xref: /openbmc/qemu/hw/pci/pci.c (revision 9cbb6362)
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_interface_info = {
174     .name          = INTERFACE_PCIE_DEVICE,
175     .parent        = TYPE_INTERFACE,
176 };
177 
178 static const TypeInfo conventional_pci_interface_info = {
179     .name          = INTERFACE_CONVENTIONAL_PCI_DEVICE,
180     .parent        = TYPE_INTERFACE,
181 };
182 
183 static const TypeInfo pcie_bus_info = {
184     .name = TYPE_PCIE_BUS,
185     .parent = TYPE_PCI_BUS,
186 };
187 
188 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
189 static void pci_update_mappings(PCIDevice *d);
190 static void pci_irq_handler(void *opaque, int irq_num, int level);
191 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom, Error **);
192 static void pci_del_option_rom(PCIDevice *pdev);
193 
194 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
195 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
196 
197 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
198 
199 int pci_bar(PCIDevice *d, int reg)
200 {
201     uint8_t type;
202 
203     if (reg != PCI_ROM_SLOT)
204         return PCI_BASE_ADDRESS_0 + reg * 4;
205 
206     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
207     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
208 }
209 
210 static inline int pci_irq_state(PCIDevice *d, int irq_num)
211 {
212 	return (d->irq_state >> irq_num) & 0x1;
213 }
214 
215 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
216 {
217 	d->irq_state &= ~(0x1 << irq_num);
218 	d->irq_state |= level << irq_num;
219 }
220 
221 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
222 {
223     PCIBus *bus;
224     for (;;) {
225         bus = pci_get_bus(pci_dev);
226         irq_num = bus->map_irq(pci_dev, irq_num);
227         if (bus->set_irq)
228             break;
229         pci_dev = bus->parent_dev;
230     }
231     bus->irq_count[irq_num] += change;
232     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
233 }
234 
235 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
236 {
237     assert(irq_num >= 0);
238     assert(irq_num < bus->nirq);
239     return !!bus->irq_count[irq_num];
240 }
241 
242 /* Update interrupt status bit in config space on interrupt
243  * state change. */
244 static void pci_update_irq_status(PCIDevice *dev)
245 {
246     if (dev->irq_state) {
247         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
248     } else {
249         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
250     }
251 }
252 
253 void pci_device_deassert_intx(PCIDevice *dev)
254 {
255     int i;
256     for (i = 0; i < PCI_NUM_PINS; ++i) {
257         pci_irq_handler(dev, i, 0);
258     }
259 }
260 
261 static void pci_do_device_reset(PCIDevice *dev)
262 {
263     int r;
264 
265     pci_device_deassert_intx(dev);
266     assert(dev->irq_state == 0);
267 
268     /* Clear all writable bits */
269     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
270                                  pci_get_word(dev->wmask + PCI_COMMAND) |
271                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
272     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
273                                  pci_get_word(dev->wmask + PCI_STATUS) |
274                                  pci_get_word(dev->w1cmask + PCI_STATUS));
275     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
276     dev->config[PCI_INTERRUPT_LINE] = 0x0;
277     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
278         PCIIORegion *region = &dev->io_regions[r];
279         if (!region->size) {
280             continue;
281         }
282 
283         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
284             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
285             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
286         } else {
287             pci_set_long(dev->config + pci_bar(dev, r), region->type);
288         }
289     }
290     pci_update_mappings(dev);
291 
292     msi_reset(dev);
293     msix_reset(dev);
294 }
295 
296 /*
297  * This function is called on #RST and FLR.
298  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
299  */
300 void pci_device_reset(PCIDevice *dev)
301 {
302     qdev_reset_all(&dev->qdev);
303     pci_do_device_reset(dev);
304 }
305 
306 /*
307  * Trigger pci bus reset under a given bus.
308  * Called via qbus_reset_all on RST# assert, after the devices
309  * have been reset qdev_reset_all-ed already.
310  */
311 static void pcibus_reset(BusState *qbus)
312 {
313     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
314     int i;
315 
316     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
317         if (bus->devices[i]) {
318             pci_do_device_reset(bus->devices[i]);
319         }
320     }
321 
322     for (i = 0; i < bus->nirq; i++) {
323         assert(bus->irq_count[i] == 0);
324     }
325 }
326 
327 static void pci_host_bus_register(DeviceState *host)
328 {
329     PCIHostState *host_bridge = PCI_HOST_BRIDGE(host);
330 
331     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
332 }
333 
334 PCIBus *pci_device_root_bus(const PCIDevice *d)
335 {
336     PCIBus *bus = pci_get_bus(d);
337 
338     while (!pci_bus_is_root(bus)) {
339         d = bus->parent_dev;
340         assert(d != NULL);
341 
342         bus = pci_get_bus(d);
343     }
344 
345     return bus;
346 }
347 
348 const char *pci_root_bus_path(PCIDevice *dev)
349 {
350     PCIBus *rootbus = pci_device_root_bus(dev);
351     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
352     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
353 
354     assert(host_bridge->bus == rootbus);
355 
356     if (hc->root_bus_path) {
357         return (*hc->root_bus_path)(host_bridge, rootbus);
358     }
359 
360     return rootbus->qbus.name;
361 }
362 
363 static void pci_root_bus_init(PCIBus *bus, DeviceState *parent,
364                               MemoryRegion *address_space_mem,
365                               MemoryRegion *address_space_io,
366                               uint8_t devfn_min)
367 {
368     assert(PCI_FUNC(devfn_min) == 0);
369     bus->devfn_min = devfn_min;
370     bus->slot_reserved_mask = 0x0;
371     bus->address_space_mem = address_space_mem;
372     bus->address_space_io = address_space_io;
373 
374     /* host bridge */
375     QLIST_INIT(&bus->child);
376 
377     pci_host_bus_register(parent);
378 }
379 
380 bool pci_bus_is_express(PCIBus *bus)
381 {
382     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
383 }
384 
385 bool pci_bus_is_root(PCIBus *bus)
386 {
387     return PCI_BUS_GET_CLASS(bus)->is_root(bus);
388 }
389 
390 void pci_root_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
391                               const char *name,
392                               MemoryRegion *address_space_mem,
393                               MemoryRegion *address_space_io,
394                               uint8_t devfn_min, const char *typename)
395 {
396     qbus_create_inplace(bus, bus_size, typename, parent, name);
397     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
398                       devfn_min);
399 }
400 
401 PCIBus *pci_root_bus_new(DeviceState *parent, const char *name,
402                          MemoryRegion *address_space_mem,
403                          MemoryRegion *address_space_io,
404                          uint8_t devfn_min, const char *typename)
405 {
406     PCIBus *bus;
407 
408     bus = PCI_BUS(qbus_create(typename, parent, name));
409     pci_root_bus_init(bus, parent, address_space_mem, address_space_io,
410                       devfn_min);
411     return bus;
412 }
413 
414 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
415                   void *irq_opaque, int nirq)
416 {
417     bus->set_irq = set_irq;
418     bus->map_irq = map_irq;
419     bus->irq_opaque = irq_opaque;
420     bus->nirq = nirq;
421     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
422 }
423 
424 PCIBus *pci_register_root_bus(DeviceState *parent, const char *name,
425                               pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
426                               void *irq_opaque,
427                               MemoryRegion *address_space_mem,
428                               MemoryRegion *address_space_io,
429                               uint8_t devfn_min, int nirq,
430                               const char *typename)
431 {
432     PCIBus *bus;
433 
434     bus = pci_root_bus_new(parent, name, address_space_mem,
435                            address_space_io, devfn_min, typename);
436     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
437     return bus;
438 }
439 
440 int pci_bus_num(PCIBus *s)
441 {
442     return PCI_BUS_GET_CLASS(s)->bus_num(s);
443 }
444 
445 int pci_bus_numa_node(PCIBus *bus)
446 {
447     return PCI_BUS_GET_CLASS(bus)->numa_node(bus);
448 }
449 
450 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size,
451                                  VMStateField *field)
452 {
453     PCIDevice *s = container_of(pv, PCIDevice, config);
454     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
455     uint8_t *config;
456     int i;
457 
458     assert(size == pci_config_size(s));
459     config = g_malloc(size);
460 
461     qemu_get_buffer(f, config, size);
462     for (i = 0; i < size; ++i) {
463         if ((config[i] ^ s->config[i]) &
464             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
465             error_report("%s: Bad config data: i=0x%x read: %x device: %x "
466                          "cmask: %x wmask: %x w1cmask:%x", __func__,
467                          i, config[i], s->config[i],
468                          s->cmask[i], s->wmask[i], s->w1cmask[i]);
469             g_free(config);
470             return -EINVAL;
471         }
472     }
473     memcpy(s->config, config, size);
474 
475     pci_update_mappings(s);
476     if (pc->is_bridge) {
477         PCIBridge *b = PCI_BRIDGE(s);
478         pci_bridge_update_mappings(b);
479     }
480 
481     memory_region_set_enabled(&s->bus_master_enable_region,
482                               pci_get_word(s->config + PCI_COMMAND)
483                               & PCI_COMMAND_MASTER);
484 
485     g_free(config);
486     return 0;
487 }
488 
489 /* just put buffer */
490 static int put_pci_config_device(QEMUFile *f, void *pv, size_t size,
491                                  VMStateField *field, QJSON *vmdesc)
492 {
493     const uint8_t **v = pv;
494     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
495     qemu_put_buffer(f, *v, size);
496 
497     return 0;
498 }
499 
500 static VMStateInfo vmstate_info_pci_config = {
501     .name = "pci config",
502     .get  = get_pci_config_device,
503     .put  = put_pci_config_device,
504 };
505 
506 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size,
507                              VMStateField *field)
508 {
509     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
510     uint32_t irq_state[PCI_NUM_PINS];
511     int i;
512     for (i = 0; i < PCI_NUM_PINS; ++i) {
513         irq_state[i] = qemu_get_be32(f);
514         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
515             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
516                     irq_state[i]);
517             return -EINVAL;
518         }
519     }
520 
521     for (i = 0; i < PCI_NUM_PINS; ++i) {
522         pci_set_irq_state(s, i, irq_state[i]);
523     }
524 
525     return 0;
526 }
527 
528 static int put_pci_irq_state(QEMUFile *f, void *pv, size_t size,
529                              VMStateField *field, QJSON *vmdesc)
530 {
531     int i;
532     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
533 
534     for (i = 0; i < PCI_NUM_PINS; ++i) {
535         qemu_put_be32(f, pci_irq_state(s, i));
536     }
537 
538     return 0;
539 }
540 
541 static VMStateInfo vmstate_info_pci_irq_state = {
542     .name = "pci irq state",
543     .get  = get_pci_irq_state,
544     .put  = put_pci_irq_state,
545 };
546 
547 static bool migrate_is_pcie(void *opaque, int version_id)
548 {
549     return pci_is_express((PCIDevice *)opaque);
550 }
551 
552 static bool migrate_is_not_pcie(void *opaque, int version_id)
553 {
554     return !pci_is_express((PCIDevice *)opaque);
555 }
556 
557 const VMStateDescription vmstate_pci_device = {
558     .name = "PCIDevice",
559     .version_id = 2,
560     .minimum_version_id = 1,
561     .fields = (VMStateField[]) {
562         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
563         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
564                                    migrate_is_not_pcie,
565                                    0, vmstate_info_pci_config,
566                                    PCI_CONFIG_SPACE_SIZE),
567         VMSTATE_BUFFER_UNSAFE_INFO_TEST(config, PCIDevice,
568                                    migrate_is_pcie,
569                                    0, vmstate_info_pci_config,
570                                    PCIE_CONFIG_SPACE_SIZE),
571         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
572 				   vmstate_info_pci_irq_state,
573 				   PCI_NUM_PINS * sizeof(int32_t)),
574         VMSTATE_END_OF_LIST()
575     }
576 };
577 
578 
579 void pci_device_save(PCIDevice *s, QEMUFile *f)
580 {
581     /* Clear interrupt status bit: it is implicit
582      * in irq_state which we are saving.
