xref: /openbmc/qemu/hw/pci/pci.c (revision 01c22f2c)
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 "hw/hw.h"
25 #include "hw/pci/pci.h"
26 #include "hw/pci/pci_bridge.h"
27 #include "hw/pci/pci_bus.h"
28 #include "hw/pci/pci_host.h"
29 #include "monitor/monitor.h"
30 #include "net/net.h"
31 #include "sysemu/sysemu.h"
32 #include "hw/loader.h"
33 #include "qemu/range.h"
34 #include "qmp-commands.h"
35 #include "hw/pci/msi.h"
36 #include "hw/pci/msix.h"
37 #include "exec/address-spaces.h"
38 #include "hw/hotplug.h"
39 
40 //#define DEBUG_PCI
41 #ifdef DEBUG_PCI
42 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
43 #else
44 # define PCI_DPRINTF(format, ...)       do { } while (0)
45 #endif
46 
47 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
48 static char *pcibus_get_dev_path(DeviceState *dev);
49 static char *pcibus_get_fw_dev_path(DeviceState *dev);
50 static void pcibus_reset(BusState *qbus);
51 
52 static Property pci_props[] = {
53     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
54     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
55     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
56     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
57                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
58     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
59                     QEMU_PCI_CAP_SERR_BITNR, true),
60     DEFINE_PROP_END_OF_LIST()
61 };
62 
63 static const VMStateDescription vmstate_pcibus = {
64     .name = "PCIBUS",
65     .version_id = 1,
66     .minimum_version_id = 1,
67     .minimum_version_id_old = 1,
68     .fields      = (VMStateField[]) {
69         VMSTATE_INT32_EQUAL(nirq, PCIBus),
70         VMSTATE_VARRAY_INT32(irq_count, PCIBus,
71                              nirq, 0, vmstate_info_int32,
72                              int32_t),
73         VMSTATE_END_OF_LIST()
74     }
75 };
76 
77 static void pci_bus_realize(BusState *qbus, Error **errp)
78 {
79     PCIBus *bus = PCI_BUS(qbus);
80 
81     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
82 }
83 
84 static void pci_bus_unrealize(BusState *qbus, Error **errp)
85 {
86     PCIBus *bus = PCI_BUS(qbus);
87 
88     vmstate_unregister(NULL, &vmstate_pcibus, bus);
89 }
90 
91 static void pci_bus_class_init(ObjectClass *klass, void *data)
92 {
93     BusClass *k = BUS_CLASS(klass);
94 
95     k->print_dev = pcibus_dev_print;
96     k->get_dev_path = pcibus_get_dev_path;
97     k->get_fw_dev_path = pcibus_get_fw_dev_path;
98     k->realize = pci_bus_realize;
99     k->unrealize = pci_bus_unrealize;
100     k->reset = pcibus_reset;
101 }
102 
103 static const TypeInfo pci_bus_info = {
104     .name = TYPE_PCI_BUS,
105     .parent = TYPE_BUS,
106     .instance_size = sizeof(PCIBus),
107     .class_init = pci_bus_class_init,
108 };
109 
110 static const TypeInfo pcie_bus_info = {
111     .name = TYPE_PCIE_BUS,
112     .parent = TYPE_PCI_BUS,
113 };
114 
115 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
116 static void pci_update_mappings(PCIDevice *d);
117 static void pci_irq_handler(void *opaque, int irq_num, int level);
118 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom);
119 static void pci_del_option_rom(PCIDevice *pdev);
120 
121 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
122 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
123 
124 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
125 
126 static int pci_bar(PCIDevice *d, int reg)
127 {
128     uint8_t type;
129 
130     if (reg != PCI_ROM_SLOT)
131         return PCI_BASE_ADDRESS_0 + reg * 4;
132 
133     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
134     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
135 }
136 
137 static inline int pci_irq_state(PCIDevice *d, int irq_num)
138 {
139 	return (d->irq_state >> irq_num) & 0x1;
140 }
141 
142 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
143 {
144 	d->irq_state &= ~(0x1 << irq_num);
145 	d->irq_state |= level << irq_num;
146 }
147 
148 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
149 {
150     PCIBus *bus;
151     for (;;) {
152         bus = pci_dev->bus;
153         irq_num = bus->map_irq(pci_dev, irq_num);
154         if (bus->set_irq)
155             break;
156         pci_dev = bus->parent_dev;
157     }
158     bus->irq_count[irq_num] += change;
159     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
160 }
161 
162 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
163 {
164     assert(irq_num >= 0);
165     assert(irq_num < bus->nirq);
166     return !!bus->irq_count[irq_num];
167 }
168 
169 /* Update interrupt status bit in config space on interrupt
170  * state change. */
171 static void pci_update_irq_status(PCIDevice *dev)
172 {
173     if (dev->irq_state) {
174         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
175     } else {
176         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
177     }
178 }
179 
180 void pci_device_deassert_intx(PCIDevice *dev)
181 {
182     int i;
183     for (i = 0; i < PCI_NUM_PINS; ++i) {
184         pci_irq_handler(dev, i, 0);
185     }
186 }
187 
188 static void pci_do_device_reset(PCIDevice *dev)
189 {
190     int r;
191 
192     dev->irq_state = 0;
193     pci_update_irq_status(dev);
194     pci_device_deassert_intx(dev);
195     /* Clear all writable bits */
196     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
197                                  pci_get_word(dev->wmask + PCI_COMMAND) |
198                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
199     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
200                                  pci_get_word(dev->wmask + PCI_STATUS) |
201                                  pci_get_word(dev->w1cmask + PCI_STATUS));
202     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
203     dev->config[PCI_INTERRUPT_LINE] = 0x0;
204     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
205         PCIIORegion *region = &dev->io_regions[r];
206         if (!region->size) {
207             continue;
208         }
209 
210         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
211             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
212             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
213         } else {
214             pci_set_long(dev->config + pci_bar(dev, r), region->type);
215         }
216     }
217     pci_update_mappings(dev);
218 
219     msi_reset(dev);
220     msix_reset(dev);
221 }
222 
223 /*
224  * This function is called on #RST and FLR.
225  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
226  */
227 void pci_device_reset(PCIDevice *dev)
228 {
229     qdev_reset_all(&dev->qdev);
230     pci_do_device_reset(dev);
231 }
232 
233 /*
234  * Trigger pci bus reset under a given bus.
235  * Called via qbus_reset_all on RST# assert, after the devices
236  * have been reset qdev_reset_all-ed already.
