xref: /openbmc/qemu/hw/pci/pci.c (revision 1b111dc1)
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 static void pci_bus_finalize(Object *obj);
51 
52 static Property pci_props[] = {
53     DEFINE_PROP_PCI_DEVFN("addr", PCIDevice, devfn, -1),
54     DEFINE_PROP_STRING("romfile", PCIDevice, romfile),
55     DEFINE_PROP_UINT32("rombar",  PCIDevice, rom_bar, 1),
56     DEFINE_PROP_BIT("multifunction", PCIDevice, cap_present,
57                     QEMU_PCI_CAP_MULTIFUNCTION_BITNR, false),
58     DEFINE_PROP_BIT("command_serr_enable", PCIDevice, cap_present,
59                     QEMU_PCI_CAP_SERR_BITNR, true),
60     DEFINE_PROP_END_OF_LIST()
61 };
62 
63 static void pci_bus_class_init(ObjectClass *klass, void *data)
64 {
65     BusClass *k = BUS_CLASS(klass);
66 
67     k->print_dev = pcibus_dev_print;
68     k->get_dev_path = pcibus_get_dev_path;
69     k->get_fw_dev_path = pcibus_get_fw_dev_path;
70     k->reset = pcibus_reset;
71 }
72 
73 static const TypeInfo pci_bus_info = {
74     .name = TYPE_PCI_BUS,
75     .parent = TYPE_BUS,
76     .instance_size = sizeof(PCIBus),
77     .instance_finalize = pci_bus_finalize,
78     .class_init = pci_bus_class_init,
79 };
80 
81 static const TypeInfo pcie_bus_info = {
82     .name = TYPE_PCIE_BUS,
83     .parent = TYPE_PCI_BUS,
84 };
85 
86 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num);
87 static void pci_update_mappings(PCIDevice *d);
88 static void pci_irq_handler(void *opaque, int irq_num, int level);
89 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom);
90 static void pci_del_option_rom(PCIDevice *pdev);
91 
92 static uint16_t pci_default_sub_vendor_id = PCI_SUBVENDOR_ID_REDHAT_QUMRANET;
93 static uint16_t pci_default_sub_device_id = PCI_SUBDEVICE_ID_QEMU;
94 
95 static QLIST_HEAD(, PCIHostState) pci_host_bridges;
96 
97 static const VMStateDescription vmstate_pcibus = {
98     .name = "PCIBUS",
99     .version_id = 1,
100     .minimum_version_id = 1,
101     .minimum_version_id_old = 1,
102     .fields      = (VMStateField []) {
103         VMSTATE_INT32_EQUAL(nirq, PCIBus),
104         VMSTATE_VARRAY_INT32(irq_count, PCIBus, nirq, 0, vmstate_info_int32, int32_t),
105         VMSTATE_END_OF_LIST()
106     }
107 };
108 static int pci_bar(PCIDevice *d, int reg)
109 {
110     uint8_t type;
111 
112     if (reg != PCI_ROM_SLOT)
113         return PCI_BASE_ADDRESS_0 + reg * 4;
114 
115     type = d->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
116     return type == PCI_HEADER_TYPE_BRIDGE ? PCI_ROM_ADDRESS1 : PCI_ROM_ADDRESS;
117 }
118 
119 static inline int pci_irq_state(PCIDevice *d, int irq_num)
120 {
121 	return (d->irq_state >> irq_num) & 0x1;
122 }
123 
124 static inline void pci_set_irq_state(PCIDevice *d, int irq_num, int level)
125 {
126 	d->irq_state &= ~(0x1 << irq_num);
127 	d->irq_state |= level << irq_num;
128 }
129 
130 static void pci_change_irq_level(PCIDevice *pci_dev, int irq_num, int change)
131 {
132     PCIBus *bus;
133     for (;;) {
134         bus = pci_dev->bus;
135         irq_num = bus->map_irq(pci_dev, irq_num);
136         if (bus->set_irq)
137             break;
138         pci_dev = bus->parent_dev;
139     }
140     bus->irq_count[irq_num] += change;
141     bus->set_irq(bus->irq_opaque, irq_num, bus->irq_count[irq_num] != 0);
142 }
143 
144 int pci_bus_get_irq_level(PCIBus *bus, int irq_num)
145 {
146     assert(irq_num >= 0);
147     assert(irq_num < bus->nirq);
148     return !!bus->irq_count[irq_num];
149 }
150 
151 /* Update interrupt status bit in config space on interrupt
152  * state change. */
153 static void pci_update_irq_status(PCIDevice *dev)
154 {
155     if (dev->irq_state) {
156         dev->config[PCI_STATUS] |= PCI_STATUS_INTERRUPT;
157     } else {
158         dev->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
159     }
160 }
161 
162 void pci_device_deassert_intx(PCIDevice *dev)
163 {
164     int i;
165     for (i = 0; i < PCI_NUM_PINS; ++i) {
166         pci_irq_handler(dev, i, 0);
167     }
168 }
169 
170 /*
171  * This function is called on #RST and FLR.