583      * This makes us compatible with old devices
584      * which never set or clear this bit. */
585     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
586     vmstate_save_state(f, &vmstate_pci_device, s, NULL);
587     /* Restore the interrupt status bit. */
588     pci_update_irq_status(s);
589 }
590 
591 int pci_device_load(PCIDevice *s, QEMUFile *f)
592 {
593     int ret;
594     ret = vmstate_load_state(f, &vmstate_pci_device, s, s->version_id);
595     /* Restore the interrupt status bit. */
596     pci_update_irq_status(s);
597     return ret;
598 }
599 
600 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
601 {
602     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
603                  pci_default_sub_vendor_id);
604     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
605                  pci_default_sub_device_id);
606 }
607 
608 /*
609  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
610  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
611  */
612 static int pci_parse_devaddr(const char *addr, int *domp, int *busp,
613                              unsigned int *slotp, unsigned int *funcp)
614 {
615     const char *p;
616     char *e;
617     unsigned long val;
618     unsigned long dom = 0, bus = 0;
619     unsigned int slot = 0;
620     unsigned int func = 0;
621 
622     p = addr;
623     val = strtoul(p, &e, 16);
624     if (e == p)
625 	return -1;
626     if (*e == ':') {
627 	bus = val;
628 	p = e + 1;
629 	val = strtoul(p, &e, 16);
630 	if (e == p)
631 	    return -1;
632 	if (*e == ':') {
633 	    dom = bus;
634 	    bus = val;
635 	    p = e + 1;
636 	    val = strtoul(p, &e, 16);
637 	    if (e == p)
638 		return -1;
639 	}
640     }
641 
642     slot = val;
643 
644     if (funcp != NULL) {
645         if (*e != '.')
646             return -1;
647 
648         p = e + 1;
649         val = strtoul(p, &e, 16);
650         if (e == p)
651             return -1;
652 
653         func = val;
654     }
655 
656     /* if funcp == NULL func is 0 */
657     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
658 	return -1;
659 
660     if (*e)
661 	return -1;
662 
663     *domp = dom;
664     *busp = bus;
665     *slotp = slot;
666     if (funcp != NULL)
667         *funcp = func;
668     return 0;
669 }
670 
671 static PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root,
672                                  const char *devaddr)
673 {
674     int dom, bus;
675     unsigned slot;
676 
677     if (!root) {
678         fprintf(stderr, "No primary PCI bus\n");
679         return NULL;
680     }
681 
682     assert(!root->parent_dev);
683 
684     if (!devaddr) {
685         *devfnp = -1;
686         return pci_find_bus_nr(root, 0);
687     }
688 
689     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
690         return NULL;
691     }
692 
693     if (dom != 0) {
694         fprintf(stderr, "No support for non-zero PCI domains\n");
695         return NULL;
696     }
697 
698     *devfnp = PCI_DEVFN(slot, 0);
699     return pci_find_bus_nr(root, bus);
700 }
701 
702 static void pci_init_cmask(PCIDevice *dev)
703 {
704     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
705     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
706     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
707     dev->cmask[PCI_REVISION_ID] = 0xff;
708     dev->cmask[PCI_CLASS_PROG] = 0xff;
709     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
710     dev->cmask[PCI_HEADER_TYPE] = 0xff;
711     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
712 }
713 
714 static void pci_init_wmask(PCIDevice *dev)
715 {
716     int config_size = pci_config_size(dev);
717 
718     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
719     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
720     pci_set_word(dev->wmask + PCI_COMMAND,
721                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
722                  PCI_COMMAND_INTX_DISABLE);
723     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
724         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
725     }
726 
727     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
728            config_size - PCI_CONFIG_HEADER_SIZE);
729 }
730 
731 static void pci_init_w1cmask(PCIDevice *dev)
732 {
733     /*
734      * Note: It's okay to set w1cmask even for readonly bits as
735      * long as their value is hardwired to 0.
736      */
737     pci_set_word(dev->w1cmask + PCI_STATUS,
738                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
739                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
740                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
741 }
742 
743 static void pci_init_mask_bridge(PCIDevice *d)
744 {
745     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
746        PCI_SEC_LETENCY_TIMER */
747     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
748 
749     /* base and limit */
750     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
751     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
752     pci_set_word(d->wmask + PCI_MEMORY_BASE,
753                  PCI_MEMORY_RANGE_MASK & 0xffff);
754     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
755                  PCI_MEMORY_RANGE_MASK & 0xffff);
756     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
757                  PCI_PREF_RANGE_MASK & 0xffff);
758     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
759                  PCI_PREF_RANGE_MASK & 0xffff);
760 
761     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
762     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
763 
764     /* Supported memory and i/o types */
765     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
766     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
767     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
768                                PCI_PREF_RANGE_TYPE_64);
769     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
770                                PCI_PREF_RANGE_TYPE_64);
771 
772     /*
773      * TODO: Bridges default to 10-bit VGA decoding but we currently only
774      * implement 16-bit decoding (no alias support).
775      */
776     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
777                  PCI_BRIDGE_CTL_PARITY |
778                  PCI_BRIDGE_CTL_SERR |
779                  PCI_BRIDGE_CTL_ISA |
780                  PCI_BRIDGE_CTL_VGA |
781                  PCI_BRIDGE_CTL_VGA_16BIT |
782                  PCI_BRIDGE_CTL_MASTER_ABORT |
783                  PCI_BRIDGE_CTL_BUS_RESET |
784                  PCI_BRIDGE_CTL_FAST_BACK |
785                  PCI_BRIDGE_CTL_DISCARD |
786                  PCI_BRIDGE_CTL_SEC_DISCARD |
787                  PCI_BRIDGE_CTL_DISCARD_SERR);
788     /* Below does not do anything as we never set this bit, put here for
789      * completeness. */
790     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
791                  PCI_BRIDGE_CTL_DISCARD_STATUS);
792     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
793     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
794     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
795                                PCI_PREF_RANGE_TYPE_MASK);
796     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
797                                PCI_PREF_RANGE_TYPE_MASK);
798 }
799 
800 static void pci_init_multifunction(PCIBus *bus, PCIDevice *dev, Error **errp)
801 {
802     uint8_t slot = PCI_SLOT(dev->devfn);
803     uint8_t func;
804 
805     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
806         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
807     }
808 
809     /*
810      * multifunction bit is interpreted in two ways as follows.
811      *   - all functions must set the bit to 1.
812      *     Example: Intel X53
813      *   - function 0 must set the bit, but the rest function (> 0)
814      *     is allowed to leave the bit to 0.
815      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
816      *
817      * So OS (at least Linux) checks the bit of only function 0,
818      * and doesn't see the bit of function > 0.
819      *
820      * The below check allows both interpretation.