237  */
238 static void pcibus_reset(BusState *qbus)
239 {
240     PCIBus *bus = DO_UPCAST(PCIBus, qbus, qbus);
241     int i;
242 
243     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
244         if (bus->devices[i]) {
245             pci_do_device_reset(bus->devices[i]);
246         }
247     }
248 
249     for (i = 0; i < bus->nirq; i++) {
250         assert(bus->irq_count[i] == 0);
251     }
252 }
253 
254 static void pci_host_bus_register(PCIBus *bus, DeviceState *parent)
255 {
256     PCIHostState *host_bridge = PCI_HOST_BRIDGE(parent);
257 
258     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
259 }
260 
261 PCIBus *pci_find_primary_bus(void)
262 {
263     PCIBus *primary_bus = NULL;
264     PCIHostState *host;
265 
266     QLIST_FOREACH(host, &pci_host_bridges, next) {
267         if (primary_bus) {
268             /* We have multiple root buses, refuse to select a primary */
269             return NULL;
270         }
271         primary_bus = host->bus;
272     }
273 
274     return primary_bus;
275 }
276 
277 PCIBus *pci_device_root_bus(const PCIDevice *d)
278 {
279     PCIBus *bus = d->bus;
280 
281     while ((d = bus->parent_dev) != NULL) {
282         bus = d->bus;
283     }
284 
285     return bus;
286 }
287 
288 const char *pci_root_bus_path(PCIDevice *dev)
289 {
290     PCIBus *rootbus = pci_device_root_bus(dev);
291     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
292     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
293 
294     assert(!rootbus->parent_dev);
295     assert(host_bridge->bus == rootbus);
296 
297     if (hc->root_bus_path) {
298         return (*hc->root_bus_path)(host_bridge, rootbus);
299     }
300 
301     return rootbus->qbus.name;
302 }
303 
304 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
305                          const char *name,
306                          MemoryRegion *address_space_mem,
307                          MemoryRegion *address_space_io,
308                          uint8_t devfn_min)
309 {
310     assert(PCI_FUNC(devfn_min) == 0);
311     bus->devfn_min = devfn_min;
312     bus->address_space_mem = address_space_mem;
313     bus->address_space_io = address_space_io;
314 
315     /* host bridge */
316     QLIST_INIT(&bus->child);
317 
318     pci_host_bus_register(bus, parent);
319 }
320 
321 bool pci_bus_is_express(PCIBus *bus)
322 {
323     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
324 }
325 
326 bool pci_bus_is_root(PCIBus *bus)
327 {
328     return !bus->parent_dev;
329 }
330 
331 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
332                          const char *name,
333                          MemoryRegion *address_space_mem,
334                          MemoryRegion *address_space_io,
335                          uint8_t devfn_min, const char *typename)
336 {
337     qbus_create_inplace(bus, bus_size, typename, parent, name);
338     pci_bus_init(bus, parent, name, address_space_mem,
339                  address_space_io, devfn_min);
340 }
341 
342 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
343                     MemoryRegion *address_space_mem,
344                     MemoryRegion *address_space_io,
345                     uint8_t devfn_min, const char *typename)
346 {
347     PCIBus *bus;
348 
349     bus = PCI_BUS(qbus_create(typename, parent, name));
350     pci_bus_init(bus, parent, name, address_space_mem,
351                  address_space_io, devfn_min);
352     return bus;
353 }
354 
355 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
356                   void *irq_opaque, int nirq)
357 {
358     bus->set_irq = set_irq;
359     bus->map_irq = map_irq;
360     bus->irq_opaque = irq_opaque;
361     bus->nirq = nirq;
362     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
363 }
364 
365 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
366                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
367                          void *irq_opaque,
368                          MemoryRegion *address_space_mem,
369                          MemoryRegion *address_space_io,
370                          uint8_t devfn_min, int nirq, const char *typename)
371 {
372     PCIBus *bus;
373 
374     bus = pci_bus_new(parent, name, address_space_mem,
375                       address_space_io, devfn_min, typename);
376     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
377     return bus;
378 }
379 
380 int pci_bus_num(PCIBus *s)
381 {
382     if (pci_bus_is_root(s))
383         return 0;       /* pci host bridge */
384     return s->parent_dev->config[PCI_SECONDARY_BUS];
385 }
386 
387 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size)
388 {
389     PCIDevice *s = container_of(pv, PCIDevice, config);
390     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
391     uint8_t *config;
392     int i;
393 
394     assert(size == pci_config_size(s));
395     config = g_malloc(size);
396 
397     qemu_get_buffer(f, config, size);
398     for (i = 0; i < size; ++i) {
399         if ((config[i] ^ s->config[i]) &
400             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
401             g_free(config);
402             return -EINVAL;
403         }
404     }
405     memcpy(s->config, config, size);
406 
407     pci_update_mappings(s);
408     if (pc->is_bridge) {
409         PCIBridge *b = PCI_BRIDGE(s);
410         pci_bridge_update_mappings(b);
411     }
412 
413     memory_region_set_enabled(&s->bus_master_enable_region,
414                               pci_get_word(s->config + PCI_COMMAND)
415                               & PCI_COMMAND_MASTER);
416 
417     g_free(config);
418     return 0;
419 }
420 
421 /* just put buffer */
422 static void put_pci_config_device(QEMUFile *f, void *pv, size_t size)
423 {
424     const uint8_t **v = pv;
425     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
426     qemu_put_buffer(f, *v, size);
427 }
428 
429 static VMStateInfo vmstate_info_pci_config = {
430     .name = "pci config",
431     .get  = get_pci_config_device,
432     .put  = put_pci_config_device,
433 };
434 
435 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size)
436 {
437     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
438     uint32_t irq_state[PCI_NUM_PINS];
439     int i;
440     for (i = 0; i < PCI_NUM_PINS; ++i) {
441         irq_state[i] = qemu_get_be32(f);
442         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
443             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
444                     irq_state[i]);
445             return -EINVAL;
446         }
447     }
448 
449     for (i = 0; i < PCI_NUM_PINS; ++i) {
450         pci_set_irq_state(s, i, irq_state[i]);
451     }
452 
453     return 0;
454 }
455 
456 static void put_pci_irq_state(QEMUFile *f, void *pv, size_t size)
457 {
458     int i;
459     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
460 
461     for (i = 0; i < PCI_NUM_PINS; ++i) {
462         qemu_put_be32(f, pci_irq_state(s, i));
463     }
464 }
465 
466 static VMStateInfo vmstate_info_pci_irq_state = {
467     .name = "pci irq state",
468     .get  = get_pci_irq_state,
469     .put  = put_pci_irq_state,
470 };
471 
472 const VMStateDescription vmstate_pci_device = {
473     .name = "PCIDevice",
474     .version_id = 2,
475     .minimum_version_id = 1,
476     .minimum_version_id_old = 1,
477     .fields      = (VMStateField []) {
478         VMSTATE_INT32_LE(version_id, PCIDevice),
479         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
480                                    vmstate_info_pci_config,
481                                    PCI_CONFIG_SPACE_SIZE),
482         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
483 				   vmstate_info_pci_irq_state,
484 				   PCI_NUM_PINS * sizeof(int32_t)),
485         VMSTATE_END_OF_LIST()
486     }
487 };
488 
489 const VMStateDescription vmstate_pcie_device = {
490     .name = "PCIEDevice",
491     .version_id = 2,
492     .minimum_version_id = 1,
493     .minimum_version_id_old = 1,
494     .fields      = (VMStateField []) {
495         VMSTATE_INT32_LE(version_id, PCIDevice),
496         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
497                                    vmstate_info_pci_config,
498                                    PCIE_CONFIG_SPACE_SIZE),
499         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
500 				   vmstate_info_pci_irq_state,
501 				   PCI_NUM_PINS * sizeof(int32_t)),
502         VMSTATE_END_OF_LIST()
503     }
504 };
505 
506 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
507 {
508     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
509 }
510 
511 void pci_device_save(PCIDevice *s, QEMUFile *f)
512 {
513     /* Clear interrupt status bit: it is implicit
514      * in irq_state which we are saving.
515      * This makes us compatible with old devices
516      * which never set or clear this bit. */
517     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
518     vmstate_save_state(f, pci_get_vmstate(s), s);
519     /* Restore the interrupt status bit. */
520     pci_update_irq_status(s);
521 }
522 
523 int pci_device_load(PCIDevice *s, QEMUFile *f)
524 {
525     int ret;
526     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
527     /* Restore the interrupt status bit. */
528     pci_update_irq_status(s);
529     return ret;
530 }
531 
532 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
533 {
534     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
535                  pci_default_sub_vendor_id);
536     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
537                  pci_default_sub_device_id);
538 }
539 
540 /*
541  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
542  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
543  */
544 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
545                       unsigned int *slotp, unsigned int *funcp)
546 {
547     const char *p;
548     char *e;
549     unsigned long val;
550     unsigned long dom = 0, bus = 0;
551     unsigned int slot = 0;
552     unsigned int func = 0;
553 
554     p = addr;
555     val = strtoul(p, &e, 16);
556     if (e == p)
557 	return -1;
558     if (*e == ':') {
559 	bus = val;
560 	p = e + 1;
561 	val = strtoul(p, &e, 16);
562 	if (e == p)
563 	    return -1;
564 	if (*e == ':') {
565 	    dom = bus;
566 	    bus = val;
567 	    p = e + 1;
568 	    val = strtoul(p, &e, 16);
569 	    if (e == p)
570 		return -1;
571 	}
572     }
573 
574     slot = val;
575 
576     if (funcp != NULL) {
577         if (*e != '.')
