xref: /openbmc/qemu/hw/pci/pci.c (revision 12a6c15e)
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 #include "qemu/osdep.h"
25 #include "hw/hw.h"
26 #include "hw/pci/pci.h"
27 #include "hw/pci/pci_bridge.h"
28 #include "hw/pci/pci_bus.h"
29 #include "hw/pci/pci_host.h"
30 #include "monitor/monitor.h"
31 #include "net/net.h"
32 #include "sysemu/sysemu.h"
33 #include "hw/loader.h"
34 #include "qemu/error-report.h"
35 #include "qemu/range.h"
36 #include "qmp-commands.h"
37 #include "trace.h"
38 #include "hw/pci/msi.h"
39 #include "hw/pci/msix.h"
40 #include "exec/address-spaces.h"
41 #include "hw/hotplug.h"
42 #include "hw/boards.h"
43 #include "qemu/cutils.h"
44 
45 //#define DEBUG_PCI
46 #ifdef DEBUG_PCI
47 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
48 #else
49 # define PCI_DPRINTF(format, ...)       do { } while (0)
50 #endif
51 
52 bool pci_available = true;
53 
54 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
55 static char *pcibus_get_dev_path(DeviceState *dev);
56 static char *pcibus_get_fw_dev_path(DeviceState *dev);
57 static void pcibus_reset(BusState *qbus);
58 
59 static Property pci_props[] = {
60     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
61     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
62     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
63     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
64                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
65     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
66                     QEMU_PCI_CAP_SERR_BITNR, true),
67     DEFINE_PROP_BIT("x-pcie-lnksta-dllla", PCIDevice, cap_present,
68                     QEMU_PCIE_LNKSTA_DLLLA_BITNR, true),
69     DEFINE_PROP_BIT("x-pcie-extcap-init", PCIDevice, cap_present,
70                     QEMU_PCIE_EXTCAP_INIT_BITNR, true),
71     DEFINE_PROP_END_OF_LIST()
72 };
73 
74 static const VMStateDescription vmstate_pcibus = {
75     .name = "PCIBUS",
76     .version_id = 1,
77     .minimum_version_id = 1,
78     .fields = (VMStateField[]) {
79         VMSTATE_INT32_EQUAL(nirq, PCIBus, NULL),
80         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
81                              nirq, 0, vmstate_info_int32,
82                              int32_t),
83         VMSTATE_END_OF_LIST()
84     }
85 };
86 
87 static void pci_init_bus_master(PCIDevice *pci_dev)
88 {
89     AddressSpace *dma_as = pci_device_iommu_address_space(pci_dev);
90 
91     memory_region_init_alias(&pci_dev->bus_master_enable_region,
92                              OBJECT(pci_dev), "bus master",
93                              dma_as->root, 0, memory_region_size(dma_as->root));
94     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
95     memory_region_add_subregion(&pci_dev->bus_master_container_region, 0,
96                                 &pci_dev->bus_master_enable_region);
97 }
98 
99 static void pcibus_machine_done(Notifier *notifier, void *data)
100 {
101     PCIBus *bus = container_of(notifier, PCIBus, machine_done);
102     int i;
103 
104     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
105         if (bus->devices[i]) {
106             pci_init_bus_master(bus->devices[i]);
107         }
108     }
109 }
110 
111 static void pci_bus_realize(BusState *qbus, Error **errp)
112 {
113     PCIBus *bus = PCI_BUS(qbus);
114 
115     bus->machine_done.notify = pcibus_machine_done;
116     qemu_add_machine_init_done_notifier(&bus->machine_done);
117 
118     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
119 }
120 
121 static void pci_bus_unrealize(BusState *qbus, Error **errp)
122 {
123     PCIBus *bus = PCI_BUS(qbus);
124 
125     qemu_remove_machine_init_done_notifier(&bus->machine_done);
126 
127     vmstate_unregister(NULL, &vmstate_pcibus, bus);
128 }
129 
130 static bool pcibus_is_root(PCIBus *bus)
131 {
132     return !bus->parent_dev;
133 }
134 
135 static int pcibus_num(PCIBus *bus)
136 {
137     if (pcibus_is_root(bus)) {
138         return 0; /* pci host bridge */
139     }
140     return bus->parent_dev->config[PCI_SECONDARY_BUS];
141 }
142 
143 static uint16_t pcibus_numa_node(PCIBus *bus)
144 {
145     return NUMA_NODE_UNASSIGNED;
146 }
147 
148 static void pci_bus_class_init(ObjectClass *klass, void *data)
149 {
150     BusClass *k = BUS_CLASS(klass);
151     PCIBusClass *pbc = PCI_BUS_CLASS(klass);
152 
153     k->print_dev = pcibus_dev_print;
154     k->get_dev_path = pcibus_get_dev_path;
155     k->get_fw_dev_path = pcibus_get_fw_dev_path;
156     k->realize = pci_bus_realize;
157     k->unrealize = pci_bus_unrealize;
158     k->reset = pcibus_reset;
159 
160     pbc->is_root = pcibus_is_root;
161     pbc->bus_num = pcibus_num;
162     pbc->numa_node = pcibus_numa_node;
163 }
164 
165 static const TypeInfo pci_bus_info = {
166     .name = TYPE_PCI_BUS,
167     .parent = TYPE_BUS,
168     .instance_size = sizeof(PCIBus),
169     .class_size = sizeof(PCIBusClass),
170     .class_init = pci_bus_class_init,
171 };
172 
173 static const TypeInfo pcie_bus_info = {
174     .name = TYPE_PCIE_BUS,
175     .parent = TYPE_PCI_BUS,
176 };
177 
178 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
179 static void pci_update_mappings(PCIDevice *d);
180 static void pci_irq_handler(void *opaque, int irq_num, int level);
181 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
182 static void pci_del_option_rom(PCIDevice *pdev);
183 
184 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
185 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
186 
187 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
188 
189 int pci_bar(PCIDevice *d, int reg)
190 {
191     uint8_t type;
192 
193     if (reg != PCI_ROM_SLOT)
194         return PCI_BASE_ADDRESS_0 + reg * 4;
195 
196     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
197     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
198 }
199 
200 static inline int pci_irq_state(PCIDevice *d, int irq_num)
201 {
202 	return (d->irq_state >> irq_num) & 0x1;
203 }
204 
205 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
206 {
207 	d->irq_state &= ~(0x1 << irq_num);
208 	d->irq_state |= level << irq_num;
209 }
210 
211 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
212 {
213     PCIBus *bus;
214     for (;;) {
215         bus = pci_dev->bus;
216         irq_num = bus->map_irq(pci_dev, irq_num);
217         if (bus->set_irq)
218             break;
219         pci_dev = bus->parent_dev;
220     }
221     bus->irq_count[irq_num] += change;
222     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
223 }
224 
225 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
226 {
227     assert(irq_num >= 0);
228     assert(irq_num < bus->nirq);
229     return !!bus->irq_count[irq_num];
230 }
231 
232 /* Update interrupt status bit in config space on interrupt
233  * state change. */
234 static void pci_update_irq_status(PCIDevice *dev)
235 {
236     if (dev->irq_state) {
237         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
238     } else {
239         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
240     }
241 }
242 
243 void pci_device_deassert_intx(PCIDevice *dev)
244 {
245     int i;
246     for (i = 0; i < PCI_NUM_PINS; ++i) {
247         pci_irq_handler(dev, i, 0);
248     }
249 }
250 
251 static void pci_do_device_reset(PCIDevice *dev)
252 {
253     int r;
254 
255     pci_device_deassert_intx(dev);
256     assert(dev->irq_state == 0);
257 
258     /* Clear all writable bits */
259     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
260                                  pci_get_word(dev->wmask + PCI_COMMAND) |
261                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
262     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
263                                  pci_get_word(dev->wmask + PCI_STATUS) |
264                                  pci_get_word(dev->w1cmask + PCI_STATUS));
265     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
266     dev->config[PCI_INTERRUPT_LINE] = 0x0;
267     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
268         PCIIORegion *region = &dev->io_regions[r];
269         if (!region->size) {
270             continue;
271         }
272 
273         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
274             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
275             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
276         } else {
277             pci_set_long(dev->config + pci_bar(dev, r), region->type);
278         }
279     }
280     pci_update_mappings(dev);
281 
282     msi_reset(dev);
283     msix_reset(dev);
284 }
285 
286 /*
287  * This function is called on #RST and FLR.
288  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
289  */
290 void pci_device_reset(PCIDevice *dev)
291 {
292     qdev_reset_all(&dev->qdev);
293     pci_do_device_reset(dev);
294 }
295 
296 /*
297  * Trigger pci bus reset under a given bus.
298  * Called via qbus_reset_all on RST# assert, after the devices
299  * have been reset qdev_reset_all-ed already.
