTX_DISABLE);
889     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
890 
891     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
892            config_size - PCI_CONFIG_HEADER_SIZE);
893 }
894 
895 static void pci_init_w1cmask(PCIDevice *dev)
896 {
897     /*
898      * Note: It's okay to set w1cmask even for readonly bits as
899      * long as their value is hardwired to 0.
900      */
901     pci_set_word(dev->w1cmask + PCI_STATUS,
902                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
903                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
904                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
905 }
906 
907 static void pci_init_mask_bridge(PCIDevice *d)
908 {
909     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
910        PCI_SEC_LATENCY_TIMER */
911     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
912 
913     /* base and limit */
914     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
915     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
916     pci_set_word(d->wmask + PCI_MEMORY_BASE,
917                  PCI_MEMORY_RANGE_MASK & 0xffff);
918     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
919                  PCI_MEMORY_RANGE_MASK & 0xffff);
920     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
921                  PCI_PREF_RANGE_MASK & 0xffff);
922     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
923                  PCI_PREF_RANGE_MASK & 0xffff);
924 
925     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
926     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
927 
928     /* Supported memory and i/o types */
929     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
930     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
931     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
932                                PCI_PREF_RANGE_TYPE_64);
933     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
934                                PCI_PREF_RANGE_TYPE_64);
935 
936     /*
937      * TODO: Bridges default to 10-bit VGA decoding but we currently only
938      * implement 16-bit decoding (no alias support).
939      */
940     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
941                  PCI_BRIDGE_CTL_PARITY |
942                  PCI_BRIDGE_CTL_SERR |
943                  PCI_BRIDGE_CTL_ISA |
944                  PCI_BRIDGE_CTL_VGA |
945                  PCI_BRIDGE_CTL_VGA_16BIT |
946                  PCI_BRIDGE_CTL_MASTER_ABORT |
947                  PCI_BRIDGE_CTL_BUS_RESET |
948                  PCI_BRIDGE_CTL_FAST_BACK |
949                  PCI_BRIDGE_CTL_DISCARD |
950                  PCI_BRIDGE_CTL_SEC_DISCARD |
951                  PCI_BRIDGE_CTL_DISCARD_SERR);
952     /* Below does not do anything as we never set this bit, put here for
953      * completeness. */
954     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
955                  PCI_BRIDGE_CTL_DISCARD_STATUS);
956     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
957     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
958     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
959                                PCI_PREF_RANGE_TYPE_MASK);
960     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
961                                PCI_PREF_RANGE_TYPE_MASK);
962 }
963 
964 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
965 {
966     uint8_t slot = PCI_SLOT(dev->devfn);
967     uint8_t func;
968 
969     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
970         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
971     }
972 
973     /*
974      * With SR/IOV and ARI, a device at function 0 need not be a multifunction
975      * device, as it may just be a VF that ended up with function 0 in
976      * the legacy PCI interpretation. Avoid failing in such cases:
977      */
978     if (pci_is_vf(dev) &&
979         dev->exp.sriov_vf.pf->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
980         return;
981     }
982 
983     /*
984      * multifunction bit is interpreted in two ways as follows.
985      *   - all functions must set the bit to 1.
986      *     Example: Intel X53
987      *   - function 0 must set the bit, but the rest function (> 0)
988      *     is allowed to leave the bit to 0.
989      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
990      *
991      * So OS (at least Linux) checks the bit of only function 0,
992      * and doesn't see the bit of function > 0.
993      *
994      * The below check allows both interpretation.
995      */
996     if (PCI_FUNC(dev->devfn)) {
997         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
998         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
999             /* function 0 should set multifunction bit */
1000             error_setg(errp, "PCI: single function device can't be populated "
1001                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
1002             return;
1003         }
1004         return;
1005     }
1006 
1007     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
1008         return;
1009     }
1010     /* function 0 indicates single function, so function > 0 must be NULL */
1011     for (func = 1; func < PCI_FUNC_MAX; ++func) {
1012         if (bus->devices[PCI_DEVFN(slot, func)]) {
1013             error_setg(errp, "PCI: %x.0 indicates single function, "
1014                        "but %x.%x is already populated.",
1015                        slot, slot, func);
1016             return;
1017         }
1018     }
1019 }
1020 
1021 static void pci_config_alloc(PCIDevice *pci_dev)
1022 {
1023     int config_size = pci_config_size(pci_dev);
1024 
1025     pci_dev->config = g_malloc0(config_size);
1026     pci_dev->cmask = g_malloc0(config_size);
1027     pci_dev->wmask = g_malloc0(config_size);
1028     pci_dev->w1cmask = g_malloc0(config_size);
1029     pci_dev->used = g_malloc0(config_size);
1030 }
1031 
1032 static void pci_config_free(PCIDevice *pci_dev)
1033 {
1034     g_free(pci_dev->config);
1035     g_free(pci_dev->cmask);
1036     g_free(pci_dev->wmask);
1037     g_free(pci_dev->w1cmask);
1038     g_free(pci_dev->used);
1039 }
1040 
1041 static void do_pci_unregister_device(PCIDevice *pci_dev)
1042 {
1043     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
1044     pci_config_free(pci_dev);
1045 
1046     if (xen_mode == XEN_EMULATE) {
1047         xen_evtchn_remove_pci_device(pci_dev);
1048     }
1049     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
1050         memory_region_del_subregion(&pci_dev->bus_master_container_region,
1051                                     &pci_dev->bus_master_enable_region);
1052     }
1053     address_space_destroy(&pci_dev->bus_master_as);
1054 }
1055 
1056 /* Extract PCIReqIDCache into BDF format */
1057 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
1058 {
1059     uint8_t bus_n;
1060     uint16_t result;
1061 
1062     switch (cache->type) {
1063     case PCI_REQ_ID_BDF:
1064         result = pci_get_bdf(cache->dev);
1065         break;
1066     case PCI_REQ_ID_SECONDARY_BUS:
1067         bus_n = pci_dev_bus_num(cache->dev);
1068         result = PCI_BUILD_BDF(bus_n, 0);
1069         break;
1070     default:
1071         error_report("Invalid PCI requester ID cache type: %d",
1072                      cache->type);
1073         exit(1);
1074         break;
1075     }
1076 
1077     return result;
1078 }
1079 
1080 /* Parse bridges up to the root complex and return requester ID
1081  * cache for specific device.  For full PCIe topology, the cache
1082  * result would be exactly the same as getting BDF of the device.
1083  * However, several tricks are required when system mixed up with
1084  * legacy PCI devices and PCIe-to-PCI bridges.
1085  *
1086  * Here we cache the proxy device (and type) not requester ID since
1087  * bus number might change from time to time.