578             return -1;
579 
580         p = e + 1;
581         val = strtoul(p, &e, 16);
582         if (e == p)
583             return -1;
584 
585         func = val;
586     }
587 
588     /* if funcp == NULL func is 0 */
589     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
590 	return -1;
591 
592     if (*e)
593 	return -1;
594 
595     *domp = dom;
596     *busp = bus;
597     *slotp = slot;
598     if (funcp != NULL)
599         *funcp = func;
600     return 0;
601 }
602 
603 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
604 {
605     int dom, bus;
606     unsigned slot;
607 
608     assert(!root->parent_dev);
609 
610     if (!root) {
611         fprintf(stderr, "No primary PCI bus\n");
612         return NULL;
613     }
614 
615     if (!devaddr) {
616         *devfnp = -1;
617         return pci_find_bus_nr(root, 0);
618     }
619 
620     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
621         return NULL;
622     }
623 
624     if (dom != 0) {
625         fprintf(stderr, "No support for non-zero PCI domains\n");
626         return NULL;
627     }
628 
629     *devfnp = PCI_DEVFN(slot, 0);
630     return pci_find_bus_nr(root, bus);
631 }
632 
633 static void pci_init_cmask(PCIDevice *dev)
634 {
635     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
636     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
637     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
638     dev->cmask[PCI_REVISION_ID] = 0xff;
639     dev->cmask[PCI_CLASS_PROG] = 0xff;
640     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
641     dev->cmask[PCI_HEADER_TYPE] = 0xff;
642     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
643 }
644 
645 static void pci_init_wmask(PCIDevice *dev)
646 {
647     int config_size = pci_config_size(dev);
648 
649     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
650     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
651     pci_set_word(dev->wmask + PCI_COMMAND,
652                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
653                  PCI_COMMAND_INTX_DISABLE);
654     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
655         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
656     }
657 
658     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
659            config_size - PCI_CONFIG_HEADER_SIZE);
660 }
661 
662 static void pci_init_w1cmask(PCIDevice *dev)
663 {
664     /*
665      * Note: It's okay to set w1cmask even for readonly bits as
666      * long as their value is hardwired to 0.
667      */
668     pci_set_word(dev->w1cmask + PCI_STATUS,
669                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
670                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
671                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
672 }
673 
674 static void pci_init_mask_bridge(PCIDevice *d)
675 {
676     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
677        PCI_SEC_LETENCY_TIMER */
678     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
679 
680     /* base and limit */
681     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
682     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
683     pci_set_word(d->wmask + PCI_MEMORY_BASE,
684                  PCI_MEMORY_RANGE_MASK & 0xffff);
685     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
686                  PCI_MEMORY_RANGE_MASK & 0xffff);
687     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
688                  PCI_PREF_RANGE_MASK & 0xffff);
689     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
690                  PCI_PREF_RANGE_MASK & 0xffff);
691 
692     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
693     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
694 
695     /* Supported memory and i/o types */
696     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
697     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
698     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
699                                PCI_PREF_RANGE_TYPE_64);
700     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
701                                PCI_PREF_RANGE_TYPE_64);
702 
703     /*
704      * TODO: Bridges default to 10-bit VGA decoding but we currently only
705      * implement 16-bit decoding (no alias support).
706      */
707     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
708                  PCI_BRIDGE_CTL_PARITY |
709                  PCI_BRIDGE_CTL_SERR |
710                  PCI_BRIDGE_CTL_ISA |
711                  PCI_BRIDGE_CTL_VGA |
712                  PCI_BRIDGE_CTL_VGA_16BIT |
713                  PCI_BRIDGE_CTL_MASTER_ABORT |
714                  PCI_BRIDGE_CTL_BUS_RESET |
715                  PCI_BRIDGE_CTL_FAST_BACK |
716                  PCI_BRIDGE_CTL_DISCARD |
717                  PCI_BRIDGE_CTL_SEC_DISCARD |
718                  PCI_BRIDGE_CTL_DISCARD_SERR);
719     /* Below does not do anything as we never set this bit, put here for
720      * completeness. */
721     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
722                  PCI_BRIDGE_CTL_DISCARD_STATUS);
723     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
724     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
725     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
726                                PCI_PREF_RANGE_TYPE_MASK);
727     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
728                                PCI_PREF_RANGE_TYPE_MASK);
729 }
730 
731 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
732 {
733     uint8_t slot = PCI_SLOT(dev->devfn);
734     uint8_t func;
735 
736     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
737         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
738     }
739 
740     /*
741      * multifunction bit is interpreted in two ways as follows.
742      *   - all functions must set the bit to 1.
743      *     Example: Intel X53
744      *   - function 0 must set the bit, but the rest function (> 0)
745      *     is allowed to leave the bit to 0.
746      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
747      *
748      * So OS (at least Linux) checks the bit of only function 0,
749      * and doesn't see the bit of function > 0.
750      *
751      * The below check allows both interpretation.
752      */
753     if (PCI_FUNC(dev->devfn)) {
754         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
755         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
756             /* function 0 should set multifunction bit */
757             error_report("PCI: single function device can't be populated "
758                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
759             return -1;
760         }
761         return 0;
762     }
763 
764     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
765         return 0;
766     }
767     /* function 0 indicates single function, so function > 0 must be NULL */
768     for (func = 1; func < PCI_FUNC_MAX; ++func) {
769         if (bus->devices[PCI_DEVFN(slot, func)]) {
770             error_report("PCI: %x.0 indicates single function, "
771                          "but %x.%x is already populated.",
772                          slot, slot, func);
773             return -1;
774         }
775     }
776     return 0;
777 }
778 
779 static void pci_config_alloc(PCIDevice *pci_dev)
780 {
781     int config_size = pci_config_size(pci_dev);
782 
783     pci_dev->config = g_malloc0(config_size);
784     pci_dev->cmask = g_malloc0(config_size);
785     pci_dev->wmask = g_malloc0(config_size);
786     pci_dev->w1cmask = g_malloc0(config_size);
787     pci_dev->used = g_malloc0(config_size);
788 }
789 
790 static void pci_config_free(PCIDevice *pci_dev)
791 {
792     g_free(pci_dev->config);
793     g_free(pci_dev->cmask);
794     g_free(pci_dev->wmask);
795     g_free(pci_dev->w1cmask);
796     g_free(pci_dev->used);
797 }
798 
799 static void do_pci_unregister_device(PCIDevice *pci_dev)
800 {
801     pci_dev->bus->devices[pci_dev->devfn] = NULL;
802     pci_config_free(pci_dev);
803 
804     address_space_destroy(&pci_dev->bus_master_as);
805     memory_region_destroy(&pci_dev->bus_master_enable_region);
806 }
807 
808 /* -1 for devfn means auto assign */
809 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
810                                          const char *name, int devfn)
811 {
812     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
813     PCIConfigReadFunc *config_read = pc->config_read;
814     PCIConfigWriteFunc *config_write = pc->config_write;
815     AddressSpace *dma_as;
816 
817     if (devfn < 0) {
818         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
819             devfn += PCI_FUNC_MAX) {
820             if (!bus->devices[devfn])
821                 goto found;
822         }
823         error_report("PCI: no slot/function available for %s, all in use", name);
824         return NULL;
825     found: ;
826     } else if (bus->devices[devfn]) {
827         error_report("PCI: slot %d function %d not available for %s, in use by %s",
828                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
829         return NULL;
830     }
831 
832     pci_dev->bus = bus;
833     dma_as = pci_device_iommu_address_space(pci_dev);
834 
835     memory_region_init_alias(&pci_dev->bus_master_enable_region,
836                              OBJECT(pci_dev), "bus master",
837                              dma_as->root, 0, memory_region_size(dma_as->root));
838     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
839     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
840                        name);
841 
842     pci_dev->devfn = devfn;
843     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
844     pci_dev->irq_state = 0;
845     pci_config_alloc(pci_dev);
846 
847     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
