xref: /openbmc/qemu/hw/pci/pci.c (revision 14a650ec)
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 
39 //#define DEBUG_PCI
40 #ifdef DEBUG_PCI
41 # define PCI_DPRINTF(format, ...)       printf(format, ## __VA_ARGS__)
42 #else
43 # define PCI_DPRINTF(format, ...)       do { } while (0)
44 #endif
45 
46 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent);
47 static char *pcibus_get_dev_path(DeviceState *dev);
48 static char *pcibus_get_fw_dev_path(DeviceState *dev);
49 static int pcibus_reset(BusState *qbus);
50 
51 static Property pci_props[] = {
52     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
53     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
54     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
55     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
56                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
57     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
58                     QEMU_PCI_CAP_SERR_BITNR, true),
59     DEFINE_PROP_END_OF_LIST()
60 };
61 
62 static void pci_bus_class_init(ObjectClass *klass, void *data)
63 {
64     BusClass *k = BUS_CLASS(klass);
65 
66     k->print_dev = pcibus_dev_print;
67     k->get_dev_path = pcibus_get_dev_path;
68     k->get_fw_dev_path = pcibus_get_fw_dev_path;
69     k->reset = pcibus_reset;
70 }
71 
72 static const TypeInfo pci_bus_info = {
73     .name = TYPE_PCI_BUS,
74     .parent = TYPE_BUS,
75     .instance_size = sizeof(PCIBus),
76     .class_init = pci_bus_class_init,
77 };
78 
79 static const TypeInfo pcie_bus_info = {
80     .name = TYPE_PCIE_BUS,
81     .parent = TYPE_PCI_BUS,
82 };
83 
84 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
85 static void pci_update_mappings(PCIDevice *d);
86 static void pci_irq_handler(void *opaque, int irq_num, int level);
87 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom);
88 static void pci_del_option_rom(PCIDevice *pdev);
89 
90 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
91 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
92 
93 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
94 
95 static const VMStateDescription vmstate_pcibus = {
96     .name = "PCIBUS",
97     .version_id = 1,
98     .minimum_version_id = 1,
99     .minimum_version_id_old = 1,
100     .fields      = (VMStateField []) {
101         VMSTATE_INT32_EQUAL(nirq, PCIBus),
102         VMSTATE_VARRAY_INT32(irq_count, PCIBus, nirq, 0, vmstate_info_int32, int32_t),
103         VMSTATE_END_OF_LIST()
104     }
105 };
106 static int pci_bar(PCIDevice *d, int reg)
107 {
108     uint8_t type;
109 
110     if (reg != PCI_ROM_SLOT)
111         return PCI_BASE_ADDRESS_0 + reg * 4;
112 
113     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
114     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
115 }
116 
117 static inline int pci_irq_state(PCIDevice *d, int irq_num)
118 {
119 	return (d->irq_state >> irq_num) & 0x1;
120 }
121 
122 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
123 {
124 	d->irq_state &= ~(0x1 << irq_num);
125 	d->irq_state |= level << irq_num;
126 }
127 
128 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
129 {
130     PCIBus *bus;
131     for (;;) {
132         bus = pci_dev->bus;
133         irq_num = bus->map_irq(pci_dev, irq_num);
134         if (bus->set_irq)
135             break;
136         pci_dev = bus->parent_dev;
137     }
138     bus->irq_count[irq_num] += change;
139     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
140 }
141 
142 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
143 {
144     assert(irq_num >= 0);
145     assert(irq_num < bus->nirq);
146     return !!bus->irq_count[irq_num];
147 }
148 
149 /* Update interrupt status bit in config space on interrupt
150  * state change. */
151 static void pci_update_irq_status(PCIDevice *dev)
152 {
153     if (dev->irq_state) {
154         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
155     } else {
156         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
157     }
158 }
159 
160 void pci_device_deassert_intx(PCIDevice *dev)
161 {
162     int i;
163     for (i = 0; i < PCI_NUM_PINS; ++i) {
164         pci_irq_handler(dev, i, 0);
165     }
166 }
167 
168 /*
169  * This function is called on #RST and FLR.
170  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
171  */
172 void pci_device_reset(PCIDevice *dev)
173 {
174     int r;
175 
176     qdev_reset_all(&dev->qdev);
177 
178     dev->irq_state = 0;
179     pci_update_irq_status(dev);
180     pci_device_deassert_intx(dev);
181     /* Clear all writable bits */
182     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
183                                  pci_get_word(dev->wmask + PCI_COMMAND) |
184                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
185     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
186                                  pci_get_word(dev->wmask + PCI_STATUS) |
187                                  pci_get_word(dev->w1cmask + PCI_STATUS));
188     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
189     dev->config[PCI_INTERRUPT_LINE] = 0x0;
190     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
191         PCIIORegion *region = &dev->io_regions[r];
192         if (!region->size) {
193             continue;
194         }
195 
196         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
197             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
198             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
199         } else {
200             pci_set_long(dev->config + pci_bar(dev, r), region->type);
201         }
202     }
203     pci_update_mappings(dev);
204 
205     msi_reset(dev);
206     msix_reset(dev);
207 }
208 
209 /*
210  * Trigger pci bus reset under a given bus.
211  * To be called on RST# assert.
212  */
213 void pci_bus_reset(PCIBus *bus)
214 {
215     int i;
216 
217     for (i = 0; i < bus->nirq; i++) {
218         bus->irq_count[i] = 0;
219     }
220     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
221         if (bus->devices[i]) {
222             pci_device_reset(bus->devices[i]);
223         }
224     }
225 }
226 
227 static int pcibus_reset(BusState *qbus)
228 {
229     pci_bus_reset(DO_UPCAST(PCIBus, qbus, qbus));
230 
231     /* topology traverse is done by pci_bus_reset().
232        Tell qbus/qdev walker not to traverse the tree */
233     return 1;
234 }
235 
236 static void pci_host_bus_register(PCIBus *bus, DeviceState *parent)
237 {
238     PCIHostState *host_bridge = PCI_HOST_BRIDGE(parent);
239 
240     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
241 }
242 
243 PCIBus *pci_find_primary_bus(void)
244 {
245     PCIBus *primary_bus = NULL;
246     PCIHostState *host;
247 
248     QLIST_FOREACH(host, &pci_host_bridges, next) {
249         if (primary_bus) {
250             /* We have multiple root buses, refuse to select a primary */
251             return NULL;
252         }
253         primary_bus = host->bus;
254     }
255 
256     return primary_bus;
257 }
258 
259 PCIBus *pci_device_root_bus(const PCIDevice *d)
260 {
261     PCIBus *bus = d->bus;
262 
263     while ((d = bus->parent_dev) != NULL) {
264         bus = d->bus;
265     }
266 
267     return bus;
268 }
269 
270 const char *pci_root_bus_path(PCIDevice *dev)
271 {
272     PCIBus *rootbus = pci_device_root_bus(dev);
273     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
274     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
275 
276     assert(!rootbus->parent_dev);
277     assert(host_bridge->bus == rootbus);
278 
279     if (hc->root_bus_path) {
280         return (*hc->root_bus_path)(host_bridge, rootbus);
281     }
282 
283     return rootbus->qbus.name;
284 }
285 
286 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
287                          const char *name,
288                          MemoryRegion *address_space_mem,
289                          MemoryRegion *address_space_io,
290                          uint8_t devfn_min)
291 {
292     assert(PCI_FUNC(devfn_min) == 0);
293     bus->devfn_min = devfn_min;
294     bus->address_space_mem = address_space_mem;
295     bus->address_space_io = address_space_io;
296 
297     /* host bridge */
298     QLIST_INIT(&bus->child);
299 
300     pci_host_bus_register(bus, parent);
301 
302     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
303 }
304 
305 bool pci_bus_is_express(PCIBus *bus)
306 {
307     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
308 }
309 
310 bool pci_bus_is_root(PCIBus *bus)
311 {
312     return !bus->parent_dev;
313 }
314 
315 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
316                          const char *name,
317                          MemoryRegion *address_space_mem,
318                          MemoryRegion *address_space_io,
319                          uint8_t devfn_min, const char *typename)
320 {
321     qbus_create_inplace(bus, bus_size, typename, parent, name);
322     pci_bus_init(bus, parent, name, address_space_mem,
323                  address_space_io, devfn_min);
324 }
325 
326 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
327                     MemoryRegion *address_space_mem,
328                     MemoryRegion *address_space_io,
329                     uint8_t devfn_min, const char *typename)
330 {
331     PCIBus *bus;
332 
333     bus = PCI_BUS(qbus_create(typename, parent, name));
334     pci_bus_init(bus, parent, name, address_space_mem,
335                  address_space_io, devfn_min);
336     return bus;
337 }
338 
339 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
340                   void *irq_opaque, int nirq)
341 {
342     bus->set_irq = set_irq;
