xref: /openbmc/qemu/hw/pci/pci.c (revision 2a886794)
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-misc.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("command_serr_enable", PCIDevice, cap_present,
73                     QEMU_PCI_CAP_SERR_BITNR, true),
74     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
75                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
76     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
77                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
78     DEFINE_PROP_STRING("failover_pair_id", PCIDevice,
79                        failover_pair_id),
80     DEFINE_PROP_END_OF_LIST()
81 };
82 
83 static const VMStateDescription vmstate_pcibus = {
84     .name = "PCIBUS",
85     .version_id = 1,
86     .minimum_version_id = 1,
87     .fields = (VMStateField[]) {
88         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
89         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
90                              nirq, 0, vmstate_info_int32,
91                              int32_t),
92         VMSTATE_END_OF_LIST()
93     }
94 };
95 
96 static void pci_init_bus_master(PCIDevice *pci_dev)
97 {
98     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
99 
100     memory_region_init_alias(&pci_dev->bus_master_enable_region,
101                              OBJECT(pci_dev), "bus master",
102                              dma_as->root, 0, memory_region_size(dma_as->root));
103     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
104     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
105                                 &pci_dev->bus_master_enable_region);
106 }
107 
108 static void pcibus_machine_done(Notifier *notifier, void *data)
109 {
110     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
111     int i;
112 
113     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
114         if (bus->devices[i]) {
115             pci_init_bus_master(bus->devices[i]);
116         }
117     }
118 }
119 
120 static void pci_bus_realize(BusState *qbus, Error **errp)
121 {
122     PCIBus *bus = PCI_BUS(qbus);
123 
124     bus->machine_done.notify = pcibus_machine_done;
125     qemu_add_machine_init_done_notifier(&bus->machine_done);
126 
127     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
128 }
129 
130 static void pcie_bus_realize(BusState *qbus, Error **errp)
131 {
132     PCIBus *bus = PCI_BUS(qbus);
133 
134     pci_bus_realize(qbus, errp);
135 
136     /*
137      * A PCI-E bus can support extended config space if it's the root
138      * bus, or if the bus/bridge above it does as well
139      */
140     if (pci_bus_is_root(bus)) {
141         bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
142     } else {
143         PCIBus *parent_bus = pci_get_bus(bus->parent_dev);
144 
145         if (pci_bus_allows_extended_config_space(parent_bus)) {
146             bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE;
147         }
148     }
149 }
150 
151 static void pci_bus_unrealize(BusState *qbus, Error **errp)
152 {
153     PCIBus *bus = PCI_BUS(qbus);
154 
155     qemu_remove_machine_init_done_notifier(&bus->machine_done);
156 
157     vmstate_unregister(NULL, &vmstate_pcibus, bus);
158 }
159 
160 static int pcibus_num(PCIBus *bus)
161 {
162     if (pci_bus_is_root(bus)) {
163         return 0; /* pci host bridge */
164     }
165     return bus->parent_dev->config[PCI_SECONDARY_BUS];
166 }
167 
168 static uint16_t pcibus_numa_node(PCIBus *bus)
169 {
170     return NUMA_NODE_UNASSIGNED;
171 }
172 
173 static void pci_bus_class_init(ObjectClass *klass, void *data)
174 {
175     BusClass *k = BUS_CLASS(klass);
176     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
177 
178     k->print_dev = pcibus_dev_print;
179     k->get_dev_path = pcibus_get_dev_path;
180     k->get_fw_dev_path = pcibus_get_fw_dev_path;
181     k->realize = pci_bus_realize;
182     k->unrealize = pci_bus_unrealize;
183     k->reset = pcibus_reset;
184 
185     pbc->bus_num = pcibus_num;
186     pbc->numa_node = pcibus_numa_node;
187 }
188 
189 static const TypeInfo pci_bus_info = {
190     .name = TYPE_PCI_BUS,
191     .parent = TYPE_BUS,
192     .instance_size = sizeof(PCIBus),
193     .class_size = sizeof(PCIBusClass),
194     .class_init = pci_bus_class_init,
195 };
196 
197 static const TypeInfo pcie_interface_info = {
198     .name          = INTERFACE_PCIE_DEVICE,
199     .parent        = TYPE_INTERFACE,
200 };
201 
202 static const TypeInfo conventional_pci_interface_info = {
203     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
204     .parent        = TYPE_INTERFACE,
205 };
206 
207 static void pcie_bus_class_init(ObjectClass *klass, void *data)
208 {
209     BusClass *k = BUS_CLASS(klass);
210 
211     k->realize = pcie_bus_realize;
212 }
213 
214 static const TypeInfo pcie_bus_info = {
215     .name = TYPE_PCIE_BUS,
216     .parent = TYPE_PCI_BUS,
217     .class_init = pcie_bus_class_init,
218 };
219 
220 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
221 static void pci_update_mappings(PCIDevice *d);
222 static void pci_irq_handler(void *opaque, int irq_num, int level);
223 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
224 static void pci_del_option_rom(PCIDevice *pdev);
225 
226 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
227 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
228 
229 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
230 
231 int pci_bar(PCIDevice *d, int reg)
232 {
233     uint8_t type;
234 
235     if (reg != PCI_ROM_SLOT)
236         return PCI_BASE_ADDRESS_0 + reg * 4;
237 
238     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
239     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
240 }
241 
242 static inline int pci_irq_state(PCIDevice *d, int irq_num)
243 {
244         return (d->irq_state >> irq_num) & 0x1;
245 }
246 
247 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
248 {
249         d->irq_state &= ~(0x1 << irq_num);
250         d->irq_state |= level << irq_num;
251 }
252 
253 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
254 {
255     PCIBus *bus;
256     for (;;) {
257         bus = pci_get_bus(pci_dev);
258         irq_num = bus->map_irq(pci_dev, irq_num);
259         if (bus->set_irq)
260             break;
261         pci_dev = bus->parent_dev;
262     }
263     bus->irq_count[irq_num] += change;
264     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
265 }
266 
267 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
268 {
269     assert(irq_num >= 0);
270     assert(irq_num < bus->nirq);
271     return !!bus->irq_count[irq_num];
272 }
273 
274 /* Update interrupt status bit in config space on interrupt
275  * state change. */
276 static void pci_update_irq_status(PCIDevice *dev)
277 {
278     if (dev->irq_state) {
279         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
280     } else {
281         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
282     }
283 }
284 
285 void pci_device_deassert_intx(PCIDevice *dev)
286 {
287     int i;
288     for (i = 0; i < PCI_NUM_PINS; ++i) {
289         pci_irq_handler(dev, i, 0);
290     }
291 }
292 
293 static void pci_do_device_reset(PCIDevice *dev)
294 {
295     int r;
296 
297     pci_device_deassert_intx(dev);
298     assert(dev->irq_state == 0);
299 
300     /* Clear all writable bits */
301     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
302                                  pci_get_word(dev->wmask + PCI_COMMAND) |
303                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
304     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
305                                  pci_get_word(dev->wmask + PCI_STATUS) |
306                                  pci_get_word(dev->w1cmask + PCI_STATUS));
307     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
308     dev->config[PCI_INTERRUPT_LINE] = 0x0;
309     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
310         PCIIORegion *region = &dev->io_regions[r];
311         if (!region->size) {
312             continue;
313         }
314 
315         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
316             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
317             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
318         } else {
319             pci_set_long(dev->config + pci_bar(dev, r), region->type);
320         }
321     }
322     pci_update_mappings(dev);
323 
324     msi_reset(dev);
325     msix_reset(dev);
326 }
327 
328 /*
329  * This function is called on #RST and FLR.
330  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
331  */
332 void pci_device_reset(PCIDevice *dev)
333 {
334     qdev_reset_all(&dev->qdev);
335     pci_do_device_reset(dev);
336 }
337 
338 /*
339  * Trigger pci bus reset under a given bus.
340  * Called via qbus_reset_all on RST# assert, after the devices
341  * have been reset qdev_reset_all-ed already.