300  */
301 static void pcibus_reset(BusState *qbus)
302 {
303     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
304     int i;
305 
306     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
307         if (bus->devices[i]) {
308             pci_do_device_reset(bus->devices[i]);
309         }
310     }
311 
312     for (i = 0; i < bus->nirq; i++) {
313         assert(bus->irq_count[i] == 0);
314     }
315 }
316 
317 static void pci_host_bus_register(DeviceState *host)
318 {
319     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
320 
321     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
322 }
323 
324 PCIBus *pci_find_primary_bus(void)
325 {
326     PCIBus *primary_bus = NULL;
327     PCIHostState *host;
328 
329     QLIST_FOREACH(host, &pci_host_bridges, next) {
330         if (primary_bus) {
331             /* We have multiple root buses, refuse to select a primary */
332             return NULL;
333         }
334         primary_bus = host->bus;
335     }
336 
337     return primary_bus;
338 }
339 
340 PCIBus *pci_device_root_bus(const PCIDevice *d)
341 {
342     PCIBus *bus = d->bus;
343 
344     while (!pci_bus_is_root(bus)) {
345         d = bus->parent_dev;
346         assert(d != NULL);
347 
348         bus = d->bus;
349     }
350 
351     return bus;
352 }
353 
354 const char *pci_root_bus_path(PCIDevice *dev)
355 {
356     PCIBus *rootbus = pci_device_root_bus(dev);
357     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
358     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
359 
360     assert(host_bridge->bus == rootbus);
361 
362     if (hc->root_bus_path) {
363         return (*hc->root_bus_path)(host_bridge, rootbus);
364     }
365 
366     return rootbus->qbus.name;
367 }
368 
369 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
370                          MemoryRegion *address_space_mem,
371                          MemoryRegion *address_space_io,
372                          uint8_t devfn_min)
373 {
374     assert(PCI_FUNC(devfn_min) == 0);
375     bus->devfn_min = devfn_min;
376     bus->address_space_mem = address_space_mem;
377     bus->address_space_io = address_space_io;
378 
379     /* host bridge */
380     QLIST_INIT(&bus->child);
381 
382     pci_host_bus_register(parent);
383 }
384 
385 bool pci_bus_is_express(PCIBus *bus)
386 {
387     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
388 }
389 
390 bool pci_bus_is_root(PCIBus *bus)
391 {
392     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
393 }
394 
395 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
396                          const char *name,
397                          MemoryRegion *address_space_mem,
398                          MemoryRegion *address_space_io,
399                          uint8_t devfn_min, const char *typename)
400 {
401     qbus_create_inplace(bus, bus_size, typename, parent, name);
402     pci_bus_init(bus, parent, address_space_mem, address_space_io, devfn_min);
403 }
404 
405 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
406                     MemoryRegion *address_space_mem,
407                     MemoryRegion *address_space_io,
408                     uint8_t devfn_min, const char *typename)
409 {
410     PCIBus *bus;
411 
412     bus = PCI_BUS(qbus_create(typename, parent, name));
413     pci_bus_init(bus, parent, address_space_mem, address_space_io, devfn_min);
414     return bus;
415 }
416 
417 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
418                   void *irq_opaque, int nirq)
419 {
420     bus->set_irq = set_irq;
421     bus->map_irq = map_irq;
422     bus->irq_opaque = irq_opaque;
423     bus->nirq = nirq;
424     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
425 }
426 
427 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
428                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
429                          void *irq_opaque,
430                          MemoryRegion *address_space_mem,
431                          MemoryRegion *address_space_io,
432                          uint8_t devfn_min, int nirq, const char *typename)
433 {
434     PCIBus *bus;
435 
436     bus = pci_bus_new(parent, name, address_space_mem,
437                       address_space_io, devfn_min, typename);
438     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
439     return bus;
440 }
441 
442 int pci_bus_num(PCIBus *s)
443 {
444     return PCI_BUS_GET_CLASS(s)->bus_num(s);
445 }
446 
447 int pci_bus_numa_node(PCIBus *bus)
448 {
449     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
450 }
451 
452 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
453                                  VMStateField *field)
454 {
455     PCIDevice *s = container_of(pv, PCIDevice, config);
456     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
457     uint8_t *config;
458     int i;
459 
460     assert(size == pci_config_size(s));
461     config = g_malloc(size);
462 
463     qemu_get_buffer(f, config, size);
464     for (i = 0; i < size; ++i) {
465         if ((config[i] ^ s->config[i]) &
466             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
467             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
468                          "cmask: %x wmask: %x w1cmask:%x", __func__,
469                          i, config[i], s->config[i],
470                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
471             g_free(config);
472             return -EINVAL;
473         }
474     }
475     memcpy(s->config, config, size);
476 
477     pci_update_mappings(s);
478     if (pc->is_bridge) {
479         PCIBridge *b = PCI_BRIDGE(s);
480         pci_bridge_update_mappings(b);
481     }
482 
483     memory_region_set_enabled(&s->bus_master_enable_region,
484                               pci_get_word(s->config + PCI_COMMAND)
485                               & PCI_COMMAND_MASTER);
486 
487     g_free(config);
488     return 0;
489 }
490 
491 /* just put buffer */
492 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
493                                  VMStateField *field, QJSON *vmdesc)
494 {
495     const uint8_t **v = pv;
496     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
497     qemu_put_buffer(f, *v, size);
498 
499     return 0;
500 }
501 
502 static VMStateInfo vmstate_info_pci_config = {
503     .name = "pci config",
504     .get  = get_pci_config_device,
505     .put  = put_pci_config_device,
506 };
507 
508 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
509                              VMStateField *field)
510 {
511     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
512     uint32_t irq_state[PCI_NUM_PINS];
513     int i;
514     for (i = 0; i < PCI_NUM_PINS; ++i) {
515         irq_state[i] = qemu_get_be32(f);
516         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
517             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
518                     irq_state[i]);
519             return -EINVAL;
520         }
521     }
522 
523     for (i = 0; i < PCI_NUM_PINS; ++i) {
524         pci_set_irq_state(s, i, irq_state[i]);
525     }
526 
527     return 0;
528 }
529 
530 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
531                              VMStateField *field, QJSON *vmdesc)
532 {
533     int i;
534     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
535 
536     for (i = 0; i < PCI_NUM_PINS; ++i) {
537         qemu_put_be32(f, pci_irq_state(s, i));
538     }
539 
540     return 0;
541 }
542 
543 static VMStateInfo vmstate_info_pci_irq_state = {
544     .name = "pci irq state",
545     .get  = get_pci_irq_state,
546     .put  = put_pci_irq_state,
547 };
548 
549 static bool migrate_is_pcie(void *opaque, int version_id)
550 {
551     return pci_is_express((PCIDevice *)opaque);
552 }
553 
554 static bool migrate_is_not_pcie(void *opaque, int version_id)
555 {
556     return !pci_is_express((PCIDevice *)opaque);
557 }
558 
559 const VMStateDescription vmstate_pci_device = {
560     .name = "PCIDevice",
561     .version_id = 2,
562     .minimum_version_id = 1,
563     .fields = (VMStateField[]) {
564         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
565         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
566                                    migrate_is_not_pcie,
567                                    0, vmstate_info_pci_config,
568                                    PCI_CONFIG_SPACE_SIZE),
569         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
570                                    migrate_is_pcie,
571                                    0, vmstate_info_pci_config,
572                                    PCIE_CONFIG_SPACE_SIZE),
573         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
574 				   vmstate_info_pci_irq_state,
575 				   PCI_NUM_PINS * sizeof(int32_t)),
576         VMSTATE_END_OF_LIST()
577     }
578 };
579 
580 
581 void pci_device_save(PCIDevice *s, QEMUFile *f)
582 {
583     /* Clear interrupt status bit: it is implicit
584      * in irq_state which we are saving.
585      * This makes us compatible with old devices
586      * which never set or clear this bit. */
587     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
588     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
589     /* Restore the interrupt status bit. */
590     pci_update_irq_status(s);
591 }
592 
593 int pci_device_load(PCIDevice *s, QEMUFile *f)
594 {
595     int ret;
596     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
597     /* Restore the interrupt status bit. */
598     pci_update_irq_status(s);
599     return ret;
600 }
601 
602 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
603 {
604     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
605                  pci_default_sub_vendor_id);
606     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
607                  pci_default_sub_device_id);
608 }
609 
610 /*
611  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
612  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
613  */
614 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
615                              unsigned int *slotp, unsigned int *funcp)
616 {
617     const char *p;
618     char *e;
619     unsigned long val;
620     unsigned long dom = 0, bus = 0;
621     unsigned int slot = 0;
622     unsigned int func = 0;
623 
624     p = addr;
625     val = strtoul(p, &e, 16);
626     if (e == p)
627 	return -1;
628     if (*e == ':') {
629 	bus = val;
630 	p = e + 1;
631 	val = strtoul(p, &e, 16);
632 	if (e == p)
633 	    return -1;
634 	if (*e == ':') {
635 	    dom = bus;
636 	    bus = val;
637 	    p = e + 1;
638 	    val = strtoul(p, &e, 16);
639 	    if (e == p)
640 		return -1;
641 	}
642     }
643 
644     slot = val;
645 
646     if (funcp != NULL) {
647         if (*e != '.')