343     bus->map_irq = map_irq;
344     bus->irq_opaque = irq_opaque;
345     bus->nirq = nirq;
346     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
347 }
348 
349 void pci_bus_hotplug(PCIBus *bus, pci_hotplug_fn hotplug, DeviceState *qdev)
350 {
351     bus->qbus.allow_hotplug = 1;
352     bus->hotplug = hotplug;
353     bus->hotplug_qdev = qdev;
354 }
355 
356 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
357                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
358                          void *irq_opaque,
359                          MemoryRegion *address_space_mem,
360                          MemoryRegion *address_space_io,
361                          uint8_t devfn_min, int nirq, const char *typename)
362 {
363     PCIBus *bus;
364 
365     bus = pci_bus_new(parent, name, address_space_mem,
366                       address_space_io, devfn_min, typename);
367     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
368     return bus;
369 }
370 
371 int pci_bus_num(PCIBus *s)
372 {
373     if (pci_bus_is_root(s))
374         return 0;       /* pci host bridge */
375     return s->parent_dev->config[PCI_SECONDARY_BUS];
376 }
377 
378 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size)
379 {
380     PCIDevice *s = container_of(pv, PCIDevice, config);
381     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
382     uint8_t *config;
383     int i;
384 
385     assert(size == pci_config_size(s));
386     config = g_malloc(size);
387 
388     qemu_get_buffer(f, config, size);
389     for (i = 0; i < size; ++i) {
390         if ((config[i] ^ s->config[i]) &
391             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
392             g_free(config);
393             return -EINVAL;
394         }
395     }
396     memcpy(s->config, config, size);
397 
398     pci_update_mappings(s);
399     if (pc->is_bridge) {
400         PCIBridge *b = PCI_BRIDGE(s);
401         pci_bridge_update_mappings(b);
402     }
403 
404     memory_region_set_enabled(&s->bus_master_enable_region,
405                               pci_get_word(s->config + PCI_COMMAND)
406                               & PCI_COMMAND_MASTER);
407 
408     g_free(config);
409     return 0;
410 }
411 
412 /* just put buffer */
413 static void put_pci_config_device(QEMUFile *f, void *pv, size_t size)
414 {
415     const uint8_t **v = pv;
416     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
417     qemu_put_buffer(f, *v, size);
418 }
419 
420 static VMStateInfo vmstate_info_pci_config = {
421     .name = "pci config",
422     .get  = get_pci_config_device,
423     .put  = put_pci_config_device,
424 };
425 
426 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size)
427 {
428     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
429     uint32_t irq_state[PCI_NUM_PINS];
430     int i;
431     for (i = 0; i < PCI_NUM_PINS; ++i) {
432         irq_state[i] = qemu_get_be32(f);
433         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
434             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
435                     irq_state[i]);
436             return -EINVAL;
437         }
438     }
439 
440     for (i = 0; i < PCI_NUM_PINS; ++i) {
441         pci_set_irq_state(s, i, irq_state[i]);
442     }
443 
444     return 0;
445 }
446 
447 static void put_pci_irq_state(QEMUFile *f, void *pv, size_t size)
448 {
449     int i;
450     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
451 
452     for (i = 0; i < PCI_NUM_PINS; ++i) {
453         qemu_put_be32(f, pci_irq_state(s, i));
454     }
455 }
456 
457 static VMStateInfo vmstate_info_pci_irq_state = {
458     .name = "pci irq state",
459     .get  = get_pci_irq_state,
460     .put  = put_pci_irq_state,
461 };
462 
463 const VMStateDescription vmstate_pci_device = {
464     .name = "PCIDevice",
465     .version_id = 2,
466     .minimum_version_id = 1,
467     .minimum_version_id_old = 1,
468     .fields      = (VMStateField []) {
469         VMSTATE_INT32_LE(version_id, PCIDevice),
470         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
471                                    vmstate_info_pci_config,
472                                    PCI_CONFIG_SPACE_SIZE),
473         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
474 				   vmstate_info_pci_irq_state,
475 				   PCI_NUM_PINS * sizeof(int32_t)),
476         VMSTATE_END_OF_LIST()
477     }
478 };
479 
480 const VMStateDescription vmstate_pcie_device = {
481     .name = "PCIEDevice",
482     .version_id = 2,
483     .minimum_version_id = 1,
484     .minimum_version_id_old = 1,
485     .fields      = (VMStateField []) {
486         VMSTATE_INT32_LE(version_id, PCIDevice),
487         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
488                                    vmstate_info_pci_config,
489                                    PCIE_CONFIG_SPACE_SIZE),
490         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
491 				   vmstate_info_pci_irq_state,
492 				   PCI_NUM_PINS * sizeof(int32_t)),
493         VMSTATE_END_OF_LIST()
494     }
495 };
496 
497 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
498 {
499     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
500 }
501 
502 void pci_device_save(PCIDevice *s, QEMUFile *f)
503 {
504     /* Clear interrupt status bit: it is implicit
505      * in irq_state which we are saving.
506      * This makes us compatible with old devices
507      * which never set or clear this bit. */
508     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
509     vmstate_save_state(f, pci_get_vmstate(s), s);
510     /* Restore the interrupt status bit. */
511     pci_update_irq_status(s);
512 }
513 
514 int pci_device_load(PCIDevice *s, QEMUFile *f)
515 {
516     int ret;
517     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
518     /* Restore the interrupt status bit. */
519     pci_update_irq_status(s);
520     return ret;
521 }
522 
523 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
524 {
525     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
526                  pci_default_sub_vendor_id);
527     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
528                  pci_default_sub_device_id);
529 }
530 
531 /*
532  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
533  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
534  */
535 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
536                       unsigned int *slotp, unsigned int *funcp)
537 {
538     const char *p;
539     char *e;
540     unsigned long val;
541     unsigned long dom = 0, bus = 0;
542     unsigned int slot = 0;
543     unsigned int func = 0;
544 
545     p = addr;
546     val = strtoul(p, &e, 16);
547     if (e == p)
548 	return -1;
549     if (*e == ':') {
550 	bus = val;
551 	p = e + 1;
552 	val = strtoul(p, &e, 16);
553 	if (e == p)
554 	    return -1;
555 	if (*e == ':') {
556 	    dom = bus;
557 	    bus = val;
558 	    p = e + 1;
559 	    val = strtoul(p, &e, 16);
560 	    if (e == p)
561 		return -1;
562 	}
563     }
564 
565     slot = val;
566 
567     if (funcp != NULL) {
568         if (*e != '.')
569             return -1;
570 
571         p = e + 1;
572         val = strtoul(p, &e, 16);
573         if (e == p)
574             return -1;
575 
576         func = val;
577     }
578 
579     /* if funcp == NULL func is 0 */
580     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
581 	return -1;
582 
583     if (*e)
584 	return -1;
585 
586     *domp = dom;
587     *busp = bus;
588     *slotp = slot;
589     if (funcp != NULL)
590         *funcp = func;
591     return 0;
592 }
593 
594 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
595 {
596     int dom, bus;
597     unsigned slot;
598 
599     assert(!root->parent_dev);
600 
601     if (!root) {
602         fprintf(stderr, "No primary PCI bus\n");
603         return NULL;
604     }
605 
606     if (!devaddr) {
607         *devfnp = -1;
608         return pci_find_bus_nr(root, 0);
609     }
610 
611     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
612         return NULL;
613     }
614 
615     if (dom != 0) {
616         fprintf(stderr, "No support for non-zero PCI domains\n");
617         return NULL;
618     }
619 
620     *devfnp = PCI_DEVFN(slot, 0);
621     return pci_find_bus_nr(root, bus);
622 }
623 
624 static void pci_init_cmask(PCIDevice *dev)
625 {
626     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
627     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
628     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
629     dev->cmask[PCI_REVISION_ID] = 0xff;
630     dev->cmask[PCI_CLASS_PROG] = 0xff;
631     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
632     dev->cmask[PCI_HEADER_TYPE] = 0xff;
633     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
634 }
635 
636 static void pci_init_wmask(PCIDevice *dev)
637 {
638     int config_size = pci_config_size(dev);
639 
640     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
641     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
642     pci_set_word(dev->wmask + PCI_COMMAND,
643                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
644                  PCI_COMMAND_INTX_DISABLE);
645     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
646         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
647     }
648 
649     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
650            config_size - PCI_CONFIG_HEADER_SIZE);
651 }
652 
653 static void pci_init_w1cmask(PCIDevice *dev)
654 {
655     /*
656      * Note: It's okay to set w1cmask even for readonly bits as
657      * long as their value is hardwired to 0.