821      */
822     if (PCI_FUNC(dev->devfn)) {
823         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
824         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
825             /* function 0 should set multifunction bit */
826             error_setg(errp, "PCI: single function device can't be populated "
827                        "in function %x.%x", slot, PCI_FUNC(dev->devfn));
828             return;
829         }
830         return;
831     }
832 
833     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
834         return;
835     }
836     /* function 0 indicates single function, so function > 0 must be NULL */
837     for (func = 1; func < PCI_FUNC_MAX; ++func) {
838         if (bus->devices[PCI_DEVFN(slot, func)]) {
839             error_setg(errp, "PCI: %x.0 indicates single function, "
840                        "but %x.%x is already populated.",
841                        slot, slot, func);
842             return;
843         }
844     }
845 }
846 
847 static void pci_config_alloc(PCIDevice *pci_dev)
848 {
849     int config_size = pci_config_size(pci_dev);
850 
851     pci_dev->config = g_malloc0(config_size);
852     pci_dev->cmask = g_malloc0(config_size);
853     pci_dev->wmask = g_malloc0(config_size);
854     pci_dev->w1cmask = g_malloc0(config_size);
855     pci_dev->used = g_malloc0(config_size);
856 }
857 
858 static void pci_config_free(PCIDevice *pci_dev)
859 {
860     g_free(pci_dev->config);
861     g_free(pci_dev->cmask);
862     g_free(pci_dev->wmask);
863     g_free(pci_dev->w1cmask);
864     g_free(pci_dev->used);
865 }
866 
867 static void do_pci_unregister_device(PCIDevice *pci_dev)
868 {
869     pci_get_bus(pci_dev)->devices[pci_dev->devfn] = NULL;
870     pci_config_free(pci_dev);
871 
872     if (memory_region_is_mapped(&pci_dev->bus_master_enable_region)) {
873         memory_region_del_subregion(&pci_dev->bus_master_container_region,
874                                     &pci_dev->bus_master_enable_region);
875     }
876     address_space_destroy(&pci_dev->bus_master_as);
877 }
878 
879 /* Extract PCIReqIDCache into BDF format */
880 static uint16_t pci_req_id_cache_extract(PCIReqIDCache *cache)
881 {
882     uint8_t bus_n;
883     uint16_t result;
884 
885     switch (cache->type) {
886     case PCI_REQ_ID_BDF:
887         result = pci_get_bdf(cache->dev);
888         break;
889     case PCI_REQ_ID_SECONDARY_BUS:
890         bus_n = pci_dev_bus_num(cache->dev);
891         result = PCI_BUILD_BDF(bus_n, 0);
892         break;
893     default:
894         error_printf("Invalid PCI requester ID cache type: %d\n",
895                      cache->type);
896         exit(1);
897         break;
898     }
899 
900     return result;
901 }
902 
903 /* Parse bridges up to the root complex and return requester ID
904  * cache for specific device.  For full PCIe topology, the cache
905  * result would be exactly the same as getting BDF of the device.
906  * However, several tricks are required when system mixed up with
907  * legacy PCI devices and PCIe-to-PCI bridges.
908  *
909  * Here we cache the proxy device (and type) not requester ID since
910  * bus number might change from time to time.
911  */
912 static PCIReqIDCache pci_req_id_cache_get(PCIDevice *dev)
913 {
914     PCIDevice *parent;
915     PCIReqIDCache cache = {
916         .dev = dev,
917         .type = PCI_REQ_ID_BDF,
918     };
919 
920     while (!pci_bus_is_root(pci_get_bus(dev))) {
921         /* We are under PCI/PCIe bridges */
922         parent = pci_get_bus(dev)->parent_dev;
923         if (pci_is_express(parent)) {
924             if (pcie_cap_get_type(parent) == PCI_EXP_TYPE_PCI_BRIDGE) {
925                 /* When we pass through PCIe-to-PCI/PCIX bridges, we
926                  * override the requester ID using secondary bus
927                  * number of parent bridge with zeroed devfn
928                  * (pcie-to-pci bridge spec chap 2.3). */
929                 cache.type = PCI_REQ_ID_SECONDARY_BUS;
930                 cache.dev = dev;
931             }
932         } else {
933             /* Legacy PCI, override requester ID with the bridge's
934              * BDF upstream.  When the root complex connects to
935              * legacy PCI devices (including buses), it can only
936              * obtain requester ID info from directly attached
937              * devices.  If devices are attached under bridges, only
938              * the requester ID of the bridge that is directly
939              * attached to the root complex can be recognized. */
940             cache.type = PCI_REQ_ID_BDF;
941             cache.dev = parent;
942         }
943         dev = parent;
944     }
945 
946     return cache;
947 }
948 
949 uint16_t pci_requester_id(PCIDevice *dev)
950 {
951     return pci_req_id_cache_extract(&dev->requester_id_cache);
952 }
953 
954 static bool pci_bus_devfn_available(PCIBus *bus, int devfn)
955 {
956     return !(bus->devices[devfn]);
957 }
958 
959 static bool pci_bus_devfn_reserved(PCIBus *bus, int devfn)
960 {
961     return bus->slot_reserved_mask & (1UL << PCI_SLOT(devfn));
962 }
963 
964 /* -1 for devfn means auto assign */
965 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev,
966                                          const char *name, int devfn,
967                                          Error **errp)
968 {
969     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
970     PCIConfigReadFunc *config_read = pc->config_read;
971     PCIConfigWriteFunc *config_write = pc->config_write;
972     Error *local_err = NULL;
973     DeviceState *dev = DEVICE(pci_dev);
974     PCIBus *bus = pci_get_bus(pci_dev);
975 
976     /* Only pci bridges can be attached to extra PCI root buses */
977     if (pci_bus_is_root(bus) && bus->parent_dev && !pc->is_bridge) {
978         error_setg(errp,
979                    "PCI: Only PCI/PCIe bridges can be plugged into %s",
980                     bus->parent_dev->name);
981         return NULL;
982     }
983 
984     if (devfn < 0) {
985         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
986             devfn += PCI_FUNC_MAX) {
987             if (pci_bus_devfn_available(bus, devfn) &&
988                    !pci_bus_devfn_reserved(bus, devfn)) {
989                 goto found;
990             }
991         }
992         error_setg(errp, "PCI: no slot/function available for %s, all in use "
993                    "or reserved", name);
994         return NULL;
995     found: ;
996     } else if (pci_bus_devfn_reserved(bus, devfn)) {
997         error_setg(errp, "PCI: slot %d function %d not available for %s,"
998                    " reserved",
999                    PCI_SLOT(devfn), PCI_FUNC(devfn), name);
1000         return NULL;
1001     } else if (!pci_bus_devfn_available(bus, devfn)) {
1002         error_setg(errp, "PCI: slot %d function %d not available for %s,"
1003                    " in use by %s",
1004                    PCI_SLOT(devfn), PCI_FUNC(devfn), name,
1005                    bus->devices[devfn]->name);
1006         return NULL;
1007     } else if (dev->hotplugged &&
1008                pci_get_function_0(pci_dev)) {
1009         error_setg(errp, "PCI: slot %d function 0 already ocuppied by %s,"
1010                    " new func %s cannot be exposed to guest.",
1011                    PCI_SLOT(pci_get_function_0(pci_dev)->devfn),
1012                    pci_get_function_0(pci_dev)->name,
1013                    name);
1014 
1015        return NULL;
1016     }
1017 
1018     pci_dev->devfn = devfn;
1019     pci_dev->requester_id_cache = pci_req_id_cache_get(pci_dev);
1020     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
1021 
1022     memory_region_init(&pci_dev->bus_master_container_region, OBJECT(pci_dev),
1023                        "bus master container", UINT64_MAX);
1024     address_space_init(&pci_dev->bus_master_as,
1025                        &pci_dev->bus_master_container_region, pci_dev->name);
1026 
1027     if (qdev_hotplug) {
1028         pci_init_bus_master(pci_dev);
1029     }
1030     pci_dev->irq_state = 0;
1031     pci_config_alloc(pci_dev);
1032 
1033     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
1034     pci_config_set_device_id(pci_dev->config, pc->device_id);
1035     pci_config_set_revision(pci_dev->config, pc->revision);
1036     pci_config_set_class(pci_dev->config, pc->class_id);
1037 
1038     if (!pc->is_bridge) {
1039         if (pc->subsystem_vendor_id || pc->subsystem_id) {
1040             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
1041                          pc->subsystem_vendor_id);
1042             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
1043                          pc->subsystem_id);
1044         } else {
1045             pci_set_default_subsystem_id(pci_dev);
1046         }
1047     } else {
1048         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
1049         assert(!pc->subsystem_vendor_id);
1050         assert(!pc->subsystem_id);
1051     }
1052     pci_init_cmask(pci_dev);
1053     pci_init_wmask(pci_dev);
1054     pci_init_w1cmask(pci_dev);
1055     if (pc->is_bridge) {
1056         pci_init_mask_bridge(pci_dev);
1057     }
1058     pci_init_multifunction(bus, pci_dev, &local_err);
1059     if (local_err) {
1060         error_propagate(errp, local_err);
1061         do_pci_unregister_device(pci_dev);
1062         return NULL;
1063     }
1064 
1065     if (!config_read)
1066         config_read = pci_default_read_config;
1067     if (!config_write)
1068         config_write = pci_default_write_config;
1069     pci_dev->config_read = config_read;
1070     pci_dev->config_write = config_write;
1071     bus->devices[devfn] = pci_dev;