342  */
343 static void pcibus_reset(BusState *qbus)
344 {
345     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
346     int i;
347 
348     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
349         if (bus->devices[i]) {
350             pci_do_device_reset(bus->devices[i]);
351         }
352     }
353 
354     for (i = 0; i < bus->nirq; i++) {
355         assert(bus->irq_count[i] == 0);
356     }
357 }
358 
359 static void pci_host_bus_register(DeviceState *host)
360 {
361     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
362 
363     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
364 }
365 
366 static void pci_host_bus_unregister(DeviceState *host)
367 {
368     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
369 
370     QLIST_REMOVE(host_bridge, next);
371 }
372 
373 PCIBus *pci_device_root_bus(const PCIDevice *d)
374 {
375     PCIBus *bus = pci_get_bus(d);
376 
377     while (!pci_bus_is_root(bus)) {
378         d = bus->parent_dev;
379         assert(d != NULL);
380 
381         bus = pci_get_bus(d);
382     }
383 
384     return bus;
385 }
386 
387 const char *pci_root_bus_path(PCIDevice *dev)
388 {
389     PCIBus *rootbus = pci_device_root_bus(dev);
390     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
391     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
392 
393     assert(host_bridge->bus == rootbus);
394 
395     if (hc->root_bus_path) {
396         return (*hc->root_bus_path)(host_bridge, rootbus);
397     }
398 
399     return rootbus->qbus.name;
400 }
401 
402 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
403                               MemoryRegion *address_space_mem,
404                               MemoryRegion *address_space_io,
405                               uint8_t devfn_min)
406 {
407     assert(PCI_FUNC(devfn_min) == 0);
408     bus->devfn_min = devfn_min;
409     bus->slot_reserved_mask = 0x0;
410     bus->address_space_mem = address_space_mem;
411     bus->address_space_io = address_space_io;
412     bus->flags |= PCI_BUS_IS_ROOT;
413 
414     /* host bridge */
415     QLIST_INIT(&bus->child);
416 
417     pci_host_bus_register(parent);
418 }
419 
420 static void pci_bus_uninit(PCIBus *bus)
421 {
422     pci_host_bus_unregister(BUS(bus)->parent);
423 }
424 
425 bool pci_bus_is_express(PCIBus *bus)
426 {
427     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
428 }
429 
430 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
431                               const char *name,
432                               MemoryRegion *address_space_mem,
433                               MemoryRegion *address_space_io,
434                               uint8_t devfn_min, const char *typename)
435 {
436     qbus_create_inplace(bus, bus_size, typename, parent, name);
437     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
438                       devfn_min);
439 }
440 
441 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
442                          MemoryRegion *address_space_mem,
443                          MemoryRegion *address_space_io,
444                          uint8_t devfn_min, const char *typename)
445 {
446     PCIBus *bus;
447 
448     bus = PCI_BUS(qbus_create(typename, parent, name));
449     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
450                       devfn_min);
451     return bus;
452 }
453 
454 void pci_root_bus_cleanup(PCIBus *bus)
455 {
456     pci_bus_uninit(bus);
457     /* the caller of the unplug hotplug handler will delete this device */
458     object_property_set_bool(OBJECT(bus), false, "realized", NULL);
459 }
460 
461 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
462                   void *irq_opaque, int nirq)
463 {
464     bus->set_irq = set_irq;
465     bus->map_irq = map_irq;
466     bus->irq_opaque = irq_opaque;
467     bus->nirq = nirq;
468     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
469 }
470 
471 void pci_bus_irqs_cleanup(PCIBus *bus)
472 {
473     bus->set_irq = NULL;
474     bus->map_irq = NULL;
475     bus->irq_opaque = NULL;
476     bus->nirq = 0;
477     g_free(bus->irq_count);
478 }
479 
480 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
481                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
482                               void *irq_opaque,
483                               MemoryRegion *address_space_mem,
484                               MemoryRegion *address_space_io,
485                               uint8_t devfn_min, int nirq,
486                               const char *typename)
487 {
488     PCIBus *bus;
489 
490     bus = pci_root_bus_new(parent, name, address_space_mem,
491                            address_space_io, devfn_min, typename);
492     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
493     return bus;
494 }
495 
496 void pci_unregister_root_bus(PCIBus *bus)
497 {
498     pci_bus_irqs_cleanup(bus);
499     pci_root_bus_cleanup(bus);
500 }
501 
502 int pci_bus_num(PCIBus *s)
503 {
504     return PCI_BUS_GET_CLASS(s)->bus_num(s);
505 }
506 
507 int pci_bus_numa_node(PCIBus *bus)
508 {
509     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
510 }
511 
512 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
513                                  const VMStateField *field)
514 {
515     PCIDevice *s = container_of(pv, PCIDevice, config);
516     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
517     uint8_t *config;
518     int i;
519 
520     assert(size == pci_config_size(s));
521     config = g_malloc(size);
522 
523     qemu_get_buffer(f, config, size);
524     for (i = 0; i < size; ++i) {
525         if ((config[i] ^ s->config[i]) &
526             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
527             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
528                          "cmask: %x wmask: %x w1cmask:%x", __func__,
529                          i, config[i], s->config[i],
530                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
531             g_free(config);
532             return -EINVAL;
533         }
534     }
535     memcpy(s->config, config, size);
536 
537     pci_update_mappings(s);
538     if (pc->is_bridge) {
539         PCIBridge *b = PCI_BRIDGE(s);
540         pci_bridge_update_mappings(b);
541     }
542 
543     memory_region_set_enabled(&s->bus_master_enable_region,
544                               pci_get_word(s->config + PCI_COMMAND)
545                               & PCI_COMMAND_MASTER);
546 
547     g_free(config);
548     return 0;
549 }
550 
551 /* just put buffer */
552 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
553                                  const VMStateField *field, QJSON *vmdesc)
554 {
555     const uint8_t **v = pv;
556     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
557     qemu_put_buffer(f, *v, size);
558 
559     return 0;
560 }
561 
562 static VMStateInfo vmstate_info_pci_config = {
563     .name = "pci config",
564     .get  = get_pci_config_device,
565     .put  = put_pci_config_device,
566 };
567 
568 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
569                              const VMStateField *field)
570 {
571     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
572     uint32_t irq_state[PCI_NUM_PINS];
573     int i;
574     for (i = 0; i < PCI_NUM_PINS; ++i) {
575         irq_state[i] = qemu_get_be32(f);
576         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
577             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
578                     irq_state[i]);
579             return -EINVAL;
580         }
581     }
582 
583     for (i = 0; i < PCI_NUM_PINS; ++i) {
584         pci_set_irq_state(s, i, irq_state[i]);
585     }
586 
587     return 0;
588 }
589 
590 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
591                              const VMStateField *field, QJSON *vmdesc)
592 {
593     int i;
594     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
595 
596     for (i = 0; i < PCI_NUM_PINS; ++i) {
597         qemu_put_be32(f, pci_irq_state(s, i));
598     }
599 
600     return 0;
601 }
602 
603 static VMStateInfo vmstate_info_pci_irq_state = {
604     .name = "pci irq state",
605     .get  = get_pci_irq_state,
606     .put  = put_pci_irq_state,
607 };
608 
609 static bool migrate_is_pcie(void *opaque, int version_id)
610 {
611     return pci_is_express((PCIDevice *)opaque);
612 }
613 
614 static bool migrate_is_not_pcie(void *opaque, int version_id)
615 {
616     return !pci_is_express((PCIDevice *)opaque);
617 }
618 
619 const VMStateDescription vmstate_pci_device = {
620     .name = "PCIDevice",
621     .version_id = 2,
622     .minimum_version_id = 1,
623     .fields = (VMStateField[]) {
624         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
625         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
626                                    migrate_is_not_pcie,
627                                    0, vmstate_info_pci_config,
628                                    PCI_CONFIG_SPACE_SIZE),
629         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
630                                    migrate_is_pcie,
631                                    0, vmstate_info_pci_config,
632                                    PCIE_CONFIG_SPACE_SIZE),
633         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
634                                    vmstate_info_pci_irq_state,
635                                    PCI_NUM_PINS * sizeof(int32_t)),
636         VMSTATE_END_OF_LIST()
637     }
638 };
639 
640 
641 void pci_device_save(PCIDevice *s, QEMUFile *f)
642 {
643     /* Clear interrupt status bit: it is implicit
644      * in irq_state which we are saving.
645      * This makes us compatible with old devices
646      * which never set or clear this bit. */
647     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
648     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
649     /* Restore the interrupt status bit. */
650     pci_update_irq_status(s);
651 }
652 
653 int pci_device_load(PCIDevice *s, QEMUFile *f)
654 {
655     int ret;
656     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
657     /* Restore the interrupt status bit. */
658     pci_update_irq_status(s);
659     return ret;
660 }
661 
662 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
663 {
664     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
665                  pci_default_sub_vendor_id);
666     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
667                  pci_default_sub_device_id);
668 }
669 
670 /*
671  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
672  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
673  */
674 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
675                              unsigned int *slotp, unsigned int *funcp)
676 {
677     const char *p;
678     char *e;
679     unsigned long val;
680     unsigned long dom = 0, bus = 0;
681     unsigned int slot = 0;
682     unsigned int func = 0;
683 
684     p = addr;
685     val = strtoul(p, &e, 16);
686     if (e == p)
687         return -1;
688     if (*e == ':') {
689         bus = val;
690         p = e + 1;
691         val = strtoul(p, &e, 16);
692         if (e == p)
693             return -1;
694         if (*e == ':') {
695             dom = bus;
696             bus = val;
697             p = e + 1;
698             val = strtoul(p, &e, 16);
699             if (e == p)
700                 return -1;
701         }
702     }
703 
704     slot = val;
705 
706     if (funcp != NULL) {
707         if (*e != '.')
708             return -1;
709 
710         p = e + 1;
711         val = strtoul(p, &e, 16);
712         if (e == p)
713             return -1;
714 
715         func = val;
716     }
717 
718     /* if funcp == NULL func is 0 */
719     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
720         return -1;
721 
722     if (*e)
723         return -1;
724 
725     *domp = dom;
726     *busp = bus;
727     *slotp = slot;
728     if (funcp != NULL)
729         *funcp = func;
730     return 0;
731 }
732 
733 static void pci_init_cmask(PCIDevice *dev)
734 {
735     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
736     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
737     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
738     dev->cmask[PCI_REVISION_ID] = 0xff;
739     dev->cmask[PCI_CLASS_PROG] = 0xff;
740     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
741     dev->cmask[PCI_HEADER_TYPE] = 0xff;
742     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
743 }
744 
745 static void pci_init_wmask(PCIDevice *dev)
746 {
747     int config_size = pci_config_size(dev);
748 
749     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
750     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
751     pci_set_word(dev->wmask + PCI_COMMAND,
752                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
753                  PCI_COMMAND_INTX_DISABLE);
754     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
755         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
756     }
757 
758     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
759            config_size - PCI_CONFIG_HEADER_SIZE);
760 }
761 
762 static void pci_init_w1cmask(PCIDevice *dev)
763 {
764     /*
765      * Note: It's okay to set w1cmask even for readonly bits as
766      * long as their value is hardwired to 0.
767      */
768     pci_set_word(dev->w1cmask + PCI_STATUS,
769                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
770                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
771                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
772 }
773 
774 static void pci_init_mask_bridge(PCIDevice *d)
775 {
776     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
777        PCI_SEC_LETENCY_TIMER */
778     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
779 
780     /* base and limit */
781     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
782     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
783     pci_set_word(d->wmask + PCI_MEMORY_BASE,
784                  PCI_MEMORY_RANGE_MASK & 0xffff);
785     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
786                  PCI_MEMORY_RANGE_MASK & 0xffff);
787     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
788                  PCI_PREF_RANGE_MASK & 0xffff);
789     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
790                  PCI_PREF_RANGE_MASK & 0xffff);
791 
792     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
793     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
794 
795     /* Supported memory and i/o types */
796     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
797     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
798     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
799                                PCI_PREF_RANGE_TYPE_64);
800     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
801                                PCI_PREF_RANGE_TYPE_64);
802 
803     /*
804      * TODO: Bridges default to 10-bit VGA decoding but we currently only
805      * implement 16-bit decoding (no alias support).