658      */
659     pci_set_word(dev->w1cmask + PCI_STATUS,
660                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
661                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
662                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
663 }
664 
665 static void pci_init_mask_bridge(PCIDevice *d)
666 {
667     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
668        PCI_SEC_LETENCY_TIMER */
669     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
670 
671     /* base and limit */
672     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
673     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
674     pci_set_word(d->wmask + PCI_MEMORY_BASE,
675                  PCI_MEMORY_RANGE_MASK & 0xffff);
676     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
677                  PCI_MEMORY_RANGE_MASK & 0xffff);
678     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
679                  PCI_PREF_RANGE_MASK & 0xffff);
680     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
681                  PCI_PREF_RANGE_MASK & 0xffff);
682 
683     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
684     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
685 
686     /* Supported memory and i/o types */
687     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
688     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
689     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
690                                PCI_PREF_RANGE_TYPE_64);
691     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
692                                PCI_PREF_RANGE_TYPE_64);
693 
694     /*
695      * TODO: Bridges default to 10-bit VGA decoding but we currently only
696      * implement 16-bit decoding (no alias support).
697      */
698     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
699                  PCI_BRIDGE_CTL_PARITY |
700                  PCI_BRIDGE_CTL_SERR |
701                  PCI_BRIDGE_CTL_ISA |
702                  PCI_BRIDGE_CTL_VGA |
703                  PCI_BRIDGE_CTL_VGA_16BIT |
704                  PCI_BRIDGE_CTL_MASTER_ABORT |
705                  PCI_BRIDGE_CTL_BUS_RESET |
706                  PCI_BRIDGE_CTL_FAST_BACK |
707                  PCI_BRIDGE_CTL_DISCARD |
708                  PCI_BRIDGE_CTL_SEC_DISCARD |
709                  PCI_BRIDGE_CTL_DISCARD_SERR);
710     /* Below does not do anything as we never set this bit, put here for
711      * completeness. */
712     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
713                  PCI_BRIDGE_CTL_DISCARD_STATUS);
714     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
715     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
716     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
717                                PCI_PREF_RANGE_TYPE_MASK);
718     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
719                                PCI_PREF_RANGE_TYPE_MASK);
720 }
721 
722 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
723 {
724     uint8_t slot = PCI_SLOT(dev->devfn);
725     uint8_t func;
726 
727     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
728         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
729     }
730 
731     /*
732      * multifunction bit is interpreted in two ways as follows.
733      *   - all functions must set the bit to 1.
734      *     Example: Intel X53
735      *   - function 0 must set the bit, but the rest function (> 0)
736      *     is allowed to leave the bit to 0.
737      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
738      *
739      * So OS (at least Linux) checks the bit of only function 0,
740      * and doesn't see the bit of function > 0.
741      *
742      * The below check allows both interpretation.
743      */
744     if (PCI_FUNC(dev->devfn)) {
745         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
746         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
747             /* function 0 should set multifunction bit */
748             error_report("PCI: single function device can't be populated "
749                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
750             return -1;
751         }
752         return 0;
753     }
754 
755     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
756         return 0;
757     }
758     /* function 0 indicates single function, so function > 0 must be NULL */
759     for (func = 1; func < PCI_FUNC_MAX; ++func) {
760         if (bus->devices[PCI_DEVFN(slot, func)]) {
761             error_report("PCI: %x.0 indicates single function, "
762                          "but %x.%x is already populated.",
763                          slot, slot, func);
764             return -1;
765         }
766     }
767     return 0;
768 }
769 
770 static void pci_config_alloc(PCIDevice *pci_dev)
771 {
772     int config_size = pci_config_size(pci_dev);
773 
774     pci_dev->config = g_malloc0(config_size);
775     pci_dev->cmask = g_malloc0(config_size);
776     pci_dev->wmask = g_malloc0(config_size);
777     pci_dev->w1cmask = g_malloc0(config_size);
778     pci_dev->used = g_malloc0(config_size);
779 }
780 
781 static void pci_config_free(PCIDevice *pci_dev)
782 {
783     g_free(pci_dev->config);
784     g_free(pci_dev->cmask);
785     g_free(pci_dev->wmask);
786     g_free(pci_dev->w1cmask);
787     g_free(pci_dev->used);
788 }
789 
790 /* -1 for devfn means auto assign */
791 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
792                                          const char *name, int devfn)
793 {
794     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
795     PCIConfigReadFunc *config_read = pc->config_read;
796     PCIConfigWriteFunc *config_write = pc->config_write;
797     AddressSpace *dma_as;
798 
799     if (devfn < 0) {
800         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
801             devfn += PCI_FUNC_MAX) {
802             if (!bus->devices[devfn])
803                 goto found;
804         }
805         error_report("PCI: no slot/function available for %s, all in use", name);
806         return NULL;
807     found: ;
808     } else if (bus->devices[devfn]) {
809         error_report("PCI: slot %d function %d not available for %s, in use by %s",
810                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
811         return NULL;
812     }
813 
814     pci_dev->bus = bus;
815     dma_as = pci_device_iommu_address_space(pci_dev);
816 
817     memory_region_init_alias(&pci_dev->bus_master_enable_region,
818                              OBJECT(pci_dev), "bus master",
819                              dma_as->root, 0, memory_region_size(dma_as->root));
820     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
821     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
822                        name);
823 
824     pci_dev->devfn = devfn;
825     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
826     pci_dev->irq_state = 0;
827     pci_config_alloc(pci_dev);
828 
829     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
830     pci_config_set_device_id(pci_dev->config, pc->device_id);
831     pci_config_set_revision(pci_dev->config, pc->revision);
832     pci_config_set_class(pci_dev->config, pc->class_id);
833 
834     if (!pc->is_bridge) {
835         if (pc->subsystem_vendor_id || pc->subsystem_id) {
836             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
837                          pc->subsystem_vendor_id);
838             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
839                          pc->subsystem_id);
840         } else {
841             pci_set_default_subsystem_id(pci_dev);
842         }
843     } else {
844         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
845         assert(!pc->subsystem_vendor_id);
846         assert(!pc->subsystem_id);
847     }
848     pci_init_cmask(pci_dev);
849     pci_init_wmask(pci_dev);
850     pci_init_w1cmask(pci_dev);
851     if (pc->is_bridge) {
852         pci_init_mask_bridge(pci_dev);
853     }
854     if (pci_init_multifunction(bus, pci_dev)) {
855         pci_config_free(pci_dev);
856         return NULL;
857     }
858 
859     if (!config_read)
860         config_read = pci_default_read_config;
861     if (!config_write)
862         config_write = pci_default_write_config;
863     pci_dev->config_read = config_read;
864     pci_dev->config_write = config_write;
865     bus->devices[devfn] = pci_dev;
866     pci_dev->version_id = 2; /* Current pci device vmstate version */
867     return pci_dev;
868 }
869 
870 static void do_pci_unregister_device(PCIDevice *pci_dev)
871 {