1072     pci_dev->version_id = 2; /* Current pci device vmstate version */
1073     return pci_dev;
1074 }
1075 
1076 static void pci_unregister_io_regions(PCIDevice *pci_dev)
1077 {
1078     PCIIORegion *r;
1079     int i;
1080 
1081     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1082         r = &pci_dev->io_regions[i];
1083         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
1084             continue;
1085         memory_region_del_subregion(r->address_space, r->memory);
1086     }
1087 
1088     pci_unregister_vga(pci_dev);
1089 }
1090 
1091 static void pci_qdev_unrealize(DeviceState *dev, Error **errp)
1092 {
1093     PCIDevice *pci_dev = PCI_DEVICE(dev);
1094     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1095 
1096     pci_unregister_io_regions(pci_dev);
1097     pci_del_option_rom(pci_dev);
1098 
1099     if (pc->exit) {
1100         pc->exit(pci_dev);
1101     }
1102 
1103     pci_device_deassert_intx(pci_dev);
1104     do_pci_unregister_device(pci_dev);
1105 }
1106 
1107 void pci_register_bar(PCIDevice *pci_dev, int region_num,
1108                       uint8_t type, MemoryRegion *memory)
1109 {
1110     PCIIORegion *r;
1111     uint32_t addr; /* offset in pci config space */
1112     uint64_t wmask;
1113     pcibus_t size = memory_region_size(memory);
1114 
1115     assert(region_num >= 0);
1116     assert(region_num < PCI_NUM_REGIONS);
1117     if (size & (size-1)) {
1118         fprintf(stderr, "ERROR: PCI region size must be pow2 "
1119                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
1120         exit(1);
1121     }
1122 
1123     r = &pci_dev->io_regions[region_num];
1124     r->addr = PCI_BAR_UNMAPPED;
1125     r->size = size;
1126     r->type = type;
1127     r->memory = memory;
1128     r->address_space = type & PCI_BASE_ADDRESS_SPACE_IO
1129                         ? pci_get_bus(pci_dev)->address_space_io
1130                         : pci_get_bus(pci_dev)->address_space_mem;
1131 
1132     wmask = ~(size - 1);
1133     if (region_num == PCI_ROM_SLOT) {
1134         /* ROM enable bit is writable */
1135         wmask |= PCI_ROM_ADDRESS_ENABLE;
1136     }
1137 
1138     addr = pci_bar(pci_dev, region_num);
1139     pci_set_long(pci_dev->config + addr, type);
1140 
1141     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
1142         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1143         pci_set_quad(pci_dev->wmask + addr, wmask);
1144         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
1145     } else {
1146         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
1147         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
1148     }
1149 }
1150 
1151 static void pci_update_vga(PCIDevice *pci_dev)
1152 {
1153     uint16_t cmd;
1154 
1155     if (!pci_dev->has_vga) {
1156         return;
1157     }
1158 
1159     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
1160 
1161     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
1162                               cmd & PCI_COMMAND_MEMORY);
1163     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
1164                               cmd & PCI_COMMAND_IO);
1165     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
1166                               cmd & PCI_COMMAND_IO);
1167 }
1168 
1169 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
1170                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1171 {
1172     PCIBus *bus = pci_get_bus(pci_dev);
1173 
1174     assert(!pci_dev->has_vga);
1175 
1176     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1177     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1178     memory_region_add_subregion_overlap(bus->address_space_mem,
1179                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1180 
1181     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1182     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1183     memory_region_add_subregion_overlap(bus->address_space_io,
1184                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1185 
1186     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1187     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1188     memory_region_add_subregion_overlap(bus->address_space_io,
1189                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1190     pci_dev->has_vga = true;
1191 
1192     pci_update_vga(pci_dev);
1193 }
1194 
1195 void pci_unregister_vga(PCIDevice *pci_dev)
1196 {
1197     PCIBus *bus = pci_get_bus(pci_dev);
1198 
1199     if (!pci_dev->has_vga) {
1200         return;
1201     }
1202 
1203     memory_region_del_subregion(bus->address_space_mem,
1204                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1205     memory_region_del_subregion(bus->address_space_io,
1206                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1207     memory_region_del_subregion(bus->address_space_io,
1208                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1209     pci_dev->has_vga = false;
1210 }
1211 
1212 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1213 {
1214     return pci_dev->io_regions[region_num].addr;
1215 }
1216 
1217 static pcibus_t pci_bar_address(PCIDevice *d,
1218 				int reg, uint8_t type, pcibus_t size)
1219 {
1220     pcibus_t new_addr, last_addr;
1221     int bar = pci_bar(d, reg);
1222     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1223     Object *machine = qdev_get_machine();
1224     ObjectClass *oc = object_get_class(machine);
1225     MachineClass *mc = MACHINE_CLASS(oc);
1226     bool allow_0_address = mc->pci_allow_0_address;
1227 
1228     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1229         if (!(cmd & PCI_COMMAND_IO)) {
1230             return PCI_BAR_UNMAPPED;
1231         }
1232         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1233         last_addr = new_addr + size - 1;
1234         /* Check if 32 bit BAR wraps around explicitly.
1235          * TODO: make priorities correct and remove this work around.
1236          */
1237         if (last_addr <= new_addr || last_addr >= UINT32_MAX ||
1238             (!allow_0_address && new_addr == 0)) {
1239             return PCI_BAR_UNMAPPED;
1240         }
1241         return new_addr;
1242     }
1243 
1244     if (!(cmd & PCI_COMMAND_MEMORY)) {
1245         return PCI_BAR_UNMAPPED;
1246     }
1247     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1248         new_addr = pci_get_quad(d->config + bar);
1249     } else {
1250         new_addr = pci_get_long(d->config + bar);
1251     }
1252     /* the ROM slot has a specific enable bit */
1253     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1254         return PCI_BAR_UNMAPPED;
1255     }
1256     new_addr &= ~(size - 1);
1257     last_addr = new_addr + size - 1;
1258     /* NOTE: we do not support wrapping */
1259     /* XXX: as we cannot support really dynamic
1260        mappings, we handle specific values as invalid
1261        mappings. */
1262     if (last_addr <= new_addr || last_addr == PCI_BAR_UNMAPPED ||
1263         (!allow_0_address && new_addr == 0)) {
1264         return PCI_BAR_UNMAPPED;
1265     }
1266 
1267     /* Now pcibus_t is 64bit.
1268      * Check if 32 bit BAR wraps around explicitly.
1269      * Without this, PC ide doesn't work well.
1270      * TODO: remove this work around.
1271      */
1272     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1273         return PCI_BAR_UNMAPPED;
1274     }
1275 
1276     /*
1277      * OS is allowed to set BAR beyond its addressable
1278      * bits. For example, 32 bit OS can set 64bit bar
1279      * to >4G. Check it. TODO: we might need to support
1280      * it in the future for e.g. PAE.
1281      */
1282     if (last_addr >= HWADDR_MAX) {
1283         return PCI_BAR_UNMAPPED;
1284     }
1285 
1286     return new_addr;
1287 }
1288 
1289 static void pci_update_mappings(PCIDevice *d)
1290 {
1291     PCIIORegion *r;
1292     int i;
1293     pcibus_t new_addr;
1294 
1295     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1296         r = &d->io_regions[i];
1297 
1298         /* this region isn't registered */
1299         if (!r->size)
1300             continue;
1301 
1302         new_addr = pci_bar_address(d, i, r->type, r->size);
1303 
1304         /* This bar isn't changed */
1305         if (new_addr == r->addr)
1306             continue;
1307 
1308         /* now do the real mapping */
1309         if (r->addr != PCI_BAR_UNMAPPED) {
1310             trace_pci_update_mappings_del(d, pci_dev_bus_num(d),
1311                                           PCI_SLOT(d->devfn),
1312                                           PCI_FUNC(d->devfn),
1313                                           i, r->addr, r->size);
1314             memory_region_del_subregion(r->address_space, r->memory);
1315         }
1316         r->addr = new_addr;
1317         if (r->addr != PCI_BAR_UNMAPPED) {
1318             trace_pci_update_mappings_add(d, pci_dev_bus_num(d),
1319                                           PCI_SLOT(d->devfn),
1320                                           PCI_FUNC(d->devfn),
1321                                           i, r->addr, r->size);
1322             memory_region_add_subregion_overlap(r->address_space,
1323                                                 r->addr, r->memory, 1);
1324         }
1325     }
1326 
1327     pci_update_vga(d);
1328 }
1329 
1330 static inline int pci_irq_disabled(PCIDevice *d)
1331 {
1332     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1333 }
1334 
1335 /* Called after interrupt disabled field update in config space,
1336  * assert/deassert interrupts if necessary.