848     pci_config_set_device_id(pci_dev->config, pc->device_id);
849     pci_config_set_revision(pci_dev->config, pc->revision);
850     pci_config_set_class(pci_dev->config, pc->class_id);
851 
852     if (!pc->is_bridge) {
853         if (pc->subsystem_vendor_id || pc->subsystem_id) {
854             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
855                          pc->subsystem_vendor_id);
856             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
857                          pc->subsystem_id);
858         } else {
859             pci_set_default_subsystem_id(pci_dev);
860         }
861     } else {
862         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
863         assert(!pc->subsystem_vendor_id);
864         assert(!pc->subsystem_id);
865     }
866     pci_init_cmask(pci_dev);
867     pci_init_wmask(pci_dev);
868     pci_init_w1cmask(pci_dev);
869     if (pc->is_bridge) {
870         pci_init_mask_bridge(pci_dev);
871     }
872     if (pci_init_multifunction(bus, pci_dev)) {
873         do_pci_unregister_device(pci_dev);
874         return NULL;
875     }
876 
877     if (!config_read)
878         config_read = pci_default_read_config;
879     if (!config_write)
880         config_write = pci_default_write_config;
881     pci_dev->config_read = config_read;
882     pci_dev->config_write = config_write;
883     bus->devices[devfn] = pci_dev;
884     pci_dev->version_id = 2; /* Current pci device vmstate version */
885     return pci_dev;
886 }
887 
888 static void pci_unregister_io_regions(PCIDevice *pci_dev)
889 {
890     PCIIORegion *r;
891     int i;
892 
893     for(i = 0; i < PCI_NUM_REGIONS; i++) {
894         r = &pci_dev->io_regions[i];
895         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
896             continue;
897         memory_region_del_subregion(r->address_space, r->memory);
898     }
899 
900     pci_unregister_vga(pci_dev);
901 }
902 
903 static int pci_unregister_device(DeviceState *dev)
904 {
905     PCIDevice *pci_dev = PCI_DEVICE(dev);
906     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
907 
908     pci_unregister_io_regions(pci_dev);
909     pci_del_option_rom(pci_dev);
910 
911     if (pc->exit) {
912         pc->exit(pci_dev);
913     }
914 
915     do_pci_unregister_device(pci_dev);
916     return 0;
917 }
918 
919 void pci_register_bar(PCIDevice *pci_dev, int region_num,
920                       uint8_t type, MemoryRegion *memory)
921 {
922     PCIIORegion *r;
923     uint32_t addr;
924     uint64_t wmask;
925     pcibus_t size = memory_region_size(memory);
926 
927     assert(region_num >= 0);
928     assert(region_num < PCI_NUM_REGIONS);
929     if (size & (size-1)) {
930         fprintf(stderr, "ERROR: PCI region size must be pow2 "
931                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
932         exit(1);
933     }
934 
935     r = &pci_dev->io_regions[region_num];
936     r->addr = PCI_BAR_UNMAPPED;
937     r->size = size;
938     r->type = type;
939     r->memory = NULL;
940 
941     wmask = ~(size - 1);
942     addr = pci_bar(pci_dev, region_num);
943     if (region_num == PCI_ROM_SLOT) {
944         /* ROM enable bit is writable */
945         wmask |= PCI_ROM_ADDRESS_ENABLE;
946     }
947     pci_set_long(pci_dev->config + addr, type);
948     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
949         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
950         pci_set_quad(pci_dev->wmask + addr, wmask);
951         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
952     } else {
953         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
954         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
955     }
956     pci_dev->io_regions[region_num].memory = memory;
957     pci_dev->io_regions[region_num].address_space
958         = type & PCI_BASE_ADDRESS_SPACE_IO
959         ? pci_dev->bus->address_space_io
960         : pci_dev->bus->address_space_mem;
961 }
962 
963 static void pci_update_vga(PCIDevice *pci_dev)
964 {
965     uint16_t cmd;
966 
967     if (!pci_dev->has_vga) {
968         return;
969     }
970 
971     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
972 
973     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
974                               cmd & PCI_COMMAND_MEMORY);
975     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
976                               cmd & PCI_COMMAND_IO);
977     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
978                               cmd & PCI_COMMAND_IO);
979 }
980 
981 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
982                       MemoryRegion *io_lo, MemoryRegion *io_hi)
983 {
984     assert(!pci_dev->has_vga);
985 
986     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
987     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
988     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
989                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
990 
991     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
992     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
993     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
994                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
995 
996     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
997     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
998     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
999                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1000     pci_dev->has_vga = true;
1001 
1002     pci_update_vga(pci_dev);
1003 }
1004 
1005 void pci_unregister_vga(PCIDevice *pci_dev)
1006 {
1007     if (!pci_dev->has_vga) {
1008         return;
1009     }
1010 
1011     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1012                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1013     memory_region_del_subregion(pci_dev->bus->address_space_io,
1014                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1015     memory_region_del_subregion(pci_dev->bus->address_space_io,
1016                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1017     pci_dev->has_vga = false;
1018 }
1019 
1020 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1021 {
1022     return pci_dev->io_regions[region_num].addr;
1023 }
1024 
1025 static pcibus_t pci_bar_address(PCIDevice *d,
1026 				int reg, uint8_t type, pcibus_t size)
1027 {
1028     pcibus_t new_addr, last_addr;
1029     int bar = pci_bar(d, reg);
1030     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1031 
1032     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1033         if (!(cmd & PCI_COMMAND_IO)) {
1034             return PCI_BAR_UNMAPPED;
1035         }
1036         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1037         last_addr = new_addr + size - 1;
1038         /* Check if 32 bit BAR wraps around explicitly.
1039          * TODO: make priorities correct and remove this work around.
1040          */
1041         if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
1042             return PCI_BAR_UNMAPPED;
1043         }
1044         return new_addr;
1045     }
1046 
1047     if (!(cmd & PCI_COMMAND_MEMORY)) {
1048         return PCI_BAR_UNMAPPED;
1049     }
1050     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1051         new_addr = pci_get_quad(d->config + bar);
1052     } else {
1053         new_addr = pci_get_long(d->config + bar);
1054     }
1055     /* the ROM slot has a specific enable bit */
1056     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1057         return PCI_BAR_UNMAPPED;
1058     }
1059     new_addr &= ~(size - 1);
1060     last_addr = new_addr + size - 1;
1061     /* NOTE: we do not support wrapping */
1062     /* XXX: as we cannot support really dynamic
1063        mappings, we handle specific values as invalid
1064        mappings. */
1065     if (last_addr <= new_addr || new_addr == 0 ||
1066         last_addr == PCI_BAR_UNMAPPED) {
1067         return PCI_BAR_UNMAPPED;
1068     }
1069 
1070     /* Now pcibus_t is 64bit.
1071      * Check if 32 bit BAR wraps around explicitly.
1072      * Without this, PC ide doesn't work well.
1073      * TODO: remove this work around.
1074      */
1075     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1076         return PCI_BAR_UNMAPPED;
1077     }
1078 
1079     /*
1080      * OS is allowed to set BAR beyond its addressable
1081      * bits. For example, 32 bit OS can set 64bit bar
1082      * to >4G. Check it. TODO: we might need to support
1083      * it in the future for e.g. PAE.
1084      */
1085     if (last_addr >= HWADDR_MAX) {
1086         return PCI_BAR_UNMAPPED;
1087     }
1088 
1089     return new_addr;
1090 }
1091 
1092 static void pci_update_mappings(PCIDevice *d)
1093 {
1094     PCIIORegion *r;
1095     int i;
1096     pcibus_t new_addr;
1097 
1098     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1099         r = &d->io_regions[i];
1100 
1101         /* this region isn't registered */
1102         if (!r->size)
1103             continue;
1104 
1105         new_addr = pci_bar_address(d, i, r->type, r->size);
1106 
1107         /* This bar isn't changed */
1108         if (new_addr == r->addr)
1109             continue;
1110 
1111         /* now do the real mapping */
1112         if (r->addr != PCI_BAR_UNMAPPED) {
1113             memory_region_del_subregion(r->address_space, r->memory);
1114         }
1115         r->addr = new_addr;
1116         if (r->addr != PCI_BAR_UNMAPPED) {
1117             memory_region_add_subregion_overlap(r->address_space,
1118                                                 r->addr, r->memory, 1);
1119         }
1120     }
1121 
1122     pci_update_vga(d);
1123 }
1124 
1125 static inline int pci_irq_disabled(PCIDevice *d)
1126 {
1127     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1128 }
1129 
1130 /* Called after interrupt disabled field update in config space,
1131  * assert/deassert interrupts if necessary.