806      */
807     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
808                  PCI_BRIDGE_CTL_PARITY |
809                  PCI_BRIDGE_CTL_SERR |
810                  PCI_BRIDGE_CTL_ISA |
811                  PCI_BRIDGE_CTL_VGA |
812                  PCI_BRIDGE_CTL_VGA_16BIT |
813                  PCI_BRIDGE_CTL_MASTER_ABORT |
814                  PCI_BRIDGE_CTL_BUS_RESET |
815                  PCI_BRIDGE_CTL_FAST_BACK |
816                  PCI_BRIDGE_CTL_DISCARD |
817                  PCI_BRIDGE_CTL_SEC_DISCARD |
818                  PCI_BRIDGE_CTL_DISCARD_SERR);
819     /* Below does not do anything as we never set this bit, put here for
820      * completeness. */
821     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
822                  PCI_BRIDGE_CTL_DISCARD_STATUS);
823     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
824     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
825     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
826                                PCI_PREF_RANGE_TYPE_MASK);
827     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
828                                PCI_PREF_RANGE_TYPE_MASK);
829 }
830 
831 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
832 {
833     uint8_t slot = PCI_SLOT(dev->devfn);
834     uint8_t func;
835 
836     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
837         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
838     }
839 
840     /*
841      * multifunction bit is interpreted in two ways as follows.
842      *   - all functions must set the bit to 1.
843      *     Example: Intel X53
844      *   - function 0 must set the bit, but the rest function (> 0)
845      *     is allowed to leave the bit to 0.
846      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
847      *
848      * So OS (at least Linux) checks the bit of only function 0,
849      * and doesn't see the bit of function > 0.
850      *
851      * The below check allows both interpretation.
852      */
853     if (PCI_FUNC(dev->devfn)) {
854         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
855         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
856             /* function 0 should set multifunction bit */
857             error_setg(errp, "PCI: single function device can't be populated "
858                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
859             return;
860         }
861         return;
862     }
863 
864     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
865         return;
866     }
867     /* function 0 indicates single function, so function > 0 must be NULL */
868     for (func = 1; func < PCI_FUNC_MAX; ++func) {
869         if (bus->devices[PCI_DEVFN(slot, func)]) {
870             error_setg(errp, "PCI: %x.0 indicates single function, "
871                        "but %x.%x is already populated.",
872                        slot, slot, func);
873             return;
874         }
875     }
876 }
877 
878 static void pci_config_alloc(PCIDevice *pci_dev)
879 {
880     int config_size = pci_config_size(pci_dev);
881 
882     pci_dev->config = g_malloc0(config_size);
883     pci_dev->cmask = g_malloc0(config_size);
884     pci_dev->wmask = g_malloc0(config_size);
885     pci_dev->w1cmask = g_malloc0(config_size);
886     pci_dev->used = g_malloc0(config_size);
887 }
888 
889 static void pci_config_free(PCIDevice *pci_dev)
890 {
891     g_free(pci_dev->config);
892     g_free(pci_dev->cmask);
893     g_free(pci_dev->wmask);
894     g_free(pci_dev->w1cmask);
895     g_free(pci_dev->used);
896 }
897 
898 static void do_pci_unregister_device(PCIDevice *pci_dev)
899 {
900     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
901     pci_config_free(pci_dev);
902 
903     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
904         memory_region_del_subregion(&pci_dev->bus_master_container_region,
905                                     &pci_dev->bus_master_enable_region);
906     }
907     address_space_destroy(&pci_dev->bus_master_as);
908 }
909 
910 /* Extract PCIReqIDCache into BDF format */
911 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
912 {
913     uint8_t bus_n;
914     uint16_t result;
915 
916     switch (cache->type) {
917     case PCI_REQ_ID_BDF:
918         result = pci_get_bdf(cache->dev);
919         break;
920     case PCI_REQ_ID_SECONDARY_BUS:
921         bus_n = pci_dev_bus_num(cache->dev);
922         result = PCI_BUILD_BDF(bus_n, 0);
923         break;
924     default:
925         error_report("Invalid PCI requester ID cache type: %d",
926                      cache->type);
927         exit(1);
928         break;
929     }
930 
931     return result;
932 }
933 
934 /* Parse bridges up to the root complex and return requester ID
935  * cache for specific device.  For full PCIe topology, the cache
936  * result would be exactly the same as getting BDF of the device.
937  * However, several tricks are required when system mixed up with
938  * legacy PCI devices and PCIe-to-PCI bridges.
939  *
940  * Here we cache the proxy device (and type) not requester ID since
941  * bus number might change from time to time.
942  */
943 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
944 {
945     PCIDevice *parent;
946     PCIReqIDCache cache = {
947         .dev = dev,
948         .type = PCI_REQ_ID_BDF,
949     };
950 
951     while (!pci_bus_is_root(pci_get_bus(dev))) {
952         /* We are under PCI/PCIe bridges */
953         parent = pci_get_bus(dev)->parent_dev;
954         if (pci_is_express(parent)) {
955             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
956                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
957                  * override the requester ID using secondary bus
958                  * number of parent bridge with zeroed devfn
959                  * (pcie-to-pci bridge spec chap 2.3). */
960                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
961                 cache.dev = dev;
962             }
963         } else {
964             /* Legacy PCI, override requester ID with the bridge's
965              * BDF upstream.  When the root complex connects to
966              * legacy PCI devices (including buses), it can only
967              * obtain requester ID info from directly attached
968              * devices.  If devices are attached under bridges, only
969              * the requester ID of the bridge that is directly
970              * attached to the root complex can be recognized. */
971             cache.type = PCI_REQ_ID_BDF;
972             cache.dev = parent;
973         }
974         dev = parent;
975     }
976 
977     return cache;
978 }
979 
980 uint16_t pci_requester_id(PCIDevice *dev)
981 {
982     return pci_req_id_cache_extract(&dev->requester_id_cache);
983 }
984 
985 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
986 {
987     return !(bus->devices[devfn]);
988 }
989 
990 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
991 {
992     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
993 }
994 
995 /* -1 for devfn means auto assign */
996 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
997                                          const char *name, int devfn,
998                                          Error **errp)
999 {
1000     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1001     PCIConfigReadFunc *config_read = pc->config_read;
1002     PCIConfigWriteFunc *config_write = pc->config_write;
1003     Error *local_err = NULL;
1004     DeviceState *dev = DEVICE(pci_dev);
1005     PCIBus *bus = pci_get_bus(pci_dev);
1006 
1007     /* Only pci bridges can be attached to extra PCI root buses */
1008     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
1009         error_setg(errp,
1010                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
1011                     bus->parent_dev->name);
1012         return NULL;
1013     }
1014 
1015     if (devfn < 0) {
1016         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
1017             devfn += PCI_FUNC_MAX) {
1018             if (pci_bus_devfn_available(bus, devfn) &&
1019                    !pci_bus_devfn_reserved(bus, devfn)) {
1020                 goto found;
1021             }
1022         }
1023         error_setg(errp, "PCI: no slot/function available for %s, all in use "
1024                    "or reserved", name);
1025         return NULL;
1026     found: ;
1027     } else if (pci_bus_devfn_reserved(bus, devfn)) {
1028         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1029                    " reserved",
1030                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1031         return NULL;
1032     } else if (!pci_bus_devfn_available(bus, devfn)) {
1033         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1034                    " in use by %s",
1035                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1036                    bus->devices[devfn]->name);
1037         return NULL;
1038     } else if (dev->hotplugged &&
1039                pci_get_function_0(pci_dev)) {
1040         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1041                    " new func %s cannot be exposed to guest.",
1042                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1043                    pci_get_function_0(pci_dev)->name,
1044                    name);
1045 
1046        return NULL;
1047     }
1048 
1049     pci_dev->devfn = devfn;
1050     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1051     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1052 
1053     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1054                        "bus master container", UINT64_MAX);
1055     address_space_init(&pci_dev->bus_master_as,
1056                        &pci_dev->bus_master_container_region, pci_dev->name);
1057 
1058     if (qdev_hotplug) {
1059         pci_init_bus_master(pci_dev);
1060     }
1061     pci_dev->irq_state = 0;
1062     pci_config_alloc(pci_dev);
1063 
1064     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1065     pci_config_set_device_id(pci_dev->config, pc->device_id);
1066     pci_config_set_revision(pci_dev->config, pc->revision);
1067     pci_config_set_class(pci_dev->config, pc->class_id);
1068 
1069     if (!pc->is_bridge) {
1070         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1071             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1072                          pc->subsystem_vendor_id);
1073             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1074                          pc->subsystem_id);
1075         } else {
1076             pci_set_default_subsystem_id(pci_dev);
1077         }
1078     } else {
1079         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1080         assert(!pc->subsystem_vendor_id);
1081         assert(!pc->subsystem_id);
1082     }
1083     pci_init_cmask(pci_dev);
1084     pci_init_wmask(pci_dev);
1085     pci_init_w1cmask(pci_dev);
1086     if (pc->is_bridge) {
1087         pci_init_mask_bridge(pci_dev);
1088     }
1089     pci_init_multifunction(bus, pci_dev, &local_err);
1090     if (local_err) {
1091         error_propagate(errp, local_err);
1092         do_pci_unregister_device(pci_dev);
1093         return NULL;
1094     }
1095 
1096     if (!config_read)
1097         config_read = pci_default_read_config;
1098     if (!config_write)
1099         config_write = pci_default_write_config;
1100     pci_dev->config_read = config_read;
1101     pci_dev->config_write = config_write;
1102     bus->devices[devfn] = pci_dev;
1103     pci_dev->version_id = 2; /* Current pci device vmstate version */
1104     return pci_dev;
1105 }
1106 
1107 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1108 {
1109     PCIIORegion *r;
1110     int i;
1111 
1112     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1113         r = &pci_dev->io_regions[i];
1114         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1115             continue;
1116         memory_region_del_subregion(r->address_space, r->memory);
1117     }
1118 
1119     pci_unregister_vga(pci_dev);
1120 }
1121 
1122 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1123 {
1124     PCIDevice *pci_dev = PCI_DEVICE(dev);
1125     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1126 
1127     pci_unregister_io_regions(pci_dev);
1128     pci_del_option_rom(pci_dev);
1129 
1130     if (pc->exit) {
1131         pc->exit(pci_dev);
1132     }
1133 
1134     pci_device_deassert_intx(pci_dev);
1135     do_pci_unregister_device(pci_dev);
1136 }
1137 
1138 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1139                       uint8_t type, MemoryRegion *memory)
1140 {
1141     PCIIORegion *r;
1142     uint32_t addr; /* offset in pci config space */
1143     uint64_t wmask;
1144     pcibus_t size = memory_region_size(memory);
1145 
1146     assert(region_num >= 0);
1147     assert(region_num < PCI_NUM_REGIONS);
1148     if (size & (size-1)) {
1149         error_report("ERROR: PCI region size must be pow2 "
1150                     "type=0x%x, size=0x%"FMT_PCIBUS"", type, size);