648             return -1;
649 
650         p = e + 1;
651         val = strtoul(p, &e, 16);
652         if (e == p)
653             return -1;
654 
655         func = val;
656     }
657 
658     /* if funcp == NULL func is 0 */
659     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
660 	return -1;
661 
662     if (*e)
663 	return -1;
664 
665     *domp = dom;
666     *busp = bus;
667     *slotp = slot;
668     if (funcp != NULL)
669         *funcp = func;
670     return 0;
671 }
672 
673 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
674                                  const char *devaddr)
675 {
676     int dom, bus;
677     unsigned slot;
678 
679     if (!root) {
680         fprintf(stderr, "No primary PCI bus\n");
681         return NULL;
682     }
683 
684     assert(!root->parent_dev);
685 
686     if (!devaddr) {
687         *devfnp = -1;
688         return pci_find_bus_nr(root, 0);
689     }
690 
691     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
692         return NULL;
693     }
694 
695     if (dom != 0) {
696         fprintf(stderr, "No support for non-zero PCI domains\n");
697         return NULL;
698     }
699 
700     *devfnp = PCI_DEVFN(slot, 0);
701     return pci_find_bus_nr(root, bus);
702 }
703 
704 static void pci_init_cmask(PCIDevice *dev)
705 {
706     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
707     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
708     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
709     dev->cmask[PCI_REVISION_ID] = 0xff;
710     dev->cmask[PCI_CLASS_PROG] = 0xff;
711     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
712     dev->cmask[PCI_HEADER_TYPE] = 0xff;
713     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
714 }
715 
716 static void pci_init_wmask(PCIDevice *dev)
717 {
718     int config_size = pci_config_size(dev);
719 
720     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
721     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
722     pci_set_word(dev->wmask + PCI_COMMAND,
723                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
724                  PCI_COMMAND_INTX_DISABLE);
725     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
726         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
727     }
728 
729     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
730            config_size - PCI_CONFIG_HEADER_SIZE);
731 }
732 
733 static void pci_init_w1cmask(PCIDevice *dev)
734 {
735     /*
736      * Note: It's okay to set w1cmask even for readonly bits as
737      * long as their value is hardwired to 0.
738      */
739     pci_set_word(dev->w1cmask + PCI_STATUS,
740                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
741                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
742                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
743 }
744 
745 static void pci_init_mask_bridge(PCIDevice *d)
746 {
747     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
748        PCI_SEC_LETENCY_TIMER */
749     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
750 
751     /* base and limit */
752     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
753     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
754     pci_set_word(d->wmask + PCI_MEMORY_BASE,
755                  PCI_MEMORY_RANGE_MASK & 0xffff);
756     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
757                  PCI_MEMORY_RANGE_MASK & 0xffff);
758     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
759                  PCI_PREF_RANGE_MASK & 0xffff);
760     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
761                  PCI_PREF_RANGE_MASK & 0xffff);
762 
763     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
764     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
765 
766     /* Supported memory and i/o types */
767     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
768     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
769     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
770                                PCI_PREF_RANGE_TYPE_64);
771     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
772                                PCI_PREF_RANGE_TYPE_64);
773 
774     /*
775      * TODO: Bridges default to 10-bit VGA decoding but we currently only
776      * implement 16-bit decoding (no alias support).
777      */
778     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
779                  PCI_BRIDGE_CTL_PARITY |
780                  PCI_BRIDGE_CTL_SERR |
781                  PCI_BRIDGE_CTL_ISA |
782                  PCI_BRIDGE_CTL_VGA |
783                  PCI_BRIDGE_CTL_VGA_16BIT |
784                  PCI_BRIDGE_CTL_MASTER_ABORT |
785                  PCI_BRIDGE_CTL_BUS_RESET |
786                  PCI_BRIDGE_CTL_FAST_BACK |
787                  PCI_BRIDGE_CTL_DISCARD |
788                  PCI_BRIDGE_CTL_SEC_DISCARD |
789                  PCI_BRIDGE_CTL_DISCARD_SERR);
790     /* Below does not do anything as we never set this bit, put here for
791      * completeness. */
792     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
793                  PCI_BRIDGE_CTL_DISCARD_STATUS);
794     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
795     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
796     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
797                                PCI_PREF_RANGE_TYPE_MASK);
798     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
799                                PCI_PREF_RANGE_TYPE_MASK);
800 }
801 
802 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
803 {
804     uint8_t slot = PCI_SLOT(dev->devfn);
805     uint8_t func;
806 
807     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
808         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
809     }
810 
811     /*
812      * multifunction bit is interpreted in two ways as follows.
813      *   - all functions must set the bit to 1.
814      *     Example: Intel X53
815      *   - function 0 must set the bit, but the rest function (> 0)
816      *     is allowed to leave the bit to 0.
817      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
818      *
819      * So OS (at least Linux) checks the bit of only function 0,
820      * and doesn't see the bit of function > 0.
821      *
822      * The below check allows both interpretation.
823      */
824     if (PCI_FUNC(dev->devfn)) {
825         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
826         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
827             /* function 0 should set multifunction bit */
828             error_setg(errp, "PCI: single function device can't be populated "
829                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
830             return;
831         }
832         return;
833     }
834 
835     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
836         return;
837     }
838     /* function 0 indicates single function, so function > 0 must be NULL */
839     for (func = 1; func < PCI_FUNC_MAX; ++func) {
840         if (bus->devices[PCI_DEVFN(slot, func)]) {
841             error_setg(errp, "PCI: %x.0 indicates single function, "
842                        "but %x.%x is already populated.",
843                        slot, slot, func);
844             return;
845         }
846     }
847 }
848 
849 static void pci_config_alloc(PCIDevice *pci_dev)
850 {
851     int config_size = pci_config_size(pci_dev);
852 
853     pci_dev->config = g_malloc0(config_size);
854     pci_dev->cmask = g_malloc0(config_size);
855     pci_dev->wmask = g_malloc0(config_size);
856     pci_dev->w1cmask = g_malloc0(config_size);
857     pci_dev->used = g_malloc0(config_size);
858 }
859 
860 static void pci_config_free(PCIDevice *pci_dev)
861 {
862     g_free(pci_dev->config);
863     g_free(pci_dev->cmask);
864     g_free(pci_dev->wmask);
865     g_free(pci_dev->w1cmask);
866     g_free(pci_dev->used);
867 }
868 
869 static void do_pci_unregister_device(PCIDevice *pci_dev)
870 {
871     pci_dev->bus->devices[pci_dev->devfn] = NULL;
872     pci_config_free(pci_dev);
873 
874     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
875         memory_region_del_subregion(&pci_dev->bus_master_container_region,
876                                     &pci_dev->bus_master_enable_region);
877     }
878     address_space_destroy(&pci_dev->bus_master_as);
879 }
880 
881 /* Extract PCIReqIDCache into BDF format */
882 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
883 {
884     uint8_t bus_n;
885     uint16_t result;
886 
887     switch (cache->type) {
888     case PCI_REQ_ID_BDF:
889         result = pci_get_bdf(cache->dev);
890         break;
891     case PCI_REQ_ID_SECONDARY_BUS:
892         bus_n = pci_bus_num(cache->dev->bus);
893         result = PCI_BUILD_BDF(bus_n, 0);
894         break;
895     default:
896         error_printf("Invalid PCI requester ID cache type: %d\n",
897                      cache->type);
898         exit(1);
899         break;
900     }
901 
902     return result;
903 }
904 
905 /* Parse bridges up to the root complex and return requester ID
906  * cache for specific device.  For full PCIe topology, the cache
907  * result would be exactly the same as getting BDF of the device.
908  * However, several tricks are required when system mixed up with
909  * legacy PCI devices and PCIe-to-PCI bridges.
910  *
911  * Here we cache the proxy device (and type) not requester ID since
912  * bus number might change from time to time.