1132  * Gets original interrupt disable bit value (before update). */
1133 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1134 {
1135     int i, disabled = pci_irq_disabled(d);
1136     if (disabled == was_irq_disabled)
1137         return;
1138     for (i = 0; i < PCI_NUM_PINS; ++i) {
1139         int state = pci_irq_state(d, i);
1140         pci_change_irq_level(d, i, disabled ? -state : state);
1141     }
1142 }
1143 
1144 uint32_t pci_default_read_config(PCIDevice *d,
1145                                  uint32_t address, int len)
1146 {
1147     uint32_t val = 0;
1148 
1149     memcpy(&val, d->config + address, len);
1150     return le32_to_cpu(val);
1151 }
1152 
1153 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
1154 {
1155     int i, was_irq_disabled = pci_irq_disabled(d);
1156 
1157     for (i = 0; i < l; val >>= 8, ++i) {
1158         uint8_t wmask = d->wmask[addr + i];
1159         uint8_t w1cmask = d->w1cmask[addr + i];
1160         assert(!(wmask & w1cmask));
1161         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1162         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1163     }
1164     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1165         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1166         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1167         range_covers_byte(addr, l, PCI_COMMAND))
1168         pci_update_mappings(d);
1169 
1170     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1171         pci_update_irq_disabled(d, was_irq_disabled);
1172         memory_region_set_enabled(&d->bus_master_enable_region,
1173                                   pci_get_word(d->config + PCI_COMMAND)
1174                                     & PCI_COMMAND_MASTER);
1175     }
1176 
1177     msi_write_config(d, addr, val, l);
1178     msix_write_config(d, addr, val, l);
1179 }
1180 
1181 /***********************************************************/
1182 /* generic PCI irq support */
1183 
1184 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1185 static void pci_irq_handler(void *opaque, int irq_num, int level)
1186 {
1187     PCIDevice *pci_dev = opaque;
1188     int change;
1189 
1190     change = level - pci_irq_state(pci_dev, irq_num);
1191     if (!change)
1192         return;
1193 
1194     pci_set_irq_state(pci_dev, irq_num, level);
1195     pci_update_irq_status(pci_dev);
1196     if (pci_irq_disabled(pci_dev))
1197         return;
1198     pci_change_irq_level(pci_dev, irq_num, change);
1199 }
1200 
1201 static inline int pci_intx(PCIDevice *pci_dev)
1202 {
1203     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1204 }
1205 
1206 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1207 {
1208     int intx = pci_intx(pci_dev);
1209 
1210     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1211 }
1212 
1213 void pci_set_irq(PCIDevice *pci_dev, int level)
1214 {
1215     int intx = pci_intx(pci_dev);
1216     pci_irq_handler(pci_dev, intx, level);
1217 }
1218 
1219 /* Special hooks used by device assignment */
1220 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1221 {
1222     assert(pci_bus_is_root(bus));
1223     bus->route_intx_to_irq = route_intx_to_irq;
1224 }
1225 
1226 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1227 {
1228     PCIBus *bus;
1229 
1230     do {
1231          bus = dev->bus;
1232          pin = bus->map_irq(dev, pin);
1233          dev = bus->parent_dev;
1234     } while (dev);
1235 
1236     if (!bus->route_intx_to_irq) {
1237         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1238                      object_get_typename(OBJECT(bus->qbus.parent)));
1239         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1240     }
1241 
1242     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1243 }
1244 
1245 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1246 {
1247     return old->mode != new->mode || old->irq != new->irq;
1248 }
1249 
1250 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1251 {
1252     PCIDevice *dev;
1253     PCIBus *sec;
1254     int i;
1255 
1256     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1257         dev = bus->devices[i];
1258         if (dev && dev->intx_routing_notifier) {
1259             dev->intx_routing_notifier(dev);
1260         }
1261     }
1262 
1263     QLIST_FOREACH(sec, &bus->child, sibling) {
1264         pci_bus_fire_intx_routing_notifier(sec);
1265     }
1266 }
1267 
1268 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1269                                           PCIINTxRoutingNotifier notifier)
1270 {
1271     dev->intx_routing_notifier = notifier;
1272 }
1273 
1274 /*
1275  * PCI-to-PCI bridge specification
1276  * 9.1: Interrupt routing. Table 9-1
1277  *
1278  * the PCI Express Base Specification, Revision 2.1
1279  * 2.2.8.1: INTx interrutp signaling - Rules
1280  *          the Implementation Note
1281  *          Table 2-20
1282  */
1283 /*
1284  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1285  * 0-origin unlike PCI interrupt pin register.
1286  */
1287 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1288 {
1289     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1290 }
1291 
1292 /***********************************************************/
1293 /* monitor info on PCI */
1294 
1295 typedef struct {
1296     uint16_t class;
1297     const char *desc;
1298     const char *fw_name;
1299     uint16_t fw_ign_bits;
1300 } pci_class_desc;
1301 
1302 static const pci_class_desc pci_class_descriptions[] =
1303 {
1304     { 0x0001, "VGA controller", "display"},
1305     { 0x0100, "SCSI controller", "scsi"},
1306     { 0x0101, "IDE controller", "ide"},
1307     { 0x0102, "Floppy controller", "fdc"},
1308     { 0x0103, "IPI controller", "ipi"},
1309     { 0x0104, "RAID controller", "raid"},
1310     { 0x0106, "SATA controller"},
1311     { 0x0107, "SAS controller"},
1312     { 0x0180, "Storage controller"},
1313     { 0x0200, "Ethernet controller", "ethernet"},
1314     { 0x0201, "Token Ring controller", "token-ring"},
1315     { 0x0202, "FDDI controller", "fddi"},
1316     { 0x0203, "ATM controller", "atm"},
1317     { 0x0280, "Network controller"},
1318     { 0x0300, "VGA controller", "display", 0x00ff},
1319     { 0x0301, "XGA controller"},
1320     { 0x0302, "3D controller"},
1321     { 0x0380, "Display controller"},
1322     { 0x0400, "Video controller", "video"},
1323     { 0x0401, "Audio controller", "sound"},
1324     { 0x0402, "Phone"},
1325     { 0x0403, "Audio controller", "sound"},
1326     { 0x0480, "Multimedia controller"},
1327     { 0x0500, "RAM controller", "memory"},
1328     { 0x0501, "Flash controller", "flash"},
1329     { 0x0580, "Memory controller"},
1330     { 0x0600, "Host bridge", "host"},
1331     { 0x0601, "ISA bridge", "isa"},
1332     { 0x0602, "EISA bridge", "eisa"},
1333     { 0x0603, "MC bridge", "mca"},
1334     { 0x0604, "PCI bridge", "pci-bridge"},
1335     { 0x0605, "PCMCIA bridge", "pcmcia"},
1336     { 0x0606, "NUBUS bridge", "nubus"},
1337     { 0x0607, "CARDBUS bridge", "cardbus"},
1338     { 0x0608, "RACEWAY bridge"},
1339     { 0x0680, "Bridge"},
1340     { 0x0700, "Serial port", "serial"},
1341     { 0x0701, "Parallel port", "parallel"},
1342     { 0x0800, "Interrupt controller", "interrupt-controller"},
1343     { 0x0801, "DMA controller", "dma-controller"},
1344     { 0x0802, "Timer", "timer"},
1345     { 0x0803, "RTC", "rtc"},
1346     { 0x0900, "Keyboard", "keyboard"},
1347     { 0x0901, "Pen", "pen"},
1348     { 0x0902, "Mouse", "mouse"},
1349     { 0x0A00, "Dock station", "dock", 0x00ff},
1350     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1351     { 0x0c00, "Fireware contorller", "fireware"},
1352     { 0x0c01, "Access bus controller", "access-bus"},
1353     { 0x0c02, "SSA controller", "ssa"},
1354     { 0x0c03, "USB controller", "usb"},
1355     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1356     { 0x0c05, "SMBus"},
1357     { 0, NULL}
1358 };
1359 
1360 static void pci_for_each_device_under_bus(PCIBus *bus,
1361                                           void (*fn)(PCIBus *b, PCIDevice *d,
1362                                                      void *opaque),
1363                                           void *opaque)
1364 {
1365     PCIDevice *d;
1366     int devfn;
1367 
1368     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1369         d = bus->devices[devfn];
1370         if (d) {
1371             fn(bus, d, opaque);
1372         }
1373     }
1374 }
1375 
1376 void pci_for_each_device(PCIBus *bus, int bus_num,
1377                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1378                          void *opaque)
1379 {
1380     bus = pci_find_bus_nr(bus, bus_num);
1381 
1382     if (bus) {
1383         pci_for_each_device_under_bus(bus, fn, opaque);
1384     }
1385 }
1386 
1387 static const pci_class_desc *get_class_desc(int class)
1388 {
1389     const pci_class_desc *desc;
1390 
1391     desc = pci_class_descriptions;
1392     while (desc->desc && class != desc->class) {
1393         desc++;
1394     }
1395 
1396     return desc;
1397 }
1398 
1399 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1400 
1401 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1402 {
1403     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1404     int i;
1405 
1406     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1407         const PCIIORegion *r = &dev->io_regions[i];
1408         PciMemoryRegionList *region;
1409 
1410         if (!r->size) {
1411             continue;
1412         }
1413 
1414         region = g_malloc0(sizeof(*region));
1415         region->value = g_malloc0(sizeof(*region->value));
1416 
1417         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1418             region->value->type = g_strdup("io");