172  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
173  */
174 void pci_device_reset(PCIDevice *dev)
175 {
176     int r;
177 
178     qdev_reset_all(&dev->qdev);
179 
180     dev->irq_state = 0;
181     pci_update_irq_status(dev);
182     pci_device_deassert_intx(dev);
183     /* Clear all writable bits */
184     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
185                                  pci_get_word(dev->wmask + PCI_COMMAND) |
186                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
187     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
188                                  pci_get_word(dev->wmask + PCI_STATUS) |
189                                  pci_get_word(dev->w1cmask + PCI_STATUS));
190     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
191     dev->config[PCI_INTERRUPT_LINE] = 0x0;
192     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
193         PCIIORegion *region = &dev->io_regions[r];
194         if (!region->size) {
195             continue;
196         }
197 
198         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
199             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
200             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
201         } else {
202             pci_set_long(dev->config + pci_bar(dev, r), region->type);
203         }
204     }
205     pci_update_mappings(dev);
206 
207     msi_reset(dev);
208     msix_reset(dev);
209 }
210 
211 /*
212  * Trigger pci bus reset under a given bus.
213  * To be called on RST# assert.
214  */
215 void pci_bus_reset(PCIBus *bus)
216 {
217     int i;
218 
219     for (i = 0; i < bus->nirq; i++) {
220         bus->irq_count[i] = 0;
221     }
222     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
223         if (bus->devices[i]) {
224             pci_device_reset(bus->devices[i]);
225         }
226     }
227 }
228 
229 static int pcibus_reset(BusState *qbus)
230 {
231     pci_bus_reset(DO_UPCAST(PCIBus, qbus, qbus));
232 
233     /* topology traverse is done by pci_bus_reset().
234        Tell qbus/qdev walker not to traverse the tree */
235     return 1;
236 }
237 
238 static void pci_host_bus_register(PCIBus *bus, DeviceState *parent)
239 {
240     PCIHostState *host_bridge = PCI_HOST_BRIDGE(parent);
241 
242     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
243 }
244 
245 PCIBus *pci_find_primary_bus(void)
246 {
247     PCIBus *primary_bus = NULL;
248     PCIHostState *host;
249 
250     QLIST_FOREACH(host, &pci_host_bridges, next) {
251         if (primary_bus) {
252             /* We have multiple root buses, refuse to select a primary */
253             return NULL;
254         }
255         primary_bus = host->bus;
256     }
257 
258     return primary_bus;
259 }
260 
261 PCIBus *pci_device_root_bus(const PCIDevice *d)
262 {
263     PCIBus *bus = d->bus;
264 
265     while ((d = bus->parent_dev) != NULL) {
266         bus = d->bus;
267     }
268 
269     return bus;
270 }
271 
272 const char *pci_root_bus_path(PCIDevice *dev)
273 {
274     PCIBus *rootbus = pci_device_root_bus(dev);
275     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
276     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
277 
278     assert(!rootbus->parent_dev);
279     assert(host_bridge->bus == rootbus);
280 
281     if (hc->root_bus_path) {
282         return (*hc->root_bus_path)(host_bridge, rootbus);
283     }
284 
285     return rootbus->qbus.name;
286 }
287 
288 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
289                          const char *name,
290                          MemoryRegion *address_space_mem,
291                          MemoryRegion *address_space_io,
292                          uint8_t devfn_min)
293 {
294     assert(PCI_FUNC(devfn_min) == 0);
295     bus->devfn_min = devfn_min;
296     bus->address_space_mem = address_space_mem;
297     bus->address_space_io = address_space_io;
298 
299     /* host bridge */
300     QLIST_INIT(&bus->child);
301 
302     pci_host_bus_register(bus, parent);
303 