913  */
914 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
915 {
916     PCIDevice *parent;
917     PCIReqIDCache cache = {
918         .dev = dev,
919         .type = PCI_REQ_ID_BDF,
920     };
921 
922     while (!pci_bus_is_root(dev->bus)) {
923         /* We are under PCI/PCIe bridges */
924         parent = dev->bus->parent_dev;
925         if (pci_is_express(parent)) {
926             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
927                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
928                  * override the requester ID using secondary bus
929                  * number of parent bridge with zeroed devfn
930                  * (pcie-to-pci bridge spec chap 2.3). */
931                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
932                 cache.dev = dev;
933             }
934         } else {
935             /* Legacy PCI, override requester ID with the bridge's
936              * BDF upstream.  When the root complex connects to
937              * legacy PCI devices (including buses), it can only
938              * obtain requester ID info from directly attached
939              * devices.  If devices are attached under bridges, only
940              * the requester ID of the bridge that is directly
941              * attached to the root complex can be recognized. */
942             cache.type = PCI_REQ_ID_BDF;
943             cache.dev = parent;
944         }
945         dev = parent;
946     }
947 
948     return cache;
949 }
950 
951 uint16_t pci_requester_id(PCIDevice *dev)
952 {
953     return pci_req_id_cache_extract(&dev->requester_id_cache);
954 }
955 
956 /* -1 for devfn means auto assign */
957 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
958                                          const char *name, int devfn,
959                                          Error **errp)
960 {
961     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
962     PCIConfigReadFunc *config_read = pc->config_read;
963     PCIConfigWriteFunc *config_write = pc->config_write;
964     Error *local_err = NULL;
965     DeviceState *dev = DEVICE(pci_dev);
966 
967     pci_dev->bus = bus;
968     /* Only pci bridges can be attached to extra PCI root buses */
969     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
970         error_setg(errp,
971                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
972                     bus->parent_dev->name);
973         return NULL;
974     }
975 
976     if (devfn < 0) {
977         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
978             devfn += PCI_FUNC_MAX) {
979             if (!bus->devices[devfn])
980                 goto found;
981         }
982         error_setg(errp, "PCI: no slot/function available for %s, all in use",
983                    name);
984         return NULL;
985     found: ;
986     } else if (bus->devices[devfn]) {
987         error_setg(errp, "PCI: slot %d function %d not available for %s,"
988                    " in use by %s",
989                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
990                    bus->devices[devfn]->name);
991         return NULL;
992     } else if (dev->hotplugged &&
993                pci_get_function_0(pci_dev)) {
994         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
995                    " new func %s cannot be exposed to guest.",
996                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
997                    pci_get_function_0(pci_dev)->name,
998                    name);
999 
1000        return NULL;
1001     }
1002 
1003     pci_dev->devfn = devfn;
1004     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1005 
1006     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1007                        "bus master container", UINT64_MAX);
1008     address_space_init(&pci_dev->bus_master_as,
1009                        &pci_dev->bus_master_container_region, pci_dev->name);
1010 
1011     if (qdev_hotplug) {
1012         pci_init_bus_master(pci_dev);
1013     }
1014     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1015     pci_dev->irq_state = 0;
1016     pci_config_alloc(pci_dev);
1017 
1018     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1019     pci_config_set_device_id(pci_dev->config, pc->device_id);
1020     pci_config_set_revision(pci_dev->config, pc->revision);
1021     pci_config_set_class(pci_dev->config, pc->class_id);
1022 
1023     if (!pc->is_bridge) {
1024         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1025             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1026                          pc->subsystem_vendor_id);
1027             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1028                          pc->subsystem_id);
1029         } else {
1030             pci_set_default_subsystem_id(pci_dev);
1031         }
1032     } else {
1033         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1034         assert(!pc->subsystem_vendor_id);
1035         assert(!pc->subsystem_id);
1036     }
1037     pci_init_cmask(pci_dev);
1038     pci_init_wmask(pci_dev);
1039     pci_init_w1cmask(pci_dev);
1040     if (pc->is_bridge) {
1041         pci_init_mask_bridge(pci_dev);
1042     }
1043     pci_init_multifunction(bus, pci_dev, &local_err);
1044     if (local_err) {
1045         error_propagate(errp, local_err);
1046         do_pci_unregister_device(pci_dev);
1047         return NULL;
1048     }
1049 
1050     if (!config_read)
1051         config_read = pci_default_read_config;
1052     if (!config_write)
1053         config_write = pci_default_write_config;
1054     pci_dev->config_read = config_read;
1055     pci_dev->config_write = config_write;
1056     bus->devices[devfn] = pci_dev;
1057     pci_dev->version_id = 2; /* Current pci device vmstate version */
1058     return pci_dev;
1059 }
1060 
1061 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1062 {
1063     PCIIORegion *r;
1064     int i;
1065 
1066     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1067         r = &pci_dev->io_regions[i];
1068         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1069             continue;
1070         memory_region_del_subregion(r->address_space, r->memory);
1071     }
1072 
1073     pci_unregister_vga(pci_dev);
1074 }
1075 
1076 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1077 {
1078     PCIDevice *pci_dev = PCI_DEVICE(dev);
1079     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1080 
1081     pci_unregister_io_regions(pci_dev);
1082     pci_del_option_rom(pci_dev);
1083 
1084     if (pc->exit) {
1085         pc->exit(pci_dev);
1086     }
1087 
1088     pci_device_deassert_intx(pci_dev);
1089     do_pci_unregister_device(pci_dev);
1090 }
1091 
1092 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1093                       uint8_t type, MemoryRegion *memory)
1094 {
1095     PCIIORegion *r;
1096     uint32_t addr; /* offset in pci config space */
1097     uint64_t wmask;
1098     pcibus_t size = memory_region_size(memory);
1099 
1100     assert(region_num >= 0);
1101     assert(region_num < PCI_NUM_REGIONS);
1102     if (size & (size-1)) {
1103         fprintf(stderr, "ERROR: PCI region size must be pow2 "
1104                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
1105         exit(1);
1106     }
1107 
1108     r = &pci_dev->io_regions[region_num];
1109     r->addr = PCI_BAR_UNMAPPED;
1110     r->size = size;
1111     r->type = type;
1112     r->memory = memory;
1113     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1114                         ? pci_dev->bus->address_space_io
1115                         : pci_dev->bus->address_space_mem;
1116 
1117     wmask = ~(size - 1);
1118     if (region_num == PCI_ROM_SLOT) {
1119         /* ROM enable bit is writable */
1120         wmask |= PCI_ROM_ADDRESS_ENABLE;
1121     }
1122 
1123     addr = pci_bar(pci_dev, region_num);
1124     pci_set_long(pci_dev->config + addr, type);
1125 
1126     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1127         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1128         pci_set_quad(pci_dev->wmask + addr, wmask);
1129         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1130     } else {
1131         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1132         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1133     }
1134 }
1135 
1136 static void pci_update_vga(PCIDevice *pci_dev)
1137 {
1138     uint16_t cmd;
1139 
1140     if (!pci_dev->has_vga) {
1141         return;
1142     }
1143 
1144     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1145 
1146     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1147                               cmd & PCI_COMMAND_MEMORY);
1148     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1149                               cmd & PCI_COMMAND_IO);
1150     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1151                               cmd & PCI_COMMAND_IO);
1152 }
1153 
1154 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1155                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1156 {
1157     assert(!pci_dev->has_vga);
1158 
1159     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1160     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1161     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
1162                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1163 
1164     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1165     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1166     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1167                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1168 
1169     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1170     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1171     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1172                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1173     pci_dev->has_vga = true;
1174 
1175     pci_update_vga(pci_dev);
1176 }
1177 
1178 void pci_unregister_vga(PCIDevice *pci_dev)
1179 {
1180     if (!pci_dev->has_vga) {
1181         return;
1182     }
1183 
1184     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1185                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1186     memory_region_del_subregion(pci_dev->bus->address_space_io,
1187                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1188     memory_region_del_subregion(pci_dev->bus->address_space_io,
1189                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1190     pci_dev->has_vga = false;
1191 }
1192 
1193 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1194 {
1195     return pci_dev->io_regions[region_num].addr;
1196 }
1197 
1198 static pcibus_t pci_bar_address(PCIDevice *d,
1199 				int reg, uint8_t type, pcibus_t size)
1200 {
1201     pcibus_t new_addr, last_addr;
1202     int bar = pci_bar(d, reg);
1203     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1204     Object *machine = qdev_get_machine();
1205     ObjectClass *oc = object_get_class(machine);
1206     MachineClass *mc = MACHINE_CLASS(oc);
1207     bool allow_0_address = mc->pci_allow_0_address;
1208 
1209     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1210         if (!(cmd & PCI_COMMAND_IO)) {
1211             return PCI_BAR_UNMAPPED;
1212         }
1213         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1214         last_addr = new_addr + size - 1;
1215         /* Check if 32 bit BAR wraps around explicitly.
1216          * TODO: make priorities correct and remove this work around.
1217          */
1218         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1219             (!allow_0_address && new_addr == 0)) {
1220             return PCI_BAR_UNMAPPED;
1221         }
1222         return new_addr;
1223     }
1224 
1225     if (!(cmd & PCI_COMMAND_MEMORY)) {
1226         return PCI_BAR_UNMAPPED;
1227     }
1228     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1229         new_addr = pci_get_quad(d->config + bar);
1230     } else {
1231         new_addr = pci_get_long(d->config + bar);
1232     }
1233     /* the ROM slot has a specific enable bit */
1234     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1235         return PCI_BAR_UNMAPPED;
1236     }
1237     new_addr &= ~(size - 1);
1238     last_addr = new_addr + size - 1;
1239     /* NOTE: we do not support wrapping */
1240     /* XXX: as we cannot support really dynamic
1241        mappings, we handle specific values as invalid
1242        mappings. */
1243     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1244         (!allow_0_address && new_addr == 0)) {
1245         return PCI_BAR_UNMAPPED;
1246     }
1247 
1248     /* Now pcibus_t is 64bit.
1249      * Check if 32 bit BAR wraps around explicitly.
1250      * Without this, PC ide doesn't work well.
1251      * TODO: remove this work around.
1252      */
1253     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1254         return PCI_BAR_UNMAPPED;
1255     }
1256 
1257     /*
1258      * OS is allowed to set BAR beyond its addressable
1259      * bits. For example, 32 bit OS can set 64bit bar
1260      * to >4G. Check it. TODO: we might need to support
1261      * it in the future for e.g. PAE.
1262      */
1263     if (last_addr >= HWADDR_MAX) {
1264         return PCI_BAR_UNMAPPED;
1265     }
1266 
1267     return new_addr;
1268 }
1269 
1270 static void pci_update_mappings(PCIDevice *d)
1271 {
1272     PCIIORegion *r;
1273     int i;
1274     pcibus_t new_addr;
1275 
1276     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1277         r = &d->io_regions[i];
1278 
1279         /* this region isn't registered */
1280         if (!r->size)
1281             continue;
1282 
1283         new_addr = pci_bar_address(d, i, r->type, r->size);
1284 
1285         /* This bar isn't changed */
1286         if (new_addr == r->addr)
1287             continue;
1288 
1289         /* now do the real mapping */
1290         if (r->addr != PCI_BAR_UNMAPPED) {
1291             trace_pci_update_mappings_del(d, pci_bus_num(d->bus),
1292                                           PCI_SLOT(d->devfn),
1293                                           PCI_FUNC(d->devfn),
1294                                           i, r->addr, r->size);
1295             memory_region_del_subregion(r->address_space, r->memory);
1296         }
1297         r->addr = new_addr;
1298         if (r->addr != PCI_BAR_UNMAPPED) {
1299             trace_pci_update_mappings_add(d, pci_bus_num(d->bus),
1300                                           PCI_SLOT(d->devfn),
1301                                           PCI_FUNC(d->devfn),
1302                                           i, r->addr, r->size);
1303             memory_region_add_subregion_overlap(r->address_space,
1304                                                 r->addr, r->memory, 1);
1305         }
1306     }
1307 
1308     pci_update_vga(d);
1309 }
1310 
1311 static inline int pci_irq_disabled(PCIDevice *d)
1312 {
1313     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1314 }
1315 
1316 /* Called after interrupt disabled field update in config space,
1317  * assert/deassert interrupts if necessary.