1151         exit(1);
1152     }
1153 
1154     r = &pci_dev->io_regions[region_num];
1155     r->addr = PCI_BAR_UNMAPPED;
1156     r->size = size;
1157     r->type = type;
1158     r->memory = memory;
1159     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1160                         ? pci_get_bus(pci_dev)->address_space_io
1161                         : pci_get_bus(pci_dev)->address_space_mem;
1162 
1163     wmask = ~(size - 1);
1164     if (region_num == PCI_ROM_SLOT) {
1165         /* ROM enable bit is writable */
1166         wmask |= PCI_ROM_ADDRESS_ENABLE;
1167     }
1168 
1169     addr = pci_bar(pci_dev, region_num);
1170     pci_set_long(pci_dev->config + addr, type);
1171 
1172     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1173         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1174         pci_set_quad(pci_dev->wmask + addr, wmask);
1175         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1176     } else {
1177         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1178         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1179     }
1180 }
1181 
1182 static void pci_update_vga(PCIDevice *pci_dev)
1183 {
1184     uint16_t cmd;
1185 
1186     if (!pci_dev->has_vga) {
1187         return;
1188     }
1189 
1190     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1191 
1192     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1193                               cmd & PCI_COMMAND_MEMORY);
1194     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1195                               cmd & PCI_COMMAND_IO);
1196     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1197                               cmd & PCI_COMMAND_IO);
1198 }
1199 
1200 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1201                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1202 {
1203     PCIBus *bus = pci_get_bus(pci_dev);
1204 
1205     assert(!pci_dev->has_vga);
1206 
1207     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1208     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1209     memory_region_add_subregion_overlap(bus->address_space_mem,
1210                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1211 
1212     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1213     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1214     memory_region_add_subregion_overlap(bus->address_space_io,
1215                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1216 
1217     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1218     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1219     memory_region_add_subregion_overlap(bus->address_space_io,
1220                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1221     pci_dev->has_vga = true;
1222 
1223     pci_update_vga(pci_dev);
1224 }
1225 
1226 void pci_unregister_vga(PCIDevice *pci_dev)
1227 {
1228     PCIBus *bus = pci_get_bus(pci_dev);
1229 
1230     if (!pci_dev->has_vga) {
1231         return;
1232     }
1233 
1234     memory_region_del_subregion(bus->address_space_mem,
1235                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1236     memory_region_del_subregion(bus->address_space_io,
1237                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1238     memory_region_del_subregion(bus->address_space_io,
1239                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1240     pci_dev->has_vga = false;
1241 }
1242 
1243 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1244 {
1245     return pci_dev->io_regions[region_num].addr;
1246 }
1247 
1248 static pcibus_t pci_bar_address(PCIDevice *d,
1249                                 int reg, uint8_t type, pcibus_t size)
1250 {
1251     pcibus_t new_addr, last_addr;
1252     int bar = pci_bar(d, reg);
1253     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1254     Object *machine = qdev_get_machine();
1255     ObjectClass *oc = object_get_class(machine);
1256     MachineClass *mc = MACHINE_CLASS(oc);
1257     bool allow_0_address = mc->pci_allow_0_address;
1258 
1259     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1260         if (!(cmd & PCI_COMMAND_IO)) {
1261             return PCI_BAR_UNMAPPED;
1262         }
1263         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1264         last_addr = new_addr + size - 1;
1265         /* Check if 32 bit BAR wraps around explicitly.
1266          * TODO: make priorities correct and remove this work around.
1267          */
1268         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1269             (!allow_0_address && new_addr == 0)) {
1270             return PCI_BAR_UNMAPPED;
1271         }
1272         return new_addr;
1273     }
1274 
1275     if (!(cmd & PCI_COMMAND_MEMORY)) {
1276         return PCI_BAR_UNMAPPED;
1277     }
1278     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1279         new_addr = pci_get_quad(d->config + bar);
1280     } else {
1281         new_addr = pci_get_long(d->config + bar);
1282     }
1283     /* the ROM slot has a specific enable bit */
1284     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1285         return PCI_BAR_UNMAPPED;
1286     }
1287     new_addr &= ~(size - 1);
1288     last_addr = new_addr + size - 1;
1289     /* NOTE: we do not support wrapping */
1290     /* XXX: as we cannot support really dynamic
1291        mappings, we handle specific values as invalid
1292        mappings. */
1293     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1294         (!allow_0_address && new_addr == 0)) {
1295         return PCI_BAR_UNMAPPED;
1296     }
1297 
1298     /* Now pcibus_t is 64bit.
1299      * Check if 32 bit BAR wraps around explicitly.
1300      * Without this, PC ide doesn't work well.
1301      * TODO: remove this work around.
1302      */
1303     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1304         return PCI_BAR_UNMAPPED;
1305     }
1306 
1307     /*
1308      * OS is allowed to set BAR beyond its addressable
1309      * bits. For example, 32 bit OS can set 64bit bar
1310      * to >4G. Check it. TODO: we might need to support
1311      * it in the future for e.g. PAE.
1312      */
1313     if (last_addr >= HWADDR_MAX) {
1314         return PCI_BAR_UNMAPPED;
1315     }
1316 
1317     return new_addr;
1318 }
1319 
1320 static void pci_update_mappings(PCIDevice *d)
1321 {
1322     PCIIORegion *r;
1323     int i;
1324     pcibus_t new_addr;
1325 
1326     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1327         r = &d->io_regions[i];
1328 
1329         /* this region isn't registered */
1330         if (!r->size)
1331             continue;
1332 
1333         new_addr = pci_bar_address(d, i, r->type, r->size);
1334 
1335         /* This bar isn't changed */
1336         if (new_addr == r->addr)
1337             continue;
1338 
1339         /* now do the real mapping */
1340         if (r->addr != PCI_BAR_UNMAPPED) {
1341             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1342                                           PCI_SLOT(d->devfn),
1343                                           PCI_FUNC(d->devfn),
1344                                           i, r->addr, r->size);
1345             memory_region_del_subregion(r->address_space, r->memory);
1346         }
1347         r->addr = new_addr;
1348         if (r->addr != PCI_BAR_UNMAPPED) {
1349             trace_pci_update_mappings_add(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_add_subregion_overlap(r->address_space,
1354                                                 r->addr, r->memory, 1);
1355         }
1356     }
1357 
1358     pci_update_vga(d);
1359 }
1360 
1361 static inline int pci_irq_disabled(PCIDevice *d)
1362 {
1363     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1364 }
1365 
1366 /* Called after interrupt disabled field update in config space,
1367  * assert/deassert interrupts if necessary.
1368  * Gets original interrupt disable bit value (before update). */
1369 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1370 {
1371     int i, disabled = pci_irq_disabled(d);
1372     if (disabled == was_irq_disabled)
1373         return;
1374     for (i = 0; i < PCI_NUM_PINS; ++i) {
1375         int state = pci_irq_state(d, i);
1376         pci_change_irq_level(d, i, disabled ? -state : state);
1377     }
1378 }
1379 
1380 uint32_t pci_default_read_config(PCIDevice *d,
1381                                  uint32_t address, int len)
1382 {
1383     uint32_t val = 0;
1384 
1385     if (pci_is_express_downstream_port(d) &&
1386         ranges_overlap(address, len, d->exp.exp_cap + PCI_EXP_LNKSTA, 2)) {
1387         pcie_sync_bridge_lnk(d);
1388     }
1389     memcpy(&val, d->config + address, len);
1390     return le32_to_cpu(val);
1391 }
1392 
1393 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1394 {
1395     int i, was_irq_disabled = pci_irq_disabled(d);
1396     uint32_t val = val_in;
1397 
1398     for (i = 0; i < l; val >>= 8, ++i) {
1399         uint8_t wmask = d->wmask[addr + i];
1400         uint8_t w1cmask = d->w1cmask[addr + i];
1401         assert(!(wmask & w1cmask));
1402         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1403         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1404     }
1405     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1406         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1407         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1408         range_covers_byte(addr, l, PCI_COMMAND))
1409         pci_update_mappings(d);
1410 
1411     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1412         pci_update_irq_disabled(d, was_irq_disabled);
1413         memory_region_set_enabled(&d->bus_master_enable_region,
1414                                   pci_get_word(d->config + PCI_COMMAND)
1415                                     & PCI_COMMAND_MASTER);
1416     }
1417 
1418     msi_write_config(d, addr, val_in, l);
1419     msix_write_config(d, addr, val_in, l);
1420 }
1421 
1422 /***********************************************************/
1423 /* generic PCI irq support */
1424 
1425 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1426 static void pci_irq_handler(void *opaque, int irq_num, int level)
1427 {
1428     PCIDevice *pci_dev = opaque;
1429     int change;
1430 
1431     change = level - pci_irq_state(pci_dev, irq_num);
1432     if (!change)
1433         return;
1434 
1435     pci_set_irq_state(pci_dev, irq_num, level);
1436     pci_update_irq_status(pci_dev);
1437     if (pci_irq_disabled(pci_dev))
1438         return;
1439     pci_change_irq_level(pci_dev, irq_num, change);
1440 }
1441 
1442 static inline int pci_intx(PCIDevice *pci_dev)
1443 {
1444     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1445 }
1446 
1447 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1448 {
1449     int intx = pci_intx(pci_dev);
1450 
1451     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1452 }
1453 
1454 void pci_set_irq(PCIDevice *pci_dev, int level)
1455 {
1456     int intx = pci_intx(pci_dev);
1457     pci_irq_handler(pci_dev, intx, level);
1458 }
1459 
1460 /* Special hooks used by device assignment */
1461 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1462 {
1463     assert(pci_bus_is_root(bus));
1464     bus->route_intx_to_irq = route_intx_to_irq;
1465 }
1466 
1467 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1468 {
1469     PCIBus *bus;
1470 
1471     do {
1472         bus = pci_get_bus(dev);
1473         pin = bus->map_irq(dev, pin);
1474         dev = bus->parent_dev;
1475     } while (dev);
1476 
1477     if (!bus->route_intx_to_irq) {
1478         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1479                      object_get_typename(OBJECT(bus->qbus.parent)));
1480         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1481     }
1482 
1483     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1484 }
1485 
1486 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1487 {
1488     return old->mode != new->mode || old->irq != new->irq;
1489 }
1490 
1491 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1492 {
1493     PCIDevice *dev;
1494     PCIBus *sec;
1495     int i;
1496 
1497     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1498         dev = bus->devices[i];
1499         if (dev && dev->intx_routing_notifier) {
1500             dev->intx_routing_notifier(dev);
1501         }
1502     }
1503 
1504     QLIST_FOREACH(sec, &bus->child, sibling) {
1505         pci_bus_fire_intx_routing_notifier(sec);
1506     }
1507 }
1508 
1509 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1510                                           PCIINTxRoutingNotifier notifier)
1511 {
1512     dev->intx_routing_notifier = notifier;
1513 }
1514 
1515 /*
1516  * PCI-to-PCI bridge specification
1517  * 9.1: Interrupt routing. Table 9-1
1518  *
1519  * the PCI Express Base Specification, Revision 2.1
1520  * 2.2.8.1: INTx interrutp signaling - Rules
1521  *          the Implementation Note
1522  *          Table 2-20
1523  */
1524 /*
1525  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1526  * 0-origin unlike PCI interrupt pin register.