1088  */
1089 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
1090 {
1091     PCIDevice *parent;
1092     PCIReqIDCache cache = {
1093         .dev = dev,
1094         .type = PCI_REQ_ID_BDF,
1095     };
1096 
1097     while (!pci_bus_is_root(pci_get_bus(dev))) {
1098         /* We are under PCI/PCIe bridges */
1099         parent = pci_get_bus(dev)->parent_dev;
1100         if (pci_is_express(parent)) {
1101             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
1102                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
1103                  * override the requester ID using secondary bus
1104                  * number of parent bridge with zeroed devfn
1105                  * (pcie-to-pci bridge spec chap 2.3). */
1106                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
1107                 cache.dev = dev;
1108             }
1109         } else {
1110             /* Legacy PCI, override requester ID with the bridge's
1111              * BDF upstream.  When the root complex connects to
1112              * legacy PCI devices (including buses), it can only
1113              * obtain requester ID info from directly attached
1114              * devices.  If devices are attached under bridges, only
1115              * the requester ID of the bridge that is directly
1116              * attached to the root complex can be recognized. */
1117             cache.type = PCI_REQ_ID_BDF;
1118             cache.dev = parent;
1119         }
1120         dev = parent;
1121     }
1122 
1123     return cache;
1124 }
1125 
1126 uint16_t pci_requester_id(PCIDevice *dev)
1127 {
1128     return pci_req_id_cache_extract(&dev->requester_id_cache);
1129 }
1130 
1131 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
1132 {
1133     return !(bus->devices[devfn]);
1134 }
1135 
1136 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
1137 {
1138     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
1139 }
1140 
1141 uint32_t pci_bus_get_slot_reserved_mask(PCIBus *bus)
1142 {
1143     return bus->slot_reserved_mask;
1144 }
1145 
1146 void pci_bus_set_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1147 {
1148     bus->slot_reserved_mask |= mask;
1149 }
1150 
1151 void pci_bus_clear_slot_reserved_mask(PCIBus *bus, uint32_t mask)
1152 {
1153     bus->slot_reserved_mask &= ~mask;
1154 }
1155 
1156 /* -1 for devfn means auto assign */
1157 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1158                                          const char *name, int devfn,
1159                                          Error **errp)
1160 {
1161     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1162     PCIConfigReadFunc *config_read = pc->config_read;
1163     PCIConfigWriteFunc *config_write = pc->config_write;
1164     Error *local_err = NULL;
1165     DeviceState *dev = DEVICE(pci_dev);
1166     PCIBus *bus = pci_get_bus(pci_dev);
1167     bool is_bridge = IS_PCI_BRIDGE(pci_dev);
1168 
1169     /* Only pci bridges can be attached to extra PCI root buses */
1170     if (pci_bus_is_root(bus) && bus->parent_dev && !is_bridge) {
1171         error_setg(errp,
1172                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1173                     bus->parent_dev->name);
1174         return NULL;
1175     }
1176 
1177     if (devfn < 0) {
1178         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1179             devfn += PCI_FUNC_MAX) {
1180             if (pci_bus_devfn_available(bus, devfn) &&
1181                    !pci_bus_devfn_reserved(bus, devfn)) {
1182                 goto found;
1183             }
1184         }
1185         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1186                    "or reserved", name);
1187         return NULL;
1188     found: ;
1189     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1190         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1191                    " reserved",
1192                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1193         return NULL;
1194     } else if (!pci_bus_devfn_available(bus, devfn)) {
1195         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1196                    " in use by %s,id=%s",
1197                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1198                    bus->devices[devfn]->name, bus->devices[devfn]->qdev.id);
1199         return NULL;
1200     }
1201 
1202     /*
1203      * Populating function 0 triggers a scan from the guest that
1204      * exposes other non-zero functions. Hence we need to ensure that
1205      * function 0 wasn't added yet.
1206      */
1207     if (dev->hotplugged && !pci_is_vf(pci_dev) &&
1208         pci_get_function_0(pci_dev)) {
1209         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1210                    " new func %s cannot be exposed to guest.",
1211                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1212                    pci_get_function_0(pci_dev)->name,
1213                    name);
1214 
1215        return NULL;
1216     }
1217 
1218     pci_dev->devfn = devfn;
1219     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1220     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1221 
1222     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1223                        "bus master container", UINT64_MAX);
1224     address_space_init(&pci_dev->bus_master_as,
1225                        &pci_dev->bus_master_container_region, pci_dev->name);
1226     pci_dev->bus_master_as.max_bounce_buffer_size =
1227         pci_dev->max_bounce_buffer_size;
1228 
1229     if (phase_check(PHASE_MACHINE_READY)) {
1230         pci_init_bus_master(pci_dev);
1231     }
1232     pci_dev->irq_state = 0;
1233     pci_config_alloc(pci_dev);
1234 
1235     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1236     pci_config_set_device_id(pci_dev->config, pc->device_id);
1237     pci_config_set_revision(pci_dev->config, pc->revision);
1238     pci_config_set_class(pci_dev->config, pc->class_id);
1239 
1240     if (!is_bridge) {
1241         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1242             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1243                          pc->subsystem_vendor_id);
1244             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1245                          pc->subsystem_id);
1246         } else {
1247             pci_set_default_subsystem_id(pci_dev);
1248         }
1249     } else {
1250         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1251         assert(!pc->subsystem_vendor_id);
1252         assert(!pc->subsystem_id);
1253     }
1254     pci_init_cmask(pci_dev);
1255     pci_init_wmask(pci_dev);
1256     pci_init_w1cmask(pci_dev);
1257     if (is_bridge) {
1258         pci_init_mask_bridge(pci_dev);
1259     }
1260     pci_init_multifunction(bus, pci_dev, &local_err);
1261     if (local_err) {
1262         error_propagate(errp, local_err);
1263         do_pci_unregister_device(pci_dev);
1264         return NULL;
1265     }
1266 
1267     if (!config_read)
1268         config_read = pci_default_read_config;
1269     if (!config_write)
1270         config_write = pci_default_write_config;
1271     pci_dev->config_read = config_read;
1272     pci_dev->config_write = config_write;
1273     bus->devices[devfn] = pci_dev;
1274     pci_dev->version_id = 2; /* Current pci device vmstate version */
1275     return pci_dev;
1276 }
1277 
1278 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1279 {
1280     PCIIORegion *r;
1281     int i;
1282 
1283     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1284         r = &pci_dev->io_regions[i];
1285         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1286             continue;
1287         memory_region_del_subregion(r->address_space, r->memory);
1288     }
1289 
1290     pci_unregister_vga(pci_dev);
1291 }
1292 
1293 static void pci_qdev_unrealize(DeviceState *dev)
1294 {
1295     PCIDevice *pci_dev = PCI_DEVICE(dev);
1296     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1297 
1298     pci_unregister_io_regions(pci_dev);
1299     pci_del_option_rom(pci_dev);
1300 
1301     if (pc->exit) {
1302         pc->exit(pci_dev);
1303     }
1304 
1305     pci_device_deassert_intx(pci_dev);
1306     do_pci_unregister_device(pci_dev);
1307 
1308     pci_dev->msi_trigger = NULL;
1309 
1310     /*
1311      * clean up acpi-index so it could reused by another device
1312      */
1313     if (pci_dev->acpi_index) {
1314         GSequence *used_indexes = pci_acpi_index_list();
1315 
1316         g_sequence_remove(g_sequence_lookup(used_indexes,
1317                           GINT_TO_POINTER(pci_dev->acpi_index),
1318                           g_cmp_uint32, NULL));
1319     }
1320 }
1321 
1322 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1323                       uint8_t type, MemoryRegion *memory)
1324 {
1325     PCIIORegion *r;
1326     uint32_t addr; /* offset in pci config space */
1327     uint64_t wmask;
1328     pcibus_t size = memory_region_size(memory);
1329     uint8_t hdr_type;
1330 
1331     assert(!pci_is_vf(pci_dev)); /* VFs must use pcie_sriov_vf_register_bar */
1332     assert(region_num >= 0);
1333     assert(region_num < PCI_NUM_REGIONS);
1334     assert(is_power_of_2(size));
1335 
1336     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1337     hdr_type =