1318  * Gets original interrupt disable bit value (before update). */
1319 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1320 {
1321     int i, disabled = pci_irq_disabled(d);
1322     if (disabled == was_irq_disabled)
1323         return;
1324     for (i = 0; i < PCI_NUM_PINS; ++i) {
1325         int state = pci_irq_state(d, i);
1326         pci_change_irq_level(d, i, disabled ? -state : state);
1327     }
1328 }
1329 
1330 uint32_t pci_default_read_config(PCIDevice *d,
1331                                  uint32_t address, int len)
1332 {
1333     uint32_t val = 0;
1334 
1335     memcpy(&val, d->config + address, len);
1336     return le32_to_cpu(val);
1337 }
1338 
1339 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1340 {
1341     int i, was_irq_disabled = pci_irq_disabled(d);
1342     uint32_t val = val_in;
1343 
1344     for (i = 0; i < l; val >>= 8, ++i) {
1345         uint8_t wmask = d->wmask[addr + i];
1346         uint8_t w1cmask = d->w1cmask[addr + i];
1347         assert(!(wmask & w1cmask));
1348         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1349         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1350     }
1351     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1352         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1353         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1354         range_covers_byte(addr, l, PCI_COMMAND))
1355         pci_update_mappings(d);
1356 
1357     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1358         pci_update_irq_disabled(d, was_irq_disabled);
1359         memory_region_set_enabled(&d->bus_master_enable_region,
1360                                   pci_get_word(d->config + PCI_COMMAND)
1361                                     & PCI_COMMAND_MASTER);
1362     }
1363 
1364     msi_write_config(d, addr, val_in, l);
1365     msix_write_config(d, addr, val_in, l);
1366 }
1367 
1368 /***********************************************************/
1369 /* generic PCI irq support */
1370 
1371 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1372 static void pci_irq_handler(void *opaque, int irq_num, int level)
1373 {
1374     PCIDevice *pci_dev = opaque;
1375     int change;
1376 
1377     change = level - pci_irq_state(pci_dev, irq_num);
1378     if (!change)
1379         return;
1380 
1381     pci_set_irq_state(pci_dev, irq_num, level);
1382     pci_update_irq_status(pci_dev);
1383     if (pci_irq_disabled(pci_dev))
1384         return;
1385     pci_change_irq_level(pci_dev, irq_num, change);
1386 }
1387 
1388 static inline int pci_intx(PCIDevice *pci_dev)
1389 {
1390     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1391 }
1392 
1393 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1394 {
1395     int intx = pci_intx(pci_dev);
1396 
1397     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1398 }
1399 
1400 void pci_set_irq(PCIDevice *pci_dev, int level)
1401 {
1402     int intx = pci_intx(pci_dev);
1403     pci_irq_handler(pci_dev, intx, level);
1404 }
1405 
1406 /* Special hooks used by device assignment */
1407 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1408 {
1409     assert(pci_bus_is_root(bus));
1410     bus->route_intx_to_irq = route_intx_to_irq;
1411 }
1412 
1413 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1414 {
1415     PCIBus *bus;
1416 
1417     do {
1418          bus = dev->bus;
1419          pin = bus->map_irq(dev, pin);
1420          dev = bus->parent_dev;
1421     } while (dev);
1422 
1423     if (!bus->route_intx_to_irq) {
1424         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1425                      object_get_typename(OBJECT(bus->qbus.parent)));
1426         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1427     }
1428 
1429     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1430 }
1431 
1432 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1433 {
1434     return old->mode != new->mode || old->irq != new->irq;
1435 }
1436 
1437 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1438 {
1439     PCIDevice *dev;
1440     PCIBus *sec;
1441     int i;
1442 
1443     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1444         dev = bus->devices[i];
1445         if (dev && dev->intx_routing_notifier) {
1446             dev->intx_routing_notifier(dev);
1447         }
1448     }
1449 
1450     QLIST_FOREACH(sec, &bus->child, sibling) {
1451         pci_bus_fire_intx_routing_notifier(sec);
1452     }
1453 }
1454 
1455 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1456                                           PCIINTxRoutingNotifier notifier)
1457 {
1458     dev->intx_routing_notifier = notifier;
1459 }
1460 
1461 /*
1462  * PCI-to-PCI bridge specification
1463  * 9.1: Interrupt routing. Table 9-1
1464  *
1465  * the PCI Express Base Specification, Revision 2.1
1466  * 2.2.8.1: INTx interrutp signaling - Rules
1467  *          the Implementation Note
1468  *          Table 2-20
1469  */
1470 /*
1471  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1472  * 0-origin unlike PCI interrupt pin register.
1473  */
1474 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1475 {
1476     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1477 }
1478 
1479 /***********************************************************/
1480 /* monitor info on PCI */
1481 
1482 typedef struct {
1483     uint16_t class;
1484     const char *desc;
1485     const char *fw_name;
1486     uint16_t fw_ign_bits;
1487 } pci_class_desc;
1488 
1489 static const pci_class_desc pci_class_descriptions[] =
1490 {
1491     { 0x0001, "VGA controller", "display"},
1492     { 0x0100, "SCSI controller", "scsi"},
1493     { 0x0101, "IDE controller", "ide"},
1494     { 0x0102, "Floppy controller", "fdc"},
1495     { 0x0103, "IPI controller", "ipi"},
1496     { 0x0104, "RAID controller", "raid"},
1497     { 0x0106, "SATA controller"},
1498     { 0x0107, "SAS controller"},
1499     { 0x0180, "Storage controller"},
1500     { 0x0200, "Ethernet controller", "ethernet"},
1501     { 0x0201, "Token Ring controller", "token-ring"},
1502     { 0x0202, "FDDI controller", "fddi"},
1503     { 0x0203, "ATM controller", "atm"},
1504     { 0x0280, "Network controller"},
1505     { 0x0300, "VGA controller", "display", 0x00ff},
1506     { 0x0301, "XGA controller"},
1507     { 0x0302, "3D controller"},
1508     { 0x0380, "Display controller"},
1509     { 0x0400, "Video controller", "video"},
1510     { 0x0401, "Audio controller", "sound"},
1511     { 0x0402, "Phone"},
1512     { 0x0403, "Audio controller", "sound"},
1513     { 0x0480, "Multimedia controller"},
1514     { 0x0500, "RAM controller", "memory"},
1515     { 0x0501, "Flash controller", "flash"},
1516     { 0x0580, "Memory controller"},
1517     { 0x0600, "Host bridge", "host"},
1518     { 0x0601, "ISA bridge", "isa"},
1519     { 0x0602, "EISA bridge", "eisa"},
1520     { 0x0603, "MC bridge", "mca"},
1521     { 0x0604, "PCI bridge", "pci-bridge"},
1522     { 0x0605, "PCMCIA bridge", "pcmcia"},
1523     { 0x0606, "NUBUS bridge", "nubus"},
1524     { 0x0607, "CARDBUS bridge", "cardbus"},
1525     { 0x0608, "RACEWAY bridge"},
1526     { 0x0680, "Bridge"},
1527     { 0x0700, "Serial port", "serial"},
1528     { 0x0701, "Parallel port", "parallel"},
1529     { 0x0800, "Interrupt controller", "interrupt-controller"},
1530     { 0x0801, "DMA controller", "dma-controller"},
1531     { 0x0802, "Timer", "timer"},
1532     { 0x0803, "RTC", "rtc"},
1533     { 0x0900, "Keyboard", "keyboard"},
1534     { 0x0901, "Pen", "pen"},
1535     { 0x0902, "Mouse", "mouse"},
1536     { 0x0A00, "Dock station", "dock", 0x00ff},
1537     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1538     { 0x0c00, "Fireware contorller", "fireware"},
1539     { 0x0c01, "Access bus controller", "access-bus"},
1540     { 0x0c02, "SSA controller", "ssa"},
1541     { 0x0c03, "USB controller", "usb"},
1542     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1543     { 0x0c05, "SMBus"},
1544     { 0, NULL}
1545 };
1546 
1547 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1548                                                   void (*fn)(PCIBus *b,
1549                                                              PCIDevice *d,
1550                                                              void *opaque),
1551                                                   void *opaque)
1552 {
1553     PCIDevice *d;
1554     int devfn;
1555 
1556     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1557         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1558         if (d) {
1559             fn(bus, d, opaque);
1560         }
1561     }
1562 }
1563 
1564 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1565                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1566                          void *opaque)
1567 {
1568     bus = pci_find_bus_nr(bus, bus_num);
1569 
1570     if (bus) {
1571         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1572     }
1573 }
1574 
1575 static void pci_for_each_device_under_bus(PCIBus *bus,
1576                                           void (*fn)(PCIBus *b, PCIDevice *d,
1577                                                      void *opaque),
1578                                           void *opaque)
1579 {
1580     PCIDevice *d;
1581     int devfn;
1582 
1583     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1584         d = bus->devices[devfn];
1585         if (d) {
1586             fn(bus, d, opaque);
1587         }
1588     }
1589 }
1590 
1591 void pci_for_each_device(PCIBus *bus, int bus_num,
1592                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1593                          void *opaque)
1594 {
1595     bus = pci_find_bus_nr(bus, bus_num);
1596 
1597     if (bus) {
1598         pci_for_each_device_under_bus(bus, fn, opaque);
1599     }
1600 }
1601 
1602 static const pci_class_desc *get_class_desc(int class)
1603 {
1604     const pci_class_desc *desc;
1605 
1606     desc = pci_class_descriptions;
1607     while (desc->desc && class != desc->class) {
1608         desc++;
1609     }
1610 
1611     return desc;
1612 }
1613 
1614 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1615 
1616 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1617 {
1618     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1619     int i;
1620 
1621     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1622         const PCIIORegion *r = &dev->io_regions[i];
1623         PciMemoryRegionList *region;
1624 
1625         if (!r->size) {
