xref: /openbmc/qemu/hw/pci/pci.c (revision 284d697c)
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-common.h"
27 #include "hw/irq.h"
28 #include "hw/pci/pci.h"
29 #include "hw/pci/pci_bridge.h"
30 #include "hw/pci/pci_bus.h"
31 #include "hw/pci/pci_host.h"
32 #include "hw/qdev-properties.h"
33 #include "migration/qemu-file-types.h"
34 #include "migration/vmstate.h"
35 #include "monitor/monitor.h"
36 #include "net/net.h"
37 #include "sysemu/numa.h"
38 #include "sysemu/sysemu.h"
39 #include "hw/loader.h"
40 #include "qemu/error-report.h"
41 #include "qemu/range.h"
42 #include "trace.h"
43 #include "hw/pci/msi.h"
44 #include "hw/pci/msix.h"
45 #include "exec/address-spaces.h"
46 #include "hw/hotplug.h"
47 #include "hw/boards.h"
48 #include "qapi/error.h"
49 #include "qapi/qapi-commands-pci.h"
50 #include "qemu/cutils.h"
51 
52 //#define DEBUG_PCI
53 #ifdef DEBUG_PCI
54 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
55 #else
56 # define PCI_DPRINTF(format, ...)       do { } while (0)
57 #endif
58 
59 bool pci_available = true;
60 
61 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
62 static char *pcibus_get_dev_path(DeviceState *dev);
63 static char *pcibus_get_fw_dev_path(DeviceState *dev);
64 static void pcibus_reset(BusState *qbus);
65 
66 static Property pci_props[] = {
67     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
68     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
69     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
70     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
71                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
72     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
73                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
74     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
75                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
76     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
77                        failover_pair_id),
78     DEFINE_PROP_END_OF_LIST()
79 };
80 
81 static const VMStateDescription vmstate_pcibus = {
82     .name = "PCIBUS",
83     .version_id = 1,
84     .minimum_version_id = 1,
85     .fields = (VMStateField[]) {
86         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
87         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
88                              nirq, 0, vmstate_info_int32,
89                              int32_t),
90         VMSTATE_END_OF_LIST()
91     }
92 };
93 
94 static void pci_init_bus_master(PCIDevice *pci_dev)
95 {
96     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
97 
98     memory_region_init_alias(&pci_dev->bus_master_enable_region,
99                              OBJECT(pci_dev), "bus master",
100                              dma_as->root, 0, memory_region_size(dma_as->root));
101     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
102     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
103                                 &pci_dev->bus_master_enable_region);
104 }
105 
106 static void pcibus_machine_done(Notifier *notifier, void *data)
107 {
108     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
109     int i;
110 
111     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
112         if (bus->devices[i]) {
113             pci_init_bus_master(bus->devices[i]);
114         }
115     }
116 }
117 
118 static void pci_bus_realize(BusState *qbus, Error **errp)
119 {
120     PCIBus *bus = PCI_BUS(qbus);
121 
122     bus->machine_done.notify = pcibus_machine_done;
123     qemu_add_machine_init_done_notifier(&bus->machine_done);
124 
125     vmstate_register(NULL, VMSTATE_INSTANCE_ID_ANY, &vmstate_pcibus, bus);
126 }
127 
128 static void pcie_bus_realize(BusState *qbus, Error **errp)
129 {
130     PCIBus *bus = PCI_BUS(qbus);
131 
132     pci_bus_realize(qbus, errp);
133 
134     /*
135      * A PCI-E bus can support extended config space if it's the root
136      * bus, or if the bus/bridge above it does as well
137      */
138     if (pci_bus_is_root(bus)) {
139         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
140     } else {
141         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
142 
143         if (pci_bus_allows_extended_config_space(parent_bus)) {
144             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
145         }
146     }
147 }
148 
149 static void pci_bus_unrealize(BusState *qbus)
150 {
151     PCIBus *bus = PCI_BUS(qbus);
152 
153     qemu_remove_machine_init_done_notifier(&bus->machine_done);
154 
155     vmstate_unregister(NULL, &vmstate_pcibus, bus);
156 }
157 
158 static int pcibus_num(PCIBus *bus)
159 {
160     if (pci_bus_is_root(bus)) {
161         return 0; /* pci host bridge */
162     }
163     return bus->parent_dev->config[PCI_SECONDARY_BUS];
164 }
165 
166 static uint16_t pcibus_numa_node(PCIBus *bus)
167 {
168     return NUMA_NODE_UNASSIGNED;
169 }
170 
171 static void pci_bus_class_init(ObjectClass *klass, void *data)
172 {
173     BusClass *k = BUS_CLASS(klass);
174     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
175 
176     k->print_dev = pcibus_dev_print;
177     k->get_dev_path = pcibus_get_dev_path;
178     k->get_fw_dev_path = pcibus_get_fw_dev_path;
179     k->realize = pci_bus_realize;
180     k->unrealize = pci_bus_unrealize;
181     k->reset = pcibus_reset;
182 
183     pbc->bus_num = pcibus_num;
184     pbc->numa_node = pcibus_numa_node;
185 }
186 
187 static const TypeInfo pci_bus_info = {
188     .name = TYPE_PCI_BUS,
189     .parent = TYPE_BUS,
190     .instance_size = sizeof(PCIBus),
191     .class_size = sizeof(PCIBusClass),
192     .class_init = pci_bus_class_init,
193 };
194 
195 static const TypeInfo pcie_interface_info = {
196     .name          = INTERFACE_PCIE_DEVICE,
197     .parent        = TYPE_INTERFACE,
198 };
199 
200 static const TypeInfo conventional_pci_interface_info = {
201     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
202     .parent        = TYPE_INTERFACE,
203 };
204 
205 static void pcie_bus_class_init(ObjectClass *klass, void *data)
206 {
207     BusClass *k = BUS_CLASS(klass);
208 
209     k->realize = pcie_bus_realize;
210 }
211 
212 static const TypeInfo pcie_bus_info = {
213     .name = TYPE_PCIE_BUS,
214     .parent = TYPE_PCI_BUS,
215     .class_init = pcie_bus_class_init,
216 };
217 
218 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
219 static void pci_update_mappings(PCIDevice *d);
220 static void pci_irq_handler(void *opaque, int irq_num, int level);
221 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
222 static void pci_del_option_rom(PCIDevice *pdev);
223 
224 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
225 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
226 
227 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
228 
229 int pci_bar(PCIDevice *d, int reg)
230 {
231     uint8_t type;
232 
233     if (reg != PCI_ROM_SLOT)
234         return PCI_BASE_ADDRESS_0 + reg * 4;
235 
236     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
237     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
238 }
239 
240 static inline int pci_irq_state(PCIDevice *d, int irq_num)
241 {
242         return (d->irq_state >> irq_num) & 0x1;
243 }
244 
245 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
246 {
247         d->irq_state &= ~(0x1 << irq_num);
248         d->irq_state |= level << irq_num;
249 }
250 
251 static void pci_bus_change_irq_level(PCIBus *bus, int irq_num, int change)
252 {
253     assert(irq_num >= 0);
254     assert(irq_num < bus->nirq);
255     bus->irq_count[irq_num] += change;
256     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
257 }
258 
259 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
260 {
261     PCIBus *bus;
262     for (;;) {
263         bus = pci_get_bus(pci_dev);
264         irq_num = bus->map_irq(pci_dev, irq_num);
265         if (bus->set_irq)
266             break;
267         pci_dev = bus->parent_dev;
268     }
269     pci_bus_change_irq_level(bus, irq_num, change);
270 }
271 
272 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
273 {
274     assert(irq_num >= 0);
275     assert(irq_num < bus->nirq);
276     return !!bus->irq_count[irq_num];
277 }
278 
279 /* Update interrupt status bit in config space on interrupt
280  * state change. */
281 static void pci_update_irq_status(PCIDevice *dev)
282 {
283     if (dev->irq_state) {
284         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
285     } else {
286         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
287     }
288 }
289 
290 void pci_device_deassert_intx(PCIDevice *dev)
291 {
292     int i;
293     for (i = 0; i < PCI_NUM_PINS; ++i) {
294         pci_irq_handler(dev, i, 0);
295     }
296 }
297 
298 static void pci_do_device_reset(PCIDevice *dev)
299 {
300     int r;
301 
302     pci_device_deassert_intx(dev);
303     assert(dev->irq_state == 0);
304 
305     /* Clear all writable bits */
306     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
307                                  pci_get_word(dev->wmask + PCI_COMMAND) |
308                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
309     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
310                                  pci_get_word(dev->wmask + PCI_STATUS) |
311                                  pci_get_word(dev->w1cmask + PCI_STATUS));
312     /* Some devices make bits of PCI_INTERRUPT_LINE read only */
313     pci_byte_test_and_clear_mask(dev->config + PCI_INTERRUPT_LINE,
314                               pci_get_word(dev->wmask + PCI_INTERRUPT_LINE) |
315                               pci_get_word(dev->w1cmask + PCI_INTERRUPT_LINE));
316     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
317     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
318         PCIIORegion *region = &dev->io_regions[r];
319         if (!region->size) {
320             continue;
321         }
322 
323         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
324             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
325             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
326         } else {
327             pci_set_long(dev->config + pci_bar(dev, r), region->type);
328         }
329     }
330     pci_update_mappings(dev);
331 
332     msi_reset(dev);
333     msix_reset(dev);
334 }
335 
336 /*
337  * This function is called on #RST and FLR.
338  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
339  */
340 void pci_device_reset(PCIDevice *dev)
341 {
342     qdev_reset_all(&dev->qdev);
343     pci_do_device_reset(dev);
344 }
345 
346 /*
347  * Trigger pci bus reset under a given bus.
348  * Called via qbus_reset_all on RST# assert, after the devices
349  * have been reset qdev_reset_all-ed already.