304     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
305 }
306 
307 bool pci_bus_is_express(PCIBus *bus)
308 {
309     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
310 }
311 
312 bool pci_bus_is_root(PCIBus *bus)
313 {
314     return !bus->parent_dev;
315 }
316 
317 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
318                          const char *name,
319                          MemoryRegion *address_space_mem,
320                          MemoryRegion *address_space_io,
321                          uint8_t devfn_min, const char *typename)
322 {
323     qbus_create_inplace(bus, bus_size, typename, parent, name);
324     pci_bus_init(bus, parent, name, address_space_mem,
325                  address_space_io, devfn_min);
326 }
327 
328 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
329                     MemoryRegion *address_space_mem,
330                     MemoryRegion *address_space_io,
331                     uint8_t devfn_min, const char *typename)
332 {
333     PCIBus *bus;
334 
335     bus = PCI_BUS(qbus_create(typename, parent, name));
336     pci_bus_init(bus, parent, name, address_space_mem,
337                  address_space_io, devfn_min);
338     return bus;
339 }
340 
341 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
342                   void *irq_opaque, int nirq)
343 {
344     bus->set_irq = set_irq;
345     bus->map_irq = map_irq;
346     bus->irq_opaque = irq_opaque;
347     bus->nirq = nirq;
348     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
349 }
350 
351 void pci_bus_hotplug(PCIBus *bus, pci_hotplug_fn hotplug, DeviceState *qdev)
352 {
353     bus->qbus.allow_hotplug = 1;
354     bus->hotplug = hotplug;
355     bus->hotplug_qdev = qdev;
356 }
357 
358 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
359                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
360                          void *irq_opaque,
361                          MemoryRegion *address_space_mem,
362                          MemoryRegion *address_space_io,
363                          uint8_t devfn_min, int nirq, const char *typename)
364 {
365     PCIBus *bus;
366 
367     bus = pci_bus_new(parent, name, address_space_mem,
368                       address_space_io, devfn_min, typename);
369     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
370     return bus;
371 }
372 
373 int pci_bus_num(PCIBus *s)
374 {
375     if (pci_bus_is_root(s))
376         return 0;       /* pci host bridge */
377     return s->parent_dev->config[PCI_SECONDARY_BUS];
378 }
379 
380 static void pci_bus_finalize(Object *obj)
381 {
382     PCIBus *bus = PCI_BUS(obj);
383     vmstate_unregister(NULL, &vmstate_pcibus, bus);
384 }
385 
386 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size)
387 {
388     PCIDevice *s = container_of(pv, PCIDevice, config);
389     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
390     uint8_t *config;
391     int i;
392 
393     assert(size == pci_config_size(s));
394     config = g_malloc(size);
395 
396     qemu_get_buffer(f, config, size);
397     for (i = 0; i < size; ++i) {
398         if ((config[i] ^ s->config[i]) &
399             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
400             g_free(config);
401             return -EINVAL;
402         }
403     }
404     memcpy(s->config, config, size);
405 
406     pci_update_mappings(s);
407     if (pc->is_bridge) {
408         PCIBridge *b = PCI_BRIDGE(s);
409         pci_bridge_update_mappings(b);
410     }
411 
412     memory_region_set_enabled(&s->bus_master_enable_region,
413                               pci_get_word(s->config + PCI_COMMAND)
414                               & PCI_COMMAND_MASTER);
415 
416     g_free(config);
417     return 0;
418 }
419 
420 /* just put buffer */
421 static void put_pci_config_device(QEMUFile *f, void *pv, size_t size)
422 {
423     const uint8_t **v = pv;
424     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
425     qemu_put_buffer(f, *v, size);
426 }
427 
428 static VMStateInfo vmstate_info_pci_config = {
429     .name = "pci config",