1337  * Gets original interrupt disable bit value (before update). */
1338 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1339 {
1340     int i, disabled = pci_irq_disabled(d);
1341     if (disabled == was_irq_disabled)
1342         return;
1343     for (i = 0; i < PCI_NUM_PINS; ++i) {
1344         int state = pci_irq_state(d, i);
1345         pci_change_irq_level(d, i, disabled ? -state : state);
1346     }
1347 }
1348 
1349 uint32_t pci_default_read_config(PCIDevice *d,
1350                                  uint32_t address, int len)
1351 {
1352     uint32_t val = 0;
1353 
1354     memcpy(&val, d->config + address, len);
1355     return le32_to_cpu(val);
1356 }
1357 
1358 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1359 {
1360     int i, was_irq_disabled = pci_irq_disabled(d);
1361     uint32_t val = val_in;
1362 
1363     for (i = 0; i < l; val >>= 8, ++i) {
1364         uint8_t wmask = d->wmask[addr + i];
1365         uint8_t w1cmask = d->w1cmask[addr + i];
1366         assert(!(wmask & w1cmask));
1367         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1368         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1369     }
1370     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1371         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1372         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1373         range_covers_byte(addr, l, PCI_COMMAND))
1374         pci_update_mappings(d);
1375 
1376     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1377         pci_update_irq_disabled(d, was_irq_disabled);
1378         memory_region_set_enabled(&d->bus_master_enable_region,
1379                                   pci_get_word(d->config + PCI_COMMAND)
1380                                     & PCI_COMMAND_MASTER);
1381     }
1382 
1383     msi_write_config(d, addr, val_in, l);
1384     msix_write_config(d, addr, val_in, l);
1385 }
1386 
1387 /***********************************************************/
1388 /* generic PCI irq support */
1389 
1390 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1391 static void pci_irq_handler(void *opaque, int irq_num, int level)
1392 {
1393     PCIDevice *pci_dev = opaque;
1394     int change;
1395 
1396     change = level - pci_irq_state(pci_dev, irq_num);
1397     if (!change)
1398         return;
1399 
1400     pci_set_irq_state(pci_dev, irq_num, level);
1401     pci_update_irq_status(pci_dev);
1402     if (pci_irq_disabled(pci_dev))
1403         return;
1404     pci_change_irq_level(pci_dev, irq_num, change);
1405 }
1406 
1407 static inline int pci_intx(PCIDevice *pci_dev)
1408 {
1409     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1410 }
1411 
1412 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1413 {
1414     int intx = pci_intx(pci_dev);
1415 
1416     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1417 }
1418 
1419 void pci_set_irq(PCIDevice *pci_dev, int level)
1420 {
1421     int intx = pci_intx(pci_dev);
1422     pci_irq_handler(pci_dev, intx, level);
1423 }
1424 
1425 /* Special hooks used by device assignment */
1426 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1427 {
1428     assert(pci_bus_is_root(bus));
1429     bus->route_intx_to_irq = route_intx_to_irq;
1430 }
1431 
1432 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1433 {
1434     PCIBus *bus;
1435 
1436     do {
1437         bus = pci_get_bus(dev);
1438         pin = bus->map_irq(dev, pin);
1439         dev = bus->parent_dev;
1440     } while (dev);
1441 
1442     if (!bus->route_intx_to_irq) {
1443         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1444                      object_get_typename(OBJECT(bus->qbus.parent)));
1445         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1446     }
1447 
1448     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1449 }
1450 
1451 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1452 {
1453     return old->mode != new->mode || old->irq != new->irq;
1454 }
1455 
1456 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1457 {
1458     PCIDevice *dev;
1459     PCIBus *sec;
1460     int i;
1461 
1462     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1463         dev = bus->devices[i];
1464         if (dev && dev->intx_routing_notifier) {
1465             dev->intx_routing_notifier(dev);
1466         }
1467     }
1468 
1469     QLIST_FOREACH(sec, &bus->child, sibling) {
1470         pci_bus_fire_intx_routing_notifier(sec);
1471     }
1472 }
1473 
1474 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1475                                           PCIINTxRoutingNotifier notifier)
1476 {
1477     dev->intx_routing_notifier = notifier;
1478 }
1479 
1480 /*
1481  * PCI-to-PCI bridge specification
1482  * 9.1: Interrupt routing. Table 9-1
1483  *
1484  * the PCI Express Base Specification, Revision 2.1
1485  * 2.2.8.1: INTx interrutp signaling - Rules
1486  *          the Implementation Note
1487  *          Table 2-20
1488  */
1489 /*
1490  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1491  * 0-origin unlike PCI interrupt pin register.
1492  */
1493 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1494 {
1495     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1496 }
1497 
1498 /***********************************************************/
1499 /* monitor info on PCI */
1500 
1501 typedef struct {
1502     uint16_t class;
1503     const char *desc;
1504     const char *fw_name;
1505     uint16_t fw_ign_bits;
1506 } pci_class_desc;
1507 
1508 static const pci_class_desc pci_class_descriptions[] =
1509 {
1510     { 0x0001, "VGA controller", "display"},
1511     { 0x0100, "SCSI controller", "scsi"},
1512     { 0x0101, "IDE controller", "ide"},
1513     { 0x0102, "Floppy controller", "fdc"},
1514     { 0x0103, "IPI controller", "ipi"},
1515     { 0x0104, "RAID controller", "raid"},
1516     { 0x0106, "SATA controller"},
1517     { 0x0107, "SAS controller"},
1518     { 0x0180, "Storage controller"},
1519     { 0x0200, "Ethernet controller", "ethernet"},
1520     { 0x0201, "Token Ring controller", "token-ring"},
1521     { 0x0202, "FDDI controller", "fddi"},
1522     { 0x0203, "ATM controller", "atm"},
1523     { 0x0280, "Network controller"},
1524     { 0x0300, "VGA controller", "display", 0x00ff},
1525     { 0x0301, "XGA controller"},
1526     { 0x0302, "3D controller"},
1527     { 0x0380, "Display controller"},
1528     { 0x0400, "Video controller", "video"},
1529     { 0x0401, "Audio controller", "sound"},
1530     { 0x0402, "Phone"},
1531     { 0x0403, "Audio controller", "sound"},
1532     { 0x0480, "Multimedia controller"},
1533     { 0x0500, "RAM controller", "memory"},
1534     { 0x0501, "Flash controller", "flash"},
1535     { 0x0580, "Memory controller"},
1536     { 0x0600, "Host bridge", "host"},
1537     { 0x0601, "ISA bridge", "isa"},
1538     { 0x0602, "EISA bridge", "eisa"},
1539     { 0x0603, "MC bridge", "mca"},
1540     { 0x0604, "PCI bridge", "pci-bridge"},
1541     { 0x0605, "PCMCIA bridge", "pcmcia"},
1542     { 0x0606, "NUBUS bridge", "nubus"},
1543     { 0x0607, "CARDBUS bridge", "cardbus"},
1544     { 0x0608, "RACEWAY bridge"},
1545     { 0x0680, "Bridge"},
1546     { 0x0700, "Serial port", "serial"},
1547     { 0x0701, "Parallel port", "parallel"},
1548     { 0x0800, "Interrupt controller", "interrupt-controller"},
1549     { 0x0801, "DMA controller", "dma-controller"},
1550     { 0x0802, "Timer", "timer"},
1551     { 0x0803, "RTC", "rtc"},
1552     { 0x0900, "Keyboard", "keyboard"},
1553     { 0x0901, "Pen", "pen"},
1554     { 0x0902, "Mouse", "mouse"},
1555     { 0x0A00, "Dock station", "dock", 0x00ff},
1556     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1557     { 0x0c00, "Fireware contorller", "fireware"},
1558     { 0x0c01, "Access bus controller", "access-bus"},
1559     { 0x0c02, "SSA controller", "ssa"},
1560     { 0x0c03, "USB controller", "usb"},
1561     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1562     { 0x0c05, "SMBus"},
1563     { 0, NULL}
1564 };
1565 
1566 static void pci_for_each_device_under_bus_reverse(PCIBus *bus,
1567                                                   void (*fn)(PCIBus *b,
1568                                                              PCIDevice *d,
1569                                                              void *opaque),
1570                                                   void *opaque)
1571 {
1572     PCIDevice *d;
1573     int devfn;
1574 
1575     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1576         d = bus->devices[ARRAY_SIZE(bus->devices) - 1 - devfn];
1577         if (d) {
1578             fn(bus, d, opaque);
1579         }
1580     }
1581 }
1582 
1583 void pci_for_each_device_reverse(PCIBus *bus, int bus_num,
1584                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1585                          void *opaque)
1586 {
1587     bus = pci_find_bus_nr(bus, bus_num);
1588 
1589     if (bus) {
1590         pci_for_each_device_under_bus_reverse(bus, fn, opaque);
1591     }
1592 }
1593 
1594 static void pci_for_each_device_under_bus(PCIBus *bus,
1595                                           void (*fn)(PCIBus *b, PCIDevice *d,
1596                                                      void *opaque),
1597                                           void *opaque)
1598 {
1599     PCIDevice *d;
1600     int devfn;
1601 
1602     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1603         d = bus->devices[devfn];
1604         if (d) {
1605             fn(bus, d, opaque);
1606         }
1607     }
1608 }
1609 
1610 void pci_for_each_device(PCIBus *bus, int bus_num,
1611                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1612                          void *opaque)
1613 {
1614     bus = pci_find_bus_nr(bus, bus_num);
1615 
1616     if (bus) {
1617         pci_for_each_device_under_bus(bus, fn, opaque);
1618     }
1619 }
1620 
1621 static const pci_class_desc *get_class_desc(int class)
1622 {
1623     const pci_class_desc *desc;
1624 
1625     desc = pci_class_descriptions;
1626     while (desc->desc && class != desc->class) {
1627         desc++;
1628     }
1629 
1630     return desc;
1631 }
1632 
1633 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1634 
1635 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1636 {
1637     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1638     int i;
1639 
1640     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1641         const PCIIORegion *r = &dev->io_regions[i];
1642         PciMemoryRegionList *region;
1643 
1644         if (!r->size) {
1645             continue;
1646         }