350  */
351 static void pcibus_reset(BusState *qbus)
352 {
353     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
354     int i;
355 
356     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
357         if (bus->devices[i]) {
358             pci_do_device_reset(bus->devices[i]);
359         }
360     }
361 
362     for (i = 0; i < bus->nirq; i++) {
363         assert(bus->irq_count[i] == 0);
364     }
365 }
366 
367 static void pci_host_bus_register(DeviceState *host)
368 {
369     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
370 
371     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
372 }
373 
374 static void pci_host_bus_unregister(DeviceState *host)
375 {
376     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
377 
378     QLIST_REMOVE(host_bridge, next);
379 }
380 
381 PCIBus *pci_device_root_bus(const PCIDevice *d)
382 {
383     PCIBus *bus = pci_get_bus(d);
384 
385     while (!pci_bus_is_root(bus)) {
386         d = bus->parent_dev;
387         assert(d != NULL);
388 
389         bus = pci_get_bus(d);
390     }
391 
392     return bus;
393 }
394 
395 const char *pci_root_bus_path(PCIDevice *dev)
396 {
397     PCIBus *rootbus = pci_device_root_bus(dev);
398     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
399     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
400 
401     assert(host_bridge->bus == rootbus);
402 
403     if (hc->root_bus_path) {
404         return (*hc->root_bus_path)(host_bridge, rootbus);
405     }
406 
407     return rootbus->qbus.name;
408 }
409 
410 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
411                               MemoryRegion *address_space_mem,
412                               MemoryRegion *address_space_io,
413                               uint8_t devfn_min)
414 {
415     assert(PCI_FUNC(devfn_min) == 0);
416     bus->devfn_min = devfn_min;
417     bus->slot_reserved_mask = 0x0;
418     bus->address_space_mem = address_space_mem;
419     bus->address_space_io = address_space_io;
420     bus->flags |= PCI_BUS_IS_ROOT;
421 
422     /* host bridge */
423     QLIST_INIT(&bus->child);
424 
425     pci_host_bus_register(parent);
426 }
427 
428 static void pci_bus_uninit(PCIBus *bus)
429 {
430     pci_host_bus_unregister(BUS(bus)->parent);
431 }
432 
433 bool pci_bus_is_express(PCIBus *bus)
434 {
435     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
436 }
437 
438 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
439                               const char *name,
440                               MemoryRegion *address_space_mem,
441                               MemoryRegion *address_space_io,
442                               uint8_t devfn_min, const char *typename)
443 {
444     qbus_create_inplace(bus, bus_size, typename, parent, name);
445     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
446                       devfn_min);
447 }
448 
449 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
450                          MemoryRegion *address_space_mem,
451                          MemoryRegion *address_space_io,
452                          uint8_t devfn_min, const char *typename)
453 {
454     PCIBus *bus;
455 
456     bus = PCI_BUS(qbus_create(typename, parent, name));
457     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
458                       devfn_min);
459     return bus;
460 }
461 
462 void pci_root_bus_cleanup(PCIBus *bus)
463 {
464     pci_bus_uninit(bus);
465     /* the caller of the unplug hotplug handler will delete this device */
466     qbus_unrealize(BUS(bus));
467 }
468 
469 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
470                   void *irq_opaque, int nirq)
471 {
472     bus->set_irq = set_irq;
473     bus->map_irq = map_irq;
474     bus->irq_opaque = irq_opaque;
475     bus->nirq = nirq;
476     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
477 }
478 
479 void pci_bus_irqs_cleanup(PCIBus *bus)
480 {
481     bus->set_irq = NULL;
482     bus->map_irq = NULL;
483     bus->irq_opaque = NULL;
484     bus->nirq = 0;
485     g_free(bus->irq_count);
486 }
487 
488 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
489                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
490                               void *irq_opaque,
491                               MemoryRegion *address_space_mem,
492                               MemoryRegion *address_space_io,
493                               uint8_t devfn_min, int nirq,
494                               const char *typename)
495 {
496     PCIBus *bus;
497 
498     bus = pci_root_bus_new(parent, name, address_space_mem,
499                            address_space_io, devfn_min, typename);
500     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
501     return bus;
502 }
503 
504 void pci_unregister_root_bus(PCIBus *bus)
505 {
506     pci_bus_irqs_cleanup(bus);
507     pci_root_bus_cleanup(bus);
508 }
509 
510 int pci_bus_num(PCIBus *s)
511 {
512     return PCI_BUS_GET_CLASS(s)->bus_num(s);
513 }
514 
515 int pci_bus_numa_node(PCIBus *bus)
516 {
517     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
518 }
519 
520 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
521                                  const VMStateField *field)
522 {
523     PCIDevice *s = container_of(pv, PCIDevice, config);
524     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
525     uint8_t *config;
526     int i;
527 
528     assert(size == pci_config_size(s));
529     config = g_malloc(size);
530 
531     qemu_get_buffer(f, config, size);
532     for (i = 0; i < size; ++i) {
533         if ((config[i] ^ s->config[i]) &
534             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
535             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
536                          "cmask: %x wmask: %x w1cmask:%x", __func__,
537                          i, config[i], s->config[i],
538                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
539             g_free(config);
540             return -EINVAL;
541         }
542     }
543     memcpy(s->config, config, size);
544 
545     pci_update_mappings(s);
546     if (pc->is_bridge) {
547         PCIBridge *b = PCI_BRIDGE(s);
548         pci_bridge_update_mappings(b);
549     }
550 
551     memory_region_set_enabled(&s->bus_master_enable_region,
552                               pci_get_word(s->config + PCI_COMMAND)
553                               & PCI_COMMAND_MASTER);
554 
555     g_free(config);
556     return 0;
557 }
558 
559 /* just put buffer */
560 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
561                                  const VMStateField *field, QJSON *vmdesc)
562 {
563     const uint8_t **v = pv;
564     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
565     qemu_put_buffer(f, *v, size);
566 
567     return 0;
568 }
569 
570 static VMStateInfo vmstate_info_pci_config = {
571     .name = "pci config",
572     .get  = get_pci_config_device,
573     .put  = put_pci_config_device,
574 };
575 
576 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
577                              const VMStateField *field)
578 {
579     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
580     uint32_t irq_state[PCI_NUM_PINS];
581     int i;
582     for (i = 0; i < PCI_NUM_PINS; ++i) {
583         irq_state[i] = qemu_get_be32(f);
584         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
585             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
586                     irq_state[i]);
587             return -EINVAL;
588         }
589     }
590 
591     for (i = 0; i < PCI_NUM_PINS; ++i) {
592         pci_set_irq_state(s, i, irq_state[i]);
593     }
594 
595     return 0;
596 }
597 
598 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
599                              const VMStateField *field, QJSON *vmdesc)
600 {
601     int i;
602     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
603 
604     for (i = 0; i < PCI_NUM_PINS; ++i) {
605         qemu_put_be32(f, pci_irq_state(s, i));
606     }
607 
608     return 0;
609 }
610 
611 static VMStateInfo vmstate_info_pci_irq_state = {
612     .name = "pci irq state",
613     .get  = get_pci_irq_state,
614     .put  = put_pci_irq_state,
615 };
616 
617 static bool migrate_is_pcie(void *opaque, int version_id)
618 {
619     return pci_is_express((PCIDevice *)opaque);
620 }
621 
622 static bool migrate_is_not_pcie(void *opaque, int version_id)
623 {
624     return !pci_is_express((PCIDevice *)opaque);
625 }
626 
627 const VMStateDescription vmstate_pci_device = {
628     .name = "PCIDevice",
629     .version_id = 2,
630     .minimum_version_id = 1,
631     .fields = (VMStateField[]) {
632         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
633         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
634                                    migrate_is_not_pcie,
635                                    0, vmstate_info_pci_config,
636                                    PCI_CONFIG_SPACE_SIZE),
637         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
638                                    migrate_is_pcie,
639                                    0, vmstate_info_pci_config,
640                                    PCIE_CONFIG_SPACE_SIZE),
641         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
642                                    vmstate_info_pci_irq_state,
643                                    PCI_NUM_PINS * sizeof(int32_t)),
644         VMSTATE_END_OF_LIST()
645     }
646 };
647 
648 
649 void pci_device_save(PCIDevice *s, QEMUFile *f)
650 {
651     /* Clear interrupt status bit: it is implicit
652      * in irq_state which we are saving.
653      * This makes us compatible with old devices
654      * which never set or clear this bit. */
655     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
656     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
657     /* Restore the interrupt status bit. */
658     pci_update_irq_status(s);
659 }
660 
661 int pci_device_load(PCIDevice *s, QEMUFile *f)
662 {
663     int ret;
664     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
665     /* Restore the interrupt status bit. */
666     pci_update_irq_status(s);
667     return ret;
668 }
669 
670 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
671 {
672     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
673                  pci_default_sub_vendor_id);
674     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
675                  pci_default_sub_device_id);
676 }
677 
678 /*
679  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
680  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
681  */
682 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
683                              unsigned int *slotp, unsigned int *funcp)
684 {
685     const char *p;
686     char *e;
687     unsigned long val;
688     unsigned long dom = 0, bus = 0;
689     unsigned int slot = 0;
690     unsigned int func = 0;
691 
692     p = addr;
693     val = strtoul(p, &e, 16);
694     if (e == p)
695         return -1;
696     if (*e == ':') {
697         bus = val;
698         p = e + 1;
699         val = strtoul(p, &e, 16);
700         if (e == p)
701             return -1;
702         if (*e == ':') {
703             dom = bus;
704             bus = val;
705             p = e + 1;
706             val = strtoul(p, &e, 16);
707             if (e == p)
708                 return -1;
709         }
710     }
711 
712     slot = val;
713 
714     if (funcp != NULL) {
715         if (*e != '.')
716             return -1;
717 
718         p = e + 1;
719         val = strtoul(p, &e, 16);
720         if (e == p)
721             return -1;
722 
723         func = val;
724     }
725 
726     /* if funcp == NULL func is 0 */
727     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
728         return -1;
729 
730     if (*e)
731         return -1;
732 
733     *domp = dom;
734     *busp = bus;
735     *slotp = slot;
736     if (funcp != NULL)
737         *funcp = func;
738     return 0;
739 }
740 
741 static void pci_init_cmask(PCIDevice *dev)
742 {
743     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
744     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
745     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
746     dev->cmask[PCI_REVISION_ID] = 0xff;
747     dev->cmask[PCI_CLASS_PROG] = 0xff;
748     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
749     dev->cmask[PCI_HEADER_TYPE] = 0xff;
750     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
751 }
752 
753 static void pci_init_wmask(PCIDevice *dev)
754 {
755     int config_size = pci_config_size(dev);
756 
757     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
758     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
759     pci_set_word(dev->wmask + PCI_COMMAND,
760                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
761                  PCI_COMMAND_INTX_DISABLE);
762     pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
763 
764     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
765            config_size - PCI_CONFIG_HEADER_SIZE);
766 }
767 
768 static void pci_init_w1cmask(PCIDevice *dev)
769 {
770     /*
771      * Note: It's okay to set w1cmask even for readonly bits as
772      * long as their value is hardwired to 0.