1527  */
1528 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1529 {
1530     return pci_swizzle(PCI_SLOT(pci_dev->devfn), pin);
1531 }
1532 
1533 /***********************************************************/
1534 /* monitor info on PCI */
1535 
1536 typedef struct {
1537     uint16_t class;
1538     const char *desc;
1539     const char *fw_name;
1540     uint16_t fw_ign_bits;
1541 } pci_class_desc;
1542 
1543 static const pci_class_desc pci_class_descriptions[] =
1544 {
1545     { 0x0001, "VGA controller", "display"},
1546     { 0x0100, "SCSI controller", "scsi"},
1547     { 0x0101, "IDE controller", "ide"},
1548     { 0x0102, "Floppy controller", "fdc"},
1549     { 0x0103, "IPI controller", "ipi"},
1550     { 0x0104, "RAID controller", "raid"},
1551     { 0x0106, "SATA controller"},
1552     { 0x0107, "SAS controller"},
1553     { 0x0180, "Storage controller"},
1554     { 0x0200, "Ethernet controller", "ethernet"},
1555     { 0x0201, "Token Ring controller", "token-ring"},
1556     { 0x0202, "FDDI controller", "fddi"},
1557     { 0x0203, "ATM controller", "atm"},
1558     { 0x0280, "Network controller"},
1559     { 0x0300, "VGA controller", "display", 0x00ff},
1560     { 0x0301, "XGA controller"},
1561     { 0x0302, "3D controller"},
1562     { 0x0380, "Display controller"},
1563     { 0x0400, "Video controller", "video"},
1564     { 0x0401, "Audio controller", "sound"},
1565     { 0x0402, "Phone"},
1566     { 0x0403, "Audio controller", "sound"},
1567     { 0x0480, "Multimedia controller"},
1568     { 0x0500, "RAM controller", "memory"},
1569     { 0x0501, "Flash controller", "flash"},
1570     { 0x0580, "Memory controller"},
1571     { 0x0600, "Host bridge", "host"},
1572     { 0x0601, "ISA bridge", "isa"},
1573     { 0x0602, "EISA bridge", "eisa"},
1574     { 0x0603, "MC bridge", "mca"},
1575     { 0x0604, "PCI bridge", "pci-bridge"},
1576     { 0x0605, "PCMCIA bridge", "pcmcia"},
1577     { 0x0606, "NUBUS bridge", "nubus"},
1578     { 0x0607, "CARDBUS bridge", "cardbus"},
1579     { 0x0608, "RACEWAY bridge"},
1580     { 0x0680, "Bridge"},
1581     { 0x0700, "Serial port", "serial"},
1582     { 0x0701, "Parallel port", "parallel"},
1583     { 0x0800, "Interrupt controller", "interrupt-controller"},
1584     { 0x0801, "DMA controller", "dma-controller"},
1585     { 0x0802, "Timer", "timer"},
1586     { 0x0803, "RTC", "rtc"},
1587     { 0x0900, "Keyboard", "keyboard"},
1588     { 0x0901, "Pen", "pen"},
1589     { 0x0902, "Mouse", "mouse"},
1590     { 0x0A00, "Dock station", "dock", 0x00ff},
1591     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1592     { 0x0c00, "Fireware contorller", "fireware"},
1593     { 0x0c01, "Access bus controller", "access-bus"},
1594     { 0x0c02, "SSA controller", "ssa"},
1595     { 0x0c03, "USB controller", "usb"},
1596     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1597     { 0x0c05, "SMBus"},
1598     { 0, NULL}
1599 };
1600 
1601 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1602                                                   void (*fn)(PCIBus *b,
1603                                                              PCIDevice *d,
1604                                                              void *opaque),
1605                                                   void *opaque)
1606 {
1607     PCIDevice *d;
1608     int devfn;
1609 
1610     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1611         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1612         if (d) {
1613             fn(bus, d, opaque);
1614         }
1615     }
1616 }
1617 
1618 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1619                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1620                          void *opaque)
1621 {
1622     bus = pci_find_bus_nr(bus, bus_num);
1623 
1624     if (bus) {
1625         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1626     }
1627 }
1628 
1629 static void pci_for_each_device_under_bus(PCIBus *bus,
1630                                           void (*fn)(PCIBus *b, PCIDevice *d,
1631                                                      void *opaque),
1632                                           void *opaque)
1633 {
1634     PCIDevice *d;
1635     int devfn;
1636 
1637     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1638         d = bus->devices[devfn];
1639         if (d) {
1640             fn(bus, d, opaque);
1641         }
1642     }
1643 }
1644 
1645 void pci_for_each_device(PCIBus *bus, int bus_num,
1646                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1647                          void *opaque)
1648 {
1649     bus = pci_find_bus_nr(bus, bus_num);
1650 
1651     if (bus) {
1652         pci_for_each_device_under_bus(bus, fn, opaque);
1653     }
1654 }
1655 
1656 static const pci_class_desc *get_class_desc(int class)
1657 {
1658     const pci_class_desc *desc;
1659 
1660     desc = pci_class_descriptions;
1661     while (desc->desc && class != desc->class) {
1662         desc++;
1663     }
1664 
1665     return desc;
1666 }
1667 
1668 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1669 
1670 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1671 {
1672     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1673     int i;
1674 
1675     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1676         const PCIIORegion *r = &dev->io_regions[i];
1677         PciMemoryRegionList *region;
1678 
1679         if (!r->size) {
1680             continue;
1681         }
1682 
1683         region = g_malloc0(sizeof(*region));
1684         region->value = g_malloc0(sizeof(*region->value));
1685 
1686         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1687             region->value->type = g_strdup("io");
1688         } else {
1689             region->value->type = g_strdup("memory");
1690             region->value->has_prefetch = true;
1691             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1692             region->value->has_mem_type_64 = true;
1693             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1694         }
1695 
1696         region->value->bar = i;
1697         region->value->address = r->addr;
1698         region->value->size = r->size;
1699 
1700         /* XXX: waiting for the qapi to support GSList */
1701         if (!cur_item) {
1702             head = cur_item = region;
1703         } else {
1704             cur_item->next = region;
1705             cur_item = region;
1706         }
1707     }
1708 
1709     return head;
1710 }
1711 
1712 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1713                                            int bus_num)
1714 {
1715     PciBridgeInfo *info;
1716     PciMemoryRange *range;
1717 
1718     info = g_new0(PciBridgeInfo, 1);
1719 
1720     info->bus = g_new0(PciBusInfo, 1);
1721     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1722     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1723     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1724 
1725     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1726     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1727     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1728 
1729     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1730     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1731     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1732 
1733     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1734     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1735     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1736 
1737     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1738         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1739         if (child_bus) {
1740             info->has_devices = true;
1741             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1742         }
1743     }
1744 
1745     return info;
1746 }
1747 
1748 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1749                                            int bus_num)
1750 {
1751     const pci_class_desc *desc;
1752     PciDeviceInfo *info;
1753     uint8_t type;
1754     int class;
1755 
1756     info = g_new0(PciDeviceInfo, 1);
1757     info->bus = bus_num;
1758     info->slot = PCI_SLOT(dev->devfn);
1759     info->function = PCI_FUNC(dev->devfn);
1760 
1761     info->class_info = g_new0(PciDeviceClass, 1);
1762     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1763     info->class_info->q_class = class;
1764     desc = get_class_desc(class);
1765     if (desc->desc) {
1766         info->class_info->has_desc = true;
1767         info->class_info->desc = g_strdup(desc->desc);
1768     }
1769 
1770     info->id = g_new0(PciDeviceId, 1);
1771     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1772     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1773     info->regions = qmp_query_pci_regions(dev);
1774     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1775 
1776     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1777         info->has_irq = true;
1778         info->irq = dev->config[PCI_INTERRUPT_LINE];
1779     }
1780 
1781     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1782     if (type == PCI_HEADER_TYPE_BRIDGE) {
1783         info->has_pci_bridge = true;
1784         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1785     } else if (type == PCI_HEADER_TYPE_NORMAL) {
1786         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1787         info->id->subsystem = pci_get_word(dev->config + PCI_SUBSYSTEM_ID);
1788         info->id->subsystem_vendor =
1789             pci_get_word(dev->config + PCI_SUBSYSTEM_VENDOR_ID);
1790     } else if (type == PCI_HEADER_TYPE_CARDBUS) {
1791         info->id->has_subsystem = info->id->has_subsystem_vendor = true;
1792         info->id->subsystem = pci_get_word(dev->config + PCI_CB_SUBSYSTEM_ID);
1793         info->id->subsystem_vendor =
1794             pci_get_word(dev->config + PCI_CB_SUBSYSTEM_VENDOR_ID);