1647 
1648         region = g_malloc0(sizeof(*region));
1649         region->value = g_malloc0(sizeof(*region->value));
1650 
1651         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1652             region->value->type = g_strdup("io");
1653         } else {
1654             region->value->type = g_strdup("memory");
1655             region->value->has_prefetch = true;
1656             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1657             region->value->has_mem_type_64 = true;
1658             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1659         }
1660 
1661         region->value->bar = i;
1662         region->value->address = r->addr;
1663         region->value->size = r->size;
1664 
1665         /* XXX: waiting for the qapi to support GSList */
1666         if (!cur_item) {
1667             head = cur_item = region;
1668         } else {
1669             cur_item->next = region;
1670             cur_item = region;
1671         }
1672     }
1673 
1674     return head;
1675 }
1676 
1677 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1678                                            int bus_num)
1679 {
1680     PciBridgeInfo *info;
1681     PciMemoryRange *range;
1682 
1683     info = g_new0(PciBridgeInfo, 1);
1684 
1685     info->bus = g_new0(PciBusInfo, 1);
1686     info->bus->number = dev->config[PCI_PRIMARY_BUS];
1687     info->bus->secondary = dev->config[PCI_SECONDARY_BUS];
1688     info->bus->subordinate = dev->config[PCI_SUBORDINATE_BUS];
1689 
1690     range = info->bus->io_range = g_new0(PciMemoryRange, 1);
1691     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1692     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1693 
1694     range = info->bus->memory_range = g_new0(PciMemoryRange, 1);
1695     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1696     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1697 
1698     range = info->bus->prefetchable_range = g_new0(PciMemoryRange, 1);
1699     range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1700     range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1701 
1702     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1703         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1704         if (child_bus) {
1705             info->has_devices = true;
1706             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1707         }
1708     }
1709 
1710     return info;
1711 }
1712 
1713 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1714                                            int bus_num)
1715 {
1716     const pci_class_desc *desc;
1717     PciDeviceInfo *info;
1718     uint8_t type;
1719     int class;
1720 
1721     info = g_new0(PciDeviceInfo, 1);
1722     info->bus = bus_num;
1723     info->slot = PCI_SLOT(dev->devfn);
1724     info->function = PCI_FUNC(dev->devfn);
1725 
1726     info->class_info = g_new0(PciDeviceClass, 1);
1727     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1728     info->class_info->q_class = class;
1729     desc = get_class_desc(class);
1730     if (desc->desc) {
1731         info->class_info->has_desc = true;
1732         info->class_info->desc = g_strdup(desc->desc);
1733     }
1734 
1735     info->id = g_new0(PciDeviceId, 1);
1736     info->id->vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1737     info->id->device = pci_get_word(dev->config + PCI_DEVICE_ID);
1738     info->regions = qmp_query_pci_regions(dev);
1739     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1740 
1741     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1742         info->has_irq = true;
1743         info->irq = dev->config[PCI_INTERRUPT_LINE];
1744     }
1745 
1746     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1747     if (type == PCI_HEADER_TYPE_BRIDGE) {
1748         info->has_pci_bridge = true;
1749         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1750     }
1751 
1752     return info;
1753 }
1754 
1755 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1756 {
1757     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1758     PCIDevice *dev;
1759     int devfn;
1760 
1761     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1762         dev = bus->devices[devfn];
1763         if (dev) {
1764             info = g_malloc0(sizeof(*info));
1765             info->value = qmp_query_pci_device(dev, bus, bus_num);
1766 
1767             /* XXX: waiting for the qapi to support GSList */
1768             if (!cur_item) {
1769                 head = cur_item = info;
1770             } else {
1771                 cur_item->next = info;
1772                 cur_item = info;
1773             }
1774         }
1775     }
1776 
1777     return head;
1778 }
1779 
1780 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1781 {
1782     PciInfo *info = NULL;
1783 
1784     bus = pci_find_bus_nr(bus, bus_num);
1785     if (bus) {
1786         info = g_malloc0(sizeof(*info));
1787         info->bus = bus_num;
1788         info->devices = qmp_query_pci_devices(bus, bus_num);
1789     }
1790 
1791     return info;
1792 }
1793 
1794 PciInfoList *qmp_query_pci(Error **errp)
1795 {
1796     PciInfoList *info, *head = NULL, *cur_item = NULL;
1797     PCIHostState *host_bridge;
1798 
1799     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1800         info = g_malloc0(sizeof(*info));
1801         info->value = qmp_query_pci_bus(host_bridge->bus,
1802                                         pci_bus_num(host_bridge->bus));
1803 
1804         /* XXX: waiting for the qapi to support GSList */
1805         if (!cur_item) {
1806             head = cur_item = info;
1807         } else {
1808             cur_item->next = info;
1809             cur_item = info;
1810         }
1811     }
1812 
1813     return head;
1814 }
1815 
1816 static const char * const pci_nic_models[] = {
1817     "ne2k_pci",
1818     "i82551",
1819     "i82557b",
1820     "i82559er",
1821     "rtl8139",
1822     "e1000",
1823     "pcnet",
1824     "virtio",
1825     "sungem",
1826     NULL
1827 };
1828 
1829 static const char * const pci_nic_names[] = {
1830     "ne2k_pci",
1831     "i82551",
1832     "i82557b",
1833     "i82559er",
1834     "rtl8139",
1835     "e1000",
1836     "pcnet",
1837     "virtio-net-pci",
1838     "sungem",
1839     NULL
1840 };
1841 
1842 /* Initialize a PCI NIC.  */
1843 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1844                                const char *default_model,
1845                                const char *default_devaddr)
1846 {
1847     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1848     PCIBus *bus;
1849     PCIDevice *pci_dev;
1850     DeviceState *dev;
1851     int devfn;
1852     int i;
1853 
1854     if (qemu_show_nic_models(nd->model, pci_nic_models)) {
1855         exit(0);
1856     }
1857 
1858     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1859     if (i < 0) {
1860         exit(1);
1861     }
1862 
1863     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1864     if (!bus) {
1865         error_report("Invalid PCI device address %s for device %s",
1866                      devaddr, pci_nic_names[i]);
1867         exit(1);
1868     }
1869 
1870     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1871     dev = &pci_dev->qdev;
1872     qdev_set_nic_properties(dev, nd);
1873     qdev_init_nofail(dev);
1874 
1875     return pci_dev;
1876 }
1877 
1878 PCIDevice *pci_vga_init(PCIBus *bus)
1879 {
1880     switch (vga_interface_type) {
1881     case VGA_CIRRUS:
1882         return pci_create_simple(bus, -1, "cirrus-vga");
1883     case VGA_QXL:
1884         return pci_create_simple(bus, -1, "qxl-vga");
1885     case VGA_STD:
1886         return pci_create_simple(bus, -1, "VGA");
1887     case VGA_VMWARE:
1888         return pci_create_simple(bus, -1, "vmware-svga");
1889     case VGA_VIRTIO:
1890         return pci_create_simple(bus, -1, "virtio-vga");
1891     case VGA_NONE:
1892     default: /* Other non-PCI types. Checking for unsupported types is already
1893                 done in vl.c. */
1894         return NULL;
1895     }
1896 }
1897 
1898 /* Whether a given bus number is in range of the secondary
1899  * bus of the given bridge device. */
1900 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1901 {
1902     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1903              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1904         dev->config[PCI_SECONDARY_BUS] <= bus_num &&
1905         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1906 }
1907 
1908 /* Whether a given bus number is in a range of a root bus */
1909 static bool pci_root_bus_in_range(PCIBus *bus, int bus_num)
1910 {
1911     int i;
1912 
1913     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1914         PCIDevice *dev = bus->devices[i];
1915 
1916         if (dev && PCI_DEVICE_GET_CLASS(dev)->is_bridge) {
1917             if (pci_secondary_bus_in_range(dev, bus_num)) {
1918                 return true;
1919             }
1920         }
1921     }
1922 
1923     return false;
1924 }
1925 
1926 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1927 {
1928     PCIBus *sec;
1929 
1930     if (!bus) {
1931         return NULL;
1932     }
1933 
1934     if (pci_bus_num(bus) == bus_num) {
1935         return bus;
1936     }
1937 
1938     /* Consider all bus numbers in range for the host pci bridge. */
1939     if (!pci_bus_is_root(bus) &&
1940         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1941         return NULL;
1942     }
1943 
1944     /* try child bus */
1945     for (; bus; bus = sec) {
1946         QLIST_FOREACH(sec, &bus->child, sibling) {
1947             if (pci_bus_num(sec) == bus_num) {
1948                 return sec;
1949             }
1950             /* PXB buses assumed to be children of bus 0 */
1951             if (pci_bus_is_root(sec)) {
1952                 if (pci_root_bus_in_range(sec, bus_num)) {
1953                     break;
1954                 }
1955             } else {
1956                 if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1957                     break;
1958                 }
1959             }
1960         }
1961     }
1962 
1963     return NULL;
1964 }
1965 
1966 void pci_for_each_bus_depth_first(PCIBus *bus,
1967                                   void *(*begin)(PCIBus *bus, void *parent_state),
1968                                   void (*end)(PCIBus *bus, void *state),
1969                                   void *parent_state)
1970 {
1971     PCIBus *sec;
1972     void *state;
1973 
1974     if (!bus) {