773      */
774     pci_set_word(dev->w1cmask + PCI_STATUS,
775                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
776                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
777                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
778 }
779 
780 static void pci_init_mask_bridge(PCIDevice *d)
781 {
782     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
783        PCI_SEC_LETENCY_TIMER */
784     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
785 
786     /* base and limit */
787     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
788     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
789     pci_set_word(d->wmask + PCI_MEMORY_BASE,
790                  PCI_MEMORY_RANGE_MASK & 0xffff);
791     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
792                  PCI_MEMORY_RANGE_MASK & 0xffff);
793     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
794                  PCI_PREF_RANGE_MASK & 0xffff);
795     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
796                  PCI_PREF_RANGE_MASK & 0xffff);
797 
798     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
799     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
800 
801     /* Supported memory and i/o types */
802     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
803     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
804     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
805                                PCI_PREF_RANGE_TYPE_64);
806     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
807                                PCI_PREF_RANGE_TYPE_64);
808 
809     /*
810      * TODO: Bridges default to 10-bit VGA decoding but we currently only
811      * implement 16-bit decoding (no alias support).
812      */
813     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
814                  PCI_BRIDGE_CTL_PARITY |
815                  PCI_BRIDGE_CTL_SERR |
816                  PCI_BRIDGE_CTL_ISA |
817                  PCI_BRIDGE_CTL_VGA |
818                  PCI_BRIDGE_CTL_VGA_16BIT |
819                  PCI_BRIDGE_CTL_MASTER_ABORT |
820                  PCI_BRIDGE_CTL_BUS_RESET |
821                  PCI_BRIDGE_CTL_FAST_BACK |
822                  PCI_BRIDGE_CTL_DISCARD |
823                  PCI_BRIDGE_CTL_SEC_DISCARD |
824                  PCI_BRIDGE_CTL_DISCARD_SERR);
825     /* Below does not do anything as we never set this bit, put here for
826      * completeness. */
827     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
828                  PCI_BRIDGE_CTL_DISCARD_STATUS);
829     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
830     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
831     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
832                                PCI_PREF_RANGE_TYPE_MASK);
833     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
834                                PCI_PREF_RANGE_TYPE_MASK);
835 }
836 
837 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
838 {
839     uint8_t slot = PCI_SLOT(dev->devfn);
840     uint8_t func;
841 
842     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
843         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
844     }
845 
846     /*
847      * multifunction bit is interpreted in two ways as follows.
848      *   - all functions must set the bit to 1.
849      *     Example: Intel X53
850      *   - function 0 must set the bit, but the rest function (> 0)
851      *     is allowed to leave the bit to 0.
852      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
853      *
854      * So OS (at least Linux) checks the bit of only function 0,
855      * and doesn't see the bit of function > 0.
856      *
857      * The below check allows both interpretation.
858      */
859     if (PCI_FUNC(dev->devfn)) {
860         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
861         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
862             /* function 0 should set multifunction bit */
863             error_setg(errp, "PCI: single function device can't be populated "
864                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
865             return;
866         }
867         return;
868     }
869 
870     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
871         return;
872     }
873     /* function 0 indicates single function, so function > 0 must be NULL */
874     for (func = 1; func < PCI_FUNC_MAX; ++func) {
875         if (bus->devices[PCI_DEVFN(slot, func)]) {
876             error_setg(errp, "PCI: %x.0 indicates single function, "
877                        "but %x.%x is already populated.",
878                        slot, slot, func);
879             return;
880         }
881     }
882 }
883 
884 static void pci_config_alloc(PCIDevice *pci_dev)
885 {
886     int config_size = pci_config_size(pci_dev);
887 
888     pci_dev->config = g_malloc0(config_size);
889     pci_dev->cmask = g_malloc0(config_size);
890     pci_dev->wmask = g_malloc0(config_size);
891     pci_dev->w1cmask = g_malloc0(config_size);
892     pci_dev->used = g_malloc0(config_size);
893 }
894 
895 static void pci_config_free(PCIDevice *pci_dev)
896 {
897     g_free(pci_dev->config);
898     g_free(pci_dev->cmask);
899     g_free(pci_dev->wmask);
900     g_free(pci_dev->w1cmask);
901     g_free(pci_dev->used);
902 }
903 
904 static void do_pci_unregister_device(PCIDevice *pci_dev)
905 {
906     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
907     pci_config_free(pci_dev);
908 
909     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
910         memory_region_del_subregion(&pci_dev->bus_master_container_region,
911                                     &pci_dev->bus_master_enable_region);
912     }
913     address_space_destroy(&pci_dev->bus_master_as);
914 }
915 
916 /* Extract PCIReqIDCache into BDF format */
917 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
918 {
919     uint8_t bus_n;
920     uint16_t result;
921 
922     switch (cache->type) {
923     case PCI_REQ_ID_BDF:
924         result = pci_get_bdf(cache->dev);
925         break;
926     case PCI_REQ_ID_SECONDARY_BUS:
927         bus_n = pci_dev_bus_num(cache->dev);
928         result = PCI_BUILD_BDF(bus_n, 0);
929         break;
930     default:
931         error_report("Invalid PCI requester ID cache type: %d",
932                      cache->type);
933         exit(1);
934         break;
935     }
936 
937     return result;
938 }
939 
940 /* Parse bridges up to the root complex and return requester ID
941  * cache for specific device.  For full PCIe topology, the cache
942  * result would be exactly the same as getting BDF of the device.
943  * However, several tricks are required when system mixed up with
944  * legacy PCI devices and PCIe-to-PCI bridges.
945  *
946  * Here we cache the proxy device (and type) not requester ID since
947  * bus number might change from time to time.
948  */
949 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
950 {
951     PCIDevice *parent;
952     PCIReqIDCache cache = {
953         .dev = dev,
954         .type = PCI_REQ_ID_BDF,
955     };
956 
957     while (!pci_bus_is_root(pci_get_bus(dev))) {
958         /* We are under PCI/PCIe bridges */
959         parent = pci_get_bus(dev)->parent_dev;
960         if (pci_is_express(parent)) {
961             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
962                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
963                  * override the requester ID using secondary bus
964                  * number of parent bridge with zeroed devfn
965                  * (pcie-to-pci bridge spec chap 2.3). */
966                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
967                 cache.dev = dev;
968             }
969         } else {
970             /* Legacy PCI, override requester ID with the bridge's
971              * BDF upstream.  When the root complex connects to
972              * legacy PCI devices (including buses), it can only
973              * obtain requester ID info from directly attached
974              * devices.  If devices are attached under bridges, only
975              * the requester ID of the bridge that is directly
976              * attached to the root complex can be recognized. */
977             cache.type = PCI_REQ_ID_BDF;
978             cache.dev = parent;
979         }
980         dev = parent;
981     }
982 
983     return cache;
984 }
985 
986 uint16_t pci_requester_id(PCIDevice *dev)
987 {
988     return pci_req_id_cache_extract(&dev->requester_id_cache);
989 }
990 
991 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
992 {
993     return !(bus->devices[devfn]);
994 }
995 
996 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
997 {
998     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
999 }
1000 
1001 /* -1 for devfn means auto assign */
1002 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
1003                                          const char *name, int devfn,
1004                                          Error **errp)
1005 {
1006     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1007     PCIConfigReadFunc *config_read = pc->config_read;
1008     PCIConfigWriteFunc *config_write = pc->config_write;
1009     Error *local_err = NULL;
1010     DeviceState *dev = DEVICE(pci_dev);
1011     PCIBus *bus = pci_get_bus(pci_dev);
1012 
1013     /* Only pci bridges can be attached to extra PCI root buses */
1014     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1015         error_setg(errp,
1016                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1017                     bus->parent_dev->name);
1018         return NULL;
1019     }
1020 
1021     if (devfn < 0) {
1022         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1023             devfn += PCI_FUNC_MAX) {
1024             if (pci_bus_devfn_available(bus, devfn) &&
1025                    !pci_bus_devfn_reserved(bus, devfn)) {
1026                 goto found;
1027             }
1028         }
1029         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1030                    "or reserved", name);
1031         return NULL;
1032     found: ;
1033     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1034         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1035                    " reserved",
1036                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1037         return NULL;
1038     } else if (!pci_bus_devfn_available(bus, devfn)) {
1039         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1040                    " in use by %s",
1041                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1042                    bus->devices[devfn]->name);
1043         return NULL;
1044     } else if (dev->hotplugged &&
1045                pci_get_function_0(pci_dev)) {
1046         error_setg(errp, "PCI: slot %d function 0 already occupied by %s,"
1047                    " new func %s cannot be exposed to guest.",
1048                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1049                    pci_get_function_0(pci_dev)->name,
1050                    name);
1051 
1052        return NULL;
1053     }
1054 
1055     pci_dev->devfn = devfn;
1056     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1057     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1058 
1059     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1060                        "bus master container", UINT64_MAX);
1061     address_space_init(&pci_dev->bus_master_as,
