xref: /openbmc/qemu/hw/pci/pci.c (revision 87a45cfe)
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 "trace.h"
36 #include "hw/pci/msi.h"
37 #include "hw/pci/msix.h"
38 #include "exec/address-spaces.h"
39 #include "hw/hotplug.h"
40 
41 //#define DEBUG_PCI
42 #ifdef DEBUG_PCI
43 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
44 #else
45 # define PCI_DPRINTF(format, ...)       do { } while (0)
46 #endif
47 
48 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
49 static char *pcibus_get_dev_path(DeviceState *dev);
50 static char *pcibus_get_fw_dev_path(DeviceState *dev);
51 static void pcibus_reset(BusState *qbus);
52 
53 static Property pci_props[] = {
54     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
55     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
56     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
57     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
58                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
59     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
60                     QEMU_PCI_CAP_SERR_BITNR, true),
61     DEFINE_PROP_END_OF_LIST()
62 };
63 
64 static const VMStateDescription vmstate_pcibus = {
65     .name = "PCIBUS",
66     .version_id = 1,
67     .minimum_version_id = 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     pci_device_deassert_intx(dev);
193     assert(dev->irq_state == 0);
194 
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     .fields = (VMStateField[]) {
477         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
478         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
479                                    vmstate_info_pci_config,
480                                    PCI_CONFIG_SPACE_SIZE),
481         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
482 				   vmstate_info_pci_irq_state,
483 				   PCI_NUM_PINS * sizeof(int32_t)),
484         VMSTATE_END_OF_LIST()
485     }
486 };
487 
488 const VMStateDescription vmstate_pcie_device = {
489     .name = "PCIEDevice",
490     .version_id = 2,
491     .minimum_version_id = 1,
492     .fields = (VMStateField[]) {
493         VMSTATE_INT32_POSITIVE_LE(version_id, PCIDevice),
494         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
495                                    vmstate_info_pci_config,
496                                    PCIE_CONFIG_SPACE_SIZE),
497         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
498 				   vmstate_info_pci_irq_state,
499 				   PCI_NUM_PINS * sizeof(int32_t)),
500         VMSTATE_END_OF_LIST()
501     }
502 };
503 
504 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
505 {
506     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
507 }
508 
509 void pci_device_save(PCIDevice *s, QEMUFile *f)
510 {
511     /* Clear interrupt status bit: it is implicit
512      * in irq_state which we are saving.
513      * This makes us compatible with old devices
514      * which never set or clear this bit. */
515     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
516     vmstate_save_state(f, pci_get_vmstate(s), s, NULL);
517     /* Restore the interrupt status bit. */
518     pci_update_irq_status(s);
519 }
520 
521 int pci_device_load(PCIDevice *s, QEMUFile *f)
522 {
523     int ret;
524     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
525     /* Restore the interrupt status bit. */
526     pci_update_irq_status(s);
527     return ret;
528 }
529 
530 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
531 {
532     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
533                  pci_default_sub_vendor_id);
534     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
535                  pci_default_sub_device_id);
536 }
537 
538 /*
539  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
540  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
541  */
542 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
543                       unsigned int *slotp, unsigned int *funcp)
544 {
545     const char *p;
546     char *e;
547     unsigned long val;
548     unsigned long dom = 0, bus = 0;
549     unsigned int slot = 0;
550     unsigned int func = 0;
551 
552     p = addr;
553     val = strtoul(p, &e, 16);
554     if (e == p)
555 	return -1;
556     if (*e == ':') {
557 	bus = val;
558 	p = e + 1;
559 	val = strtoul(p, &e, 16);
560 	if (e == p)
561 	    return -1;
562 	if (*e == ':') {
563 	    dom = bus;
564 	    bus = val;
565 	    p = e + 1;
566 	    val = strtoul(p, &e, 16);
567 	    if (e == p)
568 		return -1;
569 	}
570     }
571 
572     slot = val;
573 
574     if (funcp != NULL) {
575         if (*e != '.')
576             return -1;
577 
578         p = e + 1;
579         val = strtoul(p, &e, 16);
580         if (e == p)
581             return -1;
582 
583         func = val;
584     }
585 
586     /* if funcp == NULL func is 0 */
587     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
588 	return -1;
589 
590     if (*e)
591 	return -1;
592 
593     *domp = dom;
594     *busp = bus;
595     *slotp = slot;
596     if (funcp != NULL)
597         *funcp = func;
598     return 0;
599 }
600 
601 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
602 {
603     int dom, bus;
604     unsigned slot;
605 
606     if (!root) {
607         fprintf(stderr, "No primary PCI bus\n");
608         return NULL;
609     }
610 
611     assert(!root->parent_dev);
612 
613     if (!devaddr) {
614         *devfnp = -1;
615         return pci_find_bus_nr(root, 0);
616     }
617 
618     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
619         return NULL;
620     }
621 
622     if (dom != 0) {
623         fprintf(stderr, "No support for non-zero PCI domains\n");
624         return NULL;
625     }
626 
627     *devfnp = PCI_DEVFN(slot, 0);
628     return pci_find_bus_nr(root, bus);
629 }
630 
631 static void pci_init_cmask(PCIDevice *dev)
632 {
633     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
634     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
635     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
636     dev->cmask[PCI_REVISION_ID] = 0xff;
637     dev->cmask[PCI_CLASS_PROG] = 0xff;
638     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
639     dev->cmask[PCI_HEADER_TYPE] = 0xff;
640     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
641 }
642 
643 static void pci_init_wmask(PCIDevice *dev)
644 {
645     int config_size = pci_config_size(dev);
646 
647     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
648     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
649     pci_set_word(dev->wmask + PCI_COMMAND,
650                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
651                  PCI_COMMAND_INTX_DISABLE);
652     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
653         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
654     }
655 
656     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
657            config_size - PCI_CONFIG_HEADER_SIZE);
658 }
659 
660 static void pci_init_w1cmask(PCIDevice *dev)
661 {
662     /*
663      * Note: It's okay to set w1cmask even for readonly bits as
664      * long as their value is hardwired to 0.
665      */
666     pci_set_word(dev->w1cmask + PCI_STATUS,
667                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
668                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
669                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
670 }
671 
672 static void pci_init_mask_bridge(PCIDevice *d)
673 {
674     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
675        PCI_SEC_LETENCY_TIMER */
676     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
677 
678     /* base and limit */
679     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
680     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
681     pci_set_word(d->wmask + PCI_MEMORY_BASE,
682                  PCI_MEMORY_RANGE_MASK & 0xffff);
683     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
684                  PCI_MEMORY_RANGE_MASK & 0xffff);
685     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
686                  PCI_PREF_RANGE_MASK & 0xffff);
687     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
688                  PCI_PREF_RANGE_MASK & 0xffff);
689 
690     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
691     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
692 
693     /* Supported memory and i/o types */
694     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
695     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
696     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
697                                PCI_PREF_RANGE_TYPE_64);
698     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
699                                PCI_PREF_RANGE_TYPE_64);
700 
701     /*
702      * TODO: Bridges default to 10-bit VGA decoding but we currently only
703      * implement 16-bit decoding (no alias support).