1419         } else {
1420             region->value->type = g_strdup("memory");
1421             region->value->has_prefetch = true;
1422             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1423             region->value->has_mem_type_64 = true;
1424             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1425         }
1426 
1427         region->value->bar = i;
1428         region->value->address = r->addr;
1429         region->value->size = r->size;
1430 
1431         /* XXX: waiting for the qapi to support GSList */
1432         if (!cur_item) {
1433             head = cur_item = region;
1434         } else {
1435             cur_item->next = region;
1436             cur_item = region;
1437         }
1438     }
1439 
1440     return head;
1441 }
1442 
1443 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1444                                            int bus_num)
1445 {
1446     PciBridgeInfo *info;
1447 
1448     info = g_malloc0(sizeof(*info));
1449 
1450     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1451     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1452     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1453 
1454     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1455     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1456     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1457 
1458     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1459     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1460     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1461 
1462     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1463     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1464     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1465 
1466     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1467         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1468         if (child_bus) {
1469             info->has_devices = true;
1470             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1471         }
1472     }
1473 
1474     return info;
1475 }
1476 
1477 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1478                                            int bus_num)
1479 {
1480     const pci_class_desc *desc;
1481     PciDeviceInfo *info;
1482     uint8_t type;
1483     int class;
1484 
1485     info = g_malloc0(sizeof(*info));
1486     info->bus = bus_num;
1487     info->slot = PCI_SLOT(dev->devfn);
1488     info->function = PCI_FUNC(dev->devfn);
1489 
1490     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1491     info->class_info.q_class = class;
1492     desc = get_class_desc(class);
1493     if (desc->desc) {
1494         info->class_info.has_desc = true;
1495         info->class_info.desc = g_strdup(desc->desc);
1496     }
1497 
1498     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1499     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1500     info->regions = qmp_query_pci_regions(dev);
1501     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1502 
1503     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1504         info->has_irq = true;
1505         info->irq = dev->config[PCI_INTERRUPT_LINE];
1506     }
1507 
1508     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1509     if (type == PCI_HEADER_TYPE_BRIDGE) {
1510         info->has_pci_bridge = true;
1511         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1512     }
1513 
1514     return info;
1515 }
1516 
1517 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1518 {
1519     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1520     PCIDevice *dev;
1521     int devfn;
1522 
1523     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1524         dev = bus->devices[devfn];
1525         if (dev) {
1526             info = g_malloc0(sizeof(*info));
1527             info->value = qmp_query_pci_device(dev, bus, bus_num);
1528 
1529             /* XXX: waiting for the qapi to support GSList */
1530             if (!cur_item) {
1531                 head = cur_item = info;
1532             } else {
1533                 cur_item->next = info;
1534                 cur_item = info;
1535             }
1536         }
1537     }
1538 
1539     return head;
1540 }
1541 
1542 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1543 {
1544     PciInfo *info = NULL;
1545 
1546     bus = pci_find_bus_nr(bus, bus_num);
1547     if (bus) {
1548         info = g_malloc0(sizeof(*info));
1549         info->bus = bus_num;
1550         info->devices = qmp_query_pci_devices(bus, bus_num);
1551     }
1552 
1553     return info;
1554 }
1555 
1556 PciInfoList *qmp_query_pci(Error **errp)
1557 {
1558     PciInfoList *info, *head = NULL, *cur_item = NULL;
1559     PCIHostState *host_bridge;
1560 
1561     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1562         info = g_malloc0(sizeof(*info));
1563         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1564 
1565         /* XXX: waiting for the qapi to support GSList */
1566         if (!cur_item) {
1567             head = cur_item = info;
1568         } else {
1569             cur_item->next = info;
1570             cur_item = info;
1571         }
1572     }
1573 
1574     return head;
1575 }
1576 
1577 static const char * const pci_nic_models[] = {
1578     "ne2k_pci",
1579     "i82551",
1580     "i82557b",
1581     "i82559er",
1582     "rtl8139",
1583     "e1000",
1584     "pcnet",
1585     "virtio",
1586     NULL
1587 };
1588 
1589 static const char * const pci_nic_names[] = {
1590     "ne2k_pci",
1591     "i82551",
1592     "i82557b",
1593     "i82559er",
1594     "rtl8139",
1595     "e1000",
1596     "pcnet",
1597     "virtio-net-pci",
1598     NULL
1599 };
1600 
1601 /* Initialize a PCI NIC.  */
1602 /* FIXME callers should check for failure, but don't */
1603 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1604                         const char *default_model,
1605                         const char *default_devaddr)
1606 {
1607     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1608     PCIBus *bus;
1609     int devfn;
1610     PCIDevice *pci_dev;
1611     DeviceState *dev;
1612     int i;
1613 
1614     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1615     if (i < 0)
1616         return NULL;
1617 
1618     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1619     if (!bus) {
1620         error_report("Invalid PCI device address %s for device %s",
1621                      devaddr, pci_nic_names[i]);
1622         return NULL;
1623     }
1624 
1625     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1626     dev = &pci_dev->qdev;
1627     qdev_set_nic_properties(dev, nd);
1628     if (qdev_init(dev) < 0)
1629         return NULL;
1630     return pci_dev;
1631 }
1632 
1633 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1634                                const char *default_model,
1635                                const char *default_devaddr)
1636 {
1637     PCIDevice *res;
1638 
1639     if (qemu_show_nic_models(nd->model, pci_nic_models))
1640         exit(0);
1641 
1642     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1643     if (!res)
1644         exit(1);
1645     return res;
1646 }
1647 
1648 PCIDevice *pci_vga_init(PCIBus *bus)
1649 {
1650     switch (vga_interface_type) {
1651     case VGA_CIRRUS:
1652         return pci_create_simple(bus, -1, "cirrus-vga");
1653     case VGA_QXL:
1654         return pci_create_simple(bus, -1, "qxl-vga");
1655     case VGA_STD:
1656         return pci_create_simple(bus, -1, "VGA");
1657     case VGA_VMWARE:
1658         return pci_create_simple(bus, -1, "vmware-svga");
1659     case VGA_NONE:
1660     default: /* Other non-PCI types. Checking for unsupported types is already
1661                 done in vl.c. */
1662         return NULL;
1663     }
1664 }
1665 
1666 /* Whether a given bus number is in range of the secondary
1667  * bus of the given bridge device. */
1668 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1669 {
1670     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1671              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1672         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1673         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1674 }
1675 
1676 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1677 {
1678     PCIBus *sec;
1679 
1680     if (!bus) {
1681         return NULL;
1682     }
1683 
1684     if (pci_bus_num(bus) == bus_num) {
1685         return bus;
1686     }
1687 
1688     /* Consider all bus numbers in range for the host pci bridge. */
1689     if (!pci_bus_is_root(bus) &&
1690         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1691         return NULL;
1692     }
1693 
1694     /* try child bus */
1695     for (; bus; bus = sec) {
1696         QLIST_FOREACH(sec, &bus->child, sibling) {
1697             assert(!pci_bus_is_root(sec));
1698             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1699                 return sec;
1700             }
1701             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1702                 break;
1703             }
1704         }
1705     }
1706 
1707     return NULL;
1708 }
1709 
1710 void pci_for_each_bus_depth_first(PCIBus *bus,
1711                                   void *(*begin)(PCIBus *bus, void *parent_state),
1712                                   void (*end)(PCIBus *bus, void *state),
1713                                   void *parent_state)
1714 {
1715     PCIBus *sec;
1716     void *state;
1717 
1718     if (!bus) {
1719         return;
1720     }
1721 
1722     if (begin) {
1723         state = begin(bus, parent_state);