1062                        &pci_dev->bus_master_container_region, pci_dev->name);
1063 
1064     if (qdev_hotplug) {
1065         pci_init_bus_master(pci_dev);
1066     }
1067     pci_dev->irq_state = 0;
1068     pci_config_alloc(pci_dev);
1069 
1070     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1071     pci_config_set_device_id(pci_dev->config, pc->device_id);
1072     pci_config_set_revision(pci_dev->config, pc->revision);
1073     pci_config_set_class(pci_dev->config, pc->class_id);
1074 
1075     if (!pc->is_bridge) {
1076         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1077             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1078                          pc->subsystem_vendor_id);
1079             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1080                          pc->subsystem_id);
1081         } else {
1082             pci_set_default_subsystem_id(pci_dev);
1083         }
1084     } else {
1085         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1086         assert(!pc->subsystem_vendor_id);
1087         assert(!pc->subsystem_id);
1088     }
1089     pci_init_cmask(pci_dev);
1090     pci_init_wmask(pci_dev);
1091     pci_init_w1cmask(pci_dev);
1092     if (pc->is_bridge) {
1093         pci_init_mask_bridge(pci_dev);
1094     }
1095     pci_init_multifunction(bus, pci_dev, &local_err);
1096     if (local_err) {
1097         error_propagate(errp, local_err);
1098         do_pci_unregister_device(pci_dev);
1099         return NULL;
1100     }
1101 
1102     if (!config_read)
1103         config_read = pci_default_read_config;
1104     if (!config_write)
1105         config_write = pci_default_write_config;
1106     pci_dev->config_read = config_read;
1107     pci_dev->config_write = config_write;
1108     bus->devices[devfn] = pci_dev;
1109     pci_dev->version_id = 2; /* Current pci device vmstate version */
1110     return pci_dev;
1111 }
1112 
1113 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1114 {
1115     PCIIORegion *r;
1116     int i;
1117 
1118     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1119         r = &pci_dev->io_regions[i];
1120         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1121             continue;
1122         memory_region_del_subregion(r->address_space, r->memory);
1123     }
1124 
1125     pci_unregister_vga(pci_dev);
1126 }
1127 
1128 static void pci_qdev_unrealize(DeviceState *dev)
1129 {
1130     PCIDevice *pci_dev = PCI_DEVICE(dev);
1131     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1132 
1133     pci_unregister_io_regions(pci_dev);
1134     pci_del_option_rom(pci_dev);
1135 
1136     if (pc->exit) {
1137         pc->exit(pci_dev);
1138     }
1139 
1140     pci_device_deassert_intx(pci_dev);
1141     do_pci_unregister_device(pci_dev);
1142 }
1143 
1144 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1145                       uint8_t type, MemoryRegion *memory)
1146 {
1147     PCIIORegion *r;
1148     uint32_t addr; /* offset in pci config space */
1149     uint64_t wmask;
1150     pcibus_t size = memory_region_size(memory);
1151     uint8_t hdr_type;
1152 
1153     assert(region_num >= 0);
1154     assert(region_num < PCI_NUM_REGIONS);
1155     assert(is_power_of_2(size));
1156 
1157     /* A PCI bridge device (with Type 1 header) may only have at most 2 BARs */
1158     hdr_type =
1159         pci_dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1160     assert(hdr_type != PCI_HEADER_TYPE_BRIDGE || region_num < 2);
1161 
1162     r = &pci_dev->io_regions[region_num];
1163     r->addr = PCI_BAR_UNMAPPED;
1164     r->size = size;
1165     r->type = type;
1166     r->memory = memory;
1167     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1168                         ? pci_get_bus(pci_dev)->address_space_io
1169                         : pci_get_bus(pci_dev)->address_space_mem;
1170 
1171     wmask = ~(size - 1);
1172     if (region_num == PCI_ROM_SLOT) {
1173         /* ROM enable bit is writable */
1174         wmask |= PCI_ROM_ADDRESS_ENABLE;
1175     }
1176 
1177     addr = pci_bar(pci_dev, region_num);
1178     pci_set_long(pci_dev->config + addr, type);
1179 
1180     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1181         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1182         pci_set_quad(pci_dev->wmask + addr, wmask);
1183         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1184     } else {
1185         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1186         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1187     }
1188 }
1189 
1190 static void pci_update_vga(PCIDevice *pci_dev)
1191 {
1192     uint16_t cmd;
1193 
1194     if (!pci_dev->has_vga) {
1195         return;
1196     }
1197 
1198     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1199 
1200     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1201                               cmd & PCI_COMMAND_MEMORY);
1202     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1203                               cmd & PCI_COMMAND_IO);
1204     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1205                               cmd & PCI_COMMAND_IO);
1206 }
1207 
1208 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1209                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1210 {
1211     PCIBus *bus = pci_get_bus(pci_dev);
1212 
1213     assert(!pci_dev->has_vga);
1214 
1215     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1216     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1217     memory_region_add_subregion_overlap(bus->address_space_mem,
1218                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1219 
1220     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1221     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1222     memory_region_add_subregion_overlap(bus->address_space_io,
1223                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1224 
1225     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1226     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1227     memory_region_add_subregion_overlap(bus->address_space_io,
1228                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1229     pci_dev->has_vga = true;
1230 
1231     pci_update_vga(pci_dev);
1232 }
1233 
1234 void pci_unregister_vga(PCIDevice *pci_dev)
1235 {
1236     PCIBus *bus = pci_get_bus(pci_dev);
1237 
1238     if (!pci_dev->has_vga) {
1239         return;
1240     }
1241 
1242     memory_region_del_subregion(bus->address_space_mem,
1243                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1244     memory_region_del_subregion(bus->address_space_io,
1245                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1246     memory_region_del_subregion(bus->address_space_io,
1247                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1248     pci_dev->has_vga = false;
1249 }
1250 
1251 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1252 {
1253     return pci_dev->io_regions[region_num].addr;
1254 }
1255 
1256 static pcibus_t pci_bar_address(PCIDevice *d,
1257                                 int reg, uint8_t type, pcibus_t size)
1258 {
1259     pcibus_t new_addr, last_addr;
1260     int bar = pci_bar(d, reg);
1261     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1262     Object *machine = qdev_get_machine();
1263     ObjectClass *oc = object_get_class(machine);
1264     MachineClass *mc = MACHINE_CLASS(oc);
1265     bool allow_0_address = mc->pci_allow_0_address;
1266 
1267     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1268         if (!(cmd & PCI_COMMAND_IO)) {
1269             return PCI_BAR_UNMAPPED;
1270         }
1271         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1272         last_addr = new_addr + size - 1;
1273         /* Check if 32 bit BAR wraps around explicitly.
1274          * TODO: make priorities correct and remove this work around.
1275          */
1276         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1277             (!allow_0_address && new_addr == 0)) {
1278             return PCI_BAR_UNMAPPED;
1279         }
1280         return new_addr;
1281     }
1282 
1283     if (!(cmd & PCI_COMMAND_MEMORY)) {
1284         return PCI_BAR_UNMAPPED;
1285     }
1286     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1287         new_addr = pci_get_quad(d->config + bar);
1288     } else {
1289         new_addr = pci_get_long(d->config + bar);
1290     }
1291     /* the ROM slot has a specific enable bit */
1292     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1293         return PCI_BAR_UNMAPPED;
1294     }
1295     new_addr &= ~(size - 1);
1296     last_addr = new_addr + size - 1;
1297     /* NOTE: we do not support wrapping */
1298     /* XXX: as we cannot support really dynamic
1299        mappings, we handle specific values as invalid
1300        mappings. */
1301     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1302         (!allow_0_address && new_addr == 0)) {
1303         return PCI_BAR_UNMAPPED;
1304     }
1305 
1306     /* Now pcibus_t is 64bit.
1307      * Check if 32 bit BAR wraps around explicitly.
1308      * Without this, PC ide doesn't work well.
1309      * TODO: remove this work around.
1310      */
1311     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1312         return PCI_BAR_UNMAPPED;
1313     }
1314 
1315     /*
1316      * OS is allowed to set BAR beyond its addressable
1317      * bits. For example, 32 bit OS can set 64bit bar
1318      * to >4G. Check it. TODO: we might need to support
1319      * it in the future for e.g. PAE.
1320      */
1321     if (last_addr >= HWADDR_MAX) {
1322         return PCI_BAR_UNMAPPED;
1323     }
1324 
1325     return new_addr;
1326 }
1327 
1328 static void pci_update_mappings(PCIDevice *d)
1329 {
1330     PCIIORegion *r;
1331     int i;
1332     pcibus_t new_addr;
1333 
1334     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1335         r = &d->io_regions[i];
1336 
1337         /* this region isn't registered */
1338         if (!r->size)
1339             continue;
1340 
1341         new_addr = pci_bar_address(d, i, r->type, r->size);
1342 
1343         /* This bar isn't changed */
1344         if (new_addr == r->addr)
1345             continue;
1346 
1347         /* now do the real mapping */
1348         if (r->addr != PCI_BAR_UNMAPPED) {
1349             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1350                                           PCI_SLOT(d->devfn),
1351                                           PCI_FUNC(d->devfn),
1352                                           i, r->addr, r->size);
1353             memory_region_del_subregion(r->address_space, r->memory);
1354         }
1355         r->addr = new_addr;
1356         if (r->addr != PCI_BAR_UNMAPPED) {
1357             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1358                                           PCI_SLOT(d->devfn),
1359                                           PCI_FUNC(d->devfn),
1360                                           i, r->addr, r->size);
1361             memory_region_add_subregion_overlap(r->address_space,
1362                                                 r->addr, r->memory, 1);
1363         }
1364     }
1365 
1366     pci_update_vga(d);
1367 }
1368 
1369 static inline int pci_irq_disabled(PCIDevice *d)
1370 {
1371     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1372 }
1373 
1374 /* Called after interrupt disabled field update in config space,
1375  * assert/deassert interrupts if necessary.