1626             continue;
1627         }
1628 
1629         region = g_malloc0(sizeof(*region));
1630         region->value = g_malloc0(sizeof(*region->value));
1631 
1632         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1633             region->value->type = g_strdup("io");
1634         } else {
1635             region->value->type = g_strdup("memory");
1636             region->value->has_prefetch = true;
1637             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1638             region->value->has_mem_type_64 = true;
1639             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1640         }
1641 
1642         region->value->bar = i;
1643         region->value->address = r->addr;
1644         region->value->size = r->size;
1645 
1646         /* XXX: waiting for the qapi to support GSList */
1647         if (!cur_item) {
1648             head = cur_item = region;
1649         } else {
1650             cur_item->next = region;
1651             cur_item = region;
1652         }
1653     }
1654 
1655     return head;
1656 }
1657 
1658 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1659                                            int bus_num)
1660 {
1661     PciBridgeInfo *info;
1662     PciMemoryRange *range;
1663 
1664     info = g_new0(PciBridgeInfo, 1);
1665 
1666     info->bus = g_new0(PciBusInfo, 1);
1667     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1668     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1669     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1670 
1671     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1672     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1673     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1674 
1675     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1676     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1677     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1678 
1679     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1680     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1681     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1682 
1683     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1684         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1685         if (child_bus) {
1686             info->has_devices = true;
1687             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1688         }
1689     }
1690 
1691     return info;
1692 }
1693 
1694 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1695                                            int bus_num)
1696 {
1697     const pci_class_desc *desc;
1698     PciDeviceInfo *info;
1699     uint8_t type;
1700     int class;
1701 
1702     info = g_new0(PciDeviceInfo, 1);
1703     info->bus = bus_num;
1704     info->slot = PCI_SLOT(dev->devfn);
1705     info->function = PCI_FUNC(dev->devfn);
1706 
1707     info->class_info = g_new0(PciDeviceClass, 1);
1708     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1709     info->class_info->q_class = class;
1710     desc = get_class_desc(class);
1711     if (desc->desc) {
1712         info->class_info->has_desc = true;
1713         info->class_info->desc = g_strdup(desc->desc);
1714     }
1715 
1716     info->id = g_new0(PciDeviceId, 1);
1717     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1718     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1719     info->regions = qmp_query_pci_regions(dev);
1720     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1721 
1722     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1723         info->has_irq = true;
1724         info->irq = dev->config[PCI_INTERRUPT_LINE];
1725     }
1726 
1727     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1728     if (type == PCI_HEADER_TYPE_BRIDGE) {
1729         info->has_pci_bridge = true;
1730         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1731     }
1732 
1733     return info;
1734 }
1735 
1736 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1737 {
1738     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1739     PCIDevice *dev;
1740     int devfn;
1741 
1742     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1743         dev = bus->devices[devfn];
1744         if (dev) {
1745             info = g_malloc0(sizeof(*info));
1746             info->value = qmp_query_pci_device(dev, bus, bus_num);
1747 
1748             /* XXX: waiting for the qapi to support GSList */
1749             if (!cur_item) {
1750                 head = cur_item = info;
1751             } else {
1752                 cur_item->next = info;
1753                 cur_item = info;
1754             }
1755         }
1756     }
1757 
1758     return head;
1759 }
1760 
1761 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1762 {
1763     PciInfo *info = NULL;
1764 
1765     bus = pci_find_bus_nr(bus, bus_num);
1766     if (bus) {
1767         info = g_malloc0(sizeof(*info));
1768         info->bus = bus_num;
1769         info->devices = qmp_query_pci_devices(bus, bus_num);
1770     }
1771 
1772     return info;
1773 }
1774 
1775 PciInfoList *qmp_query_pci(Error **errp)
1776 {
1777     PciInfoList *info, *head = NULL, *cur_item = NULL;
1778     PCIHostState *host_bridge;
1779 
1780     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1781         info = g_malloc0(sizeof(*info));
1782         info->value = qmp_query_pci_bus(host_bridge->bus,
1783                                         pci_bus_num(host_bridge->bus));
1784 
1785         /* XXX: waiting for the qapi to support GSList */
1786         if (!cur_item) {
1787             head = cur_item = info;
1788         } else {
1789             cur_item->next = info;
1790             cur_item = info;
1791         }
1792     }
1793 
1794     return head;
1795 }
1796 
1797 static const char * const pci_nic_models[] = {
1798     "ne2k_pci",
1799     "i82551",
1800     "i82557b",
1801     "i82559er",
1802     "rtl8139",
1803     "e1000",
1804     "pcnet",
1805     "virtio",
1806     NULL
1807 };
1808 
1809 static const char * const pci_nic_names[] = {
1810     "ne2k_pci",
1811     "i82551",
1812     "i82557b",
1813     "i82559er",
1814     "rtl8139",
1815     "e1000",
1816     "pcnet",
1817     "virtio-net-pci",
1818     NULL
1819 };
1820 
1821 /* Initialize a PCI NIC.  */
1822 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1823                                const char *default_model,
1824                                const char *default_devaddr)
1825 {
1826     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1827     PCIBus *bus;
1828     PCIDevice *pci_dev;
1829     DeviceState *dev;
1830     int devfn;
1831     int i;
1832 
1833     if (qemu_show_nic_models(nd->model, pci_nic_models)) {
1834         exit(0);
1835     }
1836 
1837     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1838     if (i < 0) {
1839         exit(1);
1840     }
1841 
1842     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1843     if (!bus) {
1844         error_report("Invalid PCI device address %s for device %s",
1845                      devaddr, pci_nic_names[i]);
1846         exit(1);
1847     }
1848 
1849     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1850     dev = &pci_dev->qdev;
1851     qdev_set_nic_properties(dev, nd);
1852     qdev_init_nofail(dev);
1853 
1854     return pci_dev;
1855 }
1856 
1857 PCIDevice *pci_vga_init(PCIBus *bus)
1858 {
1859     switch (vga_interface_type) {
1860     case VGA_CIRRUS:
1861         return pci_create_simple(bus, -1, "cirrus-vga");
1862     case VGA_QXL:
1863         return pci_create_simple(bus, -1, "qxl-vga");
1864     case VGA_STD:
1865         return pci_create_simple(bus, -1, "VGA");
1866     case VGA_VMWARE:
1867         return pci_create_simple(bus, -1, "vmware-svga");
1868     case VGA_VIRTIO:
1869         return pci_create_simple(bus, -1, "virtio-vga");
1870     case VGA_NONE:
1871     default: /* Other non-PCI types. Checking for unsupported types is already
1872                 done in vl.c. */
1873         return NULL;
1874     }
1875 }
1876 
1877 /* Whether a given bus number is in range of the secondary
1878  * bus of the given bridge device. */
1879 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1880 {
1881     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1882              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1883         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1884         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1885 }
1886 
1887 /* Whether a given bus number is in a range of a root bus */
1888 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1889 {
1890     int i;
1891 
1892     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1893         PCIDevice *dev = bus->devices[i];
1894 
1895         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1896             if (pci_secondary_bus_in_range(dev, bus_num)) {
1897                 return true;
1898             }
1899         }
1900     }
1901 
1902     return false;
1903 }
1904 
1905 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1906 {
1907     PCIBus *sec;
1908 
1909     if (!bus) {
1910         return NULL;
1911     }
1912 
1913     if (pci_bus_num(bus) == bus_num) {
1914         return bus;
1915     }
1916 
1917     /* Consider all bus numbers in range for the host pci bridge. */
1918     if (!pci_bus_is_root(bus) &&
1919         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1920         return NULL;
1921     }
1922 
1923     /* try child bus */
1924     for (; bus; bus = sec) {
1925         QLIST_FOREACH(sec, &bus->child, sibling) {
1926             if (pci_bus_num(sec) == bus_num) {
1927                 return sec;
1928             }
1929             /* PXB buses assumed to be children of bus 0 */
1930             if (pci_bus_is_root(sec)) {
1931                 if (pci_root_bus_in_range(sec, bus_num)) {
1932                     break;
1933                 }
1934             } else {
1935                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1936                     break;
1937                 }
1938             }
1939         }
1940     }
1941 
1942     return NULL;
1943 }
1944 
1945 void pci_for_each_bus_depth_first(PCIBus *bus,
1946                                   void *(*begin)(PCIBus *bus, void *parent_state),
1947                                   void (*end)(PCIBus *bus, void *state),
1948                                   void *parent_state)
1949 {
1950     PCIBus *sec;
1951     void *state;
1952 
1953     if (!bus) {
1954         return;
1955     }