704      */
705     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
706                  PCI_BRIDGE_CTL_PARITY |
707                  PCI_BRIDGE_CTL_SERR |
708                  PCI_BRIDGE_CTL_ISA |
709                  PCI_BRIDGE_CTL_VGA |
710                  PCI_BRIDGE_CTL_VGA_16BIT |
711                  PCI_BRIDGE_CTL_MASTER_ABORT |
712                  PCI_BRIDGE_CTL_BUS_RESET |
713                  PCI_BRIDGE_CTL_FAST_BACK |
714                  PCI_BRIDGE_CTL_DISCARD |
715                  PCI_BRIDGE_CTL_SEC_DISCARD |
716                  PCI_BRIDGE_CTL_DISCARD_SERR);
717     /* Below does not do anything as we never set this bit, put here for
718      * completeness. */
719     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
720                  PCI_BRIDGE_CTL_DISCARD_STATUS);
721     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
722     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
723     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
724                                PCI_PREF_RANGE_TYPE_MASK);
725     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
726                                PCI_PREF_RANGE_TYPE_MASK);
727 }
728 
729 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
730 {
731     uint8_t slot = PCI_SLOT(dev->devfn);
732     uint8_t func;
733 
734     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
735         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
736     }
737 
738     /*
739      * multifunction bit is interpreted in two ways as follows.
740      *   - all functions must set the bit to 1.
741      *     Example: Intel X53
742      *   - function 0 must set the bit, but the rest function (> 0)
743      *     is allowed to leave the bit to 0.
744      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
745      *
746      * So OS (at least Linux) checks the bit of only function 0,
747      * and doesn't see the bit of function > 0.
748      *
749      * The below check allows both interpretation.
750      */
751     if (PCI_FUNC(dev->devfn)) {
752         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
753         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
754             /* function 0 should set multifunction bit */
755             error_report("PCI: single function device can't be populated "
756                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
757             return -1;
758         }
759         return 0;
760     }
761 
762     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
763         return 0;
764     }
765     /* function 0 indicates single function, so function > 0 must be NULL */
766     for (func = 1; func < PCI_FUNC_MAX; ++func) {
767         if (bus->devices[PCI_DEVFN(slot, func)]) {
768             error_report("PCI: %x.0 indicates single function, "
769                          "but %x.%x is already populated.",
770                          slot, slot, func);
771             return -1;
772         }
773     }
774     return 0;
775 }
776 
777 static void pci_config_alloc(PCIDevice *pci_dev)
778 {
779     int config_size = pci_config_size(pci_dev);
780 
781     pci_dev->config = g_malloc0(config_size);
782     pci_dev->cmask = g_malloc0(config_size);
783     pci_dev->wmask = g_malloc0(config_size);
784     pci_dev->w1cmask = g_malloc0(config_size);
785     pci_dev->used = g_malloc0(config_size);
786 }
787 
788 static void pci_config_free(PCIDevice *pci_dev)
789 {
790     g_free(pci_dev->config);
791     g_free(pci_dev->cmask);
792     g_free(pci_dev->wmask);
793     g_free(pci_dev->w1cmask);
794     g_free(pci_dev->used);
795 }
796 
797 static void do_pci_unregister_device(PCIDevice *pci_dev)
798 {
799     pci_dev->bus->devices[pci_dev->devfn] = NULL;
800     pci_config_free(pci_dev);
801 
802     address_space_destroy(&pci_dev->bus_master_as);
803 }
804 
805 /* -1 for devfn means auto assign */
806 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
807                                          const char *name, int devfn)
808 {
809     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
810     PCIConfigReadFunc *config_read = pc->config_read;
811     PCIConfigWriteFunc *config_write = pc->config_write;
812     AddressSpace *dma_as;
813 
814     if (devfn < 0) {
815         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
816             devfn += PCI_FUNC_MAX) {
817             if (!bus->devices[devfn])
818                 goto found;
819         }
820         error_report("PCI: no slot/function available for %s, all in use", name);
821         return NULL;
822     found: ;
823     } else if (bus->devices[devfn]) {
824         error_report("PCI: slot %d function %d not available for %s, in use by %s",
825                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
826         return NULL;
827     }
828 
829     pci_dev->bus = bus;
830     pci_dev->devfn = devfn;
831     dma_as = pci_device_iommu_address_space(pci_dev);
832 
833     memory_region_init_alias(&pci_dev->bus_master_enable_region,
834                              OBJECT(pci_dev), "bus master",
835                              dma_as->root, 0, memory_region_size(dma_as->root));
836     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
837     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
838                        name);
839 
840     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
841     pci_dev->irq_state = 0;
842     pci_config_alloc(pci_dev);
843 
844     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
845     pci_config_set_device_id(pci_dev->config, pc->device_id);
846     pci_config_set_revision(pci_dev->config, pc->revision);
847     pci_config_set_class(pci_dev->config, pc->class_id);
848 
849     if (!pc->is_bridge) {
850         if (pc->subsystem_vendor_id || pc->subsystem_id) {
851             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
852                          pc->subsystem_vendor_id);
853             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
854                          pc->subsystem_id);
855         } else {
856             pci_set_default_subsystem_id(pci_dev);
857         }
858     } else {
859         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
860         assert(!pc->subsystem_vendor_id);
861         assert(!pc->subsystem_id);
862     }
863     pci_init_cmask(pci_dev);
864     pci_init_wmask(pci_dev);
865     pci_init_w1cmask(pci_dev);
866     if (pc->is_bridge) {
867         pci_init_mask_bridge(pci_dev);
868     }
869     if (pci_init_multifunction(bus, pci_dev)) {
870         do_pci_unregister_device(pci_dev);
871         return NULL;
872     }
873 
874     if (!config_read)
875         config_read = pci_default_read_config;
876     if (!config_write)
877         config_write = pci_default_write_config;
878     pci_dev->config_read = config_read;
879     pci_dev->config_write = config_write;
880     bus->devices[devfn] = pci_dev;
881     pci_dev->version_id = 2; /* Current pci device vmstate version */
882     return pci_dev;
883 }
884 
885 static void pci_unregister_io_regions(PCIDevice *pci_dev)
886 {
887     PCIIORegion *r;
888     int i;
889 
890     for(i = 0; i < PCI_NUM_REGIONS; i++) {
891         r = &pci_dev->io_regions[i];
892         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
893             continue;
894         memory_region_del_subregion(r->address_space, r->memory);
895     }
896 
897     pci_unregister_vga(pci_dev);
898 }
899 
900 static int pci_unregister_device(DeviceState *dev)
901 {
902     PCIDevice *pci_dev = PCI_DEVICE(dev);
903     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
904 
905     pci_unregister_io_regions(pci_dev);
906     pci_del_option_rom(pci_dev);
907 
908     if (pc->exit) {
909         pc->exit(pci_dev);
910     }
911 
912     do_pci_unregister_device(pci_dev);
913     return 0;
914 }
915 
916 void pci_register_bar(PCIDevice *pci_dev, int region_num,
917                       uint8_t type, MemoryRegion *memory)
918 {
919     PCIIORegion *r;
920     uint32_t addr;