1975         return;
1976     }
1977 
1978     if (begin) {
1979         state = begin(bus, parent_state);
1980     } else {
1981         state = parent_state;
1982     }
1983 
1984     QLIST_FOREACH(sec, &bus->child, sibling) {
1985         pci_for_each_bus_depth_first(sec, begin, end, state);
1986     }
1987 
1988     if (end) {
1989         end(bus, state);
1990     }
1991 }
1992 
1993 
1994 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1995 {
1996     bus = pci_find_bus_nr(bus, bus_num);
1997 
1998     if (!bus)
1999         return NULL;
2000 
2001     return bus->devices[devfn];
2002 }
2003 
2004 static void pci_qdev_realize(DeviceState *qdev, Error **errp)
2005 {
2006     PCIDevice *pci_dev = (PCIDevice *)qdev;
2007     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
2008     Error *local_err = NULL;
2009     bool is_default_rom;
2010 
2011     /* initialize cap_present for pci_is_express() and pci_config_size() */
2012     if (pc->is_express) {
2013         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
2014     }
2015 
2016     pci_dev = do_pci_register_device(pci_dev,
2017                                      object_get_typename(OBJECT(qdev)),
2018                                      pci_dev->devfn, errp);
2019     if (pci_dev == NULL)
2020         return;
2021 
2022     if (pc->realize) {
2023         pc->realize(pci_dev, &local_err);
2024         if (local_err) {
2025             error_propagate(errp, local_err);
2026             do_pci_unregister_device(pci_dev);
2027             return;
2028         }
2029     }
2030 
2031     /* rom loading */
2032     is_default_rom = false;
2033     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
2034         pci_dev->romfile = g_strdup(pc->romfile);
2035         is_default_rom = true;
2036     }
2037 
2038     pci_add_option_rom(pci_dev, is_default_rom, &local_err);
2039     if (local_err) {
2040         error_propagate(errp, local_err);
2041         pci_qdev_unrealize(DEVICE(pci_dev), NULL);
2042         return;
2043     }
2044 }
2045 
2046 static void pci_default_realize(PCIDevice *dev, Error **errp)
2047 {
2048     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2049 
2050     if (pc->init) {
2051         if (pc->init(dev) < 0) {
2052             error_setg(errp, "Device initialization failed");
2053             return;
2054         }
2055     }
2056 }
2057 
2058 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
2059                                     const char *name)
2060 {
2061     DeviceState *dev;
2062 
2063     dev = qdev_create(&bus->qbus, name);
2064     qdev_prop_set_int32(dev, "addr", devfn);
2065     qdev_prop_set_bit(dev, "multifunction", multifunction);
2066     return PCI_DEVICE(dev);
2067 }
2068 
2069 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
2070                                            bool multifunction,
2071                                            const char *name)
2072 {
2073     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
2074     qdev_init_nofail(&dev->qdev);
2075     return dev;
2076 }
2077 
2078 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
2079 {
2080     return pci_create_multifunction(bus, devfn, false, name);
2081 }
2082 
2083 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
2084 {
2085     return pci_create_simple_multifunction(bus, devfn, false, name);
2086 }
2087 
2088 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
2089 {
2090     int offset = PCI_CONFIG_HEADER_SIZE;
2091     int i;
2092     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
2093         if (pdev->used[i])
2094             offset = i + 1;
2095         else if (i - offset + 1 == size)
2096             return offset;
2097     }
2098     return 0;
2099 }
2100 
2101 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
2102                                         uint8_t *prev_p)
2103 {
2104     uint8_t next, prev;
2105 
2106     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
2107         return 0;
2108 
2109     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2110          prev = next + PCI_CAP_LIST_NEXT)
2111         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
2112             break;
2113 
2114     if (prev_p)
2115         *prev_p = prev;
2116     return next;
2117 }
2118 
2119 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
2120 {
2121     uint8_t next, prev, found = 0;
2122 
2123     if (!(pdev->used[offset])) {
2124         return 0;
2125     }
2126 
2127     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
2128 
2129     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
2130          prev = next + PCI_CAP_LIST_NEXT) {
2131         if (next <= offset && next > found) {
2132             found = next;
2133         }
2134     }
2135     return found;
2136 }
2137 
2138 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
2139    This is needed for an option rom which is used for more than one device. */
2140 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
2141 {
2142     uint16_t vendor_id;
2143     uint16_t device_id;
2144     uint16_t rom_vendor_id;
2145     uint16_t rom_device_id;
2146     uint16_t rom_magic;
2147     uint16_t pcir_offset;
2148     uint8_t checksum;
2149 
2150     /* Words in rom data are little endian (like in PCI configuration),
2151        so they can be read / written with pci_get_word / pci_set_word. */
2152 
2153     /* Only a valid rom will be patched. */
2154     rom_magic = pci_get_word(ptr);
2155     if (rom_magic != 0xaa55) {
2156         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
2157         return;
2158     }
2159     pcir_offset = pci_get_word(ptr + 0x18);
2160     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
2161         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
2162         return;
2163     }
2164 
2165     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
2166     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
2167     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
2168     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
2169 
2170     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
2171                 vendor_id, device_id, rom_vendor_id, rom_device_id);
2172 
2173     checksum = ptr[6];
2174 
2175     if (vendor_id != rom_vendor_id) {
2176         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
2177         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
2178         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
2179         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2180         ptr[6] = checksum;
2181         pci_set_word(ptr + pcir_offset + 4, vendor_id);
2182     }
2183 
2184     if (device_id != rom_device_id) {
2185         /* Patch device id and checksum (at offset 6 for etherboot roms). */
2186         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
2187         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
2188         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
2189         ptr[6] = checksum;
2190         pci_set_word(ptr + pcir_offset + 6, device_id);
2191     }
2192 }
2193 
2194 /* Add an option rom for the device */
2195 static void pci_add_option_rom(PCIDevice *pdev, bool is_default_rom,
2196                                Error **errp)
2197 {
2198     int size;
2199     char *path;
2200     void *ptr;
2201     char name[32];
2202     const VMStateDescription *vmsd;
2203 
2204     if (!pdev->romfile)
2205         return;
2206     if (strlen(pdev->romfile) == 0)
2207         return;
2208 
2209     if (!pdev->rom_bar) {
2210         /*
2211          * Load rom via fw_cfg instead of creating a rom bar,
2212          * for 0.11 compatibility.
2213          */
2214         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
2215 
2216         /*
2217          * Hot-plugged devices can't use the option ROM
2218          * if the rom bar is disabled.
2219          */
2220         if (DEVICE(pdev)->hotplugged) {
2221             error_setg(errp, "Hot-plugged device without ROM bar"
2222                        " can't have an option ROM");
2223             return;
2224         }
2225 
2226         if (class == 0x0300) {
2227             rom_add_vga(pdev->romfile);
2228         } else {
2229             rom_add_option(pdev->romfile, -1);
2230         }
2231         return;
2232     }
2233 
2234     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
2235     if (path == NULL) {
2236         path = g_strdup(pdev->romfile);
2237     }
2238 
2239     size = get_image_size(path);
2240     if (size < 0) {
2241         error_setg(errp, "failed to find romfile \"%s\"", pdev->romfile);
2242         g_free(path);
2243         return;
2244     } else if (size == 0) {
2245         error_setg(errp, "romfile \"%s\" is empty", pdev->romfile);
2246         g_free(path);
2247         return;
2248     }
2249     size = pow2ceil(size);
2250 
2251     vmsd = qdev_get_vmsd(DEVICE(pdev));
2252 
2253     if (vmsd) {
2254         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
2255     } else {
2256         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
2257     }
2258     pdev->has_rom = true;
2259     memory_region_init_rom(&pdev->rom, OBJECT(pdev), name, size, &error_fatal);
2260     ptr = memory_region_get_ram_ptr(&pdev->rom);
2261     load_image(path, ptr);
2262     g_free(path);
2263 
2264     if (is_default_rom) {
2265         /* Only the default rom images will be patched (if needed). */
2266         pci_patch_ids(pdev, ptr, size);
2267     }
2268 
2269     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2270 }
2271 
2272 static void pci_del_option_rom(PCIDevice *pdev)
2273 {
2274     if (!pdev->has_rom)
2275         return;
2276 
2277     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2278     pdev->has_rom = false;
2279 }
2280 
2281 /*
2282  * On success, pci_add_capability() returns a positive value
2283  * that the offset of the pci capability.
2284  * On failure, it sets an error and returns a negative error
2285  * code.
2286  */
2287 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2288                        uint8_t offset, uint8_t size,
2289                        Error **errp)
2290 {
2291     uint8_t *config;
2292     int i, overlapping_cap;
2293 
2294     if (!offset) {
2295         offset = pci_find_space(pdev, size);
2296         /* out of PCI config space is programming error */
2297         assert(offset);
2298     } else {
2299         /* Verify that capabilities don't overlap.  Note: device assignment
2300          * depends on this check to verify that the device is not broken.