1338         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1339     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1340 
1341     r = &pci_dev->io_regions[region_num];
1342     r->addr = PCI_BAR_UNMAPPED;
1343     r->size = size;
1344     r->type = type;
1345     r->memory = memory;
1346     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1347                         ? pci_get_bus(pci_dev)->address_space_io
1348                         : pci_get_bus(pci_dev)->address_space_mem;
1349 
1350     wmask = ~(size - 1);
1351     if (region_num == PCI_ROM_SLOT) {
1352         /* ROM enable bit is writable */
1353         wmask |= PCI_ROM_ADDRESS_ENABLE;
1354     }
1355 
1356     addr = pci_bar(pci_dev, region_num);
1357     pci_set_long(pci_dev->config + addr, type);
1358 
1359     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1360         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1361         pci_set_quad(pci_dev->wmask + addr, wmask);
1362         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1363     } else {
1364         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1365         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1366     }
1367 }
1368 
1369 static void pci_update_vga(PCIDevice *pci_dev)
1370 {
1371     uint16_t cmd;
1372 
1373     if (!pci_dev->has_vga) {
1374         return;
1375     }
1376 
1377     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1378 
1379     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1380                               cmd & PCI_COMMAND_MEMORY);
1381     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1382                               cmd & PCI_COMMAND_IO);
1383     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1384                               cmd & PCI_COMMAND_IO);
1385 }
1386 
1387 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1388                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1389 {
1390     PCIBus *bus = pci_get_bus(pci_dev);
1391 
1392     assert(!pci_dev->has_vga);
1393 
1394     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1395     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1396     memory_region_add_subregion_overlap(bus->address_space_mem,
1397                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1398 
1399     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1400     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1401     memory_region_add_subregion_overlap(bus->address_space_io,
1402                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1403 
1404     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1405     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1406     memory_region_add_subregion_overlap(bus->address_space_io,
1407                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1408     pci_dev->has_vga = true;
1409 
1410     pci_update_vga(pci_dev);
1411 }
1412 
1413 void pci_unregister_vga(PCIDevice *pci_dev)
1414 {
1415     PCIBus *bus = pci_get_bus(pci_dev);
1416 
1417     if (!pci_dev->has_vga) {
1418         return;
1419     }
1420 
1421     memory_region_del_subregion(bus->address_space_mem,
1422                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1423     memory_region_del_subregion(bus->address_space_io,
1424                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1425     memory_region_del_subregion(bus->address_space_io,
1426                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1427     pci_dev->has_vga = false;
1428 }
1429 
1430 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1431 {
1432     return pci_dev->io_regions[region_num].addr;
1433 }
1434 
1435 static pcibus_t pci_config_get_bar_addr(PCIDevice *d, int reg,
1436                                         uint8_t type, pcibus_t size)
1437 {
1438     pcibus_t new_addr;
1439     if (!pci_is_vf(d)) {
1440         int bar = pci_bar(d, reg);
1441         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1442             new_addr = pci_get_quad(d->config + bar);
1443         } else {
1444             new_addr = pci_get_long(d->config + bar);
1445         }
1446     } else {
1447         PCIDevice *pf = d->exp.sriov_vf.pf;
1448         uint16_t sriov_cap = pf->exp.sriov_cap;
1449         int bar = sriov_cap + PCI_SRIOV_BAR + reg * 4;
1450         uint16_t vf_offset =
1451             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_OFFSET);
1452         uint16_t vf_stride =
1453             pci_get_word(pf->config + sriov_cap + PCI_SRIOV_VF_STRIDE);
1454         uint32_t vf_num = (d->devfn - (pf->devfn + vf_offset)) / vf_stride;
1455 
1456         if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1457             new_addr = pci_get_quad(pf->config + bar);
1458         } else {
1459             new_addr = pci_get_long(pf->config + bar);
1460         }
1461         new_addr += vf_num * size;
1462     }
1463     /* The ROM slot has a specific enable bit, keep it intact */
1464     if (reg != PCI_ROM_SLOT) {
1465         new_addr &= ~(size - 1);
1466     }
1467     return new_addr;
1468 }
1469 
1470 pcibus_t pci_bar_address(PCIDevice *d,
1471                          int reg, uint8_t type, pcibus_t size)
1472 {
1473     pcibus_t new_addr, last_addr;
1474     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1475     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
1476     bool allow_0_address = mc->pci_allow_0_address;
1477 
1478     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1479         if (!(cmd & PCI_COMMAND_IO)) {
1480             return PCI_BAR_UNMAPPED;
1481         }
1482         new_addr = pci_config_get_bar_addr(d, reg, type, size);
1483         last_addr = new_addr + size - 1;
1484         /* Check if 32 bit BAR wraps around explicitly.
1485          * TODO: make priorities correct and remove this work around.
1486          */
1487         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1488             (!allow_0_address && new_addr == 0)) {
1489             return PCI_BAR_UNMAPPED;
1490         }
1491         return new_addr;
1492     }
1493 
1494     if (!(cmd & PCI_COMMAND_MEMORY)) {
1495         return PCI_BAR_UNMAPPED;
1496     }
1497     new_addr = pci_config_get_bar_addr(d, reg, type, size);
1498     /* the ROM slot has a specific enable bit */
1499     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1500         return PCI_BAR_UNMAPPED;
1501     }
1502     new_addr &= ~(size - 1);
1503     last_addr = new_addr + size - 1;
1504     /* NOTE: we do not support wrapping */
1505     /* XXX: as we cannot support really dynamic
1506        mappings, we handle specific values as invalid
1507        mappings. */
1508     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1509         (!allow_0_address && new_addr == 0)) {
1510         return PCI_BAR_UNMAPPED;
1511     }
1512 
1513     /* Now pcibus_t is 64bit.
1514      * Check if 32 bit BAR wraps around explicitly.
1515      * Without this, PC ide doesn't work well.
1516      * TODO: remove this work around.
1517      */
1518     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1519         return PCI_BAR_UNMAPPED;
1520     }
1521 
1522     /*
1523      * OS is allowed to set BAR beyond its addressable
1524      * bits. For example, 32 bit OS can set 64bit bar
1525      * to >4G. Check it. TODO: we might need to support
1526      * it in the future for e.g. PAE.
1527      */
1528     if (last_addr >= HWADDR_MAX) {
1529         return PCI_BAR_UNMAPPED;
1530     }
1531 
1532     return new_addr;
1533 }
1534 
1535 static void pci_update_mappings(PCIDevice *d)
1536 {
1537     PCIIORegion *r;
1538     int i;
1539     pcibus_t new_addr;
1540 
1541     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1542         r = &d->io_regions[i];
1543 
1544         /* this region isn't registered */
1545         if (!r->size)
1546             continue;
1547 
1548         new_addr = pci_bar_address(d, i, r->type, r->size);
1549         if (!d->has_power) {
1550             new_addr = PCI_BAR_UNMAPPED;
1551         }
1552 
1553         /* This bar isn't changed */
1554         if (new_addr == r->addr)
1555             continue;
1556 
1557         /* now do the real mapping */
1558         if (r->addr != PCI_BAR_UNMAPPED) {
1559             trace_pci_update_mappings_del(d->name, pci_dev_bus_num(d),
1560                                           PCI_SLOT(d->devfn),
1561                                           PCI_FUNC(d->devfn),
1562                                           i, r->addr, r->size);
1563             memory_region_del_subregion(r->address_space, r->memory);
1564         }
1565         r->addr = new_addr;
1566         if (r->addr != PCI_BAR_UNMAPPED) {
1567             trace_pci_update_mappings_add(d->name, pci_dev_bus_num(d),
1568                                           PCI_SLOT(d->devfn),
1569                                           PCI_FUNC(d->devfn),
1570                                           i, r->addr, r->size);
1571             memory_region_add_subregion_overlap(r->address_space,
1572                                                 r->addr, r->memory, 1);
1573         }
1574     }
1575 
1576     pci_update_vga(d);
1577 }
1578 
1579 static inline int pci_irq_disabled(PCIDevice *d)
1580 {
1581     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1582 }
1583 
1584 /* Called after interrupt disabled field update in config space,
1585  * assert/deassert interrupts if necessary.