921     uint64_t wmask;
922     pcibus_t size = memory_region_size(memory);
923 
924     assert(region_num >= 0);
925     assert(region_num < PCI_NUM_REGIONS);
926     if (size & (size-1)) {
927         fprintf(stderr, "ERROR: PCI region size must be pow2 "
928                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
929         exit(1);
930     }
931 
932     r = &pci_dev->io_regions[region_num];
933     r->addr = PCI_BAR_UNMAPPED;
934     r->size = size;
935     r->type = type;
936     r->memory = NULL;
937 
938     wmask = ~(size - 1);
939     addr = pci_bar(pci_dev, region_num);
940     if (region_num == PCI_ROM_SLOT) {
941         /* ROM enable bit is writable */
942         wmask |= PCI_ROM_ADDRESS_ENABLE;
943     }
944     pci_set_long(pci_dev->config + addr, type);
945     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
946         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
947         pci_set_quad(pci_dev->wmask + addr, wmask);
948         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
949     } else {
950         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
951         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
952     }
953     pci_dev->io_regions[region_num].memory = memory;
954     pci_dev->io_regions[region_num].address_space
955         = type & PCI_BASE_ADDRESS_SPACE_IO
956         ? pci_dev->bus->address_space_io
957         : pci_dev->bus->address_space_mem;
958 }
959 
960 static void pci_update_vga(PCIDevice *pci_dev)
961 {
962     uint16_t cmd;
963 
964     if (!pci_dev->has_vga) {
965         return;
966     }
967 
968     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
969 
970     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
971                               cmd & PCI_COMMAND_MEMORY);
972     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
973                               cmd & PCI_COMMAND_IO);
974     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
975                               cmd & PCI_COMMAND_IO);
976 }
977 
978 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
979                       MemoryRegion *io_lo, MemoryRegion *io_hi)
980 {
981     assert(!pci_dev->has_vga);
982 
983     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
984     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
985     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
986                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
987 
988     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
989     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
990     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
991                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
992 
993     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
994     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
995     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
996                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
997     pci_dev->has_vga = true;
998 
999     pci_update_vga(pci_dev);
1000 }
1001 
1002 void pci_unregister_vga(PCIDevice *pci_dev)
1003 {
1004     if (!pci_dev->has_vga) {
1005         return;
1006     }
1007 
1008     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1009                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1010     memory_region_del_subregion(pci_dev->bus->address_space_io,
1011                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1012     memory_region_del_subregion(pci_dev->bus->address_space_io,
1013                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1014     pci_dev->has_vga = false;
1015 }
1016 
1017 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1018 {
1019     return pci_dev->io_regions[region_num].addr;
1020 }
1021 
1022 static pcibus_t pci_bar_address(PCIDevice *d,
1023 				int reg, uint8_t type, pcibus_t size)
1024 {
1025     pcibus_t new_addr, last_addr;
1026     int bar = pci_bar(d, reg);
1027     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1028 
1029     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1030         if (!(cmd & PCI_COMMAND_IO)) {
1031             return PCI_BAR_UNMAPPED;
1032         }
1033         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1034         last_addr = new_addr + size - 1;
1035         /* Check if 32 bit BAR wraps around explicitly.
1036          * TODO: make priorities correct and remove this work around.
1037          */
1038         if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
1039             return PCI_BAR_UNMAPPED;
1040         }
1041         return new_addr;
1042     }
1043 
1044     if (!(cmd & PCI_COMMAND_MEMORY)) {
1045         return PCI_BAR_UNMAPPED;
1046     }
1047     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1048         new_addr = pci_get_quad(d->config + bar);
1049     } else {
1050         new_addr = pci_get_long(d->config + bar);
1051     }
1052     /* the ROM slot has a specific enable bit */
1053     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1054         return PCI_BAR_UNMAPPED;
1055     }
1056     new_addr &= ~(size - 1);
1057     last_addr = new_addr + size - 1;
1058     /* NOTE: we do not support wrapping */
1059     /* XXX: as we cannot support really dynamic
1060        mappings, we handle specific values as invalid
1061        mappings. */
1062     if (last_addr <= new_addr || new_addr == 0 ||
1063         last_addr == PCI_BAR_UNMAPPED) {
1064         return PCI_BAR_UNMAPPED;
1065     }
1066 
1067     /* Now pcibus_t is 64bit.
1068      * Check if 32 bit BAR wraps around explicitly.
1069      * Without this, PC ide doesn't work well.
1070      * TODO: remove this work around.
1071      */
1072     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1073         return PCI_BAR_UNMAPPED;
1074     }
1075 
1076     /*
1077      * OS is allowed to set BAR beyond its addressable
1078      * bits. For example, 32 bit OS can set 64bit bar
1079      * to >4G. Check it. TODO: we might need to support
1080      * it in the future for e.g. PAE.
1081      */
1082     if (last_addr >= HWADDR_MAX) {
1083         return PCI_BAR_UNMAPPED;
1084     }
1085 
1086     return new_addr;
1087 }
1088 
1089 static void pci_update_mappings(PCIDevice *d)
1090 {
1091     PCIIORegion *r;
1092     int i;
1093     pcibus_t new_addr;
1094 
1095     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1096         r = &d->io_regions[i];
1097 
1098         /* this region isn't registered */
1099         if (!r->size)
1100             continue;
1101 
1102         new_addr = pci_bar_address(d, i, r->type, r->size);
1103 
1104         /* This bar isn't changed */
1105         if (new_addr == r->addr)
1106             continue;
1107 
1108         /* now do the real mapping */
1109         if (r->addr != PCI_BAR_UNMAPPED) {
1110             trace_pci_update_mappings_del(d, pci_bus_num(d->bus),
1111                                           PCI_FUNC(d->devfn),
1112                                           PCI_SLOT(d->devfn),
1113                                           i, r->addr, r->size);
1114             memory_region_del_subregion(r->address_space, r->memory);
1115         }
1116         r->addr = new_addr;
1117         if (r->addr != PCI_BAR_UNMAPPED) {
1118             trace_pci_update_mappings_add(d, pci_bus_num(d->bus),
1119                                           PCI_FUNC(d->devfn),
1120                                           PCI_SLOT(d->devfn),
1121                                           i, r->addr, r->size);
1122             memory_region_add_subregion_overlap(r->address_space,
1123                                                 r->addr, r->memory, 1);
1124         }
1125     }
1126 
1127     pci_update_vga(d);
1128 }
1129 
1130 static inline int pci_irq_disabled(PCIDevice *d)
1131 {
1132     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1133 }
1134 
1135 /* Called after interrupt disabled field update in config space,
1136  * assert/deassert interrupts if necessary.