430     .get  = get_pci_config_device,
431     .put  = put_pci_config_device,
432 };
433 
434 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size)
435 {
436     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
437     uint32_t irq_state[PCI_NUM_PINS];
438     int i;
439     for (i = 0; i < PCI_NUM_PINS; ++i) {
440         irq_state[i] = qemu_get_be32(f);
441         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
442             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
443                     irq_state[i]);
444             return -EINVAL;
445         }
446     }
447 
448     for (i = 0; i < PCI_NUM_PINS; ++i) {
449         pci_set_irq_state(s, i, irq_state[i]);
450     }
451 
452     return 0;
453 }
454 
455 static void put_pci_irq_state(QEMUFile *f, void *pv, size_t size)
456 {
457     int i;
458     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
459 
460     for (i = 0; i < PCI_NUM_PINS; ++i) {
461         qemu_put_be32(f, pci_irq_state(s, i));
462     }
463 }
464 
465 static VMStateInfo vmstate_info_pci_irq_state = {
466     .name = "pci irq state",
467     .get  = get_pci_irq_state,
468     .put  = put_pci_irq_state,
469 };
470 
471 const VMStateDescription vmstate_pci_device = {
472     .name = "PCIDevice",
473     .version_id = 2,
474     .minimum_version_id = 1,
475     .minimum_version_id_old = 1,
476     .fields      = (VMStateField []) {
477         VMSTATE_INT32_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     .minimum_version_id_old = 1,
493     .fields      = (VMStateField []) {
494         VMSTATE_INT32_LE(version_id, PCIDevice),
495         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
496                                    vmstate_info_pci_config,
497                                    PCIE_CONFIG_SPACE_SIZE),
498         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
499 				   vmstate_info_pci_irq_state,
500 				   PCI_NUM_PINS * sizeof(int32_t)),
501         VMSTATE_END_OF_LIST()
502     }
503 };
504 
505 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
506 {
507     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
508 }
509 
510 void pci_device_save(PCIDevice *s, QEMUFile *f)
511 {
512     /* Clear interrupt status bit: it is implicit
513      * in irq_state which we are saving.
514      * This makes us compatible with old devices
515      * which never set or clear this bit. */
516     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
517     vmstate_save_state(f, pci_get_vmstate(s), s);
518     /* Restore the interrupt status bit. */
519     pci_update_irq_status(s);
520 }
521 
522 int pci_device_load(PCIDevice *s, QEMUFile *f)
523 {
524     int ret;
525     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
526     /* Restore the interrupt status bit. */
527     pci_update_irq_status(s);
528     return ret;
529 }
530 
531 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
532 {
533     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
534                  pci_default_sub_vendor_id);
535     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
536                  pci_default_sub_device_id);
537 }
538 
539 /*
540  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
541  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
542  */
543 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
544                       unsigned int *slotp, unsigned int *funcp)
545 {
546     const char *p;
547     char *e;
548     unsigned long val;
549     unsigned long dom = 0, bus = 0;
550     unsigned int slot = 0;
551     unsigned int func = 0;
552 
553     p = addr;
554     val = strtoul(p, &e, 16);
555     if (e == p)
556 	return -1;
557     if (*e == ':') {
558 	bus = val;
559 	p = e + 1;
560 	val = strtoul(p, &e, 16);
561 	if (e == p)
562 	    return -1;
563 	if (*e == ':') {
564 	    dom = bus;
565 	    bus = val;
566 	    p = e + 1;
567 	    val = strtoul(p, &e, 16);
568 	    if (e == p)
569 		return -1;
570 	}
571     }
572 
573     slot = val;
574 
575     if (funcp != NULL) {
576         if (*e != '.')