1586  * Gets original interrupt disable bit value (before update). */
1587 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1588 {
1589     int i, disabled = pci_irq_disabled(d);
1590     if (disabled == was_irq_disabled)
1591         return;
1592     for (i = 0; i < PCI_NUM_PINS; ++i) {
1593         int state = pci_irq_state(d, i);
1594         pci_change_irq_level(d, i, disabled ? -state : state);
1595     }
1596 }
1597 
1598 uint32_t pci_default_read_config(PCIDevice *d,
1599                                  uint32_t address, int len)
1600 {
1601     uint32_t val = 0;
1602 
1603     assert(address + len <= pci_config_size(d));
1604 
1605     if (pci_is_express_downstream_port(d) &&
1606         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1607         pcie_sync_bridge_lnk(d);
1608     }
1609     memcpy(&val, d->config + address, len);
1610     return le32_to_cpu(val);
1611 }
1612 
1613 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1614 {
1615     int i, was_irq_disabled = pci_irq_disabled(d);
1616     uint32_t val = val_in;
1617 
1618     assert(addr + l <= pci_config_size(d));
1619 
1620     for (i = 0; i < l; val >>= 8, ++i) {
1621         uint8_t wmask = d->wmask[addr + i];
1622         uint8_t w1cmask = d->w1cmask[addr + i];
1623         assert(!(wmask & w1cmask));
1624         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1625         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1626     }
1627     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1628         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1629         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1630         range_covers_byte(addr, l, PCI_COMMAND))
1631         pci_update_mappings(d);
1632 
1633     if (ranges_overlap(addr, l, PCI_COMMAND, 2)) {
1634         pci_update_irq_disabled(d, was_irq_disabled);
1635         memory_region_set_enabled(&d->bus_master_enable_region,
1636                                   (pci_get_word(d->config + PCI_COMMAND)
1637                                    & PCI_COMMAND_MASTER) && d->has_power);
1638     }
1639 
1640     msi_write_config(d, addr, val_in, l);
1641     msix_write_config(d, addr, val_in, l);
1642     pcie_sriov_config_write(d, addr, val_in, l);
1643 }
1644 
1645 /***********************************************************/
1646 /* generic PCI irq support */
1647 
1648 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1649 static void pci_irq_handler(void *opaque, int irq_num, int level)
1650 {
1651     PCIDevice *pci_dev = opaque;
1652     int change;
1653 
1654     assert(0 <= irq_num && irq_num < PCI_NUM_PINS);
1655     assert(level == 0 || level == 1);
1656     change = level - pci_irq_state(pci_dev, irq_num);
1657     if (!change)
1658         return;
1659 
1660     pci_set_irq_state(pci_dev, irq_num, level);
1661     pci_update_irq_status(pci_dev);
1662     if (pci_irq_disabled(pci_dev))
1663         return;
1664     pci_change_irq_level(pci_dev, irq_num, change);
1665 }
1666 
1667 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1668 {
1669     int intx = pci_intx(pci_dev);
1670     assert(0 <= intx && intx < PCI_NUM_PINS);
1671 
1672     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1673 }
1674 
1675 void pci_set_irq(PCIDevice *pci_dev, int level)
1676 {
1677     int intx = pci_intx(pci_dev);
1678     pci_irq_handler(pci_dev, intx, level);
1679 }
1680 
1681 /* Special hooks used by device assignment */
1682 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1683 {
1684     assert(pci_bus_is_root(bus));
1685     bus->route_intx_to_irq = route_intx_to_irq;
1686 }
1687 
1688 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1689 {
1690     PCIBus *bus;
1691 
1692     do {
1693         int dev_irq = pin;
1694         bus = pci_get_bus(dev);
1695         pin = bus->map_irq(dev, pin);
1696         trace_pci_route_irq(dev_irq, DEVICE(dev)->canonical_path, pin,
1697                             pci_bus_is_root(bus) ? "root-complex"
1698                                     : DEVICE(bus->parent_dev)->canonical_path);
1699         dev = bus->parent_dev;
1700     } while (dev);
1701 
1702     if (!bus->route_intx_to_irq) {
1703         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1704                      object_get_typename(OBJECT(bus->qbus.parent)));
1705         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1706     }
1707 
1708     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1709 }
1710 
1711 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1712 {
1713     return old->mode != new->mode || old->irq != new->irq;
1714 }
1715 
1716 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1717 {
1718     PCIDevice *dev;
1719     PCIBus *sec;
1720     int i;
1721 
1722     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1723         dev = bus->devices[i];
1724         if (dev && dev->intx_routing_notifier) {
1725             dev->intx_routing_notifier(dev);
1726         }
1727     }
1728 
1729     QLIST_FOREACH(sec, &bus->child, sibling) {
1730         pci_bus_fire_intx_routing_notifier(sec);
1731     }
1732 }
1733 
1734 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1735                                           PCIINTxRoutingNotifier notifier)
1736 {
1737     dev->intx_routing_notifier = notifier;
1738 }
1739 
1740 /*
1741  * PCI-to-PCI bridge specification
1742  * 9.1: Interrupt routing. Table 9-1
1743  *
1744  * the PCI Express Base Specification, Revision 2.1
1745  * 2.2.8.1: INTx interrupt signaling - Rules
1746  *          the Implementation Note
1747  *          Table 2-20
1748  */
1749 /*
1750  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1751  * 0-origin unlike PCI interrupt pin register.
1752  */
1753 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1754 {
1755     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1756 }
1757 
1758 /***********************************************************/
1759 /* monitor info on PCI */
1760 
1761 static const pci_class_desc pci_class_descriptions[] =
1762 {
1763     { 0x0001, "VGA controller", "display"},
1764     { 0x0100, "SCSI controller", "scsi"},
1765     { 0x0101, "IDE controller", "ide"},
1766     { 0x0102, "Floppy controller", "fdc"},
1767     { 0x0103, "IPI controller", "ipi"},
1768     { 0x0104, "RAID controller", "raid"},
1769     { 0x0106, "SATA controller"},
1770     { 0x0107, "SAS controller"},
1771     { 0x0180, "Storage controller"},
1772     { 0x0200, "Ethernet controller", "ethernet"},
1773     { 0x0201, "Token Ring controller", "token-ring"},
1774     { 0x0202, "FDDI controller", "fddi"},
1775     { 0x0203, "ATM controller", "atm"},
1776     { 0x0280, "Network controller"},
1777     { 0x0300, "VGA controller", "display", 0x00ff},
1778     { 0x0301, "XGA controller"},
1779     { 0x0302, "3D controller"},
1780     { 0x0380, "Display controller"},
1781     { 0x0400, "Video controller", "video"},
1782     { 0x0401, "Audio controller", "sound"},
1783     { 0x0402, "Phone"},
1784     { 0x0403, "Audio controller", "sound"},
1785     { 0x0480, "Multimedia controller"},
1786     { 0x0500, "RAM controller", "memory"},
1787     { 0x0501, "Flash controller", "flash"},
1788     { 0x0580, "Memory controller"},
1789     { 0x0600, "Host bridge", "host"},
1790     { 0x0601, "ISA bridge", "isa"},
1791     { 0x0602, "EISA bridge", "eisa"},
1792     { 0x0603, "MC bridge", "mca"},
1793     { 0x0604, "PCI bridge", "pci-bridge"},
1794     { 0x0605, "PCMCIA bridge", "pcmcia"},
1795     { 0x0606, "NUBUS bridge", "nubus"},
1796     { 0x0607, "CARDBUS bridge", "cardbus"},
1797     { 0x0608, "RACEWAY bridge"},
1798     { 0x0680, "Bridge"},
1799     { 0x0700, "Serial port", "serial"},
1800     { 0x0701, "Parallel port", "parallel"},
1801     { 0x0800, "Interrupt controller", "interrupt-controller"},
1802     { 0x0801, "DMA controller", "dma-controller"},
1803     { 0x0802, "Timer", "timer"},
1804     { 0x0803, "RTC", "rtc"},
1805     { 0x0900, "Keyboard", "keyboard"},
1806     { 0x0901, "Pen", "pen"},
1807     { 0x0902, "Mouse", "mouse"},
1808     { 0x0A00, "Dock station", "dock", 0x00ff},
1809     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1810     { 0x0c00, "Firewire controller", "firewire"},
1811     { 0x0c01, "Access bus controller", "access-bus"},
1812     { 0x0c02, "SSA controller", "ssa"},
1813     { 0x0c03, "USB controller", "usb"},
1814     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1815     { 0x0c05, "SMBus"},
1816     { 0, NULL}
1817 };
1818 
1819 void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1820                                            pci_bus_dev_fn fn,
1821                                            void *opaque)
1822 {
1823     PCIDevice *d;
1824     int devfn;
1825 
1826     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1827         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1828         if (d) {
1829             fn(bus, d, opaque);
1830         }
1831     }
1832 }
1833 
1834 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1835                                  pci_bus_dev_fn fn, void *opaque)
1836 {
1837     bus = pci_find_bus_nr(bus, bus_num);
1838 
1839     if (bus) {
1840         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1841     }
1842 }
1843 
1844 void pci_for_each_device_under_bus(PCIBus *bus,
1845                                    pci_bus_dev_fn fn, void *opaque)