1376  * Gets original interrupt disable bit value (before update). */
1377 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1378 {
1379     int i, disabled = pci_irq_disabled(d);
1380     if (disabled == was_irq_disabled)
1381         return;
1382     for (i = 0; i < PCI_NUM_PINS; ++i) {
1383         int state = pci_irq_state(d, i);
1384         pci_change_irq_level(d, i, disabled ? -state : state);
1385     }
1386 }
1387 
1388 uint32_t pci_default_read_config(PCIDevice *d,
1389                                  uint32_t address, int len)
1390 {
1391     uint32_t val = 0;
1392 
1393     assert(address + len <= pci_config_size(d));
1394 
1395     if (pci_is_express_downstream_port(d) &&
1396         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1397         pcie_sync_bridge_lnk(d);
1398     }
1399     memcpy(&val, d->config + address, len);
1400     return le32_to_cpu(val);
1401 }
1402 
1403 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1404 {
1405     int i, was_irq_disabled = pci_irq_disabled(d);
1406     uint32_t val = val_in;
1407 
1408     assert(addr + l <= pci_config_size(d));
1409 
1410     for (i = 0; i < l; val >>= 8, ++i) {
1411         uint8_t wmask = d->wmask[addr + i];
1412         uint8_t w1cmask = d->w1cmask[addr + i];
1413         assert(!(wmask & w1cmask));
1414         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1415         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1416     }
1417     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1418         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1419         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1420         range_covers_byte(addr, l, PCI_COMMAND))
1421         pci_update_mappings(d);
1422 
1423     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1424         pci_update_irq_disabled(d, was_irq_disabled);
1425         memory_region_set_enabled(&d->bus_master_enable_region,
1426                                   pci_get_word(d->config + PCI_COMMAND)
1427                                     & PCI_COMMAND_MASTER);
1428     }
1429 
1430     msi_write_config(d, addr, val_in, l);
1431     msix_write_config(d, addr, val_in, l);
1432 }
1433 
1434 /***********************************************************/
1435 /* generic PCI irq support */
1436 
1437 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1438 static void pci_irq_handler(void *opaque, int irq_num, int level)
1439 {
1440     PCIDevice *pci_dev = opaque;
1441     int change;
1442 
1443     change = level - pci_irq_state(pci_dev, irq_num);
1444     if (!change)
1445         return;
1446 
1447     pci_set_irq_state(pci_dev, irq_num, level);
1448     pci_update_irq_status(pci_dev);
1449     if (pci_irq_disabled(pci_dev))
1450         return;
1451     pci_change_irq_level(pci_dev, irq_num, change);
1452 }
1453 
1454 static inline int pci_intx(PCIDevice *pci_dev)
1455 {
1456     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1457 }
1458 
1459 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1460 {
1461     int intx = pci_intx(pci_dev);
1462 
1463     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1464 }
1465 
1466 void pci_set_irq(PCIDevice *pci_dev, int level)
1467 {
1468     int intx = pci_intx(pci_dev);
1469     pci_irq_handler(pci_dev, intx, level);
1470 }
1471 
1472 /* Special hooks used by device assignment */
1473 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1474 {
1475     assert(pci_bus_is_root(bus));
1476     bus->route_intx_to_irq = route_intx_to_irq;
1477 }
1478 
1479 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1480 {
1481     PCIBus *bus;
1482 
1483     do {
1484         bus = pci_get_bus(dev);
1485         pin = bus->map_irq(dev, pin);
1486         dev = bus->parent_dev;
1487     } while (dev);
1488 
1489     if (!bus->route_intx_to_irq) {
1490         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1491                      object_get_typename(OBJECT(bus->qbus.parent)));
1492         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1493     }
1494 
1495     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1496 }
1497 
1498 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1499 {
1500     return old->mode != new->mode || old->irq != new->irq;
1501 }
1502 
1503 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1504 {
1505     PCIDevice *dev;
1506     PCIBus *sec;
1507     int i;
1508 
1509     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1510         dev = bus->devices[i];
1511         if (dev && dev->intx_routing_notifier) {
1512             dev->intx_routing_notifier(dev);
1513         }
1514     }
1515 
1516     QLIST_FOREACH(sec, &bus->child, sibling) {
1517         pci_bus_fire_intx_routing_notifier(sec);
1518     }
1519 }
1520 
1521 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1522                                           PCIINTxRoutingNotifier notifier)
1523 {
1524     dev->intx_routing_notifier = notifier;
1525 }
1526 
1527 /*
1528  * PCI-to-PCI bridge specification
1529  * 9.1: Interrupt routing. Table 9-1
1530  *
1531  * the PCI Express Base Specification, Revision 2.1
1532  * 2.2.8.1: INTx interrutp signaling - Rules
1533  *          the Implementation Note
1534  *          Table 2-20
1535  */
1536 /*
1537  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1538  * 0-origin unlike PCI interrupt pin register.
1539  */
1540 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1541 {
1542     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1543 }
1544 
1545 /***********************************************************/
1546 /* monitor info on PCI */
1547 
1548 typedef struct {
1549     uint16_t class;
1550     const char *desc;
1551     const char *fw_name;
1552     uint16_t fw_ign_bits;
1553 } pci_class_desc;
1554 
1555 static const pci_class_desc pci_class_descriptions[] =
1556 {
1557     { 0x0001, "VGA controller", "display"},
1558     { 0x0100, "SCSI controller", "scsi"},
1559     { 0x0101, "IDE controller", "ide"},
1560     { 0x0102, "Floppy controller", "fdc"},
1561     { 0x0103, "IPI controller", "ipi"},
1562     { 0x0104, "RAID controller", "raid"},
1563     { 0x0106, "SATA controller"},
1564     { 0x0107, "SAS controller"},
1565     { 0x0180, "Storage controller"},
1566     { 0x0200, "Ethernet controller", "ethernet"},
1567     { 0x0201, "Token Ring controller", "token-ring"},
1568     { 0x0202, "FDDI controller", "fddi"},
1569     { 0x0203, "ATM controller", "atm"},
1570     { 0x0280, "Network controller"},
1571     { 0x0300, "VGA controller", "display", 0x00ff},
1572     { 0x0301, "XGA controller"},
1573     { 0x0302, "3D controller"},
1574     { 0x0380, "Display controller"},
1575     { 0x0400, "Video controller", "video"},
1576     { 0x0401, "Audio controller", "sound"},
1577     { 0x0402, "Phone"},
1578     { 0x0403, "Audio controller", "sound"},
1579     { 0x0480, "Multimedia controller"},
1580     { 0x0500, "RAM controller", "memory"},
1581     { 0x0501, "Flash controller", "flash"},
1582     { 0x0580, "Memory controller"},
1583     { 0x0600, "Host bridge", "host"},
1584     { 0x0601, "ISA bridge", "isa"},
1585     { 0x0602, "EISA bridge", "eisa"},
1586     { 0x0603, "MC bridge", "mca"},
1587     { 0x0604, "PCI bridge", "pci-bridge"},
1588     { 0x0605, "PCMCIA bridge", "pcmcia"},
1589     { 0x0606, "NUBUS bridge", "nubus"},
1590     { 0x0607, "CARDBUS bridge", "cardbus"},
1591     { 0x0608, "RACEWAY bridge"},
1592     { 0x0680, "Bridge"},
1593     { 0x0700, "Serial port", "serial"},
1594     { 0x0701, "Parallel port", "parallel"},
1595     { 0x0800, "Interrupt controller", "interrupt-controller"},
1596     { 0x0801, "DMA controller", "dma-controller"},
1597     { 0x0802, "Timer", "timer"},
1598     { 0x0803, "RTC", "rtc"},
1599     { 0x0900, "Keyboard", "keyboard"},
1600     { 0x0901, "Pen", "pen"},
1601     { 0x0902, "Mouse", "mouse"},
1602     { 0x0A00, "Dock station", "dock", 0x00ff},
1603     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1604     { 0x0c00, "Fireware contorller", "fireware"},
1605     { 0x0c01, "Access bus controller", "access-bus"},
1606     { 0x0c02, "SSA controller", "ssa"},
1607     { 0x0c03, "USB controller", "usb"},
1608     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1609     { 0x0c05, "SMBus"},
1610     { 0, NULL}
1611 };
1612 
1613 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1614                                                   void (*fn)(PCIBus *b,
1615                                                              PCIDevice *d,
1616                                                              void *opaque),
1617                                                   void *opaque)
1618 {
1619     PCIDevice *d;
1620     int devfn;
1621 
1622     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1623         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1624         if (d) {
1625             fn(bus, d, opaque);
1626         }
1627     }
1628 }
1629 
1630 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1631                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1632                          void *opaque)
1633 {
1634     bus = pci_find_bus_nr(bus, bus_num);
1635 
1636     if (bus) {
1637         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1638     }
1639 }
1640 
1641 static void pci_for_each_device_under_bus(PCIBus *bus,
1642                                           void (*fn)(PCIBus *b, PCIDevice *d,
1643                                                      void *opaque),
1644                                           void *opaque)
1645 {
1646     PCIDevice *d;
1647     int devfn;
1648 
1649     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1650         d = bus->devices[devfn];
1651         if (d) {
1652             fn(bus, d, opaque);
1653         }
1654     }
1655 }
1656 
1657 void pci_for_each_device(PCIBus *bus, int bus_num,
1658                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1659                          void *opaque)
1660 {
1661     bus = pci_find_bus_nr(bus, bus_num);
1662 
1663     if (bus) {
1664         pci_for_each_device_under_bus(bus, fn, opaque);
1665     }
1666 }
1667 
1668 static const pci_class_desc *get_class_desc(int class)
1669 {
1670     const pci_class_desc *desc;
1671 
1672     desc = pci_class_descriptions;
1673     while (desc->desc && class != desc->class) {
1674         desc++;
1675     }
1676 
1677     return desc;
1678 }
1679 
1680 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1681 
1682 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1683 {
1684     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1685     int i;
1686 
1687     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1688         const PCIIORegion *r = &dev->io_regions[i];
1689         PciMemoryRegionList *region;
1690 
1691         if (!r->size) {
1692             continue;
1693         }
1694 
1695         region = g_malloc0(sizeof(*region));
1696         region->value = g_malloc0(sizeof(*region->value));
1697 
1698         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1699             region->value->type = g_strdup("io");
1700         } else {
1701             region->value->type = g_strdup("memory");
1702             region->value->has_prefetch = true;
1703             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1704             region->value->has_mem_type_64 = true;
1705             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1706         }
1707 
1708         region->value->bar = i;
1709         region->value->address = r->addr;
1710         region->value->size = r->size;
1711 
1712         /* XXX: waiting for the qapi to support GSList */
1713         if (!cur_item) {
1714             head = cur_item = region;
1715         } else {
1716             cur_item->next = region;
1717             cur_item = region;
1718         }
1719     }
1720 
1721     return head;
1722 }
1723 