1137  * Gets original interrupt disable bit value (before update). */
1138 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1139 {
1140     int i, disabled = pci_irq_disabled(d);
1141     if (disabled == was_irq_disabled)
1142         return;
1143     for (i = 0; i < PCI_NUM_PINS; ++i) {
1144         int state = pci_irq_state(d, i);
1145         pci_change_irq_level(d, i, disabled ? -state : state);
1146     }
1147 }
1148 
1149 uint32_t pci_default_read_config(PCIDevice *d,
1150                                  uint32_t address, int len)
1151 {
1152     uint32_t val = 0;
1153 
1154     memcpy(&val, d->config + address, len);
1155     return le32_to_cpu(val);
1156 }
1157 
1158 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val_in, int l)
1159 {
1160     int i, was_irq_disabled = pci_irq_disabled(d);
1161     uint32_t val = val_in;
1162 
1163     for (i = 0; i < l; val >>= 8, ++i) {
1164         uint8_t wmask = d->wmask[addr + i];
1165         uint8_t w1cmask = d->w1cmask[addr + i];
1166         assert(!(wmask & w1cmask));
1167         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1168         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1169     }
1170     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1171         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1172         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1173         range_covers_byte(addr, l, PCI_COMMAND))
1174         pci_update_mappings(d);
1175 
1176     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1177         pci_update_irq_disabled(d, was_irq_disabled);
1178         memory_region_set_enabled(&d->bus_master_enable_region,
1179                                   pci_get_word(d->config + PCI_COMMAND)
1180                                     & PCI_COMMAND_MASTER);
1181     }
1182 
1183     msi_write_config(d, addr, val_in, l);
1184     msix_write_config(d, addr, val_in, l);
1185 }
1186 
1187 /***********************************************************/
1188 /* generic PCI irq support */
1189 
1190 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1191 static void pci_irq_handler(void *opaque, int irq_num, int level)
1192 {
1193     PCIDevice *pci_dev = opaque;
1194     int change;
1195 
1196     change = level - pci_irq_state(pci_dev, irq_num);
1197     if (!change)
1198         return;
1199 
1200     pci_set_irq_state(pci_dev, irq_num, level);
1201     pci_update_irq_status(pci_dev);
1202     if (pci_irq_disabled(pci_dev))
1203         return;
1204     pci_change_irq_level(pci_dev, irq_num, change);
1205 }
1206 
1207 static inline int pci_intx(PCIDevice *pci_dev)
1208 {
1209     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1210 }
1211 
1212 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1213 {
1214     int intx = pci_intx(pci_dev);
1215 
1216     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1217 }
1218 
1219 void pci_set_irq(PCIDevice *pci_dev, int level)
1220 {
1221     int intx = pci_intx(pci_dev);
1222     pci_irq_handler(pci_dev, intx, level);
1223 }
1224 
1225 /* Special hooks used by device assignment */
1226 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1227 {
1228     assert(pci_bus_is_root(bus));
1229     bus->route_intx_to_irq = route_intx_to_irq;
1230 }
1231 
1232 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1233 {
1234     PCIBus *bus;
1235 
1236     do {
1237          bus = dev->bus;
1238          pin = bus->map_irq(dev, pin);
1239          dev = bus->parent_dev;
1240     } while (dev);
1241 
1242     if (!bus->route_intx_to_irq) {
1243         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1244                      object_get_typename(OBJECT(bus->qbus.parent)));
1245         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1246     }
1247 
1248     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1249 }
1250 
1251 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1252 {
1253     return old->mode != new->mode || old->irq != new->irq;
1254 }
1255 
1256 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1257 {
1258     PCIDevice *dev;
1259     PCIBus *sec;
1260     int i;
1261 
1262     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1263         dev = bus->devices[i];
1264         if (dev && dev->intx_routing_notifier) {
1265             dev->intx_routing_notifier(dev);
1266         }
1267     }
1268 
1269     QLIST_FOREACH(sec, &bus->child, sibling) {
1270         pci_bus_fire_intx_routing_notifier(sec);
1271     }
1272 }
1273 
1274 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1275                                           PCIINTxRoutingNotifier notifier)
1276 {
1277     dev->intx_routing_notifier = notifier;
1278 }
1279 
1280 /*
1281  * PCI-to-PCI bridge specification
1282  * 9.1: Interrupt routing. Table 9-1
1283  *
1284  * the PCI Express Base Specification, Revision 2.1
1285  * 2.2.8.1: INTx interrutp signaling - Rules
1286  *          the Implementation Note
1287  *          Table 2-20
1288  */
1289 /*
1290  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1291  * 0-origin unlike PCI interrupt pin register.
1292  */
1293 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1294 {
1295     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1296 }
1297 
1298 /***********************************************************/
1299 /* monitor info on PCI */
1300 
1301 typedef struct {
1302     uint16_t class;
1303     const char *desc;
1304     const char *fw_name;
1305     uint16_t fw_ign_bits;
1306 } pci_class_desc;
1307 
1308 static const pci_class_desc pci_class_descriptions[] =
1309 {
1310     { 0x0001, "VGA controller", "display"},
1311     { 0x0100, "SCSI controller", "scsi"},
1312     { 0x0101, "IDE controller", "ide"},
1313     { 0x0102, "Floppy controller", "fdc"},
1314     { 0x0103, "IPI controller", "ipi"},
1315     { 0x0104, "RAID controller", "raid"},
1316     { 0x0106, "SATA controller"},
1317     { 0x0107, "SAS controller"},
1318     { 0x0180, "Storage controller"},
1319     { 0x0200, "Ethernet controller", "ethernet"},
1320     { 0x0201, "Token Ring controller", "token-ring"},
1321     { 0x0202, "FDDI controller", "fddi"},
1322     { 0x0203, "ATM controller", "atm"},
1323     { 0x0280, "Network controller"},
1324     { 0x0300, "VGA controller", "display", 0x00ff},
1325     { 0x0301, "XGA controller"},
1326     { 0x0302, "3D controller"},
1327     { 0x0380, "Display controller"},
1328     { 0x0400, "Video controller", "video"},
1329     { 0x0401, "Audio controller", "sound"},
1330     { 0x0402, "Phone"},
1331     { 0x0403, "Audio controller", "sound"},
1332     { 0x0480, "Multimedia controller"},
1333     { 0x0500, "RAM controller", "memory"},
1334     { 0x0501, "Flash controller", "flash"},
1335     { 0x0580, "Memory controller"},
1336     { 0x0600, "Host bridge", "host"},
1337     { 0x0601, "ISA bridge", "isa"},
1338     { 0x0602, "EISA bridge", "eisa"},
1339     { 0x0603, "MC bridge", "mca"},
1340     { 0x0604, "PCI bridge", "pci-bridge"},
1341     { 0x0605, "PCMCIA bridge", "pcmcia"},
1342     { 0x0606, "NUBUS bridge", "nubus"},
1343     { 0x0607, "CARDBUS bridge", "cardbus"},
1344     { 0x0608, "RACEWAY bridge"},
1345     { 0x0680, "Bridge"},
1346     { 0x0700, "Serial port", "serial"},
1347     { 0x0701, "Parallel port", "parallel"},
1348     { 0x0800, "Interrupt controller", "interrupt-controller"},
1349     { 0x0801, "DMA controller", "dma-controller"},
1350     { 0x0802, "Timer", "timer"},
1351     { 0x0803, "RTC", "rtc"},
1352     { 0x0900, "Keyboard", "keyboard"},
1353     { 0x0901, "Pen", "pen"},
1354     { 0x0902, "Mouse", "mouse"},
1355     { 0x0A00, "Dock station", "dock", 0x00ff},
1356     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1357     { 0x0c00, "Fireware contorller", "fireware"},
1358     { 0x0c01, "Access bus controller", "access-bus"},
1359     { 0x0c02, "SSA controller", "ssa"},
1360     { 0x0c03, "USB controller", "usb"},
1361     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1362     { 0x0c05, "SMBus"},
1363     { 0, NULL}
1364 };
1365 
1366 static void pci_for_each_device_under_bus(PCIBus *bus,
1367                                           void (*fn)(PCIBus *b, PCIDevice *d,
1368                                                      void *opaque),
1369                                           void *opaque)
1370 {
1371     PCIDevice *d;
1372     int devfn;
1373 
1374     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1375         d = bus->devices[devfn];
1376         if (d) {
1377             fn(bus, d, opaque);
1378         }
1379     }
1380 }
1381 
1382 void pci_for_each_device(PCIBus *bus, int bus_num,
1383                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1384                          void *opaque)
1385 {
1386     bus = pci_find_bus_nr(bus, bus_num);
1387 
1388     if (bus) {
1389         pci_for_each_device_under_bus(bus, fn, opaque);
1390     }
1391 }
1392 