1846 {
1847     PCIDevice *d;
1848     int devfn;
1849 
1850     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1851         d = bus->devices[devfn];
1852         if (d) {
1853             fn(bus, d, opaque);
1854         }
1855     }
1856 }
1857 
1858 void pci_for_each_device(PCIBus *bus, int bus_num,
1859                          pci_bus_dev_fn fn, void *opaque)
1860 {
1861     bus = pci_find_bus_nr(bus, bus_num);
1862 
1863     if (bus) {
1864         pci_for_each_device_under_bus(bus, fn, opaque);
1865     }
1866 }
1867 
1868 const pci_class_desc *get_class_desc(int class)
1869 {
1870     const pci_class_desc *desc;
1871 
1872     desc = pci_class_descriptions;
1873     while (desc->desc && class != desc->class) {
1874         desc++;
1875     }
1876 
1877     return desc;
1878 }
1879 
1880 void pci_init_nic_devices(PCIBus *bus, const char *default_model)
1881 {
1882     qemu_create_nic_bus_devices(&bus->qbus, TYPE_PCI_DEVICE, default_model,
1883                                 "virtio", "virtio-net-pci");
1884 }
1885 
1886 bool pci_init_nic_in_slot(PCIBus *rootbus, const char *model,
1887                           const char *alias, const char *devaddr)
1888 {
1889     NICInfo *nd = qemu_find_nic_info(model, true, alias);
1890     int dom, busnr, devfn;
1891     PCIDevice *pci_dev;
1892     unsigned slot;
1893     PCIBus *bus;
1894 
1895     if (!nd) {
1896         return false;
1897     }
1898 
1899     if (!devaddr || pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1900         error_report("Invalid PCI device address %s for device %s",
1901                      devaddr, model);
1902         exit(1);
1903     }
1904 
1905     if (dom != 0) {
1906         error_report("No support for non-zero PCI domains");
1907         exit(1);
1908     }
1909 
1910     devfn = PCI_DEVFN(slot, 0);
1911 
1912     bus = pci_find_bus_nr(rootbus, busnr);
1913     if (!bus) {
1914         error_report("Invalid PCI device address %s for device %s",
1915                      devaddr, model);
1916         exit(1);
1917     }
1918 
1919     pci_dev = pci_new(devfn, model);
1920     qdev_set_nic_properties(&pci_dev->qdev, nd);
1921     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1922     return true;
1923 }
1924 
1925 PCIDevice *pci_vga_init(PCIBus *bus)
1926 {
1927     vga_interface_created = true;
1928     switch (vga_interface_type) {
1929     case VGA_CIRRUS:
1930         return pci_create_simple(bus, -1, "cirrus-vga");
1931     case VGA_QXL:
1932         return pci_create_simple(bus, -1, "qxl-vga");
1933     case VGA_STD:
1934         return pci_create_simple(bus, -1, "VGA");
1935     case VGA_VMWARE:
1936         return pci_create_simple(bus, -1, "vmware-svga");
1937     case VGA_VIRTIO:
1938         return pci_create_simple(bus, -1, "virtio-vga");
1939     case VGA_NONE:
1940     default: /* Other non-PCI types. Checking for unsupported types is already
1941                 done in vl.c. */
1942         return NULL;
1943     }
1944 }
1945 
1946 /* Whether a given bus number is in range of the secondary
1947  * bus of the given bridge device. */
1948 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1949 {
1950     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1951              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1952         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1953         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1954 }
1955 
1956 /* Whether a given bus number is in a range of a root bus */
1957 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1958 {
1959     int i;
1960 
1961     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1962         PCIDevice *dev = bus->devices[i];
1963 
1964         if (dev && IS_PCI_BRIDGE(dev)) {
1965             if (pci_secondary_bus_in_range(dev, bus_num)) {
1966                 return true;
1967             }
1968         }
1969     }
1970 
1971     return false;
1972 }
1973 
1974 PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1975 {
1976     PCIBus *sec;
1977 
1978     if (!bus) {
1979         return NULL;
1980     }
1981 
1982     if (pci_bus_num(bus) == bus_num) {
1983         return bus;
1984     }
1985 
1986     /* Consider all bus numbers in range for the host pci bridge. */
1987     if (!pci_bus_is_root(bus) &&
1988         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1989         return NULL;
1990     }
1991 
1992     /* try child bus */
1993     for (; bus; bus = sec) {
1994         QLIST_FOREACH(sec, &bus->child, sibling) {
1995             if (pci_bus_num(sec) == bus_num) {
1996                 return sec;
1997             }
1998             /* PXB buses assumed to be children of bus 0 */
1999             if (pci_bus_is_root(sec)) {
2000                 if (pci_root_bus_in_range(sec, bus_num)) {
2001                     break;
2002                 }
2003             } else {
2004                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2005                     break;
2006                 }
2007             }
2008         }
2009     }
2010 
2011     return NULL;
2012 }
2013 
2014 void pci_for_each_bus_depth_first(PCIBus *bus, pci_bus_ret_fn begin,
2015                                   pci_bus_fn end, void *parent_state)
2016 {
2017     PCIBus *sec;
2018     void *state;
2019 
2020     if (!bus) {
2021         return;
2022     }
2023 
2024     if (begin) {
2025         state = begin(bus, parent_state);
2026     } else {
2027         state = parent_state;
2028     }
2029 
2030     QLIST_FOREACH(sec, &bus->child, sibling) {
2031         pci_for_each_bus_depth_first(sec, begin, end, state);
2032     }
2033 
2034     if (end) {
2035         end(bus, state);
2036     }
2037 }
2038 
2039 
2040 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2041 {
2042     bus = pci_find_bus_nr(bus, bus_num);
2043 
2044     if (!bus)
2045         return NULL;
2046 
2047     return bus->devices[devfn];
2048 }
2049 
2050 #define ONBOARD_INDEX_MAX (16 * 1024 - 1)
2051 
2052 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2053 {
2054     PCIDevice *pci_dev = (PCIDevice *)qdev;
2055     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2056     ObjectClass *klass = OBJECT_CLASS(pc);
2057     Error *local_err = NULL;
2058     bool is_default_rom;
2059     uint16_t class_id;
2060 
2061     /*
2062      * capped by systemd (see: udev-builtin-net_id.c)
2063      * as it's the only known user honor it to avoid users
2064      * misconfigure QEMU and then wonder why acpi-index doesn't work
2065      */
2066     if (pci_dev->acpi_index > ONBOARD_INDEX_MAX) {
2067         error_setg(errp, "acpi-index should be less or equal to %u",
2068                    ONBOARD_INDEX_MAX);
2069         return;
2070     }
2071 
2072     /*
2073      * make sure that acpi-index is unique across all present PCI devices
2074      */
2075     if (pci_dev->acpi_index) {
2076         GSequence *used_indexes = pci_acpi_index_list();
2077 
2078         if (g_sequence_lookup(used_indexes,
2079                               GINT_TO_POINTER(pci_dev->acpi_index),
2080                               g_cmp_uint32, NULL)) {
2081             error_setg(errp, "a PCI device with acpi-index = %" PRIu32
2082                        " already exist", pci_dev->acpi_index);
2083             return;
2084         }
2085         g_sequence_insert_sorted(used_indexes,
2086                                  GINT_TO_POINTER(pci_dev->acpi_index),
2087                                  g_cmp_uint32, NULL);
2088     }
2089 
2090     if (pci_dev->romsize != UINT32_MAX && !is_power_of_2(pci_dev->romsize)) {
2091         error_setg(errp, "ROM size %u is not a power of two", pci_dev->romsize);
2092         return;
2093     }
2094 
2095     /* initialize cap_present for pci_is_express() and pci_config_size(),
2096      * Note that hybrid PCIs are not set automatically and need to manage
2097      * QEMU_PCI_CAP_EXPRESS manually */
2098     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2099        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2100         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2101     }
2102 
2103     if (object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE)) {
2104         pci_dev->cap_present |= QEMU_PCIE_CAP_CXL;
2105     }
2106 
2107     pci_dev = do_pci_register_device(pci_dev,
2108                                      object_get_typename(OBJECT(qdev)),
2109                                      pci_dev->devfn, errp);
2110     if (pci_dev == NULL)
2111         return;
2112 
2113     if (pc->realize) {
2114         pc->realize(pci_dev, &local_err);
2115         if (local_err) {
2116             error_propagate(errp, local_err);
2117             do_pci_unregister_device(pci_dev);
2118             return;
2119         }
2120     }
2121 
2122     /*
2123      * A PCIe Downstream Port that do not have ARI Forwarding enabled must
2124      * associate only Device 0 with the device attached to the bus
2125      * representing the Link from the Port (PCIe base spec rev 4.0 ver 0.3,
2126      * sec 7.3.1).
2127      * With ARI, PCI_SLOT() can return non-zero value as the traditional
2128      * 5-bit Device Number and 3-bit Function Number fields in its associated
2129      * Routing IDs, Requester IDs and Completer IDs are interpreted as a
2130      * single 8-bit Function Number. Hence, ignore ARI capable devices.