1795     }
1796 
1797     return info;
1798 }
1799 
1800 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1801 {
1802     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1803     PCIDevice *dev;
1804     int devfn;
1805 
1806     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1807         dev = bus->devices[devfn];
1808         if (dev) {
1809             info = g_malloc0(sizeof(*info));
1810             info->value = qmp_query_pci_device(dev, bus, bus_num);
1811 
1812             /* XXX: waiting for the qapi to support GSList */
1813             if (!cur_item) {
1814                 head = cur_item = info;
1815             } else {
1816                 cur_item->next = info;
1817                 cur_item = info;
1818             }
1819         }
1820     }
1821 
1822     return head;
1823 }
1824 
1825 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1826 {
1827     PciInfo *info = NULL;
1828 
1829     bus = pci_find_bus_nr(bus, bus_num);
1830     if (bus) {
1831         info = g_malloc0(sizeof(*info));
1832         info->bus = bus_num;
1833         info->devices = qmp_query_pci_devices(bus, bus_num);
1834     }
1835 
1836     return info;
1837 }
1838 
1839 PciInfoList *qmp_query_pci(Error **errp)
1840 {
1841     PciInfoList *info, *head = NULL, *cur_item = NULL;
1842     PCIHostState *host_bridge;
1843 
1844     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1845         info = g_malloc0(sizeof(*info));
1846         info->value = qmp_query_pci_bus(host_bridge->bus,
1847                                         pci_bus_num(host_bridge->bus));
1848 
1849         /* XXX: waiting for the qapi to support GSList */
1850         if (!cur_item) {
1851             head = cur_item = info;
1852         } else {
1853             cur_item->next = info;
1854             cur_item = info;
1855         }
1856     }
1857 
1858     return head;
1859 }
1860 
1861 /* Initialize a PCI NIC.  */
1862 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1863                                const char *default_model,
1864                                const char *default_devaddr)
1865 {
1866     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1867     GSList *list;
1868     GPtrArray *pci_nic_models;
1869     PCIBus *bus;
1870     PCIDevice *pci_dev;
1871     DeviceState *dev;
1872     int devfn;
1873     int i;
1874     int dom, busnr;
1875     unsigned slot;
1876 
1877     if (nd->model && !strcmp(nd->model, "virtio")) {
1878         g_free(nd->model);
1879         nd->model = g_strdup("virtio-net-pci");
1880     }
1881 
1882     list = object_class_get_list_sorted(TYPE_PCI_DEVICE, false);
1883     pci_nic_models = g_ptr_array_new();
1884     while (list) {
1885         DeviceClass *dc = OBJECT_CLASS_CHECK(DeviceClass, list->data,
1886                                              TYPE_DEVICE);
1887         GSList *next;
1888         if (test_bit(DEVICE_CATEGORY_NETWORK, dc->categories) &&
1889             dc->user_creatable) {
1890             const char *name = object_class_get_name(list->data);
1891             g_ptr_array_add(pci_nic_models, (gpointer)name);
1892         }
1893         next = list->next;
1894         g_slist_free_1(list);
1895         list = next;
1896     }
1897     g_ptr_array_add(pci_nic_models, NULL);
1898 
1899     if (qemu_show_nic_models(nd->model, (const char **)pci_nic_models->pdata)) {
1900         exit(0);
1901     }
1902 
1903     i = qemu_find_nic_model(nd, (const char **)pci_nic_models->pdata,
1904                             default_model);
1905     if (i < 0) {
1906         exit(1);
1907     }
1908 
1909     if (!rootbus) {
1910         error_report("No primary PCI bus");
1911         exit(1);
1912     }
1913 
1914     assert(!rootbus->parent_dev);
1915 
1916     if (!devaddr) {
1917         devfn = -1;
1918         busnr = 0;
1919     } else {
1920         if (pci_parse_devaddr(devaddr, &dom, &busnr, &slot, NULL) < 0) {
1921             error_report("Invalid PCI device address %s for device %s",
1922                          devaddr, nd->model);
1923             exit(1);
1924         }
1925 
1926         if (dom != 0) {
1927             error_report("No support for non-zero PCI domains");
1928             exit(1);
1929         }
1930 
1931         devfn = PCI_DEVFN(slot, 0);
1932     }
1933 
1934     bus = pci_find_bus_nr(rootbus, busnr);
1935     if (!bus) {
1936         error_report("Invalid PCI device address %s for device %s",
1937                      devaddr, nd->model);
1938         exit(1);
1939     }
1940 
1941     pci_dev = pci_create(bus, devfn, nd->model);
1942     dev = &pci_dev->qdev;
1943     qdev_set_nic_properties(dev, nd);
1944     qdev_init_nofail(dev);
1945     g_ptr_array_free(pci_nic_models, true);
1946     return pci_dev;
1947 }
1948 
1949 PCIDevice *pci_vga_init(PCIBus *bus)
1950 {
1951     switch (vga_interface_type) {
1952     case VGA_CIRRUS:
1953         return pci_create_simple(bus, -1, "cirrus-vga");
1954     case VGA_QXL:
1955         return pci_create_simple(bus, -1, "qxl-vga");
1956     case VGA_STD:
1957         return pci_create_simple(bus, -1, "VGA");
1958     case VGA_VMWARE:
1959         return pci_create_simple(bus, -1, "vmware-svga");
1960     case VGA_VIRTIO:
1961         return pci_create_simple(bus, -1, "virtio-vga");
1962     case VGA_NONE:
1963     default: /* Other non-PCI types. Checking for unsupported types is already
1964                 done in vl.c. */
1965         return NULL;
1966     }
1967 }
1968 
1969 /* Whether a given bus number is in range of the secondary
1970  * bus of the given bridge device. */
1971 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1972 {
1973     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1974              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1975         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1976         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1977 }
1978 
1979 /* Whether a given bus number is in a range of a root bus */
1980 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1981 {
1982     int i;
1983 
1984     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1985         PCIDevice *dev = bus->devices[i];
1986 
1987         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1988             if (pci_secondary_bus_in_range(dev, bus_num)) {
1989                 return true;
1990             }
1991         }
1992     }
1993 
1994     return false;
1995 }
1996 
1997 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1998 {
1999     PCIBus *sec;
2000 
2001     if (!bus) {
2002         return NULL;
2003     }
2004 
2005     if (pci_bus_num(bus) == bus_num) {
2006         return bus;
2007     }
2008 
2009     /* Consider all bus numbers in range for the host pci bridge. */
2010     if (!pci_bus_is_root(bus) &&
2011         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
2012         return NULL;
2013     }
2014 
2015     /* try child bus */
2016     for (; bus; bus = sec) {
2017         QLIST_FOREACH(sec, &bus->child, sibling) {
2018             if (pci_bus_num(sec) == bus_num) {
2019                 return sec;
2020             }
2021             /* PXB buses assumed to be children of bus 0 */
2022             if (pci_bus_is_root(sec)) {
2023                 if (pci_root_bus_in_range(sec, bus_num)) {
2024                     break;
2025                 }
2026             } else {
2027                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
2028                     break;
2029                 }
2030             }
2031         }
2032     }
2033 
2034     return NULL;
2035 }
2036 
2037 void pci_for_each_bus_depth_first(PCIBus *bus,
2038                                   void *(*begin)(PCIBus *bus, void *parent_state),
2039                                   void (*end)(PCIBus *bus, void *state),
2040                                   void *parent_state)
2041 {
2042     PCIBus *sec;
2043     void *state;
2044 
2045     if (!bus) {
2046         return;
2047     }
2048 
2049     if (begin) {
2050         state = begin(bus, parent_state);
2051     } else {
2052         state = parent_state;
2053     }
2054 
2055     QLIST_FOREACH(sec, &bus->child, sibling) {
2056         pci_for_each_bus_depth_first(sec, begin, end, state);
2057     }
2058 
2059     if (end) {
2060         end(bus, state);
2061     }
2062 }
2063 
2064 
2065 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
2066 {
2067     bus = pci_find_bus_nr(bus, bus_num);
2068 
2069     if (!bus)
2070         return NULL;
2071 
2072     return bus->devices[devfn];
2073 }
2074 
2075 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2076 {
2077     PCIDevice *pci_dev = (PCIDevice *)qdev;
2078     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2079     ObjectClass *klass = OBJECT_CLASS(pc);
2080     Error *local_err = NULL;
2081     bool is_default_rom;
2082     uint16_t class_id;
2083 
2084     /* initialize cap_present for pci_is_express() and pci_config_size(),
2085      * Note that hybrid PCIs are not set automatically and need to manage
2086      * QEMU_PCI_CAP_EXPRESS manually */
2087     if (object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE) &&
2088        !object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE)) {
2089         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2090     }
2091 
2092     pci_dev = do_pci_register_device(pci_dev,
2093                                      object_get_typename(OBJECT(qdev)),
2094                                      pci_dev->devfn, errp);
2095     if (pci_dev == NULL)
2096         return;
2097 
2098     if (pc->realize) {
2099         pc->realize(pci_dev, &local_err);
2100         if (local_err) {
2101             error_propagate(errp, local_err);
2102             do_pci_unregister_device(pci_dev);
2103             return;
2104         }
2105     }
2106 
2107     if (pci_dev->failover_pair_id) {
2108         if (!pci_bus_is_express(pci_get_bus(pci_dev))) {
2109             error_setg(errp, "failover primary device must be on "