1956 
1957     if (begin) {
1958         state = begin(bus, parent_state);
1959     } else {
1960         state = parent_state;
1961     }
1962 
1963     QLIST_FOREACH(sec, &bus->child, sibling) {
1964         pci_for_each_bus_depth_first(sec, begin, end, state);
1965     }
1966 
1967     if (end) {
1968         end(bus, state);
1969     }
1970 }
1971 
1972 
1973 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1974 {
1975     bus = pci_find_bus_nr(bus, bus_num);
1976 
1977     if (!bus)
1978         return NULL;
1979 
1980     return bus->devices[devfn];
1981 }
1982 
1983 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
1984 {
1985     PCIDevice *pci_dev = (PCIDevice *)qdev;
1986     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1987     Error *local_err = NULL;
1988     PCIBus *bus;
1989     bool is_default_rom;
1990 
1991     /* initialize cap_present for pci_is_express() and pci_config_size() */
1992     if (pc->is_express) {
1993         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1994     }
1995 
1996     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1997     pci_dev = do_pci_register_device(pci_dev, bus,
1998                                      object_get_typename(OBJECT(qdev)),
1999                                      pci_dev->devfn, errp);
2000     if (pci_dev == NULL)
2001         return;
2002 
2003     if (pc->realize) {
2004         pc->realize(pci_dev, &local_err);
2005         if (local_err) {
2006             error_propagate(errp, local_err);
2007             do_pci_unregister_device(pci_dev);
2008             return;
2009         }
2010     }
2011 
2012     /* rom loading */
2013     is_default_rom = false;
2014     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2015         pci_dev->romfile = g_strdup(pc->romfile);
2016         is_default_rom = true;
2017     }
2018 
2019     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2020     if (local_err) {
2021         error_propagate(errp, local_err);
2022         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2023         return;
2024     }
2025 }
2026 
2027 static void pci_default_realize(PCIDevice *dev, Error **errp)
2028 {
2029     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2030 
2031     if (pc->init) {
2032         if (pc->init(dev) < 0) {
2033             error_setg(errp, "Device initialization failed");
2034             return;
2035         }
2036     }
2037 }
2038 
2039 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2040                                     const char *name)
2041 {
2042     DeviceState *dev;
2043 
2044     dev = qdev_create(&bus->qbus, name);
2045     qdev_prop_set_int32(dev, "addr", devfn);
2046     qdev_prop_set_bit(dev, "multifunction", multifunction);
2047     return PCI_DEVICE(dev);
2048 }
2049 
2050 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2051                                            bool multifunction,
2052                                            const char *name)
2053 {
2054     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2055     qdev_init_nofail(&dev->qdev);
2056     return dev;
2057 }
2058 
2059 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2060 {
2061     return pci_create_multifunction(bus, devfn, false, name);
2062 }
2063 
2064 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2065 {
2066     return pci_create_simple_multifunction(bus, devfn, false, name);
2067 }
2068 
2069 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2070 {
2071     int offset = PCI_CONFIG_HEADER_SIZE;
2072     int i;
2073     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2074         if (pdev->used[i])
2075             offset = i + 1;
2076         else if (i - offset + 1 == size)
2077             return offset;
2078     }
2079     return 0;
2080 }
2081 
2082 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2083                                         uint8_t *prev_p)
2084 {
2085     uint8_t next, prev;
2086 
2087     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2088         return 0;
2089 
2090     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2091          prev = next + PCI_CAP_LIST_NEXT)
2092         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2093             break;
2094 
2095     if (prev_p)
2096         *prev_p = prev;
2097     return next;
2098 }
2099 
2100 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2101 {
2102     uint8_t next, prev, found = 0;
2103 
2104     if (!(pdev->used[offset])) {
2105         return 0;
2106     }
2107 
2108     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2109 
2110     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2111          prev = next + PCI_CAP_LIST_NEXT) {
2112         if (next <= offset && next > found) {
2113             found = next;
2114         }
2115     }
2116     return found;
2117 }
2118 
2119 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2120    This is needed for an option rom which is used for more than one device. */
2121 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2122 {
2123     uint16_t vendor_id;
2124     uint16_t device_id;
2125     uint16_t rom_vendor_id;
2126     uint16_t rom_device_id;
2127     uint16_t rom_magic;
2128     uint16_t pcir_offset;
2129     uint8_t checksum;
2130 
2131     /* Words in rom data are little endian (like in PCI configuration),
2132        so they can be read / written with pci_get_word / pci_set_word. */
2133 
2134     /* Only a valid rom will be patched. */
2135     rom_magic = pci_get_word(ptr);
2136     if (rom_magic != 0xaa55) {
2137         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2138         return;
2139     }
2140     pcir_offset = pci_get_word(ptr + 0x18);
2141     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2142         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2143         return;
2144     }
2145 
2146     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2147     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2148     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2149     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2150 
2151     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2152                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2153 
2154     checksum = ptr[6];
2155 
2156     if (vendor_id != rom_vendor_id) {
2157         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2158         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2159         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2160         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2161         ptr[6] = checksum;
2162         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2163     }
2164 
2165     if (device_id != rom_device_id) {
2166         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2167         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2168         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2169         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2170         ptr[6] = checksum;
2171         pci_set_word(ptr + pcir_offset + 6, device_id);
2172     }
2173 }
2174 
2175 /* Add an option rom for the device */
2176 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2177                                Error **errp)
2178 {
2179     int size;
2180     char *path;
2181     void *ptr;
2182     char name[32];
2183     const VMStateDescription *vmsd;
2184 
2185     if (!pdev->romfile)
2186         return;
2187     if (strlen(pdev->romfile) == 0)
2188         return;
2189 
2190     if (!pdev->rom_bar) {
2191         /*
2192          * Load rom via fw_cfg instead of creating a rom bar,
2193          * for 0.11 compatibility.
2194          */
2195         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2196 
2197         /*
2198          * Hot-plugged devices can't use the option ROM
2199          * if the rom bar is disabled.
2200          */
2201         if (DEVICE(pdev)->hotplugged) {
2202             error_setg(errp, "Hot-plugged device without ROM bar"
2203                        " can't have an option ROM");
2204             return;
2205         }
2206 
2207         if (class == 0x0300) {
2208             rom_add_vga(pdev->romfile);
2209         } else {
2210             rom_add_option(pdev->romfile, -1);
2211         }
2212         return;
2213     }
2214 
2215     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2216     if (path == NULL) {
2217         path = g_strdup(pdev->romfile);
2218     }
2219 
2220     size = get_image_size(path);
2221     if (size < 0) {
2222         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2223         g_free(path);
2224         return;
2225     } else if (size == 0) {
2226         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2227         g_free(path);
2228         return;
2229     }
2230     size = pow2ceil(size);
2231 
2232     vmsd = qdev_get_vmsd(DEVICE(pdev));
2233 
2234     if (vmsd) {
2235         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2236     } else {
2237         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2238     }
2239     pdev->has_rom = true;
2240     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2241     ptr = memory_region_get_ram_ptr(&pdev->rom);
2242     load_image(path, ptr);
2243     g_free(path);
2244 
2245     if (is_default_rom) {
2246         /* Only the default rom images will be patched (if needed). */
2247         pci_patch_ids(pdev, ptr, size);
2248     }
2249 
2250     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2251 }
2252 
2253 static void pci_del_option_rom(PCIDevice *pdev)
2254 {
2255     if (!pdev->has_rom)
2256         return;
2257 
2258     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2259     pdev->has_rom = false;
2260 }
2261 
2262 /*
2263  * On success, pci_add_capability() returns a positive value
2264  * that the offset of the pci capability.
2265  * On failure, it sets an error and returns a negative error
2266  * code.
2267  */
2268 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2269                        uint8_t offset, uint8_t size,
2270                        Error **errp)
2271 {
2272     uint8_t *config;
2273     int i, overlapping_cap;
2274 
2275     if (!offset) {
2276         offset = pci_find_space(pdev, size);
2277         /* out of PCI config space is programming error */
2278         assert(offset);
2279     } else {
2280         /* Verify that capabilities don't overlap.  Note: device assignment
2281          * depends on this check to verify that the device is not broken.