2131      */
2132     if (pci_is_express(pci_dev) &&
2133         !pcie_find_capability(pci_dev, PCI_EXT_CAP_ID_ARI) &&
2134         pcie_has_upstream_port(pci_dev) &&
2135         PCI_SLOT(pci_dev->devfn)) {
2136         warn_report("PCI: slot %d is not valid for %s,"
2137                     " parent device only allows plugging into slot 0.",
2138                     PCI_SLOT(pci_dev->devfn), pci_dev->name);
2139     }
2140 
2141     if (pci_dev->failover_pair_id) {
2142         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2143             error_setg(errp, "failover primary device must be on "
2144                              "PCIExpress bus");
2145             pci_qdev_unrealize(DEVICE(pci_dev));
2146             return;
2147         }
2148         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2149         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2150             error_setg(errp, "failover primary device is not an "
2151                              "Ethernet device");
2152             pci_qdev_unrealize(DEVICE(pci_dev));
2153             return;
2154         }
2155         if ((pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2156             || (PCI_FUNC(pci_dev->devfn) != 0)) {
2157             error_setg(errp, "failover: primary device must be in its own "
2158                               "PCI slot");
2159             pci_qdev_unrealize(DEVICE(pci_dev));
2160             return;
2161         }
2162         qdev->allow_unplug_during_migration = true;
2163     }
2164 
2165     /* rom loading */
2166     is_default_rom = false;
2167     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2168         pci_dev->romfile = g_strdup(pc->romfile);
2169         is_default_rom = true;
2170     }
2171 
2172     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2173     if (local_err) {
2174         error_propagate(errp, local_err);
2175         pci_qdev_unrealize(DEVICE(pci_dev));
2176         return;
2177     }
2178 
2179     pci_set_power(pci_dev, true);
2180 
2181     pci_dev->msi_trigger = pci_msi_trigger;
2182 }
2183 
2184 static PCIDevice *pci_new_internal(int devfn, bool multifunction,
2185                                    const char *name)
2186 {
2187     DeviceState *dev;
2188 
2189     dev = qdev_new(name);
2190     qdev_prop_set_int32(dev, "addr", devfn);
2191     qdev_prop_set_bit(dev, "multifunction", multifunction);
2192     return PCI_DEVICE(dev);
2193 }
2194 
2195 PCIDevice *pci_new_multifunction(int devfn, const char *name)
2196 {
2197     return pci_new_internal(devfn, true, name);
2198 }
2199 
2200 PCIDevice *pci_new(int devfn, const char *name)
2201 {
2202     return pci_new_internal(devfn, false, name);
2203 }
2204 
2205 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2206 {
2207     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2208 }
2209 
2210 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2211                                            const char *name)
2212 {
2213     PCIDevice *dev = pci_new_multifunction(devfn, name);
2214     pci_realize_and_unref(dev, bus, &error_fatal);
2215     return dev;
2216 }
2217 
2218 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2219 {
2220     PCIDevice *dev = pci_new(devfn, name);
2221     pci_realize_and_unref(dev, bus, &error_fatal);
2222     return dev;
2223 }
2224 
2225 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2226 {
2227     int offset = PCI_CONFIG_HEADER_SIZE;
2228     int i;
2229     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2230         if (pdev->used[i])
2231             offset = i + 1;
2232         else if (i - offset + 1 == size)
2233             return offset;
2234     }
2235     return 0;
2236 }
2237 
2238 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2239                                         uint8_t *prev_p)
2240 {
2241     uint8_t next, prev;
2242 
2243     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2244         return 0;
2245 
2246     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2247          prev = next + PCI_CAP_LIST_NEXT)
2248         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2249             break;
2250 
2251     if (prev_p)
2252         *prev_p = prev;
2253     return next;
2254 }
2255 
2256 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2257 {
2258     uint8_t next, prev, found = 0;
2259 
2260     if (!(pdev->used[offset])) {
2261         return 0;
2262     }
2263 
2264     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2265 
2266     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2267          prev = next + PCI_CAP_LIST_NEXT) {
2268         if (next <= offset && next > found) {
2269             found = next;
2270         }
2271     }
2272     return found;
2273 }
2274 
2275 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2276    This is needed for an option rom which is used for more than one device. */
2277 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, uint32_t size)
2278 {
2279     uint16_t vendor_id;
2280     uint16_t device_id;
2281     uint16_t rom_vendor_id;
2282     uint16_t rom_device_id;
2283     uint16_t rom_magic;
2284     uint16_t pcir_offset;
2285     uint8_t checksum;
2286 
2287     /* Words in rom data are little endian (like in PCI configuration),
2288        so they can be read / written with pci_get_word / pci_set_word. */
2289 
2290     /* Only a valid rom will be patched. */
2291     rom_magic = pci_get_word(ptr);
2292     if (rom_magic != 0xaa55) {
2293         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2294         return;
2295     }
2296     pcir_offset = pci_get_word(ptr + 0x18);
2297     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2298         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2299         return;
2300     }
2301 
2302     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2303     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2304     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2305     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2306 
2307     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2308                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2309 
2310     checksum = ptr[6];
2311 
2312     if (vendor_id != rom_vendor_id) {
2313         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2314         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2315         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2316         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2317         ptr[6] = checksum;
2318         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2319     }
2320 
2321     if (device_id != rom_device_id) {
2322         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2323         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2324         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2325         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2326         ptr[6] = checksum;
2327         pci_set_word(ptr + pcir_offset + 6, device_id);
2328     }
2329 }
2330 
2331 /* Add an option rom for the device */
2332 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2333                                Error **errp)
2334 {
2335     int64_t size = 0;
2336     g_autofree char *path = NULL;
2337     char name[32];
2338     const VMStateDescription *vmsd;
2339 
2340     /*
2341      * In case of incoming migration ROM will come with migration stream, no
2342      * reason to load the file.  Neither we want to fail if local ROM file
2343      * mismatches with specified romsize.
2344      */
2345     bool load_file = !runstate_check(RUN_STATE_INMIGRATE);
2346 
2347     if (!pdev->romfile || !strlen(pdev->romfile)) {
2348         return;
2349     }
2350 
2351     if (!pdev->rom_bar) {
2352         /*
2353          * Load rom via fw_cfg instead of creating a rom bar,
2354          * for 0.11 compatibility.
2355          */
2356         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2357 
2358         /*
2359          * Hot-plugged devices can't use the option ROM
2360          * if the rom bar is disabled.
2361          */
2362         if (DEVICE(pdev)->hotplugged) {
2363             error_setg(errp, "Hot-plugged device without ROM bar"
2364                        " can't have an option ROM");
2365             return;
2366         }
2367 
2368         if (class == 0x0300) {
2369             rom_add_vga(pdev->romfile);
2370         } else {
2371             rom_add_option(pdev->romfile, -1);
2372         }
2373         return;
2374     }
2375 
2376     if (load_file || pdev->romsize == UINT32_MAX) {
2377         path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2378         if (path == NULL) {
2379             path = g_strdup(pdev->romfile);
2380         }
2381 
2382         size = get_image_size(path);
2383         if (size < 0) {
2384             error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2385             return;
2386         } else if (size == 0) {
2387             error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2388             return;
2389         } else if (size > 2 * GiB) {
2390             error_setg(errp,
2391                        "romfile \"%s\" too large (size cannot exceed 2 GiB)",
2392                        pdev->romfile);
2393             return;
2394         }
2395         if (pdev->romsize != UINT_MAX) {
2396             if (size > pdev->romsize) {
2397                 error_setg(errp, "romfile \"%s\" (%u bytes) "
2398                            "is too large for ROM size %u",
2399                            pdev->romfile, (uint32_t)size, pdev->romsize);
2400                 return;
2401             }
2402         } else {
2403             pdev->romsize = pow2ceil(size);
2404         }
2405     }
2406 
2407     vmsd = qdev_get_vmsd(DEVICE(pdev));
2408     snprintf(name, sizeof(name), "%s.rom",
2409              vmsd ? vmsd->name : object_get_typename(OBJECT(pdev)));
2410 
2411     pdev->has_rom = true;
2412     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, pdev->romsize,
2413                            &error_fatal);
2414 
2415     if (load_file) {
2416         void *ptr = memory_region_get_ram_ptr(&pdev->rom);
2417 
2418         if (load_image_size(path, ptr, size) < 0) {
2419             error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2420             return;
2421         }
2422 
2423         if (is_default_rom) {
2424             /* Only the default rom images will be patched (if needed). */
2425             pci_patch_ids(pdev, ptr, size);
2426         }
2427     }
2428 
2429     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2430 }
2431 
2432 static void pci_del_option_rom(PCIDevice *pdev)
2433 {
2434     if (!pdev->has_rom)
2435         return;
2436 
2437     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2438     pdev->has_rom = false;
2439 }
2440 
2441 /*
2442  * On success, pci_add_capability() returns a positive value
2443  * that the offset of the pci capability.