577             return -1;
578 
579         p = e + 1;
580         val = strtoul(p, &e, 16);
581         if (e == p)
582             return -1;
583 
584         func = val;
585     }
586 
587     /* if funcp == NULL func is 0 */
588     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
589 	return -1;
590 
591     if (*e)
592 	return -1;
593 
594     *domp = dom;
595     *busp = bus;
596     *slotp = slot;
597     if (funcp != NULL)
598         *funcp = func;
599     return 0;
600 }
601 
602 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
603 {
604     int dom, bus;
605     unsigned slot;
606 
607     assert(!root->parent_dev);
608 
609     if (!root) {
610         fprintf(stderr, "No primary PCI bus\n");
611         return NULL;
612     }
613 
614     if (!devaddr) {
615         *devfnp = -1;
616         return pci_find_bus_nr(root, 0);
617     }
618 
619     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
620         return NULL;
621     }
622 
623     if (dom != 0) {
624         fprintf(stderr, "No support for non-zero PCI domains\n");
625         return NULL;
626     }
627 
628     *devfnp = PCI_DEVFN(slot, 0);
629     return pci_find_bus_nr(root, bus);
630 }
631 
632 static void pci_init_cmask(PCIDevice *dev)
633 {
634     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
635     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
636     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
637     dev->cmask[PCI_REVISION_ID] = 0xff;
638     dev->cmask[PCI_CLASS_PROG] = 0xff;
639     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
640     dev->cmask[PCI_HEADER_TYPE] = 0xff;
641     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
642 }
643 
644 static void pci_init_wmask(PCIDevice *dev)
645 {
646     int config_size = pci_config_size(dev);
647 
648     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
649     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
650     pci_set_word(dev->wmask + PCI_COMMAND,
651                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
652                  PCI_COMMAND_INTX_DISABLE);
653     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
654         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
655     }
656 
657     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
658            config_size - PCI_CONFIG_HEADER_SIZE);
659 }
660 
661 static void pci_init_w1cmask(PCIDevice *dev)
662 {
663     /*
664      * Note: It's okay to set w1cmask even for readonly bits as
665      * long as their value is hardwired to 0.
666      */
667     pci_set_word(dev->w1cmask + PCI_STATUS,
668                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
669                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
670                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
671 }
672 
673 static void pci_init_mask_bridge(PCIDevice *d)
674 {
675     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
676        PCI_SEC_LETENCY_TIMER */
677     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
678 
679     /* base and limit */
680     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
681     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
682     pci_set_word(d->wmask + PCI_MEMORY_BASE,
683                  PCI_MEMORY_RANGE_MASK & 0xffff);
684     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
685                  PCI_MEMORY_RANGE_MASK & 0xffff);
686     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
687                  PCI_PREF_RANGE_MASK & 0xffff);
688     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
689                  PCI_PREF_RANGE_MASK & 0xffff);
690 
691     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
692     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
693 
694     /* Supported memory and i/o types */
695     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
696     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
697     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
698                                PCI_PREF_RANGE_TYPE_64);
699     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
700                                PCI_PREF_RANGE_TYPE_64);
701 
702     /*
703      * TODO: Bridges default to 10-bit VGA decoding but we currently only
704      * implement 16-bit decoding (no alias support).
705      */
706     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
707                  PCI_BRIDGE_CTL_PARITY |
708                  PCI_BRIDGE_CTL_SERR |
709                  PCI_BRIDGE_CTL_ISA |
710                  PCI_BRIDGE_CTL_VGA |
711                  PCI_BRIDGE_CTL_VGA_16BIT |
712                  PCI_BRIDGE_CTL_MASTER_ABORT |
713                  PCI_BRIDGE_CTL_BUS_RESET |
714                  PCI_BRIDGE_CTL_FAST_BACK |
715                  PCI_BRIDGE_CTL_DISCARD |
716                  PCI_BRIDGE_CTL_SEC_DISCARD |
717                  PCI_BRIDGE_CTL_DISCARD_SERR);
718     /* Below does not do anything as we never set this bit, put here for
719      * completeness. */
720     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
721                  PCI_BRIDGE_CTL_DISCARD_STATUS);
722     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
723     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
724     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
725                                PCI_PREF_RANGE_TYPE_MASK);
726     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
727                                PCI_PREF_RANGE_TYPE_MASK);
728 }
729 
730 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
731 {
732     uint8_t slot = PCI_SLOT(dev->devfn);
733     uint8_t func;
734 
735     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
736         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
737     }
738 
739     /*
740      * multifunction bit is interpreted in two ways as follows.
741      *   - all functions must set the bit to 1.
742      *     Example: Intel X53
743      *   - function 0 must set the bit, but the rest function (> 0)
744      *     is allowed to leave the bit to 0.