872     pci_dev->bus->devices[pci_dev->devfn] = NULL;
873     pci_config_free(pci_dev);
874 
875     address_space_destroy(&pci_dev->bus_master_as);
876     memory_region_destroy(&pci_dev->bus_master_enable_region);
877 }
878 
879 static void pci_unregister_io_regions(PCIDevice *pci_dev)
880 {
881     PCIIORegion *r;
882     int i;
883 
884     for(i = 0; i < PCI_NUM_REGIONS; i++) {
885         r = &pci_dev->io_regions[i];
886         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
887             continue;
888         memory_region_del_subregion(r->address_space, r->memory);
889     }
890 
891     pci_unregister_vga(pci_dev);
892 }
893 
894 static int pci_unregister_device(DeviceState *dev)
895 {
896     PCIDevice *pci_dev = PCI_DEVICE(dev);
897     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
898 
899     pci_unregister_io_regions(pci_dev);
900     pci_del_option_rom(pci_dev);
901 
902     if (pc->exit) {
903         pc->exit(pci_dev);
904     }
905 
906     do_pci_unregister_device(pci_dev);
907     return 0;
908 }
909 
910 void pci_register_bar(PCIDevice *pci_dev, int region_num,
911                       uint8_t type, MemoryRegion *memory)
912 {
913     PCIIORegion *r;
914     uint32_t addr;
915     uint64_t wmask;
916     pcibus_t size = memory_region_size(memory);
917 
918     assert(region_num >= 0);
919     assert(region_num < PCI_NUM_REGIONS);
920     if (size & (size-1)) {
921         fprintf(stderr, "ERROR: PCI region size must be pow2 "
922                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
923         exit(1);
924     }
925 
926     r = &pci_dev->io_regions[region_num];
927     r->addr = PCI_BAR_UNMAPPED;
928     r->size = size;
929     r->type = type;
930     r->memory = NULL;
931 
932     wmask = ~(size - 1);
933     addr = pci_bar(pci_dev, region_num);
934     if (region_num == PCI_ROM_SLOT) {
935         /* ROM enable bit is writable */
936         wmask |= PCI_ROM_ADDRESS_ENABLE;
937     }
938     pci_set_long(pci_dev->config + addr, type);
939     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
940         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
941         pci_set_quad(pci_dev->wmask + addr, wmask);
942         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
943     } else {
944         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
945         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
946     }
947     pci_dev->io_regions[region_num].memory = memory;
948     pci_dev->io_regions[region_num].address_space
949         = type & PCI_BASE_ADDRESS_SPACE_IO
950         ? pci_dev->bus->address_space_io
951         : pci_dev->bus->address_space_mem;
952 }
953 
954 static void pci_update_vga(PCIDevice *pci_dev)
955 {
956     uint16_t cmd;
957 
958     if (!pci_dev->has_vga) {
959         return;
960     }
961 
962     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
963 
964     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
965                               cmd & PCI_COMMAND_MEMORY);
966     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
967                               cmd & PCI_COMMAND_IO);
968     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
969                               cmd & PCI_COMMAND_IO);
970 }
971 
972 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
973                       MemoryRegion *io_lo, MemoryRegion *io_hi)
974 {
975     assert(!pci_dev->has_vga);
976 
977     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
978     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
979     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
980                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
981 
982     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
983     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
984     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
985                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
986 
987     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
988     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
989     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
990                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
991     pci_dev->has_vga = true;
992 
993     pci_update_vga(pci_dev);
994 }
995 
996 void pci_unregister_vga(PCIDevice *pci_dev)
997 {
998     if (!pci_dev->has_vga) {
999         return;
1000     }
1001 
1002     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1003                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1004     memory_region_del_subregion(pci_dev->bus->address_space_io,
1005                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1006     memory_region_del_subregion(pci_dev->bus->address_space_io,
1007                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1008     pci_dev->has_vga = false;
1009 }
1010 
1011 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1012 {
1013     return pci_dev->io_regions[region_num].addr;
1014 }
1015 
1016 static pcibus_t pci_bar_address(PCIDevice *d,
1017 				int reg, uint8_t type, pcibus_t size)
1018 {
1019     pcibus_t new_addr, last_addr;
1020     int bar = pci_bar(d, reg);
1021     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1022 
1023     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1024         if (!(cmd & PCI_COMMAND_IO)) {
1025             return PCI_BAR_UNMAPPED;
1026         }
1027         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1028         last_addr = new_addr + size - 1;
1029         /* Check if 32 bit BAR wraps around explicitly.
1030          * TODO: make priorities correct and remove this work around.
1031          */
1032         if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
1033             return PCI_BAR_UNMAPPED;
1034         }
1035         return new_addr;
1036     }
1037 
1038     if (!(cmd & PCI_COMMAND_MEMORY)) {
1039         return PCI_BAR_UNMAPPED;
1040     }
1041     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1042         new_addr = pci_get_quad(d->config + bar);
1043     } else {
1044         new_addr = pci_get_long(d->config + bar);
1045     }
1046     /* the ROM slot has a specific enable bit */
1047     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1048         return PCI_BAR_UNMAPPED;
1049     }
1050     new_addr &= ~(size - 1);
1051     last_addr = new_addr + size - 1;
1052     /* NOTE: we do not support wrapping */
1053     /* XXX: as we cannot support really dynamic
1054        mappings, we handle specific values as invalid
1055        mappings. */
1056     if (last_addr <= new_addr || new_addr == 0 ||
1057         last_addr == PCI_BAR_UNMAPPED) {
1058         return PCI_BAR_UNMAPPED;
1059     }
1060 
1061     /* Now pcibus_t is 64bit.
1062      * Check if 32 bit BAR wraps around explicitly.
1063      * Without this, PC ide doesn't work well.
1064      * TODO: remove this work around.
1065      */
1066     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1067         return PCI_BAR_UNMAPPED;
1068     }
1069 
1070     /*
1071      * OS is allowed to set BAR beyond its addressable
1072      * bits. For example, 32 bit OS can set 64bit bar
1073      * to >4G. Check it. TODO: we might need to support
1074      * it in the future for e.g. PAE.
1075      */
1076     if (last_addr >= HWADDR_MAX) {
1077         return PCI_BAR_UNMAPPED;
1078     }
1079 
1080     return new_addr;
1081 }
1082 
1083 static void pci_update_mappings(PCIDevice *d)
1084 {
1085     PCIIORegion *r;
1086     int i;
1087     pcibus_t new_addr;
1088 
1089     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1090         r = &d->io_regions[i];
1091 
1092         /* this region isn't registered */
1093         if (!r->size)
1094             continue;
1095 
1096         new_addr = pci_bar_address(d, i, r->type, r->size);
1097 
1098         /* This bar isn't changed */
1099         if (new_addr == r->addr)
1100             continue;
1101 
1102         /* now do the real mapping */
1103         if (r->addr != PCI_BAR_UNMAPPED) {
1104             memory_region_del_subregion(r->address_space, r->memory);
1105         }
1106         r->addr = new_addr;
1107         if (r->addr != PCI_BAR_UNMAPPED) {
1108             memory_region_add_subregion_overlap(r->address_space,
1109                                                 r->addr, r->memory, 1);
1110         }
1111     }
1112 
1113     pci_update_vga(d);
1114 }
1115 
1116 static inline int pci_irq_disabled(PCIDevice *d)
1117 {
1118     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1119 }
1120 
1121 /* Called after interrupt disabled field update in config space,
1122  * assert/deassert interrupts if necessary.