2444  * On failure, it sets an error and returns a negative error
2445  * code.
2446  */
2447 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2448                        uint8_t offset, uint8_t size,
2449                        Error **errp)
2450 {
2451     uint8_t *config;
2452     int i, overlapping_cap;
2453 
2454     if (!offset) {
2455         offset = pci_find_space(pdev, size);
2456         /* out of PCI config space is programming error */
2457         assert(offset);
2458     } else {
2459         /* Verify that capabilities don't overlap.  Note: device assignment
2460          * depends on this check to verify that the device is not broken.
2461          * Should never trigger for emulated devices, but it's helpful
2462          * for debugging these. */
2463         for (i = offset; i < offset + size; i++) {
2464             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2465             if (overlapping_cap) {
2466                 error_setg(errp, "%s:%02x:%02x.%x "
2467                            "Attempt to add PCI capability %x at offset "
2468                            "%x overlaps existing capability %x at offset %x",
2469                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2470                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2471                            cap_id, offset, overlapping_cap, i);
2472                 return -EINVAL;
2473             }
2474         }
2475     }
2476 
2477     config = pdev->config + offset;
2478     config[PCI_CAP_LIST_ID] = cap_id;
2479     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2480     pdev->config[PCI_CAPABILITY_LIST] = offset;
2481     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2482     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2483     /* Make capability read-only by default */
2484     memset(pdev->wmask + offset, 0, size);
2485     /* Check capability by default */
2486     memset(pdev->cmask + offset, 0xFF, size);
2487     return offset;
2488 }
2489 
2490 /* Unlink capability from the pci config space. */
2491 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2492 {
2493     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2494     if (!offset)
2495         return;
2496     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2497     /* Make capability writable again */
2498     memset(pdev->wmask + offset, 0xff, size);
2499     memset(pdev->w1cmask + offset, 0, size);
2500     /* Clear cmask as device-specific registers can't be checked */
2501     memset(pdev->cmask + offset, 0, size);
2502     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2503 
2504     if (!pdev->config[PCI_CAPABILITY_LIST])
2505         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2506 }
2507 
2508 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2509 {
2510     return pci_find_capability_list(pdev, cap_id, NULL);
2511 }
2512 
2513 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2514 {
2515     PCIDevice *d = (PCIDevice *)dev;
2516     const char *name = NULL;
2517     const pci_class_desc *desc =  pci_class_descriptions;
2518     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2519 
2520     while (desc->desc &&
2521           (class & ~desc->fw_ign_bits) !=
2522           (desc->class & ~desc->fw_ign_bits)) {
2523         desc++;
2524     }
2525 
2526     if (desc->desc) {
2527         name = desc->fw_name;
2528     }
2529 
2530     if (name) {
2531         pstrcpy(buf, len, name);
2532     } else {
2533         snprintf(buf, len, "pci%04x,%04x",
2534                  pci_get_word(d->config + PCI_VENDOR_ID),
2535                  pci_get_word(d->config + PCI_DEVICE_ID));
2536     }
2537 
2538     return buf;
2539 }
2540 
2541 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2542 {
2543     PCIDevice *d = (PCIDevice *)dev;
2544     char name[33];
2545     int has_func = !!PCI_FUNC(d->devfn);
2546 
2547     return g_strdup_printf("%s@%x%s%.*x",
2548                            pci_dev_fw_name(dev, name, sizeof(name)),
2549                            PCI_SLOT(d->devfn),
2550                            has_func ? "," : "",
2551                            has_func,
2552                            PCI_FUNC(d->devfn));
2553 }
2554 
2555 static char *pcibus_get_dev_path(DeviceState *dev)
2556 {
2557     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2558     PCIDevice *t;
2559     int slot_depth;
2560     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2561      * 00 is added here to make this format compatible with
2562      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2563      * Slot.Function list specifies the slot and function numbers for all
2564      * devices on the path from root to the specific device. */
2565     const char *root_bus_path;
2566     int root_bus_len;
2567     char slot[] = ":SS.F";
2568     int slot_len = sizeof slot - 1 /* For '\0' */;
2569     int path_len;
2570     char *path, *p;
2571     int s;
2572 
2573     root_bus_path = pci_root_bus_path(d);
2574     root_bus_len = strlen(root_bus_path);
2575 
2576     /* Calculate # of slots on path between device and root. */;
2577     slot_depth = 0;
2578     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2579         ++slot_depth;
2580     }
2581 
2582     path_len = root_bus_len + slot_len * slot_depth;
2583 
2584     /* Allocate memory, fill in the terminating null byte. */
2585     path = g_malloc(path_len + 1 /* For '\0' */);
2586     path[path_len] = '\0';
2587 
2588     memcpy(path, root_bus_path, root_bus_len);
2589 
2590     /* Fill in slot numbers. We walk up from device to root, so need to print
2591      * them in the reverse order, last to first. */
2592     p = path + path_len;
2593     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2594         p -= slot_len;
2595         s = snprintf(slot, sizeof slot, ":%02x.%x",
2596                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2597         assert(s == slot_len);
2598         memcpy(p, slot, slot_len);
2599     }
2600 
2601     return path;
2602 }
2603 
2604 static int pci_qdev_find_recursive(PCIBus *bus,
2605                                    const char *id, PCIDevice **pdev)
2606 {
2607     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2608     if (!qdev) {
2609         return -ENODEV;
2610     }
2611 
2612     /* roughly check if given qdev is pci device */
2613     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2614         *pdev = PCI_DEVICE(qdev);
2615         return 0;
2616     }
2617     return -EINVAL;
2618 }
2619 
2620 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2621 {
2622     PCIHostState *host_bridge;
2623     int rc = -ENODEV;
2624 
2625     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2626         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2627         if (!tmp) {
2628             rc = 0;
2629             break;
2630         }
2631         if (tmp != -ENODEV) {
2632             rc = tmp;
2633         }
2634     }
2635 
2636     return rc;
2637 }
2638 
2639 MemoryRegion *pci_address_space(PCIDevice *dev)
2640 {
2641     return pci_get_bus(dev)->address_space_mem;
2642 }
2643 
2644 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2645 {
2646     return pci_get_bus(dev)->address_space_io;
2647 }
2648 
2649 static void pci_device_class_init(ObjectClass *klass, void *data)
2650 {
2651     DeviceClass *k = DEVICE_CLASS(klass);
2652 
2653     k->realize = pci_qdev_realize;
2654     k->unrealize = pci_qdev_unrealize;
2655     k->bus_type = TYPE_PCI_BUS;
2656     device_class_set_props(k, pci_props);
2657     object_class_property_set_description(
2658         klass, "x-max-bounce-buffer-size",
2659         "Maximum buffer size allocated for bounce buffers used for mapped "
2660         "access to indirect DMA memory");
2661 }
2662 
2663 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2664 {
2665     if (!object_class_is_abstract(klass)) {
2666         ObjectClass *conventional =
2667             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2668         ObjectClass *pcie =
2669             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2670         ObjectClass *cxl =
2671             object_class_dynamic_cast(klass, INTERFACE_CXL_DEVICE);
2672         assert(conventional || pcie || cxl);
2673     }
2674 }
2675 
2676 /*
2677  * Get IOMMU root bus, aliased bus and devfn of a PCI device
2678  *
2679  * IOMMU root bus is needed by all call sites to call into iommu_ops.
2680  * For call sites which don't need aliased BDF, passing NULL to
2681  * aliased_[bus|devfn] is allowed.
2682  *
2683  * @piommu_bus: return root #PCIBus backed by an IOMMU for the PCI device.
2684  *
2685  * @aliased_bus: return aliased #PCIBus of the PCI device, optional.
2686  *
2687  * @aliased_devfn: return aliased devfn of the PCI device, optional.