1393 static const pci_class_desc *get_class_desc(int class)
1394 {
1395     const pci_class_desc *desc;
1396 
1397     desc = pci_class_descriptions;
1398     while (desc->desc && class != desc->class) {
1399         desc++;
1400     }
1401 
1402     return desc;
1403 }
1404 
1405 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1406 
1407 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1408 {
1409     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1410     int i;
1411 
1412     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1413         const PCIIORegion *r = &dev->io_regions[i];
1414         PciMemoryRegionList *region;
1415 
1416         if (!r->size) {
1417             continue;
1418         }
1419 
1420         region = g_malloc0(sizeof(*region));
1421         region->value = g_malloc0(sizeof(*region->value));
1422 
1423         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1424             region->value->type = g_strdup("io");
1425         } else {
1426             region->value->type = g_strdup("memory");
1427             region->value->has_prefetch = true;
1428             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1429             region->value->has_mem_type_64 = true;
1430             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1431         }
1432 
1433         region->value->bar = i;
1434         region->value->address = r->addr;
1435         region->value->size = r->size;
1436 
1437         /* XXX: waiting for the qapi to support GSList */
1438         if (!cur_item) {
1439             head = cur_item = region;
1440         } else {
1441             cur_item->next = region;
1442             cur_item = region;
1443         }
1444     }
1445 
1446     return head;
1447 }
1448 
1449 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1450                                            int bus_num)
1451 {
1452     PciBridgeInfo *info;
1453 
1454     info = g_malloc0(sizeof(*info));
1455 
1456     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1457     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1458     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1459 
1460     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1461     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1462     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1463 
1464     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1465     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1466     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1467 
1468     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1469     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1470     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1471 
1472     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1473         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1474         if (child_bus) {
1475             info->has_devices = true;
1476             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1477         }
1478     }
1479 
1480     return info;
1481 }
1482 
1483 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1484                                            int bus_num)
1485 {
1486     const pci_class_desc *desc;
1487     PciDeviceInfo *info;
1488     uint8_t type;
1489     int class;
1490 
1491     info = g_malloc0(sizeof(*info));
1492     info->bus = bus_num;
1493     info->slot = PCI_SLOT(dev->devfn);
1494     info->function = PCI_FUNC(dev->devfn);
1495 
1496     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1497     info->class_info.q_class = class;
1498     desc = get_class_desc(class);
1499     if (desc->desc) {
1500         info->class_info.has_desc = true;
1501         info->class_info.desc = g_strdup(desc->desc);
1502     }
1503 
1504     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1505     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1506     info->regions = qmp_query_pci_regions(dev);
1507     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1508 
1509     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1510         info->has_irq = true;
1511         info->irq = dev->config[PCI_INTERRUPT_LINE];
1512     }
1513 
1514     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1515     if (type == PCI_HEADER_TYPE_BRIDGE) {
1516         info->has_pci_bridge = true;
1517         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1518     }
1519 
1520     return info;
1521 }
1522 
1523 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1524 {
1525     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1526     PCIDevice *dev;
1527     int devfn;
1528 
1529     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1530         dev = bus->devices[devfn];
1531         if (dev) {
1532             info = g_malloc0(sizeof(*info));
1533             info->value = qmp_query_pci_device(dev, bus, bus_num);
1534 
1535             /* XXX: waiting for the qapi to support GSList */
1536             if (!cur_item) {
1537                 head = cur_item = info;
1538             } else {
1539                 cur_item->next = info;
1540                 cur_item = info;
1541             }
1542         }
1543     }
1544 
1545     return head;
1546 }
1547 
1548 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1549 {
1550     PciInfo *info = NULL;
1551 
1552     bus = pci_find_bus_nr(bus, bus_num);
1553     if (bus) {
1554         info = g_malloc0(sizeof(*info));
1555         info->bus = bus_num;
1556         info->devices = qmp_query_pci_devices(bus, bus_num);
1557     }
1558 
1559     return info;
1560 }
1561 
1562 PciInfoList *qmp_query_pci(Error **errp)
1563 {
1564     PciInfoList *info, *head = NULL, *cur_item = NULL;
1565     PCIHostState *host_bridge;
1566 
1567     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1568         info = g_malloc0(sizeof(*info));
1569         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1570 
1571         /* XXX: waiting for the qapi to support GSList */
1572         if (!cur_item) {
1573             head = cur_item = info;
1574         } else {
1575             cur_item->next = info;
1576             cur_item = info;
1577         }
1578     }
1579 
1580     return head;
1581 }
1582 
1583 static const char * const pci_nic_models[] = {
1584     "ne2k_pci",
1585     "i82551",
1586     "i82557b",
1587     "i82559er",
1588     "rtl8139",
1589     "e1000",
1590     "pcnet",
1591     "virtio",
1592     NULL
1593 };
1594 
1595 static const char * const pci_nic_names[] = {
1596     "ne2k_pci",
1597     "i82551",
1598     "i82557b",
1599     "i82559er",
1600     "rtl8139",
1601     "e1000",
1602     "pcnet",
1603     "virtio-net-pci",
1604     NULL
1605 };
1606 
1607 /* Initialize a PCI NIC.  */
1608 /* FIXME callers should check for failure, but don't */
1609 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1610                         const char *default_model,
1611                         const char *default_devaddr)
1612 {
1613     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1614     PCIBus *bus;
1615     int devfn;
1616     PCIDevice *pci_dev;
1617     DeviceState *dev;
1618     int i;
1619 
1620     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1621     if (i < 0)
1622         return NULL;
1623 
1624     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1625     if (!bus) {
1626         error_report("Invalid PCI device address %s for device %s",
1627                      devaddr, pci_nic_names[i]);
1628         return NULL;
1629     }
1630 
1631     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1632     dev = &pci_dev->qdev;
1633     qdev_set_nic_properties(dev, nd);
1634     if (qdev_init(dev) < 0)
1635         return NULL;
1636     return pci_dev;
1637 }
1638 
1639 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1640                                const char *default_model,
1641                                const char *default_devaddr)
1642 {
1643     PCIDevice *res;
1644 
1645     if (qemu_show_nic_models(nd->model, pci_nic_models))
1646         exit(0);
1647 
1648     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1649     if (!res)
1650         exit(1);
1651     return res;
1652 }
1653 
1654 PCIDevice *pci_vga_init(PCIBus *bus)
1655 {
1656     switch (vga_interface_type) {
1657     case VGA_CIRRUS:
1658         return pci_create_simple(bus, -1, "cirrus-vga");
1659     case VGA_QXL:
1660         return pci_create_simple(bus, -1, "qxl-vga");
1661     case VGA_STD:
1662         return pci_create_simple(bus, -1, "VGA");
1663     case VGA_VMWARE:
1664         return pci_create_simple(bus, -1, "vmware-svga");
1665     case VGA_NONE:
1666     default: /* Other non-PCI types. Checking for unsupported types is already
1667                 done in vl.c. */
1668         return NULL;
1669     }
1670 }
1671 
1672 /* Whether a given bus number is in range of the secondary