1123  * Gets original interrupt disable bit value (before update). */
1124 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1125 {
1126     int i, disabled = pci_irq_disabled(d);
1127     if (disabled == was_irq_disabled)
1128         return;
1129     for (i = 0; i < PCI_NUM_PINS; ++i) {
1130         int state = pci_irq_state(d, i);
1131         pci_change_irq_level(d, i, disabled ? -state : state);
1132     }
1133 }
1134 
1135 uint32_t pci_default_read_config(PCIDevice *d,
1136                                  uint32_t address, int len)
1137 {
1138     uint32_t val = 0;
1139 
1140     memcpy(&val, d->config + address, len);
1141     return le32_to_cpu(val);
1142 }
1143 
1144 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
1145 {
1146     int i, was_irq_disabled = pci_irq_disabled(d);
1147 
1148     for (i = 0; i < l; val >>= 8, ++i) {
1149         uint8_t wmask = d->wmask[addr + i];
1150         uint8_t w1cmask = d->w1cmask[addr + i];
1151         assert(!(wmask & w1cmask));
1152         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1153         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1154     }
1155     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1156         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1157         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1158         range_covers_byte(addr, l, PCI_COMMAND))
1159         pci_update_mappings(d);
1160 
1161     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1162         pci_update_irq_disabled(d, was_irq_disabled);
1163         memory_region_set_enabled(&d->bus_master_enable_region,
1164                                   pci_get_word(d->config + PCI_COMMAND)
1165                                     & PCI_COMMAND_MASTER);
1166     }
1167 
1168     msi_write_config(d, addr, val, l);
1169     msix_write_config(d, addr, val, l);
1170 }
1171 
1172 /***********************************************************/
1173 /* generic PCI irq support */
1174 
1175 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1176 static void pci_irq_handler(void *opaque, int irq_num, int level)
1177 {
1178     PCIDevice *pci_dev = opaque;
1179     int change;
1180 
1181     change = level - pci_irq_state(pci_dev, irq_num);
1182     if (!change)
1183         return;
1184 
1185     pci_set_irq_state(pci_dev, irq_num, level);
1186     pci_update_irq_status(pci_dev);
1187     if (pci_irq_disabled(pci_dev))
1188         return;
1189     pci_change_irq_level(pci_dev, irq_num, change);
1190 }
1191 
1192 static inline int pci_intx(PCIDevice *pci_dev)
1193 {
1194     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1195 }
1196 
1197 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1198 {
1199     int intx = pci_intx(pci_dev);
1200 
1201     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1202 }
1203 
1204 void pci_set_irq(PCIDevice *pci_dev, int level)
1205 {
1206     int intx = pci_intx(pci_dev);
1207     pci_irq_handler(pci_dev, intx, level);
1208 }
1209 
1210 /* Special hooks used by device assignment */
1211 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1212 {
1213     assert(pci_bus_is_root(bus));
1214     bus->route_intx_to_irq = route_intx_to_irq;
1215 }
1216 
1217 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1218 {
1219     PCIBus *bus;
1220 
1221     do {
1222          bus = dev->bus;
1223          pin = bus->map_irq(dev, pin);
1224          dev = bus->parent_dev;
1225     } while (dev);
1226 
1227     if (!bus->route_intx_to_irq) {
1228         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1229                      object_get_typename(OBJECT(bus->qbus.parent)));
1230         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1231     }
1232 
1233     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1234 }
1235 
1236 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1237 {
1238     return old->mode != new->mode || old->irq != new->irq;
1239 }
1240 
1241 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1242 {
1243     PCIDevice *dev;
1244     PCIBus *sec;
1245     int i;
1246 
1247     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1248         dev = bus->devices[i];
1249         if (dev && dev->intx_routing_notifier) {
1250             dev->intx_routing_notifier(dev);
1251         }
1252     }
1253 
1254     QLIST_FOREACH(sec, &bus->child, sibling) {
1255         pci_bus_fire_intx_routing_notifier(sec);
1256     }
1257 }
1258 
1259 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1260                                           PCIINTxRoutingNotifier notifier)
1261 {
1262     dev->intx_routing_notifier = notifier;
1263 }
1264 
1265 /*
1266  * PCI-to-PCI bridge specification
1267  * 9.1: Interrupt routing. Table 9-1
1268  *
1269  * the PCI Express Base Specification, Revision 2.1
1270  * 2.2.8.1: INTx interrutp signaling - Rules
1271  *          the Implementation Note
1272  *          Table 2-20
1273  */
1274 /*
1275  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1276  * 0-origin unlike PCI interrupt pin register.
1277  */
1278 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1279 {
1280     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1281 }
1282 
1283 /***********************************************************/
1284 /* monitor info on PCI */
1285 
1286 typedef struct {
1287     uint16_t class;
1288     const char *desc;
1289     const char *fw_name;
1290     uint16_t fw_ign_bits;
1291 } pci_class_desc;
1292 
1293 static const pci_class_desc pci_class_descriptions[] =
1294 {
1295     { 0x0001, "VGA controller", "display"},
1296     { 0x0100, "SCSI controller", "scsi"},
1297     { 0x0101, "IDE controller", "ide"},
1298     { 0x0102, "Floppy controller", "fdc"},
1299     { 0x0103, "IPI controller", "ipi"},
1300     { 0x0104, "RAID controller", "raid"},
1301     { 0x0106, "SATA controller"},
1302     { 0x0107, "SAS controller"},
1303     { 0x0180, "Storage controller"},
1304     { 0x0200, "Ethernet controller", "ethernet"},
1305     { 0x0201, "Token Ring controller", "token-ring"},
1306     { 0x0202, "FDDI controller", "fddi"},
1307     { 0x0203, "ATM controller", "atm"},
1308     { 0x0280, "Network controller"},
1309     { 0x0300, "VGA controller", "display", 0x00ff},
1310     { 0x0301, "XGA controller"},
1311     { 0x0302, "3D controller"},
1312     { 0x0380, "Display controller"},
1313     { 0x0400, "Video controller", "video"},
1314     { 0x0401, "Audio controller", "sound"},
1315     { 0x0402, "Phone"},
1316     { 0x0403, "Audio controller", "sound"},
1317     { 0x0480, "Multimedia controller"},
1318     { 0x0500, "RAM controller", "memory"},
1319     { 0x0501, "Flash controller", "flash"},
1320     { 0x0580, "Memory controller"},
1321     { 0x0600, "Host bridge", "host"},
1322     { 0x0601, "ISA bridge", "isa"},
1323     { 0x0602, "EISA bridge", "eisa"},
1324     { 0x0603, "MC bridge", "mca"},
1325     { 0x0604, "PCI bridge", "pci"},
1326     { 0x0605, "PCMCIA bridge", "pcmcia"},
1327     { 0x0606, "NUBUS bridge", "nubus"},
1328     { 0x0607, "CARDBUS bridge", "cardbus"},
1329     { 0x0608, "RACEWAY bridge"},
1330     { 0x0680, "Bridge"},
1331     { 0x0700, "Serial port", "serial"},
1332     { 0x0701, "Parallel port", "parallel"},
1333     { 0x0800, "Interrupt controller", "interrupt-controller"},
1334     { 0x0801, "DMA controller", "dma-controller"},
1335     { 0x0802, "Timer", "timer"},
1336     { 0x0803, "RTC", "rtc"},
1337     { 0x0900, "Keyboard", "keyboard"},
1338     { 0x0901, "Pen", "pen"},
1339     { 0x0902, "Mouse", "mouse"},
1340     { 0x0A00, "Dock station", "dock", 0x00ff},
1341     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1342     { 0x0c00, "Fireware contorller", "fireware"},
1343     { 0x0c01, "Access bus controller", "access-bus"},
1344     { 0x0c02, "SSA controller", "ssa"},
1345     { 0x0c03, "USB controller", "usb"},
1346     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1347     { 0x0c05, "SMBus"},
1348     { 0, NULL}
1349 };
1350 
1351 static void pci_for_each_device_under_bus(PCIBus *bus,
1352                                           void (*fn)(PCIBus *b, PCIDevice *d,
1353                                                      void *opaque),
1354                                           void *opaque)
1355 {
1356     PCIDevice *d;
1357     int devfn;
1358 
1359     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1360         d = bus->devices[devfn];
1361         if (d) {
1362             fn(bus, d, opaque);
1363         }
1364     }
1365 }
1366 
1367 void pci_for_each_device(PCIBus *bus, int bus_num,
1368                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1369                          void *opaque)
1370 {
1371     bus = pci_find_bus_nr(bus, bus_num);
1372 
1373     if (bus) {
1374         pci_for_each_device_under_bus(bus, fn, opaque);
1375     }
1376 }
1377 
1378 static const pci_class_desc *get_class_desc(int class)
1379 {
1380     const pci_class_desc *desc;
1381 
1382     desc = pci_class_descriptions;
1383     while (desc->desc && class != desc->class) {
1384         desc++;
1385     }
1386 
1387     return desc;
1388 }
1389 
1390 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1391 
1392 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1393 {
1394     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1395     int i;
1396 
1397     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1398         const PCIIORegion *r = &dev->io_regions[i];
1399         PciMemoryRegionList *region;
1400 
1401         if (!r->size) {
1402             continue;
1403         }
1404 
1405         region = g_malloc0(sizeof(*region));
1406         region->value = g_malloc0(sizeof(*region->value));
1407 
1408         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1409             region->value->type = g_strdup("io");
1410         } else {
1411             region->value->type = g_strdup("memory");