2688  */
2689 static void pci_device_get_iommu_bus_devfn(PCIDevice *dev,
2690                                            PCIBus **piommu_bus,
2691                                            PCIBus **aliased_bus,
2692                                            int *aliased_devfn)
2693 {
2694     PCIBus *bus = pci_get_bus(dev);
2695     PCIBus *iommu_bus = bus;
2696     int devfn = dev->devfn;
2697 
2698     while (iommu_bus && !iommu_bus->iommu_ops && iommu_bus->parent_dev) {
2699         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2700 
2701         /*
2702          * The requester ID of the provided device may be aliased, as seen from
2703          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2704          * requester ID to provide a unique AddressSpace for devices, but
2705          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2706          * to-PCI bridge creates and accepts transactions on behalf of down-
2707          * stream devices.  When doing so, all downstream devices are masked
2708          * (aliased) behind a single requester ID.  The requester ID used
2709          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2710          * bridges, with a PCIe capability indicating such, follow the
2711          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2712          * where the bridge uses the seconary bus as the bridge portion of the
2713          * requester ID and devfn of 00.0.  For other bridges, typically those
2714          * found on the root complex such as the dmi-to-pci-bridge, we follow
2715          * the convention of typical bare-metal hardware, which uses the
2716          * requester ID of the bridge itself.  There are device specific
2717          * exceptions to these rules, but these are the defaults that the
2718          * Linux kernel uses when determining DMA aliases itself and believed
2719          * to be true for the bare metal equivalents of the devices emulated
2720          * in QEMU.
2721          */
2722         if (!pci_bus_is_express(iommu_bus)) {
2723             PCIDevice *parent = iommu_bus->parent_dev;
2724 
2725             if (pci_is_express(parent) &&
2726                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2727                 devfn = PCI_DEVFN(0, 0);
2728                 bus = iommu_bus;
2729             } else {
2730                 devfn = parent->devfn;
2731                 bus = parent_bus;
2732             }
2733         }
2734 
2735         iommu_bus = parent_bus;
2736     }
2737 
2738     assert(0 <= devfn && devfn < PCI_DEVFN_MAX);
2739     assert(iommu_bus);
2740 
2741     if (pci_bus_bypass_iommu(bus) || !iommu_bus->iommu_ops) {
2742         iommu_bus = NULL;
2743     }
2744 
2745     *piommu_bus = iommu_bus;
2746 
2747     if (aliased_bus) {
2748         *aliased_bus = bus;
2749     }
2750 
2751     if (aliased_devfn) {
2752         *aliased_devfn = devfn;
2753     }
2754 }
2755 
2756 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2757 {
2758     PCIBus *bus;
2759     PCIBus *iommu_bus;
2760     int devfn;
2761 
2762     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, &bus, &devfn);
2763     if (iommu_bus) {
2764         return iommu_bus->iommu_ops->get_address_space(bus,
2765                                  iommu_bus->iommu_opaque, devfn);
2766     }
2767     return &address_space_memory;
2768 }
2769 
2770 bool pci_device_set_iommu_device(PCIDevice *dev, HostIOMMUDevice *hiod,
2771                                  Error **errp)
2772 {
2773     PCIBus *iommu_bus, *aliased_bus;
2774     int aliased_devfn;
2775 
2776     /* set_iommu_device requires device's direct BDF instead of aliased BDF */
2777     pci_device_get_iommu_bus_devfn(dev, &iommu_bus,
2778                                    &aliased_bus, &aliased_devfn);
2779     if (iommu_bus && iommu_bus->iommu_ops->set_iommu_device) {
2780         hiod->aliased_bus = aliased_bus;
2781         hiod->aliased_devfn = aliased_devfn;
2782         return iommu_bus->iommu_ops->set_iommu_device(pci_get_bus(dev),
2783                                                       iommu_bus->iommu_opaque,
2784                                                       dev->devfn, hiod, errp);
2785     }
2786     return true;
2787 }
2788 
2789 void pci_device_unset_iommu_device(PCIDevice *dev)
2790 {
2791     PCIBus *iommu_bus;
2792 
2793     pci_device_get_iommu_bus_devfn(dev, &iommu_bus, NULL, NULL);
2794     if (iommu_bus && iommu_bus->iommu_ops->unset_iommu_device) {
2795         return iommu_bus->iommu_ops->unset_iommu_device(pci_get_bus(dev),
2796                                                         iommu_bus->iommu_opaque,
2797                                                         dev->devfn);
2798     }
2799 }
2800 
2801 void pci_setup_iommu(PCIBus *bus, const PCIIOMMUOps *ops, void *opaque)
2802 {
2803     /*
2804      * If called, pci_setup_iommu() should provide a minimum set of
2805      * useful callbacks for the bus.
2806      */
2807     assert(ops);
2808     assert(ops->get_address_space);
2809 
2810     bus->iommu_ops = ops;
2811     bus->iommu_opaque = opaque;
2812 }
2813 
2814 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2815 {
2816     Range *range = opaque;
2817     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2818     int i;
2819 
2820     if (!(cmd & PCI_COMMAND_MEMORY)) {
2821         return;
2822     }
2823 
2824     if (IS_PCI_BRIDGE(dev)) {
2825         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2826         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2827 
2828         base = MAX(base, 0x1ULL << 32);
2829 
2830         if (limit >= base) {
2831             Range pref_range;
2832             range_set_bounds(&pref_range, base, limit);
2833             range_extend(range, &pref_range);
2834         }
2835     }
2836     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2837         PCIIORegion *r = &dev->io_regions[i];
2838         pcibus_t lob, upb;
2839         Range region_range;
2840 
2841         if (!r->size ||
2842             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2843             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2844             continue;
2845         }
2846 
2847         lob = pci_bar_address(dev, i, r->type, r->size);
2848         upb = lob + r->size - 1;
2849         if (lob == PCI_BAR_UNMAPPED) {
2850             continue;
2851         }
2852 
2853         lob = MAX(lob, 0x1ULL << 32);
2854 
2855         if (upb >= lob) {
2856             range_set_bounds(&region_range, lob, upb);
2857             range_extend(range, &region_range);
2858         }
2859     }
2860 }
2861 
2862 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2863 {
2864     range_make_empty(range);
2865     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2866 }
2867 
2868 static bool pcie_has_upstream_port(PCIDevice *dev)
2869 {
2870     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2871 
2872     /* Device associated with an upstream port.
2873      * As there are several types of these, it's easier to check the
2874      * parent device: upstream ports are always connected to
2875      * root or downstream ports.
2876      */
2877     return parent_dev &&
2878         pci_is_express(parent_dev) &&
2879         parent_dev->exp.exp_cap &&
2880         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2881          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2882 }
2883 
2884 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2885 {
2886     PCIBus *bus = pci_get_bus(pci_dev);
2887 
2888     if(pcie_has_upstream_port(pci_dev)) {
2889         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2890         return bus->devices[0];
2891     } else {
2892         /* Other bus types might support multiple devices at slots 0-31 */
2893         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2894     }
2895 }
2896 
2897 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2898 {
2899     MSIMessage msg;
2900     if (msix_enabled(dev)) {
2901         msg = msix_get_message(dev, vector);
2902     } else if (msi_enabled(dev)) {
2903         msg = msi_get_message(dev, vector);
2904     } else {
2905         /* Should never happen */
2906         error_report("%s: unknown interrupt type", __func__);
2907         abort();
2908     }
2909     return msg;
2910 }
2911 
2912 void pci_set_power(PCIDevice *d, bool state)
2913 {
2914     if (d->has_power == state) {
2915         return;
2916     }
2917 
2918     d->has_power = state;
2919     pci_update_mappings(d);
2920     memory_region_set_enabled(&d->bus_master_enable_region,
2921                               (pci_get_word(d->config + PCI_COMMAND)
2922                                & PCI_COMMAND_MASTER) && d->has_power);
2923     if (!d->has_power) {
2924         pci_device_reset(d);
2925     }
2926 }
2927 
2928 static const TypeInfo pci_device_type_info = {
2929     .name = TYPE_PCI_DEVICE,
2930     .parent = TYPE_DEVICE,
2931     .instance_size = sizeof(PCIDevice),
2932     .abstract = true,
2933     .class_size = sizeof(PCIDeviceClass),
2934     .class_init = pci_device_class_init,
2935     .class_base_init = pci_device_class_base_init,
2936 };
2937 
2938 static void pci_register_types(void)
2939 {
2940     type_register_static(&pci_bus_info);
2941     type_register_static(&pcie_bus_info);
2942     type_register_static(&cxl_bus_info);
2943     type_register_static(&conventional_pci_interface_info);
2944     type_register_static(&cxl_interface_info);
2945     type_register_static(&pcie_interface_info);
2946     type_register_static(&pci_device_type_info);
2947 }
2948 
2949 type_init(pci_register_types)
2950