1724 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1725                                            int bus_num)
1726 {
1727     PciBridgeInfo *info;
1728     PciMemoryRange *range;
1729 
1730     info = g_new0(PciBridgeInfo, 1);
1731 
1732     info->bus = g_new0(PciBusInfo, 1);
1733     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1734     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1735     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1736 
1737     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1738     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1739     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1740 
1741     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1742     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1743     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1744 
1745     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1746     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1747     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1748 
1749     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1750         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1751         if (child_bus) {
1752             info->has_devices = true;
1753             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1754         }
1755     }
1756 
1757     return info;
1758 }
1759 
1760 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1761                                            int bus_num)
1762 {
1763     const pci_class_desc *desc;
1764     PciDeviceInfo *info;
1765     uint8_t type;
1766     int class;
1767 
1768     info = g_new0(PciDeviceInfo, 1);
1769     info->bus = bus_num;
1770     info->slot = PCI_SLOT(dev->devfn);
1771     info->function = PCI_FUNC(dev->devfn);
1772 
1773     info->class_info = g_new0(PciDeviceClass, 1);
1774     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1775     info->class_info->q_class = class;
1776     desc = get_class_desc(class);
1777     if (desc->desc) {
1778         info->class_info->has_desc = true;
1779         info->class_info->desc = g_strdup(desc->desc);
1780     }
1781 
1782     info->id = g_new0(PciDeviceId, 1);
1783     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1784     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1785     info->regions = qmp_query_pci_regions(dev);
1786     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1787 
1788     info->irq_pin = dev->config[PCI_INTERRUPT_PIN];
1789     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1790         info->has_irq = true;
1791         info->irq = dev->config[PCI_INTERRUPT_LINE];
1792     }
1793 
1794     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1795     if (type == PCI_HEADER_TYPE_BRIDGE) {
1796         info->has_pci_bridge = true;
1797         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1798     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1799         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1800         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1801         info->id->subsystem_vendor =
1802             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1803     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1804         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1805         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1806         info->id->subsystem_vendor =
1807             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1808     }
1809 
1810     return info;
1811 }
1812 
1813 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1814 {
1815     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1816     PCIDevice *dev;
1817     int devfn;
1818 
1819     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1820         dev = bus->devices[devfn];
1821         if (dev) {
1822             info = g_malloc0(sizeof(*info));
1823             info->value = qmp_query_pci_device(dev, bus, bus_num);
1824 
1825             /* XXX: waiting for the qapi to support GSList */
1826             if (!cur_item) {
1827                 head = cur_item = info;
1828             } else {
1829                 cur_item->next = info;
1830                 cur_item = info;
1831             }
1832         }
1833     }
1834 
1835     return head;
1836 }
1837 
1838 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1839 {
1840     PciInfo *info = NULL;
1841 
1842     bus = pci_find_bus_nr(bus, bus_num);
1843     if (bus) {
1844         info = g_malloc0(sizeof(*info));
1845         info->bus = bus_num;
1846         info->devices = qmp_query_pci_devices(bus, bus_num);
1847     }
1848 
1849     return info;
1850 }
1851 
1852 PciInfoList *qmp_query_pci(Error **errp)
1853 {
1854     PciInfoList *info, *head = NULL, *cur_item = NULL;
1855     PCIHostState *host_bridge;
1856 
1857     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1858         info = g_malloc0(sizeof(*info));
1859         info->value = qmp_query_pci_bus(host_bridge->bus,
1860                                         pci_bus_num(host_bridge->bus));
1861 
1862         /* XXX: waiting for the qapi to support GSList */
1863         if (!cur_item) {
1864             head = cur_item = info;
1865         } else {
1866             cur_item->next = info;
1867             cur_item = info;
1868         }
1869     }
1870 
1871     return head;
1872 }
1873 
1874 /* Initialize a PCI NIC.  */
1875 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1876                                const char *default_model,
1877                                const char *default_devaddr)
1878 {
1879     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1880     GSList *list;
1881     GPtrArray *pci_nic_models;
1882     PCIBus *bus;
1883     PCIDevice *pci_dev;
1884     DeviceState *dev;
1885     int devfn;
1886     int i;
1887     int dom, busnr;
1888     unsigned slot;
1889 
1890     if (nd->model && !strcmp(nd->model, "virtio")) {
1891         g_free(nd->model);
1892         nd->model = g_strdup("virtio-net-pci");
1893     }
1894 
1895     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1896     pci_nic_models = g_ptr_array_new();
1897     while (list) {
1898         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1899                                              TYPE_DEVICE);
1900         GSList *next;
1901         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1902             dc->user_creatable) {
1903             const char *name = object_class_get_name(list->data);
1904             /*
1905              * A network device might also be something else than a NIC, see
1906              * e.g. the "rocker" device. Thus we have to look for the "netdev"
1907              * property, too. Unfortunately, some devices like virtio-net only
1908              * create this property during instance_init, so we have to create
1909              * a temporary instance here to be able to check it.
1910              */
1911             Object *obj = object_new_with_class(OBJECT_CLASS(dc));
1912             if (object_property_find(obj, "netdev")) {
1913                 g_ptr_array_add(pci_nic_models, (gpointer)name);
1914             }
1915             object_unref(obj);
1916         }
1917         next = list->next;
1918         g_slist_free_1(list);
1919         list = next;
1920     }
1921     g_ptr_array_add(pci_nic_models, NULL);
1922 
1923     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1924         exit(0);
1925     }
1926 
1927     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1928                             default_model);
1929     if (i < 0) {
1930         exit(1);
1931     }
1932 
1933     if (!rootbus) {
1934         error_report("No primary PCI bus");
1935         exit(1);
1936     }
1937 
1938     assert(!rootbus->parent_dev);
1939 
1940     if (!devaddr) {
1941         devfn = -1;
1942         busnr = 0;
1943     } else {
1944         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1945             error_report("Invalid PCI device address %s for device %s",
1946                          devaddr, nd->model);
1947             exit(1);
1948         }
1949 
1950         if (dom != 0) {
1951             error_report("No support for non-zero PCI domains");
1952             exit(1);
1953         }
1954 
1955         devfn = PCI_DEVFN(slot, 0);
1956     }
1957 
1958     bus = pci_find_bus_nr(rootbus, busnr);
1959     if (!bus) {
1960         error_report("Invalid PCI device address %s for device %s",
1961                      devaddr, nd->model);
1962         exit(1);
1963     }
1964 
1965     pci_dev = pci_new(devfn, nd->model);
1966     dev = &pci_dev->qdev;
1967     qdev_set_nic_properties(dev, nd);
1968     pci_realize_and_unref(pci_dev, bus, &error_fatal);
1969     g_ptr_array_free(pci_nic_models, true);
1970     return pci_dev;
1971 }
1972 
1973 PCIDevice *pci_vga_init(PCIBus *bus)
1974 {
1975     switch (vga_interface_type) {
1976     case VGA_CIRRUS:
1977         return pci_create_simple(bus, -1, "cirrus-vga");
1978     case VGA_QXL:
1979         return pci_create_simple(bus, -1, "qxl-vga");
1980     case VGA_STD:
1981         return pci_create_simple(bus, -1, "VGA");
1982     case VGA_VMWARE:
1983         return pci_create_simple(bus, -1, "vmware-svga");
1984     case VGA_VIRTIO:
1985         return pci_create_simple(bus, -1, "virtio-vga");
1986     case VGA_NONE:
1987     default: /* Other non-PCI types. Checking for unsupported types is already
1988                 done in vl.c. */
1989         return NULL;
1990     }
1991 }
1992 
1993 /* Whether a given bus number is in range of the secondary
1994  * bus of the given bridge device. */
1995 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1996 {
1997     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1998              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1999         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
2000         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
2001 }
2002 
2003 /* Whether a given bus number is in a range of a root bus */
2004 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
2005 {
2006     int i;
2007 
2008     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
2009         PCIDevice *dev = bus->devices[i];
2010 
2011         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
2012             if (pci_secondary_bus_in_range(dev, bus_num)) {
2013                 return true;
2014             }
2015         }
2016     }
2017 
2018     return false;
2019 }
2020 
2021 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
2022 {
2023     PCIBus *sec;
2024 
2025     if (!bus) {
2026         return NULL;
2027     }
2028 
2029     if (pci_bus_num(bus) == bus_num) {
2030         return bus;
2031     }
2032 
2033     /* Consider all bus numbers in range for the host pci bridge. */
2034     if (!pci_bus_is_root(bus) &&
2035         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2036         return NULL;
2037     }
2038 
2039     /* try child bus */
2040     for (; bus; bus = sec) {
2041         QLIST_FOREACH(sec, &bus->child, sibling) {
2042             if (pci_bus_num(sec) == bus_num) {
2043                 return sec;
2044             }
2045             /* PXB buses assumed to be children of bus 0 */
2046             if (pci_bus_is_root(sec)) {
2047                 if (pci_root_bus_in_range(sec, bus_num)) {
2048                     break;
2049                 }
2050             } else {
2051                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2052                     break;
2053                 }
2054             }
2055         }
2056     }
2057 
2058     return NULL;
2059 }
2060 
2061 void pci_for_each_bus_depth_first(PCIBus *bus,
2062                                   void *(*begin)(PCIBus *bus, void *parent_state),
2063                                   void (*end)(PCIBus *bus, void *state),
2064                                   void *parent_state)
2065 {
2066     PCIBus *sec;
2067     void *state;
2068 