1412             region->value->has_prefetch = true;
1413             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1414             region->value->has_mem_type_64 = true;
1415             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1416         }
1417 
1418         region->value->bar = i;
1419         region->value->address = r->addr;
1420         region->value->size = r->size;
1421 
1422         /* XXX: waiting for the qapi to support GSList */
1423         if (!cur_item) {
1424             head = cur_item = region;
1425         } else {
1426             cur_item->next = region;
1427             cur_item = region;
1428         }
1429     }
1430 
1431     return head;
1432 }
1433 
1434 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1435                                            int bus_num)
1436 {
1437     PciBridgeInfo *info;
1438 
1439     info = g_malloc0(sizeof(*info));
1440 
1441     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1442     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1443     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1444 
1445     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1446     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1447     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1448 
1449     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1450     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1451     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1452 
1453     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1454     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1455     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1456 
1457     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1458         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1459         if (child_bus) {
1460             info->has_devices = true;
1461             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1462         }
1463     }
1464 
1465     return info;
1466 }
1467 
1468 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1469                                            int bus_num)
1470 {
1471     const pci_class_desc *desc;
1472     PciDeviceInfo *info;
1473     uint8_t type;
1474     int class;
1475 
1476     info = g_malloc0(sizeof(*info));
1477     info->bus = bus_num;
1478     info->slot = PCI_SLOT(dev->devfn);
1479     info->function = PCI_FUNC(dev->devfn);
1480 
1481     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1482     info->class_info.q_class = class;
1483     desc = get_class_desc(class);
1484     if (desc->desc) {
1485         info->class_info.has_desc = true;
1486         info->class_info.desc = g_strdup(desc->desc);
1487     }
1488 
1489     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1490     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1491     info->regions = qmp_query_pci_regions(dev);
1492     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1493 
1494     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1495         info->has_irq = true;
1496         info->irq = dev->config[PCI_INTERRUPT_LINE];
1497     }
1498 
1499     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1500     if (type == PCI_HEADER_TYPE_BRIDGE) {
1501         info->has_pci_bridge = true;
1502         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1503     }
1504 
1505     return info;
1506 }
1507 
1508 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1509 {
1510     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1511     PCIDevice *dev;
1512     int devfn;
1513 
1514     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1515         dev = bus->devices[devfn];
1516         if (dev) {
1517             info = g_malloc0(sizeof(*info));
1518             info->value = qmp_query_pci_device(dev, bus, bus_num);
1519 
1520             /* XXX: waiting for the qapi to support GSList */
1521             if (!cur_item) {
1522                 head = cur_item = info;
1523             } else {
1524                 cur_item->next = info;
1525                 cur_item = info;
1526             }
1527         }
1528     }
1529 
1530     return head;
1531 }
1532 
1533 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1534 {
1535     PciInfo *info = NULL;
1536 
1537     bus = pci_find_bus_nr(bus, bus_num);
1538     if (bus) {
1539         info = g_malloc0(sizeof(*info));
1540         info->bus = bus_num;
1541         info->devices = qmp_query_pci_devices(bus, bus_num);
1542     }
1543 
1544     return info;
1545 }
1546 
1547 PciInfoList *qmp_query_pci(Error **errp)
1548 {
1549     PciInfoList *info, *head = NULL, *cur_item = NULL;
1550     PCIHostState *host_bridge;
1551 
1552     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1553         info = g_malloc0(sizeof(*info));
1554         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1555 
1556         /* XXX: waiting for the qapi to support GSList */
1557         if (!cur_item) {
1558             head = cur_item = info;
1559         } else {
1560             cur_item->next = info;
1561             cur_item = info;
1562         }
1563     }
1564 
1565     return head;
1566 }
1567 
1568 static const char * const pci_nic_models[] = {
1569     "ne2k_pci",
1570     "i82551",
1571     "i82557b",
1572     "i82559er",
1573     "rtl8139",
1574     "e1000",
1575     "pcnet",
1576     "virtio",
1577     NULL
1578 };
1579 
1580 static const char * const pci_nic_names[] = {
1581     "ne2k_pci",
1582     "i82551",
1583     "i82557b",
1584     "i82559er",
1585     "rtl8139",
1586     "e1000",
1587     "pcnet",
1588     "virtio-net-pci",
1589     NULL
1590 };
1591 
1592 /* Initialize a PCI NIC.  */
1593 /* FIXME callers should check for failure, but don't */
1594 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1595                         const char *default_model,
1596                         const char *default_devaddr)
1597 {
1598     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1599     PCIBus *bus;
1600     int devfn;
1601     PCIDevice *pci_dev;
1602     DeviceState *dev;
1603     int i;
1604 
1605     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1606     if (i < 0)
1607         return NULL;
1608 
1609     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1610     if (!bus) {
1611         error_report("Invalid PCI device address %s for device %s",
1612                      devaddr, pci_nic_names[i]);
1613         return NULL;
1614     }
1615 
1616     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1617     dev = &pci_dev->qdev;
1618     qdev_set_nic_properties(dev, nd);
1619     if (qdev_init(dev) < 0)
1620         return NULL;
1621     return pci_dev;
1622 }
1623 
1624 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1625                                const char *default_model,
1626                                const char *default_devaddr)
1627 {
1628     PCIDevice *res;
1629 
1630     if (qemu_show_nic_models(nd->model, pci_nic_models))
1631         exit(0);
1632 
1633     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1634     if (!res)
1635         exit(1);
1636     return res;
1637 }
1638 
1639 PCIDevice *pci_vga_init(PCIBus *bus)
1640 {
1641     switch (vga_interface_type) {
1642     case VGA_CIRRUS:
1643         return pci_create_simple(bus, -1, "cirrus-vga");
1644     case VGA_QXL:
1645         return pci_create_simple(bus, -1, "qxl-vga");
1646     case VGA_STD:
1647         return pci_create_simple(bus, -1, "VGA");
1648     case VGA_VMWARE:
1649         return pci_create_simple(bus, -1, "vmware-svga");
1650     case VGA_NONE:
1651     default: /* Other non-PCI types. Checking for unsupported types is already
1652                 done in vl.c. */
1653         return NULL;
1654     }
1655 }
1656 
1657 /* Whether a given bus number is in range of the secondary
1658  * bus of the given bridge device. */
1659 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1660 {
1661     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1662              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1663         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1664         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1665 }
1666 
1667 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1668 {
1669     PCIBus *sec;
1670 
1671     if (!bus) {
1672         return NULL;
1673     }
1674 
1675     if (pci_bus_num(bus) == bus_num) {
1676         return bus;
1677     }
1678 
1679     /* Consider all bus numbers in range for the host pci bridge. */
1680     if (!pci_bus_is_root(bus) &&
1681         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1682         return NULL;
1683     }
1684 
1685     /* try child bus */
1686     for (; bus; bus = sec) {
1687         QLIST_FOREACH(sec, &bus->child, sibling) {
1688             assert(!pci_bus_is_root(sec));
1689             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1690                 return sec;
1691             }
1692             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1693                 break;
1694             }
1695         }
1696     }
1697 
1698     return NULL;
1699 }
1700 
1701 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1702 {
1703     bus = pci_find_bus_nr(bus, bus_num);
1704 
1705     if (!bus)
1706         return NULL;
1707 
1708     return bus->devices[devfn];
1709 }
1710 
1711 static int pci_qdev_init(DeviceState *qdev)
1712 {
1713     PCIDevice *pci_dev = (PCIDevice *)qdev;
1714     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1715     PCIBus *bus;
1716     int rc;
1717     bool is_default_rom;
1718 
1719     /* initialize cap_present for pci_is_express() and pci_config_size() */
1720     if (pc->is_express) {
1721         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1722     }
1723 
1724     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1725     pci_dev = do_pci_register_device(pci_dev, bus,