2110                              "PCIExpress bus");
2111             error_propagate(errp, local_err);
2112             pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2113             return;
2114         }
2115         class_id = pci_get_word(pci_dev->config + PCI_CLASS_DEVICE);
2116         if (class_id != PCI_CLASS_NETWORK_ETHERNET) {
2117             error_setg(errp, "failover primary device is not an "
2118                              "Ethernet device");
2119             error_propagate(errp, local_err);
2120             pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2121             return;
2122         }
2123         if (!(pci_dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)
2124             && (PCI_FUNC(pci_dev->devfn) == 0)) {
2125             qdev->allow_unplug_during_migration = true;
2126         } else {
2127             error_setg(errp, "failover: primary device must be in its own "
2128                               "PCI slot");
2129             error_propagate(errp, local_err);
2130             pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2131             return;
2132         }
2133         qdev->allow_unplug_during_migration = true;
2134     }
2135 
2136     /* rom loading */
2137     is_default_rom = false;
2138     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2139         pci_dev->romfile = g_strdup(pc->romfile);
2140         is_default_rom = true;
2141     }
2142 
2143     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2144     if (local_err) {
2145         error_propagate(errp, local_err);
2146         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2147         return;
2148     }
2149 }
2150 
2151 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2152                                     const char *name)
2153 {
2154     DeviceState *dev;
2155 
2156     dev = qdev_create(&bus->qbus, name);
2157     qdev_prop_set_int32(dev, "addr", devfn);
2158     qdev_prop_set_bit(dev, "multifunction", multifunction);
2159     return PCI_DEVICE(dev);
2160 }
2161 
2162 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2163                                            bool multifunction,
2164                                            const char *name)
2165 {
2166     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2167     qdev_init_nofail(&dev->qdev);
2168     return dev;
2169 }
2170 
2171 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2172 {
2173     return pci_create_multifunction(bus, devfn, false, name);
2174 }
2175 
2176 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2177 {
2178     return pci_create_simple_multifunction(bus, devfn, false, name);
2179 }
2180 
2181 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2182 {
2183     int offset = PCI_CONFIG_HEADER_SIZE;
2184     int i;
2185     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2186         if (pdev->used[i])
2187             offset = i + 1;
2188         else if (i - offset + 1 == size)
2189             return offset;
2190     }
2191     return 0;
2192 }
2193 
2194 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2195                                         uint8_t *prev_p)
2196 {
2197     uint8_t next, prev;
2198 
2199     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2200         return 0;
2201 
2202     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2203          prev = next + PCI_CAP_LIST_NEXT)
2204         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2205             break;
2206 
2207     if (prev_p)
2208         *prev_p = prev;
2209     return next;
2210 }
2211 
2212 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2213 {
2214     uint8_t next, prev, found = 0;
2215 
2216     if (!(pdev->used[offset])) {
2217         return 0;
2218     }
2219 
2220     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2221 
2222     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2223          prev = next + PCI_CAP_LIST_NEXT) {
2224         if (next <= offset && next > found) {
2225             found = next;
2226         }
2227     }
2228     return found;
2229 }
2230 
2231 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2232    This is needed for an option rom which is used for more than one device. */
2233 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2234 {
2235     uint16_t vendor_id;
2236     uint16_t device_id;
2237     uint16_t rom_vendor_id;
2238     uint16_t rom_device_id;
2239     uint16_t rom_magic;
2240     uint16_t pcir_offset;
2241     uint8_t checksum;
2242 
2243     /* Words in rom data are little endian (like in PCI configuration),
2244        so they can be read / written with pci_get_word / pci_set_word. */
2245 
2246     /* Only a valid rom will be patched. */
2247     rom_magic = pci_get_word(ptr);
2248     if (rom_magic != 0xaa55) {
2249         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2250         return;
2251     }
2252     pcir_offset = pci_get_word(ptr + 0x18);
2253     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2254         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2255         return;
2256     }
2257 
2258     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2259     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2260     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2261     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2262 
2263     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2264                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2265 
2266     checksum = ptr[6];
2267 
2268     if (vendor_id != rom_vendor_id) {
2269         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2270         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2271         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2272         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2273         ptr[6] = checksum;
2274         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2275     }
2276 
2277     if (device_id != rom_device_id) {
2278         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2279         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2280         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2281         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2282         ptr[6] = checksum;
2283         pci_set_word(ptr + pcir_offset + 6, device_id);
2284     }
2285 }
2286 
2287 /* Add an option rom for the device */
2288 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2289                                Error **errp)
2290 {
2291     int size;
2292     char *path;
2293     void *ptr;
2294     char name[32];
2295     const VMStateDescription *vmsd;
2296 
2297     if (!pdev->romfile)
2298         return;
2299     if (strlen(pdev->romfile) == 0)
2300         return;
2301 
2302     if (!pdev->rom_bar) {
2303         /*
2304          * Load rom via fw_cfg instead of creating a rom bar,
2305          * for 0.11 compatibility.
2306          */
2307         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2308 
2309         /*
2310          * Hot-plugged devices can't use the option ROM
2311          * if the rom bar is disabled.
2312          */
2313         if (DEVICE(pdev)->hotplugged) {
2314             error_setg(errp, "Hot-plugged device without ROM bar"
2315                        " can't have an option ROM");
2316             return;
2317         }
2318 
2319         if (class == 0x0300) {
2320             rom_add_vga(pdev->romfile);
2321         } else {
2322             rom_add_option(pdev->romfile, -1);
2323         }
2324         return;
2325     }
2326 
2327     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2328     if (path == NULL) {
2329         path = g_strdup(pdev->romfile);
2330     }
2331 
2332     size = get_image_size(path);
2333     if (size < 0) {
2334         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2335         g_free(path);
2336         return;
2337     } else if (size == 0) {
2338         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2339         g_free(path);
2340         return;
2341     }
2342     size = pow2ceil(size);
2343 
2344     vmsd = qdev_get_vmsd(DEVICE(pdev));
2345 
2346     if (vmsd) {
2347         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2348     } else {
2349         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2350     }
2351     pdev->has_rom = true;
2352     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2353     ptr = memory_region_get_ram_ptr(&pdev->rom);
2354     if (load_image_size(path, ptr, size) < 0) {
2355         error_setg(errp, "failed to load romfile \"%s\"", pdev->romfile);
2356         g_free(path);
2357         return;
2358     }
2359     g_free(path);
2360 
2361     if (is_default_rom) {
2362         /* Only the default rom images will be patched (if needed). */
2363         pci_patch_ids(pdev, ptr, size);
2364     }
2365 
2366     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2367 }
2368 
2369 static void pci_del_option_rom(PCIDevice *pdev)
2370 {
2371     if (!pdev->has_rom)
2372         return;
2373 
2374     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2375     pdev->has_rom = false;
2376 }
2377 
2378 /*
2379  * On success, pci_add_capability() returns a positive value
2380  * that the offset of the pci capability.
2381  * On failure, it sets an error and returns a negative error
2382  * code.
2383  */
2384 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2385                        uint8_t offset, uint8_t size,
2386                        Error **errp)
2387 {
2388     uint8_t *config;
2389     int i, overlapping_cap;
2390 
2391     if (!offset) {
2392         offset = pci_find_space(pdev, size);
2393         /* out of PCI config space is programming error */
2394         assert(offset);
2395     } else {
2396         /* Verify that capabilities don't overlap.  Note: device assignment
2397          * depends on this check to verify that the device is not broken.