1724     } else {
1725         state = parent_state;
1726     }
1727 
1728     QLIST_FOREACH(sec, &bus->child, sibling) {
1729         pci_for_each_bus_depth_first(sec, begin, end, state);
1730     }
1731 
1732     if (end) {
1733         end(bus, state);
1734     }
1735 }
1736 
1737 
1738 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1739 {
1740     bus = pci_find_bus_nr(bus, bus_num);
1741 
1742     if (!bus)
1743         return NULL;
1744 
1745     return bus->devices[devfn];
1746 }
1747 
1748 static int pci_qdev_init(DeviceState *qdev)
1749 {
1750     PCIDevice *pci_dev = (PCIDevice *)qdev;
1751     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1752     PCIBus *bus;
1753     int rc;
1754     bool is_default_rom;
1755 
1756     /* initialize cap_present for pci_is_express() and pci_config_size() */
1757     if (pc->is_express) {
1758         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1759     }
1760 
1761     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1762     pci_dev = do_pci_register_device(pci_dev, bus,
1763                                      object_get_typename(OBJECT(qdev)),
1764                                      pci_dev->devfn);
1765     if (pci_dev == NULL)
1766         return -1;
1767 
1768     if (pc->init) {
1769         rc = pc->init(pci_dev);
1770         if (rc != 0) {
1771             do_pci_unregister_device(pci_dev);
1772             return rc;
1773         }
1774     }
1775 
1776     /* rom loading */
1777     is_default_rom = false;
1778     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1779         pci_dev->romfile = g_strdup(pc->romfile);
1780         is_default_rom = true;
1781     }
1782     pci_add_option_rom(pci_dev, is_default_rom);
1783 
1784     return 0;
1785 }
1786 
1787 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1788                                     const char *name)
1789 {
1790     DeviceState *dev;
1791 
1792     dev = qdev_create(&bus->qbus, name);
1793     qdev_prop_set_int32(dev, "addr", devfn);
1794     qdev_prop_set_bit(dev, "multifunction", multifunction);
1795     return PCI_DEVICE(dev);
1796 }
1797 
1798 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1799                                            bool multifunction,
1800                                            const char *name)
1801 {
1802     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1803     qdev_init_nofail(&dev->qdev);
1804     return dev;
1805 }
1806 
1807 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1808 {
1809     return pci_create_multifunction(bus, devfn, false, name);
1810 }
1811 
1812 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1813 {
1814     return pci_create_simple_multifunction(bus, devfn, false, name);
1815 }
1816 
1817 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1818 {
1819     int offset = PCI_CONFIG_HEADER_SIZE;
1820     int i;
1821     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1822         if (pdev->used[i])
1823             offset = i + 1;
1824         else if (i - offset + 1 == size)
1825             return offset;
1826     }
1827     return 0;
1828 }
1829 
1830 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1831                                         uint8_t *prev_p)
1832 {
1833     uint8_t next, prev;
1834 
1835     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1836         return 0;
1837 
1838     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1839          prev = next + PCI_CAP_LIST_NEXT)
1840         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1841             break;
1842 
1843     if (prev_p)
1844         *prev_p = prev;
1845     return next;
1846 }
1847 
1848 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1849 {
1850     uint8_t next, prev, found = 0;
1851 
1852     if (!(pdev->used[offset])) {
1853         return 0;
1854     }
1855 
1856     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1857 
1858     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1859          prev = next + PCI_CAP_LIST_NEXT) {
1860         if (next <= offset && next > found) {
1861             found = next;
1862         }
1863     }
1864     return found;
1865 }
1866 
1867 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1868    This is needed for an option rom which is used for more than one device. */
1869 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1870 {
1871     uint16_t vendor_id;
1872     uint16_t device_id;
1873     uint16_t rom_vendor_id;
1874     uint16_t rom_device_id;
1875     uint16_t rom_magic;
1876     uint16_t pcir_offset;
1877     uint8_t checksum;
1878 
1879     /* Words in rom data are little endian (like in PCI configuration),
1880        so they can be read / written with pci_get_word / pci_set_word. */
1881 
1882     /* Only a valid rom will be patched. */
1883     rom_magic = pci_get_word(ptr);
1884     if (rom_magic != 0xaa55) {
1885         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1886         return;
1887     }
1888     pcir_offset = pci_get_word(ptr + 0x18);
1889     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1890         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1891         return;
1892     }
1893 
1894     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1895     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1896     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1897     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1898 
1899     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1900                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1901 
1902     checksum = ptr[6];
1903 
1904     if (vendor_id != rom_vendor_id) {
1905         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1906         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1907         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1908         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1909         ptr[6] = checksum;
1910         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1911     }
1912 
1913     if (device_id != rom_device_id) {
1914         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1915         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1916         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1917         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1918         ptr[6] = checksum;
1919         pci_set_word(ptr + pcir_offset + 6, device_id);
1920     }
1921 }
1922 
1923 /* Add an option rom for the device */
1924 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1925 {
1926     int size;
1927     char *path;
1928     void *ptr;
1929     char name[32];
1930     const VMStateDescription *vmsd;
1931 
1932     if (!pdev->romfile)
1933         return 0;
1934     if (strlen(pdev->romfile) == 0)
1935         return 0;
1936 
1937     if (!pdev->rom_bar) {
1938         /*
1939          * Load rom via fw_cfg instead of creating a rom bar,
1940          * for 0.11 compatibility.
1941          */
1942         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1943         if (class == 0x0300) {
1944             rom_add_vga(pdev->romfile);
1945         } else {
1946             rom_add_option(pdev->romfile, -1);
1947         }
1948         return 0;
1949     }
1950 
1951     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1952     if (path == NULL) {
1953         path = g_strdup(pdev->romfile);
1954     }
1955 
1956     size = get_image_size(path);
1957     if (size < 0) {
1958         error_report("%s: failed to find romfile \"%s\"",
1959                      __func__, pdev->romfile);
1960         g_free(path);
1961         return -1;
1962     } else if (size == 0) {
1963         error_report("%s: ignoring empty romfile \"%s\"",
1964                      __func__, pdev->romfile);
1965         g_free(path);
1966         return -1;
1967     }
1968     if (size & (size - 1)) {
1969         size = 1 << qemu_fls(size);
1970     }
1971 
1972     vmsd = qdev_get_vmsd(DEVICE(pdev));
1973 
1974     if (vmsd) {
1975         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1976     } else {
1977         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1978     }
1979     pdev->has_rom = true;
1980     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size);
1981     vmstate_register_ram(&pdev->rom, &pdev->qdev);
1982     ptr = memory_region_get_ram_ptr(&pdev->rom);
1983     load_image(path, ptr);
1984     g_free(path);
1985 
1986     if (is_default_rom) {
1987         /* Only the default rom images will be patched (if needed). */
1988         pci_patch_ids(pdev, ptr, size);
1989     }
1990 
1991     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
1992 
1993     return 0;
1994 }
1995 
1996 static void pci_del_option_rom(PCIDevice *pdev)
1997 {
1998     if (!pdev->has_rom)
1999         return;
2000 
2001     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2002     memory_region_destroy(&pdev->rom);
2003     pdev->has_rom = false;
2004 }
2005 
2006 /*
2007  * if !offset
2008  * Reserve space and add capability to the linked list in pci config space
2009  *
2010  * if offset = 0,
2011  * Find and reserve space and add capability to the linked list
2012  * in pci config space */
2013 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2014                        uint8_t offset, uint8_t size)
2015 {
2016     uint8_t *config;
2017     int i, overlapping_cap;
2018 
2019     if (!offset) {
2020         offset = pci_find_space(pdev, size);
2021         if (!offset) {
2022             return -ENOSPC;
2023         }
2024     } else {
2025         /* Verify that capabilities don't overlap.  Note: device assignment
2026          * depends on this check to verify that the device is not broken.