1726                                      object_get_typename(OBJECT(qdev)),
1727                                      pci_dev->devfn);
1728     if (pci_dev == NULL)
1729         return -1;
1730     if (qdev->hotplugged && pc->no_hotplug) {
1731         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(pci_dev)));
1732         do_pci_unregister_device(pci_dev);
1733         return -1;
1734     }
1735     if (pc->init) {
1736         rc = pc->init(pci_dev);
1737         if (rc != 0) {
1738             do_pci_unregister_device(pci_dev);
1739             return rc;
1740         }
1741     }
1742 
1743     /* rom loading */
1744     is_default_rom = false;
1745     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1746         pci_dev->romfile = g_strdup(pc->romfile);
1747         is_default_rom = true;
1748     }
1749     pci_add_option_rom(pci_dev, is_default_rom);
1750 
1751     if (bus->hotplug) {
1752         /* Let buses differentiate between hotplug and when device is
1753          * enabled during qemu machine creation. */
1754         rc = bus->hotplug(bus->hotplug_qdev, pci_dev,
1755                           qdev->hotplugged ? PCI_HOTPLUG_ENABLED:
1756                           PCI_COLDPLUG_ENABLED);
1757         if (rc != 0) {
1758             int r = pci_unregister_device(&pci_dev->qdev);
1759             assert(!r);
1760             return rc;
1761         }
1762     }
1763     return 0;
1764 }
1765 
1766 static int pci_unplug_device(DeviceState *qdev)
1767 {
1768     PCIDevice *dev = PCI_DEVICE(qdev);
1769     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1770 
1771     if (pc->no_hotplug) {
1772         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(dev)));
1773         return -1;
1774     }
1775     return dev->bus->hotplug(dev->bus->hotplug_qdev, dev,
1776                              PCI_HOTPLUG_DISABLED);
1777 }
1778 
1779 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1780                                     const char *name)
1781 {
1782     DeviceState *dev;
1783 
1784     dev = qdev_create(&bus->qbus, name);
1785     qdev_prop_set_int32(dev, "addr", devfn);
1786     qdev_prop_set_bit(dev, "multifunction", multifunction);
1787     return PCI_DEVICE(dev);
1788 }
1789 
1790 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1791                                            bool multifunction,
1792                                            const char *name)
1793 {
1794     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1795     qdev_init_nofail(&dev->qdev);
1796     return dev;
1797 }
1798 
1799 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1800 {
1801     return pci_create_multifunction(bus, devfn, false, name);
1802 }
1803 
1804 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1805 {
1806     return pci_create_simple_multifunction(bus, devfn, false, name);
1807 }
1808 
1809 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1810 {
1811     int offset = PCI_CONFIG_HEADER_SIZE;
1812     int i;
1813     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1814         if (pdev->used[i])
1815             offset = i + 1;
1816         else if (i - offset + 1 == size)
1817             return offset;
1818     }
1819     return 0;
1820 }
1821 
1822 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1823                                         uint8_t *prev_p)
1824 {
1825     uint8_t next, prev;
1826 
1827     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1828         return 0;
1829 
1830     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1831          prev = next + PCI_CAP_LIST_NEXT)
1832         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1833             break;
1834 
1835     if (prev_p)
1836         *prev_p = prev;
1837     return next;
1838 }
1839 
1840 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1841 {
1842     uint8_t next, prev, found = 0;
1843 
1844     if (!(pdev->used[offset])) {
1845         return 0;
1846     }
1847 
1848     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1849 
1850     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1851          prev = next + PCI_CAP_LIST_NEXT) {
1852         if (next <= offset && next > found) {
1853             found = next;
1854         }
1855     }
1856     return found;
1857 }
1858 
1859 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1860    This is needed for an option rom which is used for more than one device. */
1861 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1862 {
1863     uint16_t vendor_id;
1864     uint16_t device_id;
1865     uint16_t rom_vendor_id;
1866     uint16_t rom_device_id;
1867     uint16_t rom_magic;
1868     uint16_t pcir_offset;
1869     uint8_t checksum;
1870 
1871     /* Words in rom data are little endian (like in PCI configuration),
1872        so they can be read / written with pci_get_word / pci_set_word. */
1873 
1874     /* Only a valid rom will be patched. */
1875     rom_magic = pci_get_word(ptr);
1876     if (rom_magic != 0xaa55) {
1877         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1878         return;
1879     }
1880     pcir_offset = pci_get_word(ptr + 0x18);
1881     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1882         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1883         return;
1884     }
1885 
1886     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1887     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1888     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1889     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1890 
1891     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1892                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1893 
1894     checksum = ptr[6];
1895 
1896     if (vendor_id != rom_vendor_id) {
1897         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1898         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1899         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1900         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1901         ptr[6] = checksum;
1902         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1903     }
1904 
1905     if (device_id != rom_device_id) {
1906         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1907         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1908         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1909         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1910         ptr[6] = checksum;
1911         pci_set_word(ptr + pcir_offset + 6, device_id);
1912     }
1913 }
1914 
1915 /* Add an option rom for the device */
1916 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1917 {
1918     int size;
1919     char *path;
1920     void *ptr;
1921     char name[32];
1922     const VMStateDescription *vmsd;
1923 
1924     if (!pdev->romfile)
1925         return 0;
1926     if (strlen(pdev->romfile) == 0)
1927         return 0;
1928 
1929     if (!pdev->rom_bar) {
1930         /*
1931          * Load rom via fw_cfg instead of creating a rom bar,
1932          * for 0.11 compatibility.
1933          */
1934         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1935         if (class == 0x0300) {
1936             rom_add_vga(pdev->romfile);
1937         } else {
1938             rom_add_option(pdev->romfile, -1);
1939         }
1940         return 0;
1941     }
1942 
1943     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1944     if (path == NULL) {
1945         path = g_strdup(pdev->romfile);
1946     }
1947 
1948     size = get_image_size(path);
1949     if (size < 0) {
1950         error_report("%s: failed to find romfile \"%s\"",
1951                      __func__, pdev->romfile);
1952         g_free(path);
1953         return -1;
1954     } else if (size == 0) {
1955         error_report("%s: ignoring empty romfile \"%s\"",
1956                      __func__, pdev->romfile);
1957         g_free(path);
1958         return -1;
1959     }
1960     if (size & (size - 1)) {
1961         size = 1 << qemu_fls(size);
1962     }
1963 
1964     vmsd = qdev_get_vmsd(DEVICE(pdev));
1965 
1966     if (vmsd) {
1967         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1968     } else {
1969         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1970     }
1971     pdev->has_rom = true;
1972     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size);
1973     vmstate_register_ram(&pdev->rom, &pdev->qdev);
1974     ptr = memory_region_get_ram_ptr(&pdev->rom);
1975     load_image(path, ptr);
1976     g_free(path);
1977 
1978     if (is_default_rom) {
1979         /* Only the default rom images will be patched (if needed). */
1980         pci_patch_ids(pdev, ptr, size);
1981     }
1982 
1983     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
1984 
1985     return 0;
1986 }
1987 
1988 static void pci_del_option_rom(PCIDevice *pdev)
1989 {
1990     if (!pdev->has_rom)
1991         return;
1992 
1993     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
1994     memory_region_destroy(&pdev->rom);
1995     pdev->has_rom = false;
1996 }
1997 
1998 /*
1999  * if !offset
2000  * Reserve space and add capability to the linked list in pci config space
2001  *
2002  * if offset = 0,
2003  * Find and reserve space and add capability to the linked list
2004  * in pci config space */
2005 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2006                        uint8_t offset, uint8_t size)
2007 {
2008     uint8_t *config;
2009     int i, overlapping_cap;
2010 
2011     if (!offset) {
2012         offset = pci_find_space(pdev, size);
2013         if (!offset) {
2014             return -ENOSPC;
2015         }
2016     } else {
2017         /* Verify that capabilities don't overlap.  Note: device assignment
2018          * depends on this check to verify that the device is not broken.