2282          * Should never trigger for emulated devices, but it's helpful
2283          * for debugging these. */
2284         for (i = offset; i < offset + size; i++) {
2285             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2286             if (overlapping_cap) {
2287                 error_setg(errp, "%s:%02x:%02x.%x "
2288                            "Attempt to add PCI capability %x at offset "
2289                            "%x overlaps existing capability %x at offset %x",
2290                            pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2291                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2292                            cap_id, offset, overlapping_cap, i);
2293                 return -EINVAL;
2294             }
2295         }
2296     }
2297 
2298     config = pdev->config + offset;
2299     config[PCI_CAP_LIST_ID] = cap_id;
2300     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2301     pdev->config[PCI_CAPABILITY_LIST] = offset;
2302     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2303     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2304     /* Make capability read-only by default */
2305     memset(pdev->wmask + offset, 0, size);
2306     /* Check capability by default */
2307     memset(pdev->cmask + offset, 0xFF, size);
2308     return offset;
2309 }
2310 
2311 /* Unlink capability from the pci config space. */
2312 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2313 {
2314     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2315     if (!offset)
2316         return;
2317     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2318     /* Make capability writable again */
2319     memset(pdev->wmask + offset, 0xff, size);
2320     memset(pdev->w1cmask + offset, 0, size);
2321     /* Clear cmask as device-specific registers can't be checked */
2322     memset(pdev->cmask + offset, 0, size);
2323     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2324 
2325     if (!pdev->config[PCI_CAPABILITY_LIST])
2326         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2327 }
2328 
2329 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2330 {
2331     return pci_find_capability_list(pdev, cap_id, NULL);
2332 }
2333 
2334 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2335 {
2336     PCIDevice *d = (PCIDevice *)dev;
2337     const pci_class_desc *desc;
2338     char ctxt[64];
2339     PCIIORegion *r;
2340     int i, class;
2341 
2342     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2343     desc = pci_class_descriptions;
2344     while (desc->desc && class != desc->class)
2345         desc++;
2346     if (desc->desc) {
2347         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2348     } else {
2349         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2350     }
2351 
2352     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2353                    "pci id %04x:%04x (sub %04x:%04x)\n",
2354                    indent, "", ctxt, pci_bus_num(d->bus),
2355                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2356                    pci_get_word(d->config + PCI_VENDOR_ID),
2357                    pci_get_word(d->config + PCI_DEVICE_ID),
2358                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2359                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2360     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2361         r = &d->io_regions[i];
2362         if (!r->size)
2363             continue;
2364         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2365                        " [0x%"FMT_PCIBUS"]\n",
2366                        indent, "",
2367                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2368                        r->addr, r->addr + r->size - 1);
2369     }
2370 }
2371 
2372 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2373 {
2374     PCIDevice *d = (PCIDevice *)dev;
2375     const char *name = NULL;
2376     const pci_class_desc *desc =  pci_class_descriptions;
2377     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2378 
2379     while (desc->desc &&
2380           (class & ~desc->fw_ign_bits) !=
2381           (desc->class & ~desc->fw_ign_bits)) {
2382         desc++;
2383     }
2384 
2385     if (desc->desc) {
2386         name = desc->fw_name;
2387     }
2388 
2389     if (name) {
2390         pstrcpy(buf, len, name);
2391     } else {
2392         snprintf(buf, len, "pci%04x,%04x",
2393                  pci_get_word(d->config + PCI_VENDOR_ID),
2394                  pci_get_word(d->config + PCI_DEVICE_ID));
2395     }
2396 
2397     return buf;
2398 }
2399 
2400 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2401 {
2402     PCIDevice *d = (PCIDevice *)dev;
2403     char path[50], name[33];
2404     int off;
2405 
2406     off = snprintf(path, sizeof(path), "%s@%x",
2407                    pci_dev_fw_name(dev, name, sizeof name),
2408                    PCI_SLOT(d->devfn));
2409     if (PCI_FUNC(d->devfn))
2410         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2411     return g_strdup(path);
2412 }
2413 
2414 static char *pcibus_get_dev_path(DeviceState *dev)
2415 {
2416     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2417     PCIDevice *t;
2418     int slot_depth;
2419     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2420      * 00 is added here to make this format compatible with
2421      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2422      * Slot.Function list specifies the slot and function numbers for all
2423      * devices on the path from root to the specific device. */
2424     const char *root_bus_path;
2425     int root_bus_len;
2426     char slot[] = ":SS.F";
2427     int slot_len = sizeof slot - 1 /* For '\0' */;
2428     int path_len;
2429     char *path, *p;
2430     int s;
2431 
2432     root_bus_path = pci_root_bus_path(d);
2433     root_bus_len = strlen(root_bus_path);
2434 
2435     /* Calculate # of slots on path between device and root. */;
2436     slot_depth = 0;
2437     for (t = d; t; t = t->bus->parent_dev) {
2438         ++slot_depth;
2439     }
2440 
2441     path_len = root_bus_len + slot_len * slot_depth;
2442 
2443     /* Allocate memory, fill in the terminating null byte. */
2444     path = g_malloc(path_len + 1 /* For '\0' */);
2445     path[path_len] = '\0';
2446 
2447     memcpy(path, root_bus_path, root_bus_len);
2448 
2449     /* Fill in slot numbers. We walk up from device to root, so need to print
2450      * them in the reverse order, last to first. */
2451     p = path + path_len;
2452     for (t = d; t; t = t->bus->parent_dev) {
2453         p -= slot_len;
2454         s = snprintf(slot, sizeof slot, ":%02x.%x",
2455                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2456         assert(s == slot_len);
2457         memcpy(p, slot, slot_len);
2458     }
2459 
2460     return path;
2461 }
2462 
2463 static int pci_qdev_find_recursive(PCIBus *bus,
2464                                    const char *id, PCIDevice **pdev)
2465 {
2466     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2467     if (!qdev) {
2468         return -ENODEV;
2469     }
2470 
2471     /* roughly check if given qdev is pci device */
2472     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2473         *pdev = PCI_DEVICE(qdev);
2474         return 0;
2475     }
2476     return -EINVAL;
2477 }
2478 
2479 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2480 {
2481     PCIHostState *host_bridge;
2482     int rc = -ENODEV;
2483 
2484     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2485         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2486         if (!tmp) {
2487             rc = 0;
2488             break;
2489         }
2490         if (tmp != -ENODEV) {
2491             rc = tmp;
2492         }
2493     }
2494 
2495     return rc;
2496 }
2497 
2498 MemoryRegion *pci_address_space(PCIDevice *dev)
2499 {
2500     return dev->bus->address_space_mem;
2501 }
2502 
2503 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2504 {
2505     return dev->bus->address_space_io;
2506 }
2507 
2508 static void pci_device_class_init(ObjectClass *klass, void *data)
2509 {
2510     DeviceClass *k = DEVICE_CLASS(klass);
2511     PCIDeviceClass *pc = PCI_DEVICE_CLASS(klass);
2512 
2513     k->realize = pci_qdev_realize;
2514     k->unrealize = pci_qdev_unrealize;
2515     k->bus_type = TYPE_PCI_BUS;
2516     k->props = pci_props;
2517     pc->realize = pci_default_realize;
2518 }
2519 
2520 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2521 {
2522     PCIBus *bus = PCI_BUS(dev->bus);
2523     PCIBus *iommu_bus = bus;
2524 
2525     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2526         iommu_bus = PCI_BUS(iommu_bus->parent_dev->bus);
2527     }
2528     if (iommu_bus && iommu_bus->iommu_fn) {
2529         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2530     }
2531     return &address_space_memory;
2532 }
2533 
2534 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2535 {
2536     bus->iommu_fn = fn;
2537     bus->iommu_opaque = opaque;
2538 }
2539 
2540 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2541 {
2542     Range *range = opaque;
2543     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2544     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2545     int i;
2546 
2547     if (!(cmd & PCI_COMMAND_MEMORY)) {
2548         return;
2549     }
2550 
2551     if (pc->is_bridge) {
2552         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2553         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2554 
2555         base = MAX(base, 0x1ULL << 32);
2556 
2557         if (limit >= base) {
2558             Range pref_range;
2559             range_set_bounds(&pref_range, base, limit);
2560             range_extend(range, &pref_range);
2561         }
2562     }
2563     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2564         PCIIORegion *r = &dev->io_regions[i];
2565         pcibus_t lob, upb;
2566         Range region_range;
2567 
2568         if (!r->size ||
2569             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2570             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2571             continue;
2572         }
2573 
2574         lob = pci_bar_address(dev, i, r->type, r->size);
2575         upb = lob + r->size - 1;
2576         if (lob == PCI_BAR_UNMAPPED) {
2577             continue;
2578         }
2579 
2580         lob = MAX(lob, 0x1ULL << 32);
2581 
2582         if (upb >= lob) {
2583             range_set_bounds(&region_range, lob, upb);
2584             range_extend(range, &region_range);
2585         }
2586     }
2587 }
2588 
2589 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2590 {
2591     range_make_empty(range);
2592     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2593 }
2594 
2595 static bool pcie_has_upstream_port(PCIDevice *dev)
2596 {
2597     PCIDevice *parent_dev = pci_bridge_get_device(dev->bus);
2598 
2599     /* Device associated with an upstream port.
2600      * As there are several types of these, it's easier to check the
2601      * parent device: upstream ports are always connected to
2602      * root or downstream ports.
2603      */
2604     return parent_dev &&
2605         pci_is_express(parent_dev) &&
2606         parent_dev->exp.exp_cap &&
2607         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2608          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2609 }
2610 
2611 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2612 {
2613     if(pcie_has_upstream_port(pci_dev)) {
2614         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2615         return pci_dev->bus->devices[0];
2616     } else {
2617         /* Other bus types might support multiple devices at slots 0-31 */
2618         return pci_dev->bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2619     }
2620 }
2621 
2622 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2623 {
2624     MSIMessage msg;
2625     if (msix_enabled(dev)) {
2626         msg = msix_get_message(dev, vector);
2627     } else if (msi_enabled(dev)) {
2628         msg = msi_get_message(dev, vector);
2629     } else {
2630         /* Should never happen */
2631         error_report("%s: unknown interrupt type", __func__);
2632         abort();
2633     }
2634     return msg;
2635 }
2636 
2637 static const TypeInfo pci_device_type_info = {
2638     .name = TYPE_PCI_DEVICE,
2639     .parent = TYPE_DEVICE,
2640     .instance_size = sizeof(PCIDevice),
2641     .abstract = true,
2642     .class_size = sizeof(PCIDeviceClass),
2643     .class_init = pci_device_class_init,
2644 };
2645 
2646 static void pci_register_types(void)
2647 {
2648     type_register_static(&pci_bus_info);
2649     type_register_static(&pcie_bus_info);
2650     type_register_static(&pci_device_type_info);
2651 }
2652 
2653 type_init(pci_register_types)
2654