1673  * bus of the given bridge device. */
1674 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1675 {
1676     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1677              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1678         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1679         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1680 }
1681 
1682 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1683 {
1684     PCIBus *sec;
1685 
1686     if (!bus) {
1687         return NULL;
1688     }
1689 
1690     if (pci_bus_num(bus) == bus_num) {
1691         return bus;
1692     }
1693 
1694     /* Consider all bus numbers in range for the host pci bridge. */
1695     if (!pci_bus_is_root(bus) &&
1696         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1697         return NULL;
1698     }
1699 
1700     /* try child bus */
1701     for (; bus; bus = sec) {
1702         QLIST_FOREACH(sec, &bus->child, sibling) {
1703             assert(!pci_bus_is_root(sec));
1704             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1705                 return sec;
1706             }
1707             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1708                 break;
1709             }
1710         }
1711     }
1712 
1713     return NULL;
1714 }
1715 
1716 void pci_for_each_bus_depth_first(PCIBus *bus,
1717                                   void *(*begin)(PCIBus *bus, void *parent_state),
1718                                   void (*end)(PCIBus *bus, void *state),
1719                                   void *parent_state)
1720 {
1721     PCIBus *sec;
1722     void *state;
1723 
1724     if (!bus) {
1725         return;
1726     }
1727 
1728     if (begin) {
1729         state = begin(bus, parent_state);
1730     } else {
1731         state = parent_state;
1732     }
1733 
1734     QLIST_FOREACH(sec, &bus->child, sibling) {
1735         pci_for_each_bus_depth_first(sec, begin, end, state);
1736     }
1737 
1738     if (end) {
1739         end(bus, state);
1740     }
1741 }
1742 
1743 
1744 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1745 {
1746     bus = pci_find_bus_nr(bus, bus_num);
1747 
1748     if (!bus)
1749         return NULL;
1750 
1751     return bus->devices[devfn];
1752 }
1753 
1754 static int pci_qdev_init(DeviceState *qdev)
1755 {
1756     PCIDevice *pci_dev = (PCIDevice *)qdev;
1757     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1758     PCIBus *bus;
1759     int rc;
1760     bool is_default_rom;
1761 
1762     /* initialize cap_present for pci_is_express() and pci_config_size() */
1763     if (pc->is_express) {
1764         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1765     }
1766 
1767     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1768     pci_dev = do_pci_register_device(pci_dev, bus,
1769                                      object_get_typename(OBJECT(qdev)),
1770                                      pci_dev->devfn);
1771     if (pci_dev == NULL)
1772         return -1;
1773 
1774     if (pc->init) {
1775         rc = pc->init(pci_dev);
1776         if (rc != 0) {
1777             do_pci_unregister_device(pci_dev);
1778             return rc;
1779         }
1780     }
1781 
1782     /* rom loading */
1783     is_default_rom = false;
1784     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1785         pci_dev->romfile = g_strdup(pc->romfile);
1786         is_default_rom = true;
1787     }
1788 
1789     rc = pci_add_option_rom(pci_dev, is_default_rom);
1790     if (rc != 0) {
1791         pci_unregister_device(DEVICE(pci_dev));
1792         return rc;
1793     }
1794 
1795     return 0;
1796 }
1797 
1798 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1799                                     const char *name)
1800 {
1801     DeviceState *dev;
1802 
1803     dev = qdev_create(&bus->qbus, name);
1804     qdev_prop_set_int32(dev, "addr", devfn);
1805     qdev_prop_set_bit(dev, "multifunction", multifunction);
1806     return PCI_DEVICE(dev);
1807 }
1808 
1809 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1810                                            bool multifunction,
1811                                            const char *name)
1812 {
1813     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1814     qdev_init_nofail(&dev->qdev);
1815     return dev;
1816 }
1817 
1818 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1819 {
1820     return pci_create_multifunction(bus, devfn, false, name);
1821 }
1822 
1823 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1824 {
1825     return pci_create_simple_multifunction(bus, devfn, false, name);
1826 }
1827 
1828 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1829 {
1830     int offset = PCI_CONFIG_HEADER_SIZE;
1831     int i;
1832     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1833         if (pdev->used[i])
1834             offset = i + 1;
1835         else if (i - offset + 1 == size)
1836             return offset;
1837     }
1838     return 0;
1839 }
1840 
1841 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1842                                         uint8_t *prev_p)
1843 {
1844     uint8_t next, prev;
1845 
1846     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1847         return 0;
1848 
1849     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1850          prev = next + PCI_CAP_LIST_NEXT)
1851         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1852             break;
1853 
1854     if (prev_p)
1855         *prev_p = prev;
1856     return next;
1857 }
1858 
1859 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1860 {
1861     uint8_t next, prev, found = 0;
1862 
1863     if (!(pdev->used[offset])) {
1864         return 0;
1865     }
1866 
1867     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1868 
1869     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1870          prev = next + PCI_CAP_LIST_NEXT) {
1871         if (next <= offset && next > found) {
1872             found = next;
1873         }
1874     }
1875     return found;
1876 }
1877 
1878 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1879    This is needed for an option rom which is used for more than one device. */
1880 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1881 {
1882     uint16_t vendor_id;
1883     uint16_t device_id;
1884     uint16_t rom_vendor_id;
1885     uint16_t rom_device_id;
1886     uint16_t rom_magic;
1887     uint16_t pcir_offset;
1888     uint8_t checksum;
1889 
1890     /* Words in rom data are little endian (like in PCI configuration),
1891        so they can be read / written with pci_get_word / pci_set_word. */
1892 
1893     /* Only a valid rom will be patched. */
1894     rom_magic = pci_get_word(ptr);
1895     if (rom_magic != 0xaa55) {
1896         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1897         return;
1898     }
1899     pcir_offset = pci_get_word(ptr + 0x18);
1900     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1901         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1902         return;
1903     }
1904 
1905     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1906     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1907     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1908     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1909 
1910     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1911                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1912 
1913     checksum = ptr[6];
1914 
1915     if (vendor_id != rom_vendor_id) {
1916         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1917         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1918         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1919         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1920         ptr[6] = checksum;
1921         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1922     }
1923 
1924     if (device_id != rom_device_id) {
1925         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1926         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1927         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1928         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1929         ptr[6] = checksum;
1930         pci_set_word(ptr + pcir_offset + 6, device_id);
1931     }
1932 }
1933 
1934 /* Add an option rom for the device */
1935 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1936 {
1937     int size;
1938     char *path;
1939     void *ptr;
1940     char name[32];
1941     const VMStateDescription *vmsd;
1942 
1943     if (!pdev->romfile)
1944         return 0;
1945     if (strlen(pdev->romfile) == 0)
1946         return 0;
1947 
1948     if (!pdev->rom_bar) {
1949         /*
1950          * Load rom via fw_cfg instead of creating a rom bar,
1951          * for 0.11 compatibility.