2301          * Should never trigger for emulated devices, but it's helpful
2302          * for debugging these. */
2303         for (i = offset; i < offset + size; i++) {
2304             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2305             if (overlapping_cap) {
2306                 error_setg(errp, "%s:%02x:%02x.%x "
2307                            "Attempt to add PCI capability %x at offset "
2308                            "%x overlaps existing capability %x at offset %x",
2309                            pci_root_bus_path(pdev), pci_dev_bus_num(pdev),
2310                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2311                            cap_id, offset, overlapping_cap, i);
2312                 return -EINVAL;
2313             }
2314         }
2315     }
2316 
2317     config = pdev->config + offset;
2318     config[PCI_CAP_LIST_ID] = cap_id;
2319     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2320     pdev->config[PCI_CAPABILITY_LIST] = offset;
2321     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2322     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2323     /* Make capability read-only by default */
2324     memset(pdev->wmask + offset, 0, size);
2325     /* Check capability by default */
2326     memset(pdev->cmask + offset, 0xFF, size);
2327     return offset;
2328 }
2329 
2330 /* Unlink capability from the pci config space. */
2331 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2332 {
2333     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2334     if (!offset)
2335         return;
2336     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2337     /* Make capability writable again */
2338     memset(pdev->wmask + offset, 0xff, size);
2339     memset(pdev->w1cmask + offset, 0, size);
2340     /* Clear cmask as device-specific registers can't be checked */
2341     memset(pdev->cmask + offset, 0, size);
2342     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2343 
2344     if (!pdev->config[PCI_CAPABILITY_LIST])
2345         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2346 }
2347 
2348 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2349 {
2350     return pci_find_capability_list(pdev, cap_id, NULL);
2351 }
2352 
2353 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2354 {
2355     PCIDevice *d = (PCIDevice *)dev;
2356     const pci_class_desc *desc;
2357     char ctxt[64];
2358     PCIIORegion *r;
2359     int i, class;
2360 
2361     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2362     desc = pci_class_descriptions;
2363     while (desc->desc && class != desc->class)
2364         desc++;
2365     if (desc->desc) {
2366         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2367     } else {
2368         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2369     }
2370 
2371     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2372                    "pci id %04x:%04x (sub %04x:%04x)\n",
2373                    indent, "", ctxt, pci_dev_bus_num(d),
2374                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2375                    pci_get_word(d->config + PCI_VENDOR_ID),
2376                    pci_get_word(d->config + PCI_DEVICE_ID),
2377                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2378                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2379     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2380         r = &d->io_regions[i];
2381         if (!r->size)
2382             continue;
2383         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2384                        " [0x%"FMT_PCIBUS"]\n",
2385                        indent, "",
2386                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2387                        r->addr, r->addr + r->size - 1);
2388     }
2389 }
2390 
2391 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2392 {
2393     PCIDevice *d = (PCIDevice *)dev;
2394     const char *name = NULL;
2395     const pci_class_desc *desc =  pci_class_descriptions;
2396     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2397 
2398     while (desc->desc &&
2399           (class & ~desc->fw_ign_bits) !=
2400           (desc->class & ~desc->fw_ign_bits)) {
2401         desc++;
2402     }
2403 
2404     if (desc->desc) {
2405         name = desc->fw_name;
2406     }
2407 
2408     if (name) {
2409         pstrcpy(buf, len, name);
2410     } else {
2411         snprintf(buf, len, "pci%04x,%04x",
2412                  pci_get_word(d->config + PCI_VENDOR_ID),
2413                  pci_get_word(d->config + PCI_DEVICE_ID));
2414     }
2415 
2416     return buf;
2417 }
2418 
2419 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2420 {
2421     PCIDevice *d = (PCIDevice *)dev;
2422     char path[50], name[33];
2423     int off;
2424 
2425     off = snprintf(path, sizeof(path), "%s@%x",
2426                    pci_dev_fw_name(dev, name, sizeof name),
2427                    PCI_SLOT(d->devfn));
2428     if (PCI_FUNC(d->devfn))
2429         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2430     return g_strdup(path);
2431 }
2432 
2433 static char *pcibus_get_dev_path(DeviceState *dev)
2434 {
2435     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2436     PCIDevice *t;
2437     int slot_depth;
2438     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2439      * 00 is added here to make this format compatible with
2440      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2441      * Slot.Function list specifies the slot and function numbers for all
2442      * devices on the path from root to the specific device. */
2443     const char *root_bus_path;
2444     int root_bus_len;
2445     char slot[] = ":SS.F";
2446     int slot_len = sizeof slot - 1 /* For '\0' */;
2447     int path_len;
2448     char *path, *p;
2449     int s;
2450 
2451     root_bus_path = pci_root_bus_path(d);
2452     root_bus_len = strlen(root_bus_path);
2453 
2454     /* Calculate # of slots on path between device and root. */;
2455     slot_depth = 0;
2456     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2457         ++slot_depth;
2458     }
2459 
2460     path_len = root_bus_len + slot_len * slot_depth;
2461 
2462     /* Allocate memory, fill in the terminating null byte. */
2463     path = g_malloc(path_len + 1 /* For '\0' */);
2464     path[path_len] = '\0';
2465 
2466     memcpy(path, root_bus_path, root_bus_len);
2467 
2468     /* Fill in slot numbers. We walk up from device to root, so need to print
2469      * them in the reverse order, last to first. */
2470     p = path + path_len;
2471     for (t = d; t; t = pci_get_bus(t)->parent_dev) {
2472         p -= slot_len;
2473         s = snprintf(slot, sizeof slot, ":%02x.%x",
2474                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2475         assert(s == slot_len);
2476         memcpy(p, slot, slot_len);
2477     }
2478 
2479     return path;
2480 }
2481 
2482 static int pci_qdev_find_recursive(PCIBus *bus,
2483                                    const char *id, PCIDevice **pdev)
2484 {
2485     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2486     if (!qdev) {
2487         return -ENODEV;
2488     }
2489 
2490     /* roughly check if given qdev is pci device */
2491     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2492         *pdev = PCI_DEVICE(qdev);
2493         return 0;
2494     }
2495     return -EINVAL;
2496 }
2497 
2498 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2499 {
2500     PCIHostState *host_bridge;
2501     int rc = -ENODEV;
2502 
2503     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2504         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2505         if (!tmp) {
2506             rc = 0;
2507             break;
2508         }
2509         if (tmp != -ENODEV) {
2510             rc = tmp;
2511         }
2512     }
2513 
2514     return rc;
2515 }
2516 
2517 MemoryRegion *pci_address_space(PCIDevice *dev)
2518 {
2519     return pci_get_bus(dev)->address_space_mem;
2520 }
2521 
2522 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2523 {
2524     return pci_get_bus(dev)->address_space_io;
2525 }
2526 
2527 static void pci_device_class_init(ObjectClass *klass, void *data)
2528 {
2529     DeviceClass *k = DEVICE_CLASS(klass);
2530     PCIDeviceClass *pc = PCI_DEVICE_CLASS(klass);
2531 
2532     k->realize = pci_qdev_realize;
2533     k->unrealize = pci_qdev_unrealize;
2534     k->bus_type = TYPE_PCI_BUS;
2535     k->props = pci_props;
2536     pc->realize = pci_default_realize;
2537 }
2538 
2539 static void pci_device_class_base_init(ObjectClass *klass, void *data)
2540 {
2541     if (!object_class_is_abstract(klass)) {
2542         ObjectClass *conventional =
2543             object_class_dynamic_cast(klass, INTERFACE_CONVENTIONAL_PCI_DEVICE);
2544         ObjectClass *pcie =
2545             object_class_dynamic_cast(klass, INTERFACE_PCIE_DEVICE);
2546         assert(conventional || pcie);
2547     }
2548 }
2549 
2550 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2551 {
2552     PCIBus *bus = pci_get_bus(dev);
2553     PCIBus *iommu_bus = bus;
2554 
2555     while(iommu_bus && !iommu_bus->iommu_fn && iommu_bus->parent_dev) {
2556         iommu_bus = pci_get_bus(iommu_bus->parent_dev);
2557     }
2558     if (iommu_bus && iommu_bus->iommu_fn) {
2559         return iommu_bus->iommu_fn(bus, iommu_bus->iommu_opaque, dev->devfn);
2560     }
2561     return &address_space_memory;
2562 }
2563 
2564 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2565 {
2566     bus->iommu_fn = fn;
2567     bus->iommu_opaque = opaque;
2568 }
2569 
2570 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2571 {
2572     Range *range = opaque;
2573     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2574     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2575     int i;
2576 
2577     if (!(cmd & PCI_COMMAND_MEMORY)) {
2578         return;
2579     }
2580 
2581     if (pc->is_bridge) {
2582         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2583         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2584 
2585         base = MAX(base, 0x1ULL << 32);
2586 
2587         if (limit >= base) {
2588             Range pref_range;
2589             range_set_bounds(&pref_range, base, limit);
2590             range_extend(range, &pref_range);
2591         }
2592     }
2593     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2594         PCIIORegion *r = &dev->io_regions[i];
2595         pcibus_t lob, upb;
2596         Range region_range;
2597 
2598         if (!r->size ||
2599             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2600             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2601             continue;
2602         }
2603 
2604         lob = pci_bar_address(dev, i, r->type, r->size);
2605         upb = lob + r->size - 1;
2606         if (lob == PCI_BAR_UNMAPPED) {
2607             continue;
2608         }
2609 
2610         lob = MAX(lob, 0x1ULL << 32);
2611 
2612         if (upb >= lob) {
2613             range_set_bounds(&region_range, lob, upb);
2614             range_extend(range, &region_range);
2615         }
2616     }
2617 }
2618 
2619 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2620 {
2621     range_make_empty(range);
2622     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2623 }
2624 
2625 static bool pcie_has_upstream_port(PCIDevice *dev)
2626 {
2627     PCIDevice *parent_dev = pci_bridge_get_device(pci_get_bus(dev));
2628 
2629     /* Device associated with an upstream port.
2630      * As there are several types of these, it's easier to check the
2631      * parent device: upstream ports are always connected to
2632      * root or downstream ports.
2633      */
2634     return parent_dev &&
2635         pci_is_express(parent_dev) &&
2636         parent_dev->exp.exp_cap &&
2637         (pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_ROOT_PORT ||
2638          pcie_cap_get_type(parent_dev) == PCI_EXP_TYPE_DOWNSTREAM);
2639 }
2640 
2641 PCIDevice *pci_get_function_0(PCIDevice *pci_dev)
2642 {
2643     PCIBus *bus = pci_get_bus(pci_dev);
2644 
2645     if(pcie_has_upstream_port(pci_dev)) {
2646         /* With an upstream PCIe port, we only support 1 device at slot 0 */
2647         return bus->devices[0];
2648     } else {
2649         /* Other bus types might support multiple devices at slots 0-31 */
2650         return bus->devices[PCI_DEVFN(PCI_SLOT(pci_dev->devfn), 0)];
2651     }
2652 }
2653 
2654 MSIMessage pci_get_msi_message(PCIDevice *dev, int vector)
2655 {
2656     MSIMessage msg;
2657     if (msix_enabled(dev)) {
2658         msg = msix_get_message(dev, vector);
2659     } else if (msi_enabled(dev)) {
2660         msg = msi_get_message(dev, vector);
2661     } else {
2662         /* Should never happen */
2663         error_report("%s: unknown interrupt type", __func__);
2664         abort();
2665     }
2666     return msg;
2667 }
2668 
2669 static const TypeInfo pci_device_type_info = {
2670     .name = TYPE_PCI_DEVICE,
2671     .parent = TYPE_DEVICE,
2672     .instance_size = sizeof(PCIDevice),
2673     .abstract = true,
2674     .class_size = sizeof(PCIDeviceClass),
2675     .class_init = pci_device_class_init,
2676     .class_base_init = pci_device_class_base_init,
2677 };
2678 
2679 static void pci_register_types(void)
2680 {
2681     type_register_static(&pci_bus_info);
2682     type_register_static(&pcie_bus_info);
2683     type_register_static(&conventional_pci_interface_info);
2684     type_register_static(&pcie_interface_info);
2685     type_register_static(&pci_device_type_info);
2686 }
2687 
2688 type_init(pci_register_types)
2689