2398          * Should never trigger for emulated devices, but it's helpful
2399          * for debugging these. */
2400         for (i = offset; i < offset + size; i++) {
2401             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2402             if (overlapping_cap) {
2403                 error_setg(errp, "%s:%02x:%02x.%x "
2404                            "Attempt to add PCI capability %x at offset "
2405                            "%x overlaps existing capability %x at offset %x",
2406                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2407                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2408                            cap_id, offset, overlapping_cap, i);
2409                 return -EINVAL;
2410             }
2411         }
2412     }
2413 
2414     config = pdev->config + offset;
2415     config[PCI_CAP_LIST_ID] = cap_id;
2416     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2417     pdev->config[PCI_CAPABILITY_LIST] = offset;
2418     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2419     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2420     /* Make capability read-only by default */
2421     memset(pdev->wmask + offset, 0, size);
2422     /* Check capability by default */
2423     memset(pdev->cmask + offset, 0xFF, size);
2424     return offset;
2425 }
2426 
2427 /* Unlink capability from the pci config space. */
2428 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2429 {
2430     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2431     if (!offset)
2432         return;
2433     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2434     /* Make capability writable again */
2435     memset(pdev->wmask + offset, 0xff, size);
2436     memset(pdev->w1cmask + offset, 0, size);
2437     /* Clear cmask as device-specific registers can't be checked */
2438     memset(pdev->cmask + offset, 0, size);
2439     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2440 
2441     if (!pdev->config[PCI_CAPABILITY_LIST])
2442         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2443 }
2444 
2445 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2446 {
2447     return pci_find_capability_list(pdev, cap_id, NULL);
2448 }
2449 
2450 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2451 {
2452     PCIDevice *d = (PCIDevice *)dev;
2453     const pci_class_desc *desc;
2454     char ctxt[64];
2455     PCIIORegion *r;
2456     int i, class;
2457 
2458     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2459     desc = pci_class_descriptions;
2460     while (desc->desc && class != desc->class)
2461         desc++;
2462     if (desc->desc) {
2463         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2464     } else {
2465         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2466     }
2467 
2468     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2469                    "pci id %04x:%04x (sub %04x:%04x)\n",
2470                    indent, "", ctxt, pci_dev_bus_num(d),
2471                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2472                    pci_get_word(d->config + PCI_VENDOR_ID),
2473                    pci_get_word(d->config + PCI_DEVICE_ID),
2474                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2475                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2476     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2477         r = &d->io_regions[i];
2478         if (!r->size)
2479             continue;
2480         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2481                        " [0x%"FMT_PCIBUS"]\n",
2482                        indent, "",
2483                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2484                        r->addr, r->addr + r->size - 1);
2485     }
2486 }
2487 
2488 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2489 {
2490     PCIDevice *d = (PCIDevice *)dev;
2491     const char *name = NULL;
2492     const pci_class_desc *desc =  pci_class_descriptions;
2493     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2494 
2495     while (desc->desc &&
2496           (class & ~desc->fw_ign_bits) !=
2497           (desc->class & ~desc->fw_ign_bits)) {
2498         desc++;
2499     }
2500 
2501     if (desc->desc) {
2502         name = desc->fw_name;
2503     }
2504 
2505     if (name) {
2506         pstrcpy(buf, len, name);
2507     } else {
2508         snprintf(buf, len, "pci%04x,%04x",
2509                  pci_get_word(d->config + PCI_VENDOR_ID),
2510                  pci_get_word(d->config + PCI_DEVICE_ID));
2511     }
2512 
2513     return buf;
2514 }
2515 
2516 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2517 {
2518     PCIDevice *d = (PCIDevice *)dev;
2519     char path[50], name[33];
2520     int off;
2521 
2522     off = snprintf(path, sizeof(path), "%s@%x",
2523                    pci_dev_fw_name(dev, name, sizeof name),
2524                    PCI_SLOT(d->devfn));
2525     if (PCI_FUNC(d->devfn))
2526         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2527     return g_strdup(path);
2528 }
2529 
2530 static char *pcibus_get_dev_path(DeviceState *dev)
2531 {
2532     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2533     PCIDevice *t;
2534     int slot_depth;
2535     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2536      * 00 is added here to make this format compatible with
2537      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2538      * Slot.Function list specifies the slot and function numbers for all
2539      * devices on the path from root to the specific device. */
2540     const char *root_bus_path;
2541     int root_bus_len;
2542     char slot[] = ":SS.F";
2543     int slot_len = sizeof slot - 1 /* For '\0' */;
2544     int path_len;
2545     char *path, *p;
2546     int s;
2547 
2548     root_bus_path = pci_root_bus_path(d);
2549     root_bus_len = strlen(root_bus_path);
2550 
2551     /* Calculate # of slots on path between device and root. */;
2552     slot_depth = 0;
2553     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2554         ++slot_depth;
2555     }
2556 
2557     path_len = root_bus_len + slot_len * slot_depth;
2558 
2559     /* Allocate memory, fill in the terminating null byte. */
2560     path = g_malloc(path_len + 1 /* For '\0' */);
2561     path[path_len] = '\0';
2562 
2563     memcpy(path, root_bus_path, root_bus_len);
2564 
2565     /* Fill in slot numbers. We walk up from device to root, so need to print
2566      * them in the reverse order, last to first. */
2567     p = path + path_len;
2568     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2569         p -= slot_len;
2570         s = snprintf(slot, sizeof slot, ":%02x.%x",
2571                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2572         assert(s == slot_len);
2573         memcpy(p, slot, slot_len);
2574     }
2575 
2576     return path;
2577 }
2578 
2579 static int pci_qdev_find_recursive(PCIBus *bus,
2580                                    const char *id, PCIDevice **pdev)
2581 {
2582     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2583     if (!qdev) {
2584         return -ENODEV;
2585     }
2586 
2587     /* roughly check if given qdev is pci device */
2588     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2589         *pdev = PCI_DEVICE(qdev);
2590         return 0;
2591     }
2592     return -EINVAL;
2593 }
2594 
2595 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2596 {
2597     PCIHostState *host_bridge;
2598     int rc = -ENODEV;
2599 
2600     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2601         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2602         if (!tmp) {
2603             rc = 0;
2604             break;
2605         }
2606         if (tmp != -ENODEV) {
2607             rc = tmp;
2608         }
2609     }
2610 
2611     return rc;
2612 }
2613 
2614 MemoryRegion *pci_address_space(PCIDevice *dev)
2615 {
2616     return pci_get_bus(dev)->address_space_mem;
2617 }
2618 
2619 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2620 {
2621     return pci_get_bus(dev)->address_space_io;
2622 }
2623 
2624 static void pci_device_class_init(ObjectClass *klass, void *data)
2625 {
2626     DeviceClass *k = DEVICE_CLASS(klass);
2627 
2628     k->realize = pci_qdev_realize;
2629     k->unrealize = pci_qdev_unrealize;
2630     k->bus_type = TYPE_PCI_BUS;
2631     k->props = pci_props;
2632 }
2633 
2634 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2635 {
2636     if (!object_class_is_abstract(klass)) {
2637         ObjectClass *conventional =
2638             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2639         ObjectClass *pcie =
2640             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2641         assert(conventional || pcie);
2642     }
2643 }
2644 
2645 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2646 {
2647     PCIBus *bus = pci_get_bus(dev);
2648     PCIBus *iommu_bus = bus;
2649     uint8_t devfn = dev->devfn;
2650 
2651     while (iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2652         PCIBus *parent_bus = pci_get_bus(iommu_bus->parent_dev);
2653 
2654         /*
2655          * The requester ID of the provided device may be aliased, as seen from
2656          * the IOMMU, due to topology limitations.  The IOMMU relies on a
2657          * requester ID to provide a unique AddressSpace for devices, but
2658          * conventional PCI buses pre-date such concepts.  Instead, the PCIe-
2659          * to-PCI bridge creates and accepts transactions on behalf of down-
2660          * stream devices.  When doing so, all downstream devices are masked
2661          * (aliased) behind a single requester ID.  The requester ID used
2662          * depends on the format of the bridge devices.  Proper PCIe-to-PCI
2663          * bridges, with a PCIe capability indicating such, follow the
2664          * guidelines of chapter 2.3 of the PCIe-to-PCI/X bridge specification,
2665          * where the bridge uses the seconary bus as the bridge portion of the
2666          * requester ID and devfn of 00.0.  For other bridges, typically those
2667          * found on the root complex such as the dmi-to-pci-bridge, we follow
2668          * the convention of typical bare-metal hardware, which uses the
2669          * requester ID of the bridge itself.  There are device specific
2670          * exceptions to these rules, but these are the defaults that the
2671          * Linux kernel uses when determining DMA aliases itself and believed
2672          * to be true for the bare metal equivalents of the devices emulated
2673          * in QEMU.
2674          */
2675         if (!pci_bus_is_express(iommu_bus)) {
2676             PCIDevice *parent = iommu_bus->parent_dev;
2677 
2678             if (pci_is_express(parent) &&
2679                 pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
2680                 devfn = PCI_DEVFN(0, 0);
2681                 bus = iommu_bus;
2682             } else {
2683                 devfn = parent->devfn;
2684                 bus = parent_bus;
2685             }
2686         }
2687 
2688         iommu_bus = parent_bus;
2689     }
2690     if (iommu_bus && iommu_bus->iommu_fn) {
2691         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, devfn);
2692     }
2693     return &address_space_memory;
2694 }
2695 
2696 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2697 {
2698     bus->iommu_fn = fn;
2699     bus->iommu_opaque = opaque;
2700 }
2701 
2702 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2703 {
2704     Range *range = opaque;
2705     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2706     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2707     int i;
2708 
2709     if (!(cmd & PCI_COMMAND_MEMORY)) {
2710         return;
2711     }
2712 
2713     if (pc->is_bridge) {
2714         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2715         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2716 
2717         base = MAX(base, 0x1ULL << 32);
2718 
2719         if (limit >= base) {
2720             Range pref_range;
2721             range_set_bounds(&pref_range, base, limit);
2722             range_extend(range, &pref_range);
2723         }
2724     }
2725     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2726         PCIIORegion *r = &dev->io_regions[i];
2727         pcibus_t lob, upb;
2728         Range region_range;
2729 
2730         if (!r->size ||
2731             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2732             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2733             continue;
2734         }
2735 
2736         lob = pci_bar_address(dev, i, r->type, r->size);
2737         upb = lob + r->size - 1;
2738         if (lob == PCI_BAR_UNMAPPED) {
2739             continue;
2740         }
2741 
2742         lob = MAX(lob, 0x1ULL << 32);
2743 
2744         if (upb >= lob) {
2745             range_set_bounds(&region_range, lob, upb);
2746             range_extend(range, &region_range);
2747         }
2748     }
2749 }
2750 
2751 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2752 {
2753     range_make_empty(range);
2754     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2755 }
2756 
2757 static bool pcie_has_upstream_port(PCIDevice *dev)
2758 {
2759     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2760 
2761     /* Device associated with an upstream port.
2762      * As there are several types of these, it's easier to check the
2763      * parent device: upstream ports are always connected to
2764      * root or downstream ports.
2765      */
2766     return parent_dev &&
2767         pci_is_express(parent_dev) &&
2768         parent_dev->exp.exp_cap &&
2769         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2770          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2771 }
2772 
2773 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2774 {
2775     PCIBus *bus = pci_get_bus(pci_dev);
2776 
2777     if(pcie_has_upstream_port(pci_dev)) {
2778         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2779         return bus->devices[0];
2780     } else {
2781         /* Other bus types might support multiple devices at slots 0-31 */
2782         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2783     }
2784 }
2785 
2786 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2787 {
2788     MSIMessage msg;
2789     if (msix_enabled(dev)) {
2790         msg = msix_get_message(dev, vector);
2791     } else if (msi_enabled(dev)) {
2792         msg = msi_get_message(dev, vector);
2793     } else {
2794         /* Should never happen */
2795         error_report("%s: unknown interrupt type", __func__);
2796         abort();
2797     }
2798     return msg;
2799 }
2800 
2801 static const TypeInfo pci_device_type_info = {
2802     .name = TYPE_PCI_DEVICE,
2803     .parent = TYPE_DEVICE,
2804     .instance_size = sizeof(PCIDevice),
2805     .abstract = true,
2806     .class_size = sizeof(PCIDeviceClass),
2807     .class_init = pci_device_class_init,
2808     .class_base_init = pci_device_class_base_init,
2809 };
2810 
2811 static void pci_register_types(void)
2812 {
2813     type_register_static(&pci_bus_info);
2814     type_register_static(&pcie_bus_info);
2815     type_register_static(&conventional_pci_interface_info);
2816     type_register_static(&pcie_interface_info);
2817     type_register_static(&pci_device_type_info);
2818 }
2819 
2820 type_init(pci_register_types)
2821