745      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
746      *
747      * So OS (at least Linux) checks the bit of only function 0,
748      * and doesn't see the bit of function > 0.
749      *
750      * The below check allows both interpretation.
751      */
752     if (PCI_FUNC(dev->devfn)) {
753         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
754         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
755             /* function 0 should set multifunction bit */
756             error_report("PCI: single function device can't be populated "
757                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
758             return -1;
759         }
760         return 0;
761     }
762 
763     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
764         return 0;
765     }
766     /* function 0 indicates single function, so function > 0 must be NULL */
767     for (func = 1; func < PCI_FUNC_MAX; ++func) {
768         if (bus->devices[PCI_DEVFN(slot, func)]) {
769             error_report("PCI: %x.0 indicates single function, "
770                          "but %x.%x is already populated.",
771                          slot, slot, func);
772             return -1;
773         }
774     }
775     return 0;
776 }
777 
778 static void pci_config_alloc(PCIDevice *pci_dev)
779 {
780     int config_size = pci_config_size(pci_dev);
781 
782     pci_dev->config = g_malloc0(config_size);
783     pci_dev->cmask = g_malloc0(config_size);
784     pci_dev->wmask = g_malloc0(config_size);
785     pci_dev->w1cmask = g_malloc0(config_size);
786     pci_dev->used = g_malloc0(config_size);
787 }
788 
789 static void pci_config_free(PCIDevice *pci_dev)
790 {
791     g_free(pci_dev->config);
792     g_free(pci_dev->cmask);
793     g_free(pci_dev->wmask);
794     g_free(pci_dev->w1cmask);
795     g_free(pci_dev->used);
796 }
797 
798 /* -1 for devfn means auto assign */
799 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
800                                          const char *name, int devfn)
801 {
802     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
803     PCIConfigReadFunc *config_read = pc->config_read;
804     PCIConfigWriteFunc *config_write = pc->config_write;
805     AddressSpace *dma_as;
806 
807     if (devfn < 0) {
808         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
809             devfn += PCI_FUNC_MAX) {
810             if (!bus->devices[devfn])
811                 goto found;
812         }
813         error_report("PCI: no slot/function available for %s, all in use", name);
814         return NULL;
815     found: ;
816     } else if (bus->devices[devfn]) {
817         error_report("PCI: slot %d function %d not available for %s, in use by %s",
818                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
819         return NULL;
820     }
821 
822     pci_dev->bus = bus;
823     dma_as = pci_device_iommu_address_space(pci_dev);
824 
825     memory_region_init_alias(&pci_dev->bus_master_enable_region,
826                              OBJECT(pci_dev), "bus master",
827                              dma_as->root, 0, memory_region_size(dma_as->root));
828     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
829     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
830                        name);
831 
832     pci_dev->devfn = devfn;
833     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
834     pci_dev->irq_state = 0;
835     pci_config_alloc(pci_dev);
836 
837     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
838     pci_config_set_device_id(pci_dev->config, pc->device_id);
839     pci_config_set_revision(pci_dev->config, pc->revision);
840     pci_config_set_class(pci_dev->config, pc->class_id);
841 
842     if (!pc->is_bridge) {
843         if (pc->subsystem_vendor_id || pc->subsystem_id) {
844             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
845                          pc->subsystem_vendor_id);
846             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
847                          pc->subsystem_id);
848         } else {
849             pci_set_default_subsystem_id(pci_dev);
850         }
851     } else {
852         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
853         assert(!pc->subsystem_vendor_id);
854         assert(!pc->subsystem_id);
855     }
856     pci_init_cmask(pci_dev);
857     pci_init_wmask(pci_dev);
858     pci_init_w1cmask(pci_dev);
859     if (pc->is_bridge) {
860         pci_init_mask_bridge(pci_dev);
861     }
862     if (pci_init_multifunction(bus, pci_dev)) {
863         pci_config_free(pci_dev);
864         return NULL;
865     }
866 
867     if (!config_read)
868         config_read = pci_default_read_config;
869     if (!config_write)
870         config_write = pci_default_write_config;
871     pci_dev->config_read = config_read;
872     pci_dev->config_write = config_write;
873     bus->devices[devfn] = pci_dev;
874     pci_dev->version_id = 2; /* Current pci device vmstate version */