1952          */
1953         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1954 
1955         /*
1956          * Hot-plugged devices can't use the option ROM
1957          * if the rom bar is disabled.
1958          */
1959         if (DEVICE(pdev)->hotplugged) {
1960             return -1;
1961         }
1962 
1963         if (class == 0x0300) {
1964             rom_add_vga(pdev->romfile);
1965         } else {
1966             rom_add_option(pdev->romfile, -1);
1967         }
1968         return 0;
1969     }
1970 
1971     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1972     if (path == NULL) {
1973         path = g_strdup(pdev->romfile);
1974     }
1975 
1976     size = get_image_size(path);
1977     if (size < 0) {
1978         error_report("%s: failed to find romfile \"%s\"",
1979                      __func__, pdev->romfile);
1980         g_free(path);
1981         return -1;
1982     } else if (size == 0) {
1983         error_report("%s: ignoring empty romfile \"%s\"",
1984                      __func__, pdev->romfile);
1985         g_free(path);
1986         return -1;
1987     }
1988     if (size & (size - 1)) {
1989         size = 1 << qemu_fls(size);
1990     }
1991 
1992     vmsd = qdev_get_vmsd(DEVICE(pdev));
1993 
1994     if (vmsd) {
1995         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1996     } else {
1997         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1998     }
1999     pdev->has_rom = true;
2000     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size, &error_abort);
2001     vmstate_register_ram(&pdev->rom, &pdev->qdev);
2002     ptr = memory_region_get_ram_ptr(&pdev->rom);
2003     load_image(path, ptr);
2004     g_free(path);
2005 
2006     if (is_default_rom) {
2007         /* Only the default rom images will be patched (if needed). */
2008         pci_patch_ids(pdev, ptr, size);
2009     }
2010 
2011     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2012 
2013     return 0;
2014 }
2015 
2016 static void pci_del_option_rom(PCIDevice *pdev)
2017 {
2018     if (!pdev->has_rom)
2019         return;
2020 
2021     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2022     pdev->has_rom = false;
2023 }
2024 
2025 /*
2026  * if !offset
2027  * Reserve space and add capability to the linked list in pci config space
2028  *
2029  * if offset = 0,
2030  * Find and reserve space and add capability to the linked list
2031  * in pci config space */
2032 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2033                        uint8_t offset, uint8_t size)
2034 {
2035     int ret;
2036     Error *local_err = NULL;
2037 
2038     ret = pci_add_capability2(pdev, cap_id, offset, size, &local_err);
2039     if (local_err) {
2040         assert(ret < 0);
2041         error_report_err(local_err);
2042     } else {
2043         /* success implies a positive offset in config space */
2044         assert(ret > 0);
2045     }
2046     return ret;
2047 }
2048 
2049 int pci_add_capability2(PCIDevice *pdev, uint8_t cap_id,
2050                        uint8_t offset, uint8_t size,
2051                        Error **errp)
2052 {
2053     uint8_t *config;
2054     int i, overlapping_cap;
2055 
2056     if (!offset) {
2057         offset = pci_find_space(pdev, size);
2058         if (!offset) {
2059             error_setg(errp, "out of PCI config space");
2060             return -ENOSPC;
2061         }
2062     } else {
2063         /* Verify that capabilities don't overlap.  Note: device assignment
2064          * depends on this check to verify that the device is not broken.
2065          * Should never trigger for emulated devices, but it's helpful
2066          * for debugging these. */
2067         for (i = offset; i < offset + size; i++) {
2068             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2069             if (overlapping_cap) {
2070                 error_setg(errp, "%s:%02x:%02x.%x "
2071                            "Attempt to add PCI capability %x at offset "
2072                            "%x overlaps existing capability %x at offset %x",
2073                            pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2074                            PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2075                            cap_id, offset, overlapping_cap, i);
2076                 return -EINVAL;
2077             }
2078         }
2079     }
2080 
2081     config = pdev->config + offset;
2082     config[PCI_CAP_LIST_ID] = cap_id;
2083     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2084     pdev->config[PCI_CAPABILITY_LIST] = offset;
2085     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2086     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2087     /* Make capability read-only by default */
2088     memset(pdev->wmask + offset, 0, size);
2089     /* Check capability by default */
2090     memset(pdev->cmask + offset, 0xFF, size);
2091     return offset;
2092 }
2093 
2094 /* Unlink capability from the pci config space. */
2095 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2096 {
2097     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2098     if (!offset)
2099         return;
2100     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2101     /* Make capability writable again */
2102     memset(pdev->wmask + offset, 0xff, size);
2103     memset(pdev->w1cmask + offset, 0, size);
2104     /* Clear cmask as device-specific registers can't be checked */
2105     memset(pdev->cmask + offset, 0, size);
2106     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2107 
2108     if (!pdev->config[PCI_CAPABILITY_LIST])
2109         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2110 }
2111 
2112 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2113 {
2114     return pci_find_capability_list(pdev, cap_id, NULL);
2115 }
2116 
2117 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2118 {
2119     PCIDevice *d = (PCIDevice *)dev;
2120     const pci_class_desc *desc;
2121     char ctxt[64];
2122     PCIIORegion *r;
2123     int i, class;
2124 
2125     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2126     desc = pci_class_descriptions;
2127     while (desc->desc && class != desc->class)
2128         desc++;
2129     if (desc->desc) {
2130         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2131     } else {
2132         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2133     }
2134 
2135     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2136                    "pci id %04x:%04x (sub %04x:%04x)\n",
2137                    indent, "", ctxt, pci_bus_num(d->bus),
2138                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2139                    pci_get_word(d->config + PCI_VENDOR_ID),
2140                    pci_get_word(d->config + PCI_DEVICE_ID),
2141                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2142                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2143     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2144         r = &d->io_regions[i];
2145         if (!r->size)
2146             continue;
2147         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2148                        " [0x%"FMT_PCIBUS"]\n",
2149                        indent, "",
2150                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2151                        r->addr, r->addr + r->size - 1);
2152     }
2153 }
2154 
2155 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2156 {