2027          * Should never trigger for emulated devices, but it's helpful
2028          * for debugging these. */
2029         for (i = offset; i < offset + size; i++) {
2030             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2031             if (overlapping_cap) {
2032                 fprintf(stderr, "ERROR: %s:%02x:%02x.%x "
2033                         "Attempt to add PCI capability %x at offset "
2034                         "%x overlaps existing capability %x at offset %x\n",
2035                         pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2036                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2037                         cap_id, offset, overlapping_cap, i);
2038                 return -EINVAL;
2039             }
2040         }
2041     }
2042 
2043     config = pdev->config + offset;
2044     config[PCI_CAP_LIST_ID] = cap_id;
2045     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2046     pdev->config[PCI_CAPABILITY_LIST] = offset;
2047     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2048     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2049     /* Make capability read-only by default */
2050     memset(pdev->wmask + offset, 0, size);
2051     /* Check capability by default */
2052     memset(pdev->cmask + offset, 0xFF, size);
2053     return offset;
2054 }
2055 
2056 /* Unlink capability from the pci config space. */
2057 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2058 {
2059     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2060     if (!offset)
2061         return;
2062     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2063     /* Make capability writable again */
2064     memset(pdev->wmask + offset, 0xff, size);
2065     memset(pdev->w1cmask + offset, 0, size);
2066     /* Clear cmask as device-specific registers can't be checked */
2067     memset(pdev->cmask + offset, 0, size);
2068     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2069 
2070     if (!pdev->config[PCI_CAPABILITY_LIST])
2071         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2072 }
2073 
2074 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2075 {
2076     return pci_find_capability_list(pdev, cap_id, NULL);
2077 }
2078 
2079 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2080 {
2081     PCIDevice *d = (PCIDevice *)dev;
2082     const pci_class_desc *desc;
2083     char ctxt[64];
2084     PCIIORegion *r;
2085     int i, class;
2086 
2087     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2088     desc = pci_class_descriptions;
2089     while (desc->desc && class != desc->class)
2090         desc++;
2091     if (desc->desc) {
2092         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2093     } else {
2094         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2095     }
2096 
2097     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2098                    "pci id %04x:%04x (sub %04x:%04x)\n",
2099                    indent, "", ctxt, pci_bus_num(d->bus),
2100                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2101                    pci_get_word(d->config + PCI_VENDOR_ID),
2102                    pci_get_word(d->config + PCI_DEVICE_ID),
2103                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2104                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2105     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2106         r = &d->io_regions[i];
2107         if (!r->size)
2108             continue;
2109         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2110                        " [0x%"FMT_PCIBUS"]\n",
2111                        indent, "",
2112                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2113                        r->addr, r->addr + r->size - 1);
2114     }
2115 }
2116 
2117 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2118 {
2119     PCIDevice *d = (PCIDevice *)dev;
2120     const char *name = NULL;
2121     const pci_class_desc *desc =  pci_class_descriptions;
2122     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2123 
2124     while (desc->desc &&
2125           (class & ~desc->fw_ign_bits) !=
2126           (desc->class & ~desc->fw_ign_bits)) {
2127         desc++;
2128     }
2129 
2130     if (desc->desc) {
2131         name = desc->fw_name;
2132     }
2133 
2134     if (name) {
2135         pstrcpy(buf, len, name);
2136     } else {
2137         snprintf(buf, len, "pci%04x,%04x",
2138                  pci_get_word(d->config + PCI_VENDOR_ID),
2139                  pci_get_word(d->config + PCI_DEVICE_ID));
2140     }
2141 
2142     return buf;
2143 }
2144 
2145 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2146 {
2147     PCIDevice *d = (PCIDevice *)dev;
2148     char path[50], name[33];
2149     int off;
2150 
2151     off = snprintf(path, sizeof(path), "%s@%x",
2152                    pci_dev_fw_name(dev, name, sizeof name),
2153                    PCI_SLOT(d->devfn));
2154     if (PCI_FUNC(d->devfn))
2155         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2156     return g_strdup(path);
2157 }
2158 
2159 static char *pcibus_get_dev_path(DeviceState *dev)
2160 {
2161     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2162     PCIDevice *t;
2163     int slot_depth;
2164     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2165      * 00 is added here to make this format compatible with
2166      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2167      * Slot.Function list specifies the slot and function numbers for all
2168      * devices on the path from root to the specific device. */
2169     const char *root_bus_path;
2170     int root_bus_len;
2171     char slot[] = ":SS.F";
2172     int slot_len = sizeof slot - 1 /* For '\0' */;
2173     int path_len;
2174     char *path, *p;
2175     int s;
2176 
2177     root_bus_path = pci_root_bus_path(d);
2178     root_bus_len = strlen(root_bus_path);
2179 
2180     /* Calculate # of slots on path between device and root. */;
2181     slot_depth = 0;
2182     for (t = d; t; t = t->bus->parent_dev) {
2183         ++slot_depth;
2184     }
2185 
2186     path_len = root_bus_len + slot_len * slot_depth;
2187 
2188     /* Allocate memory, fill in the terminating null byte. */
2189     path = g_malloc(path_len + 1 /* For '\0' */);
2190     path[path_len] = '\0';
2191 
2192     memcpy(path, root_bus_path, root_bus_len);
2193 
2194     /* Fill in slot numbers. We walk up from device to root, so need to print
2195      * them in the reverse order, last to first. */
2196     p = path + path_len;
2197     for (t = d; t; t = t->bus->parent_dev) {
2198         p -= slot_len;
2199         s = snprintf(slot, sizeof slot, ":%02x.%x",
2200                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2201         assert(s == slot_len);
2202         memcpy(p, slot, slot_len);
2203     }
2204 
2205     return path;
2206 }
2207 
2208 static int pci_qdev_find_recursive(PCIBus *bus,
2209                                    const char *id, PCIDevice **pdev)
2210 {
2211     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2212     if (!qdev) {
2213         return -ENODEV;
2214     }
2215 
2216     /* roughly check if given qdev is pci device */
2217     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2218         *pdev = PCI_DEVICE(qdev);
2219         return 0;
2220     }
2221     return -EINVAL;
2222 }
2223 
2224 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2225 {
2226     PCIHostState *host_bridge;
2227     int rc = -ENODEV;
2228 
2229     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2230         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2231         if (!tmp) {
2232             rc = 0;
2233             break;
2234         }
2235         if (tmp != -ENODEV) {
2236             rc = tmp;
2237         }
2238     }
2239 
2240     return rc;
2241 }
2242 
2243 MemoryRegion *pci_address_space(PCIDevice *dev)
2244 {
2245     return dev->bus->address_space_mem;
2246 }
2247 
2248 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2249 {
2250     return dev->bus->address_space_io;
2251 }
2252 
2253 static void pci_device_class_init(ObjectClass *klass, void *data)
2254 {
2255     DeviceClass *k = DEVICE_CLASS(klass);
2256     k->init = pci_qdev_init;
2257     k->exit = pci_unregister_device;
2258     k->bus_type = TYPE_PCI_BUS;
2259     k->props = pci_props;
2260 }
2261 
2262 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2263 {
2264     PCIBus *bus = PCI_BUS(dev->bus);
2265 
2266     if (bus->iommu_fn) {
2267         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2268     }
2269 
2270     if (bus->parent_dev) {
2271         /** We are ignoring the bus master DMA bit of the bridge
2272          *  as it would complicate things such as VFIO for no good reason */
2273         return pci_device_iommu_address_space(bus->parent_dev);
2274     }
2275 
2276     return &address_space_memory;
2277 }
2278 
2279 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2280 {
2281     bus->iommu_fn = fn;
2282     bus->iommu_opaque = opaque;
2283 }
2284 
2285 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2286 {
2287     Range *range = opaque;
2288     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2289     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2290     int i;
2291 
2292     if (!(cmd & PCI_COMMAND_MEMORY)) {
2293         return;
2294     }
2295 
2296     if (pc->is_bridge) {
2297         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2298         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2299 
2300         base = MAX(base, 0x1ULL << 32);
2301 
2302         if (limit >= base) {
2303             Range pref_range;
2304             pref_range.begin = base;
2305             pref_range.end = limit + 1;
2306             range_extend(range, &pref_range);
2307         }
2308     }
2309     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2310         PCIIORegion *r = &dev->io_regions[i];
2311         Range region_range;
2312 
2313         if (!r->size ||
2314             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2315             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2316             continue;
2317         }
2318         region_range.begin = pci_bar_address(dev, i, r->type, r->size);
2319         region_range.end = region_range.begin + r->size;
2320 
2321         if (region_range.begin == PCI_BAR_UNMAPPED) {
2322             continue;
2323         }
2324 
2325         region_range.begin = MAX(region_range.begin, 0x1ULL << 32);
2326 
2327         if (region_range.end - 1 >= region_range.begin) {
2328             range_extend(range, &region_range);
2329         }
2330     }
2331 }
2332 
2333 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2334 {
2335     range->begin = range->end = 0;
2336     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2337 }
2338 
2339 static const TypeInfo pci_device_type_info = {
2340     .name = TYPE_PCI_DEVICE,
2341     .parent = TYPE_DEVICE,
2342     .instance_size = sizeof(PCIDevice),
2343     .abstract = true,
2344     .class_size = sizeof(PCIDeviceClass),
2345     .class_init = pci_device_class_init,
2346 };
2347 
2348 static void pci_register_types(void)
2349 {
2350     type_register_static(&pci_bus_info);
2351     type_register_static(&pcie_bus_info);
2352     type_register_static(&pci_device_type_info);
2353 }
2354 
2355 type_init(pci_register_types)
2356