2069     if (!bus) {
2070         return;
2071     }
2072 
2073     if (begin) {
2074         state = begin(bus, parent_state);
2075     } else {
2076         state = parent_state;
2077     }
2078 
2079     QLIST_FOREACH(sec, &bus->child, sibling) {
2080         pci_for_each_bus_depth_first(sec, begin, end, state);
2081     }
2082 
2083     if (end) {
2084         end(bus, state);
2085     }
2086 }
2087 
2088 
2089 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2090 {
2091     bus = pci_find_bus_nr(bus, bus_num);
2092 
2093     if (!bus)
2094         return NULL;
2095 
2096     return bus->devices[devfn];
2097 }
2098 
2099 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2100 {
2101     PCIDevice *pci_dev = (PCIDevice *)qdev;
2102     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2103     ObjectClass *klass = OBJECT_CLASS(pc);
2104     Error *local_err = NULL;
2105     bool is_default_rom;
2106     uint16_t class_id;
2107 
2108     /* initialize cap_present for pci_is_express() and pci_config_size(),
2109      * Note that hybrid PCIs are not set automatically and need to manage
2110      * QEMU_PCI_CAP_EXPRESS manually */
2111     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2112        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2113         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2114     }
2115 
2116     pci_dev = do_pci_register_device(pci_dev,
2117                                      object_get_typename(OBJECT(qdev)),
2118                                      pci_dev->devfn, errp);
2119     if (pci_dev == NULL)
2120         return;
2121 
2122     if (pc->realize) {
2123         pc->realize(pci_dev, &local_err);
2124         if (local_err) {
2125             error_propagate(errp, local_err);
2126             do_pci_unregister_device(pci_dev);
2127             return;
2128         }
2129     }
2130 
2131     if (pci_dev->failover_pair_id) {
2132         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2133             error_setg(errp, "failover primary device must be on "
2134                              "PCIExpress bus");
2135             pci_qdev_unrealize(DEVICE(pci_dev));
2136             return;
2137         }
2138         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2139         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2140             error_setg(errp, "failover primary device is not an "
2141                              "Ethernet device");
2142             pci_qdev_unrealize(DEVICE(pci_dev));
2143             return;
2144         }
2145         if (!(pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2146             && (PCI_FUNC(pci_dev->devfn) == 0)) {
2147             qdev->allow_unplug_during_migration = true;
2148         } else {
2149             error_setg(errp, "failover: primary device must be in its own "
2150                               "PCI slot");
2151             pci_qdev_unrealize(DEVICE(pci_dev));
2152             return;
2153         }
2154         qdev->allow_unplug_during_migration = true;
2155     }
2156 
2157     /* rom loading */
2158     is_default_rom = false;
2159     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2160         pci_dev->romfile = g_strdup(pc->romfile);
2161         is_default_rom = true;
2162     }
2163 
2164     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2165     if (local_err) {
2166         error_propagate(errp, local_err);
2167         pci_qdev_unrealize(DEVICE(pci_dev));
2168         return;
2169     }
2170 }
2171 
2172 PCIDevice *pci_new_multifunction(int devfn, bool multifunction,
2173                                  const char *name)
2174 {
2175     DeviceState *dev;
2176 
2177     dev = qdev_new(name);
2178     qdev_prop_set_int32(dev, "addr", devfn);
2179     qdev_prop_set_bit(dev, "multifunction", multifunction);
2180     return PCI_DEVICE(dev);
2181 }
2182 
2183 PCIDevice *pci_new(int devfn, const char *name)
2184 {
2185     return pci_new_multifunction(devfn, false, name);
2186 }
2187 
2188 bool pci_realize_and_unref(PCIDevice *dev, PCIBus *bus, Error **errp)
2189 {
2190     return qdev_realize_and_unref(&dev->qdev, &bus->qbus, errp);
2191 }
2192 
2193 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2194                                            bool multifunction,
2195                                            const char *name)
2196 {
2197     PCIDevice *dev = pci_new_multifunction(devfn, multifunction, name);
2198     pci_realize_and_unref(dev, bus, &error_fatal);
2199     return dev;
2200 }
2201 
2202 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2203 {
2204     return pci_create_simple_multifunction(bus, devfn, false, name);
2205 }
2206 
2207 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2208 {
2209     int offset = PCI_CONFIG_HEADER_SIZE;
2210     int i;
2211     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2212         if (pdev->used[i])
2213             offset = i + 1;
2214         else if (i - offset + 1 == size)
2215             return offset;
2216     }
2217     return 0;
2218 }
2219 
2220 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2221                                         uint8_t *prev_p)
2222 {
2223     uint8_t next, prev;
2224 
2225     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2226         return 0;
2227 
2228     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2229          prev = next + PCI_CAP_LIST_NEXT)
2230         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2231             break;
2232 
2233     if (prev_p)
2234         *prev_p = prev;
2235     return next;
2236 }
2237 
2238 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2239 {
2240     uint8_t next, prev, found = 0;
2241 
2242     if (!(pdev->used[offset])) {
2243         return 0;
2244     }
2245 
2246     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2247 
2248     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2249          prev = next + PCI_CAP_LIST_NEXT) {
2250         if (next <= offset && next > found) {
2251             found = next;
2252         }
2253     }
2254     return found;
2255 }
2256 
2257 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2258    This is needed for an option rom which is used for more than one device. */
2259 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2260 {
2261     uint16_t vendor_id;
2262     uint16_t device_id;
2263     uint16_t rom_vendor_id;
2264     uint16_t rom_device_id;
2265     uint16_t rom_magic;
2266     uint16_t pcir_offset;
2267     uint8_t checksum;
2268 
2269     /* Words in rom data are little endian (like in PCI configuration),
2270        so they can be read / written with pci_get_word / pci_set_word. */
2271 
2272     /* Only a valid rom will be patched. */
2273     rom_magic = pci_get_word(ptr);
2274     if (rom_magic != 0xaa55) {
2275         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2276         return;
2277     }
2278     pcir_offset = pci_get_word(ptr + 0x18);
2279     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2280         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2281         return;
2282     }
2283 
2284     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2285     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2286     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2287     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2288 
2289     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2290                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2291 
2292     checksum = ptr[6];
2293 
2294     if (vendor_id != rom_vendor_id) {
2295         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2296         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2297         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2298         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2299         ptr[6] = checksum;
2300         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2301     }
2302 
2303     if (device_id != rom_device_id) {
2304         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2305         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2306         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2307         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2308         ptr[6] = checksum;
2309         pci_set_word(ptr + pcir_offset + 6, device_id);
2310     }
2311 }
2312 
2313 /* Add an option rom for the device */
2314 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2315                                Error **errp)
2316 {
2317     int size;
2318     char *path;
2319     void *ptr;
2320     char name[32];
2321     const VMStateDescription *vmsd;
2322 
2323     if (!pdev->romfile)
2324         return;
2325     if (strlen(pdev->romfile) == 0)
2326         return;
2327 
2328     if (!pdev->rom_bar) {
2329         /*
2330          * Load rom via fw_cfg instead of creating a rom bar,
2331          * for 0.11 compatibility.
2332          */
2333         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2334 
2335         /*
2336          * Hot-plugged devices can't use the option ROM
2337          * if the rom bar is disabled.
2338          */
2339         if (DEVICE(pdev)->hotplugged) {
2340             error_setg(errp, "Hot-plugged device without ROM bar"
2341                        " can't have an option ROM");
2342             return;
2343         }
2344 
2345         if (class == 0x0300) {
2346             rom_add_vga(pdev->romfile);
2347         } else {
2348             rom_add_option(pdev->romfile, -1);
2349         }
2350         return;
2351     }
2352 
2353     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2354     if (path == NULL) {
2355         path = g_strdup(pdev->romfile);
2356     }
2357 
2358     size = get_image_size(path);
2359     if (size < 0) {
2360         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2361         g_free(path);
2362         return;
2363     } else if (size == 0) {
2364         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2365         g_free(path);
2366         return;
2367     }
2368     size = pow2ceil(size);
2369 
2370     vmsd = qdev_get_vmsd(DEVICE(pdev));
2371 
2372     if (vmsd) {
2373         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2374     } else {
2375         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2376     }
2377     pdev->has_rom = true;
2378     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2379     ptr = memory_region_get_ram_ptr(&pdev->rom);
2380     if (load_image_size(path, ptr, size) < 0) {
2381         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2382         g_free(path);
2383         return;
2384     }
2385     g_free(path);
2386 
2387     if (is_default_rom) {
2388         /* Only the default rom images will be patched (if needed). */
2389         pci_patch_ids(pdev, ptr, size);
2390     }
2391 
2392     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2393 }
2394 
2395 static void pci_del_option_rom(PCIDevice *pdev)
2396 {
2397     if (!pdev->has_rom)
2398         return;
2399 
2400     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2401     pdev->has_rom = false;
2402 }
2403 
2404 /*
2405  * On success, pci_add_capability() returns a positive value
2406  * that the offset of the pci capability.
2407  * On failure, it sets an error and returns a negative error
2408  * code.
2409  */
2410 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2411                        uint8_t offset, uint8_t size,
2412                        Error **errp)
2413 {
2414     uint8_t *config;
2415     int i, overlapping_cap;
2416 
2417     if (!offset) {
2418         offset = pci_find_space(pdev, size);
2419         /* out of PCI config space is programming error */
2420         assert(offset);
2421     } else {
2422         /* Verify that capabilities don't overlap.  Note: device assignment
2423          * depends on this check to verify that the device is not broken.