2157     PCIDevice *d = (PCIDevice *)dev;
2158     const char *name = NULL;
2159     const pci_class_desc *desc =  pci_class_descriptions;
2160     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2161 
2162     while (desc->desc &&
2163           (class & ~desc->fw_ign_bits) !=
2164           (desc->class & ~desc->fw_ign_bits)) {
2165         desc++;
2166     }
2167 
2168     if (desc->desc) {
2169         name = desc->fw_name;
2170     }
2171 
2172     if (name) {
2173         pstrcpy(buf, len, name);
2174     } else {
2175         snprintf(buf, len, "pci%04x,%04x",
2176                  pci_get_word(d->config + PCI_VENDOR_ID),
2177                  pci_get_word(d->config + PCI_DEVICE_ID));
2178     }
2179 
2180     return buf;
2181 }
2182 
2183 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2184 {
2185     PCIDevice *d = (PCIDevice *)dev;
2186     char path[50], name[33];
2187     int off;
2188 
2189     off = snprintf(path, sizeof(path), "%s@%x",
2190                    pci_dev_fw_name(dev, name, sizeof name),
2191                    PCI_SLOT(d->devfn));
2192     if (PCI_FUNC(d->devfn))
2193         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2194     return g_strdup(path);
2195 }
2196 
2197 static char *pcibus_get_dev_path(DeviceState *dev)
2198 {
2199     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2200     PCIDevice *t;
2201     int slot_depth;
2202     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2203      * 00 is added here to make this format compatible with
2204      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2205      * Slot.Function list specifies the slot and function numbers for all
2206      * devices on the path from root to the specific device. */
2207     const char *root_bus_path;
2208     int root_bus_len;
2209     char slot[] = ":SS.F";
2210     int slot_len = sizeof slot - 1 /* For '\0' */;
2211     int path_len;
2212     char *path, *p;
2213     int s;
2214 
2215     root_bus_path = pci_root_bus_path(d);
2216     root_bus_len = strlen(root_bus_path);
2217 
2218     /* Calculate # of slots on path between device and root. */;
2219     slot_depth = 0;
2220     for (t = d; t; t = t->bus->parent_dev) {
2221         ++slot_depth;
2222     }
2223 
2224     path_len = root_bus_len + slot_len * slot_depth;
2225 
2226     /* Allocate memory, fill in the terminating null byte. */
2227     path = g_malloc(path_len + 1 /* For '\0' */);
2228     path[path_len] = '\0';
2229 
2230     memcpy(path, root_bus_path, root_bus_len);
2231 
2232     /* Fill in slot numbers. We walk up from device to root, so need to print
2233      * them in the reverse order, last to first. */
2234     p = path + path_len;
2235     for (t = d; t; t = t->bus->parent_dev) {
2236         p -= slot_len;
2237         s = snprintf(slot, sizeof slot, ":%02x.%x",
2238                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2239         assert(s == slot_len);
2240         memcpy(p, slot, slot_len);
2241     }
2242 
2243     return path;
2244 }
2245 
2246 static int pci_qdev_find_recursive(PCIBus *bus,
2247                                    const char *id, PCIDevice **pdev)
2248 {
2249     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2250     if (!qdev) {
2251         return -ENODEV;
2252     }
2253 
2254     /* roughly check if given qdev is pci device */
2255     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2256         *pdev = PCI_DEVICE(qdev);
2257         return 0;
2258     }
2259     return -EINVAL;
2260 }
2261 
2262 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2263 {
2264     PCIHostState *host_bridge;
2265     int rc = -ENODEV;
2266 
2267     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2268         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2269         if (!tmp) {
2270             rc = 0;
2271             break;
2272         }
2273         if (tmp != -ENODEV) {
2274             rc = tmp;
2275         }
2276     }
2277 
2278     return rc;
2279 }
2280 
2281 MemoryRegion *pci_address_space(PCIDevice *dev)
2282 {
2283     return dev->bus->address_space_mem;
2284 }
2285 
2286 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2287 {
2288     return dev->bus->address_space_io;
2289 }
2290 
2291 static void pci_device_class_init(ObjectClass *klass, void *data)
2292 {
2293     DeviceClass *k = DEVICE_CLASS(klass);
2294     k->init = pci_qdev_init;
2295     k->exit = pci_unregister_device;
2296     k->bus_type = TYPE_PCI_BUS;
2297     k->props = pci_props;
2298 }
2299 
2300 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2301 {
2302     PCIBus *bus = PCI_BUS(dev->bus);
2303 
2304     if (bus->iommu_fn) {
2305         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2306     }
2307 
2308     if (bus->parent_dev) {
2309         /** We are ignoring the bus master DMA bit of the bridge
2310          *  as it would complicate things such as VFIO for no good reason */
2311         return pci_device_iommu_address_space(bus->parent_dev);
2312     }
2313 
2314     return &address_space_memory;
2315 }
2316 
2317 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2318 {
2319     bus->iommu_fn = fn;
2320     bus->iommu_opaque = opaque;
2321 }
2322 
2323 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2324 {
2325     Range *range = opaque;
2326     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2327     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2328     int i;
2329 
2330     if (!(cmd & PCI_COMMAND_MEMORY)) {
2331         return;
2332     }
2333 
2334     if (pc->is_bridge) {
2335         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2336         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2337 
2338         base = MAX(base, 0x1ULL << 32);
2339 
2340         if (limit >= base) {
2341             Range pref_range;
2342             pref_range.begin = base;
2343             pref_range.end = limit + 1;
2344             range_extend(range, &pref_range);
2345         }
2346     }
2347     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2348         PCIIORegion *r = &dev->io_regions[i];
2349         Range region_range;
2350 
2351         if (!r->size ||
2352             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2353             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2354             continue;
2355         }
2356         region_range.begin = pci_bar_address(dev, i, r->type, r->size);
2357         region_range.end = region_range.begin + r->size;
2358 
2359         if (region_range.begin == PCI_BAR_UNMAPPED) {
2360             continue;
2361         }
2362 
2363         region_range.begin = MAX(region_range.begin, 0x1ULL << 32);
2364 
2365         if (region_range.end - 1 >= region_range.begin) {
2366             range_extend(range, &region_range);
2367         }
2368     }
2369 }
2370 
2371 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2372 {
2373     range->begin = range->end = 0;
2374     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2375 }
2376 
2377 static const TypeInfo pci_device_type_info = {
2378     .name = TYPE_PCI_DEVICE,
2379     .parent = TYPE_DEVICE,
2380     .instance_size = sizeof(PCIDevice),
2381     .abstract = true,
2382     .class_size = sizeof(PCIDeviceClass),
2383     .class_init = pci_device_class_init,
2384 };
2385 
2386 static void pci_register_types(void)
2387 {
2388     type_register_static(&pci_bus_info);
2389     type_register_static(&pcie_bus_info);
2390     type_register_static(&pci_device_type_info);
2391 }
2392 
2393 type_init(pci_register_types)
2394