2019          * Should never trigger for emulated devices, but it's helpful
2020          * for debugging these. */
2021         for (i = offset; i < offset + size; i++) {
2022             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2023             if (overlapping_cap) {
2024                 fprintf(stderr, "ERROR: %s:%02x:%02x.%x "
2025                         "Attempt to add PCI capability %x at offset "
2026                         "%x overlaps existing capability %x at offset %x\n",
2027                         pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2028                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2029                         cap_id, offset, overlapping_cap, i);
2030                 return -EINVAL;
2031             }
2032         }
2033     }
2034 
2035     config = pdev->config + offset;
2036     config[PCI_CAP_LIST_ID] = cap_id;
2037     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2038     pdev->config[PCI_CAPABILITY_LIST] = offset;
2039     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2040     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2041     /* Make capability read-only by default */
2042     memset(pdev->wmask + offset, 0, size);
2043     /* Check capability by default */
2044     memset(pdev->cmask + offset, 0xFF, size);
2045     return offset;
2046 }
2047 
2048 /* Unlink capability from the pci config space. */
2049 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2050 {
2051     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2052     if (!offset)
2053         return;
2054     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2055     /* Make capability writable again */
2056     memset(pdev->wmask + offset, 0xff, size);
2057     memset(pdev->w1cmask + offset, 0, size);
2058     /* Clear cmask as device-specific registers can't be checked */
2059     memset(pdev->cmask + offset, 0, size);
2060     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2061 
2062     if (!pdev->config[PCI_CAPABILITY_LIST])
2063         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2064 }
2065 
2066 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2067 {
2068     return pci_find_capability_list(pdev, cap_id, NULL);
2069 }
2070 
2071 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2072 {
2073     PCIDevice *d = (PCIDevice *)dev;
2074     const pci_class_desc *desc;
2075     char ctxt[64];
2076     PCIIORegion *r;
2077     int i, class;
2078 
2079     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2080     desc = pci_class_descriptions;
2081     while (desc->desc && class != desc->class)
2082         desc++;
2083     if (desc->desc) {
2084         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2085     } else {
2086         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2087     }
2088 
2089     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2090                    "pci id %04x:%04x (sub %04x:%04x)\n",
2091                    indent, "", ctxt, pci_bus_num(d->bus),
2092                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2093                    pci_get_word(d->config + PCI_VENDOR_ID),
2094                    pci_get_word(d->config + PCI_DEVICE_ID),
2095                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2096                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2097     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2098         r = &d->io_regions[i];
2099         if (!r->size)
2100             continue;
2101         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2102                        " [0x%"FMT_PCIBUS"]\n",
2103                        indent, "",
2104                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2105                        r->addr, r->addr + r->size - 1);
2106     }
2107 }
2108 
2109 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2110 {
2111     PCIDevice *d = (PCIDevice *)dev;
2112     const char *name = NULL;
2113     const pci_class_desc *desc =  pci_class_descriptions;
2114     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2115 
2116     while (desc->desc &&
2117           (class & ~desc->fw_ign_bits) !=
2118           (desc->class & ~desc->fw_ign_bits)) {
2119         desc++;
2120     }
2121 
2122     if (desc->desc) {
2123         name = desc->fw_name;
2124     }
2125 
2126     if (name) {
2127         pstrcpy(buf, len, name);
2128     } else {
2129         snprintf(buf, len, "pci%04x,%04x",
2130                  pci_get_word(d->config + PCI_VENDOR_ID),
2131                  pci_get_word(d->config + PCI_DEVICE_ID));
2132     }
2133 
2134     return buf;
2135 }
2136 
2137 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2138 {
2139     PCIDevice *d = (PCIDevice *)dev;
2140     char path[50], name[33];
2141     int off;
2142 
2143     off = snprintf(path, sizeof(path), "%s@%x",
2144                    pci_dev_fw_name(dev, name, sizeof name),
2145                    PCI_SLOT(d->devfn));
2146     if (PCI_FUNC(d->devfn))
2147         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2148     return g_strdup(path);
2149 }
2150 
2151 static char *pcibus_get_dev_path(DeviceState *dev)
2152 {
2153     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2154     PCIDevice *t;
2155     int slot_depth;
2156     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2157      * 00 is added here to make this format compatible with
2158      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2159      * Slot.Function list specifies the slot and function numbers for all
2160      * devices on the path from root to the specific device. */
2161     const char *root_bus_path;
2162     int root_bus_len;
2163     char slot[] = ":SS.F";
2164     int slot_len = sizeof slot - 1 /* For '\0' */;
2165     int path_len;
2166     char *path, *p;
2167     int s;
2168 
2169     root_bus_path = pci_root_bus_path(d);
2170     root_bus_len = strlen(root_bus_path);
2171 
2172     /* Calculate # of slots on path between device and root. */;
2173     slot_depth = 0;
2174     for (t = d; t; t = t->bus->parent_dev) {
2175         ++slot_depth;
2176     }
2177 
2178     path_len = root_bus_len + slot_len * slot_depth;
2179 
2180     /* Allocate memory, fill in the terminating null byte. */
2181     path = g_malloc(path_len + 1 /* For '\0' */);
2182     path[path_len] = '\0';
2183 
2184     memcpy(path, root_bus_path, root_bus_len);
2185 
2186     /* Fill in slot numbers. We walk up from device to root, so need to print
2187      * them in the reverse order, last to first. */
2188     p = path + path_len;
2189     for (t = d; t; t = t->bus->parent_dev) {
2190         p -= slot_len;
2191         s = snprintf(slot, sizeof slot, ":%02x.%x",
2192                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2193         assert(s == slot_len);
2194         memcpy(p, slot, slot_len);
2195     }
2196 
2197     return path;
2198 }
2199 
2200 static int pci_qdev_find_recursive(PCIBus *bus,
2201                                    const char *id, PCIDevice **pdev)
2202 {
2203     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2204     if (!qdev) {
2205         return -ENODEV;
2206     }
2207 
2208     /* roughly check if given qdev is pci device */
2209     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2210         *pdev = PCI_DEVICE(qdev);
2211         return 0;
2212     }
2213     return -EINVAL;
2214 }
2215 
2216 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2217 {
2218     PCIHostState *host_bridge;
2219     int rc = -ENODEV;
2220 
2221     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2222         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2223         if (!tmp) {
2224             rc = 0;
2225             break;
2226         }
2227         if (tmp != -ENODEV) {
2228             rc = tmp;
2229         }
2230     }
2231 
2232     return rc;
2233 }
2234 
2235 MemoryRegion *pci_address_space(PCIDevice *dev)
2236 {
2237     return dev->bus->address_space_mem;
2238 }
2239 
2240 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2241 {
2242     return dev->bus->address_space_io;
2243 }
2244 
2245 static void pci_device_class_init(ObjectClass *klass, void *data)
2246 {
2247     DeviceClass *k = DEVICE_CLASS(klass);
2248     k->init = pci_qdev_init;
2249     k->unplug = pci_unplug_device;
2250     k->exit = pci_unregister_device;
2251     k->bus_type = TYPE_PCI_BUS;
2252     k->props = pci_props;
2253 }
2254 
2255 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2256 {
2257     PCIBus *bus = PCI_BUS(dev->bus);
2258 
2259     if (bus->iommu_fn) {
2260         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2261     }
2262 
2263     if (bus->parent_dev) {
2264         /** We are ignoring the bus master DMA bit of the bridge
2265          *  as it would complicate things such as VFIO for no good reason */
2266         return pci_device_iommu_address_space(bus->parent_dev);
2267     }
2268 
2269     return &address_space_memory;
2270 }
2271 
2272 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2273 {
2274     bus->iommu_fn = fn;
2275     bus->iommu_opaque = opaque;
2276 }
2277 
2278 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2279 {
2280     Range *range = opaque;
2281     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2282     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2283     int i;
2284 
2285     if (!(cmd & PCI_COMMAND_MEMORY)) {
2286         return;
2287     }
2288 
2289     if (pc->is_bridge) {
2290         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2291         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2292 
2293         base = MAX(base, 0x1ULL << 32);
2294 
2295         if (limit >= base) {
2296             Range pref_range;
2297             pref_range.begin = base;
2298             pref_range.end = limit + 1;
2299             range_extend(range, &pref_range);
2300         }
2301     }
2302     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2303         PCIIORegion *r = &dev->io_regions[i];
2304         Range region_range;
2305 
2306         if (!r->size ||
2307             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2308             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2309             continue;
2310         }
2311         region_range.begin = pci_bar_address(dev, i, r->type, r->size);
2312         region_range.end = region_range.begin + r->size;
2313 
2314         if (region_range.begin == PCI_BAR_UNMAPPED) {
2315             continue;
2316         }
2317 
2318         region_range.begin = MAX(region_range.begin, 0x1ULL << 32);
2319 
2320         if (region_range.end - 1 >= region_range.begin) {
2321             range_extend(range, &region_range);
2322         }
2323     }
2324 }
2325 
2326 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2327 {
2328     range->begin = range->end = 0;
2329     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2330 }
2331 
2332 static const TypeInfo pci_device_type_info = {
2333     .name = TYPE_PCI_DEVICE,
2334     .parent = TYPE_DEVICE,
2335     .instance_size = sizeof(PCIDevice),
2336     .abstract = true,
2337     .class_size = sizeof(PCIDeviceClass),
2338     .class_init = pci_device_class_init,
2339 };
2340 
2341 static void pci_register_types(void)
2342 {
2343     type_register_static(&pci_bus_info);
2344     type_register_static(&pcie_bus_info);
2345     type_register_static(&pci_device_type_info);
2346 }
2347 
2348 type_init(pci_register_types)
2349