2424          * Should never trigger for emulated devices, but it's helpful
2425          * for debugging these. */
2426         for (i = offset; i < offset + size; i++) {
2427             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2428             if (overlapping_cap) {
2429                 error_setg(errp, "%s:%02x:%02x.%x "
2430                            "Attempt to add PCI capability %x at offset "
2431                            "%x overlaps existing capability %x at offset %x",
2432                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2433                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2434                            cap_id, offset, overlapping_cap, i);
2435                 return -EINVAL;
2436             }
2437         }
2438     }
2439 
2440     config = pdev->config + offset;
2441     config[PCI_CAP_LIST_ID] = cap_id;
2442     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2443     pdev->config[PCI_CAPABILITY_LIST] = offset;
2444     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2445     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2446     /* Make capability read-only by default */
2447     memset(pdev->wmask + offset, 0, size);
2448     /* Check capability by default */
2449     memset(pdev->cmask + offset, 0xFF, size);
2450     return offset;
2451 }
2452 
2453 /* Unlink capability from the pci config space. */
2454 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2455 {
2456     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2457     if (!offset)
2458         return;
2459     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2460     /* Make capability writable again */
2461     memset(pdev->wmask + offset, 0xff, size);
2462     memset(pdev->w1cmask + offset, 0, size);
2463     /* Clear cmask as device-specific registers can't be checked */
2464     memset(pdev->cmask + offset, 0, size);
2465     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2466 
2467     if (!pdev->config[PCI_CAPABILITY_LIST])
2468         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2469 }
2470 
2471 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2472 {
2473     return pci_find_capability_list(pdev, cap_id, NULL);
2474 }
2475 
2476 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2477 {
2478     PCIDevice *d = (PCIDevice *)dev;
2479     const pci_class_desc *desc;
2480     char ctxt[64];
2481     PCIIORegion *r;
2482     int i, class;
2483 
2484     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2485     desc = pci_class_descriptions;
2486     while (desc->desc && class != desc->class)
2487         desc++;
2488     if (desc->desc) {
2489         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2490     } else {
2491         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2492     }
2493 
2494     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2495                    "pci id %04x:%04x (sub %04x:%04x)\n",
2496                    indent, "", ctxt, pci_dev_bus_num(d),
2497                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2498                    pci_get_word(d->config + PCI_VENDOR_ID),
2499                    pci_get_word(d->config + PCI_DEVICE_ID),
2500                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2501                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2502     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2503         r = &d->io_regions[i];
2504         if (!r->size)
2505             continue;
2506         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2507                        " [0x%"FMT_PCIBUS"]\n",
2508                        indent, "",
2509                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2510                        r->addr, r->addr + r->size - 1);
2511     }
2512 }
2513 
2514 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2515 {
2516     PCIDevice *d = (PCIDevice *)dev;
2517     const char *name = NULL;
2518     const pci_class_desc *desc =  pci_class_descriptions;
2519     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2520 
2521     while (desc->desc &&
2522           (class & ~desc->fw_ign_bits) !=
2523           (desc->class & ~desc->fw_ign_bits)) {
2524         desc++;
2525     }
2526 
2527     if (desc->desc) {
2528         name = desc->fw_name;
2529     }
2530 
2531     if (name) {
2532         pstrcpy(buf, len, name);
2533     } else {
2534         snprintf(buf, len, "pci%04x,%04x",
2535                  pci_get_word(d->config + PCI_VENDOR_ID),
2536                  pci_get_word(d->config + PCI_DEVICE_ID));
2537     }
2538 
2539     return buf;
2540 }
2541 
2542 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2543 {
2544     PCIDevice *d = (PCIDevice *)dev;
2545     char path[50], name[33];
2546     int off;
2547 
2548     off = snprintf(path, sizeof(path), "%s@%x",
2549                    pci_dev_fw_name(dev, name, sizeof name),
2550                    PCI_SLOT(d->devfn));
2551     if (PCI_FUNC(d->devfn))
2552         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2553     return g_strdup(path);
2554 }
2555 
2556 static char *pcibus_get_dev_path(DeviceState *dev)
2557 {
2558     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2559     PCIDevice *t;
2560     int slot_depth;
2561     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2562      * 00 is added here to make this format compatible with
2563      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2564      * Slot.Function list specifies the slot and function numbers for all
2565      * devices on the path from root to the specific device. */
2566     const char *root_bus_path;
2567     int root_bus_len;
2568     char slot[] = ":SS.F";
2569     int slot_len = sizeof slot - 1 /* For '\0' */;
2570     int path_len;
2571     char *path, *p;
2572     int s;
2573 
2574     root_bus_path = pci_root_bus_path(d);
2575     root_bus_len = strlen(root_bus_path);
2576 
2577     /* Calculate # of slots on path between device and root. */;
2578     slot_depth = 0;
2579     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2580         ++slot_depth;
2581     }
2582 
2583     path_len = root_bus_len + slot_len * slot_depth;
2584 
2585     /* Allocate memory, fill in the terminating null byte. */
2586     path = g_malloc(path_len + 1 /* For '\0' */);
2587     path[path_len] = '\0';
2588 
2589     memcpy(path, root_bus_path, root_bus_len);
2590 
2591     /* Fill in slot numbers. We walk up from device to root, so need to print
2592      * them in the reverse order, last to first. */
2593     p = path + path_len;
2594     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2595         p -= slot_len;
2596         s = snprintf(slot, sizeof slot, ":%02x.%x",
2597                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2598         assert(s == slot_len);
2599         memcpy(p, slot, slot_len);
2600     }
2601 
2602     return path;
2603 }
2604 
2605 static int pci_qdev_find_recursive(PCIBus *bus,
2606                                    const char *id, PCIDevice **pdev)
2607 {
2608     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2609     if (!qdev) {
2610         return -ENODEV;
2611     }
2612 
2613     /* roughly check if given qdev is pci device */
2614     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2615         *pdev = PCI_DEVICE(qdev);
2616         return 0;
2617     }
2618     return -EINVAL;
2619 }
2620 
2621 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2622 {
2623     PCIHostState *host_bridge;
2624     int rc = -ENODEV;
2625 
2626     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2627         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2628         if (!tmp) {
2629             rc = 0;
2630             break;
2631         }
2632         if (tmp != -ENODEV) {
2633             rc = tmp;
2634         }
2635     }
2636 
2637     return rc;
2638 }
2639 
2640 MemoryRegion *pci_address_space(PCIDevice *dev)
2641 {
2642     return pci_get_bus(dev)->address_space_mem;
2643 }
2644 
2645 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2646 {
2647     return pci_get_bus(dev)->address_space_io;
2648 }
2649 
2650 static void pci_device_class_init(ObjectClass *klass, void *data)
2651 {
2652     DeviceClass *k = DEVICE_CLASS(klass);
2653 
2654     k->realize = pci_qdev_realize;
2655     k->unrealize = pci_qdev_unrealize;
2656     k->bus_type = TYPE_PCI_BUS;
2657     device_class_set_props(k, pci_props);
2658 }
2659 
2660 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2661 {
2662     if (!object_class_is_abstract(klass)) {
2663         ObjectClass *conventional =
2664             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2665         ObjectClass *pcie =
2666             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2667         assert(conventional || pcie);
2668     }
2669 }
2670 
2671 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2672 {
2673     PCIBus *bus = pci_get_bus(dev);
2674     PCIBus *iommu_bus = bus;
2675     uint8_t devfn = dev->devfn;
2676 
2677     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2678         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2679 
2680         /*
2681          * The requester ID of the provided device may be aliased, as seen from
2682          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2683          * requester ID to provide a unique AddressSpace for devices, but
2684          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2685          * to-PCI bridge creates and accepts transactions on behalf of down-
2686          * stream devices.  When doing so, all downstream devices are masked
2687          * (aliased) behind a single requester ID.  The requester ID used
2688          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2689          * bridges, with a PCIe capability indicating such, follow the
2690          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2691          * where the bridge uses the seconary bus as the bridge portion of the
2692          * requester ID and devfn of 00.0.  For other bridges, typically those
2693          * found on the root complex such as the dmi-to-pci-bridge, we follow
2694          * the convention of typical bare-metal hardware, which uses the
2695          * requester ID of the bridge itself.  There are device specific
2696          * exceptions to these rules, but these are the defaults that the
2697          * Linux kernel uses when determining DMA aliases itself and believed
2698          * to be true for the bare metal equivalents of the devices emulated
2699          * in QEMU.
2700          */
2701         if (!pci_bus_is_express(iommu_bus)) {
2702             PCIDevice *parent = iommu_bus->parent_dev;
2703 
2704             if (pci_is_express(parent) &&
2705                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2706                 devfn = PCI_DEVFN(0, 0);
2707                 bus = iommu_bus;
2708             } else {
2709                 devfn = parent->devfn;
2710                 bus = parent_bus;
2711             }
2712         }
2713 
2714         iommu_bus = parent_bus;
2715     }
2716     if (iommu_bus && iommu_bus->iommu_fn) {
2717         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2718     }
2719     return &address_space_memory;
2720 }
2721 
2722 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2723 {
2724     bus->iommu_fn = fn;
2725     bus->iommu_opaque = opaque;
2726 }
2727 
2728 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2729 {
2730     Range *range = opaque;
2731     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2732     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2733     int i;
2734 
2735     if (!(cmd & PCI_COMMAND_MEMORY)) {
2736         return;
2737     }
2738 
2739     if (pc->is_bridge) {
2740         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2741         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2742 
2743         base = MAX(base, 0x1ULL << 32);
2744 
2745         if (limit >= base) {
2746             Range pref_range;
2747             range_set_bounds(&pref_range, base, limit);
2748             range_extend(range, &pref_range);
2749         }
2750     }
2751     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2752         PCIIORegion *r = &dev->io_regions[i];
2753         pcibus_t lob, upb;
2754         Range region_range;
2755 
2756         if (!r->size ||
2757             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2758             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2759             continue;
2760         }
2761 
2762         lob = pci_bar_address(dev, i, r->type, r->size);
2763         upb = lob + r->size - 1;
2764         if (lob == PCI_BAR_UNMAPPED) {
2765             continue;
2766         }
2767 
2768         lob = MAX(lob, 0x1ULL << 32);
2769 
2770         if (upb >= lob) {
2771             range_set_bounds(&region_range, lob, upb);
2772             range_extend(range, &region_range);
2773         }
2774     }
2775 }
2776 
2777 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2778 {
2779     range_make_empty(range);
2780     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2781 }
2782 
2783 static bool pcie_has_upstream_port(PCIDevice *dev)
2784 {
2785     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2786 
2787     /* Device associated with an upstream port.
2788      * As there are several types of these, it's easier to check the
2789      * parent device: upstream ports are always connected to
2790      * root or downstream ports.
2791      */
2792     return parent_dev &&
2793         pci_is_express(parent_dev) &&
2794         parent_dev->exp.exp_cap &&
2795         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2796          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2797 }
2798 
2799 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2800 {
2801     PCIBus *bus = pci_get_bus(pci_dev);
2802 
2803     if(pcie_has_upstream_port(pci_dev)) {
2804         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2805         return bus->devices[0];
2806     } else {
2807         /* Other bus types might support multiple devices at slots 0-31 */
2808         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2809     }
2810 }
2811 
2812 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2813 {
2814     MSIMessage msg;
2815     if (msix_enabled(dev)) {
2816         msg = msix_get_message(dev, vector);
2817     } else if (msi_enabled(dev)) {
2818         msg = msi_get_message(dev, vector);
2819     } else {
2820         /* Should never happen */
2821         error_report("%s: unknown interrupt type", __func__);
2822         abort();
2823     }
2824     return msg;
2825 }
2826 
2827 static const TypeInfo pci_device_type_info = {
2828     .name = TYPE_PCI_DEVICE,
2829     .parent = TYPE_DEVICE,
2830     .instance_size = sizeof(PCIDevice),
2831     .abstract = true,
2832     .class_size = sizeof(PCIDeviceClass),
2833     .class_init = pci_device_class_init,
2834     .class_base_init = pci_device_class_base_init,
2835 };
2836 
2837 static void pci_register_types(void)
2838 {
2839     type_register_static(&pci_bus_info);
2840     type_register_static(&pcie_bus_info);
2841     type_register_static(&conventional_pci_interface_info);
2842     type_register_static(&pcie_interface_info);
2843     type_register_static(&pci_device_type_info);
2844 }
2845 
2846 type_init(pci_register_types)
2847