875     return pci_dev;
876 }
877 
878 static void do_pci_unregister_device(PCIDevice *pci_dev)
879 {
880     pci_dev->bus->devices[pci_dev->devfn] = NULL;
881     pci_config_free(pci_dev);
882 
883     address_space_destroy(&pci_dev->bus_master_as);
884     memory_region_destroy(&pci_dev->bus_master_enable_region);
885 }
886 
887 static void pci_unregister_io_regions(PCIDevice *pci_dev)
888 {
889     PCIIORegion *r;
890     int i;
891 
892     for(i = 0; i < PCI_NUM_REGIONS; i++) {
893         r = &pci_dev->io_regions[i];
894         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
895             continue;
896         memory_region_del_subregion(r->address_space, r->memory);
897     }
898 
899     pci_unregister_vga(pci_dev);
900 }
901 
902 static int pci_unregister_device(DeviceState *dev)
903 {
904     PCIDevice *pci_dev = PCI_DEVICE(dev);
905     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
906 
907     pci_unregister_io_regions(pci_dev);
908     pci_del_option_rom(pci_dev);
909 
910     if (pc->exit) {
911         pc->exit(pci_dev);
912     }
913 
914     do_pci_unregister_device(pci_dev);
915     return 0;
916 }
917 
918 void pci_register_bar(PCIDevice *pci_dev, int region_num,
919                       uint8_t type, MemoryRegion *memory)
920 {
921     PCIIORegion *r;
922     uint32_t addr;
923     uint64_t wmask;
924     pcibus_t size = memory_region_size(memory);
925 
926     assert(region_num >= 0);
927     assert(region_num < PCI_NUM_REGIONS);
928     if (size & (size-1)) {
929         fprintf(stderr, "ERROR: PCI region size must be pow2 "
930                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
931         exit(1);
932     }
933 
934     r = &pci_dev->io_regions[region_num];
935     r->addr = PCI_BAR_UNMAPPED;
936     r->size = size;
937     r->type = type;
938     r->memory = NULL;
939 
940     wmask = ~(size - 1);
941     addr = pci_bar(pci_dev, region_num);
942     if (region_num == PCI_ROM_SLOT) {
943         /* ROM enable bit is writable */
944         wmask |= PCI_ROM_ADDRESS_ENABLE;
945     }
946     pci_set_long(pci_dev->config + addr, type);
947     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
948         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
949         pci_set_quad(pci_dev->wmask + addr, wmask);
950         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
951     } else {
952         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
953         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
954     }
955     pci_dev->io_regions[region_num].memory = memory;
956     pci_dev->io_regions[region_num].address_space
957         = type & PCI_BASE_ADDRESS_SPACE_IO
958         ? pci_dev->bus->address_space_io
959         : pci_dev->bus->address_space_mem;
960 }
961 
962 static void pci_update_vga(PCIDevice *pci_dev)
963 {
964     uint16_t cmd;
965 
966     if (!pci_dev->has_vga) {
967         return;
968     }
969 
970     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
971 
972     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
973                               cmd & PCI_COMMAND_MEMORY);
974     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
975                               cmd & PCI_COMMAND_IO);
976     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
977                               cmd & PCI_COMMAND_IO);
978 }
979 
980 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
981                       MemoryRegion *io_lo, MemoryRegion *io_hi)
982 {
983     assert(!pci_dev->has_vga);
984 
985     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
986     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
987     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
988                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
989 
990     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
991     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
992     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
993                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
994 
995     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
996     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
997     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
998                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
999     pci_dev->has_vga = true;
1000 
1001     pci_update_vga(pci_dev);
1002 }
1003 
1004 void pci_unregister_vga(PCIDevice *pci_dev)
1005 {
1006     if (!pci_dev->has_vga) {
1007         return;
1008     }
1009 
1010     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1011                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1012     memory_region_del_subregion(pci_dev->bus->address_space_io,
1013                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1014     memory_region_del_subregion(pci_dev->bus->address_space_io,
1015                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1016     pci_dev->has_vga = false;
1017 }
1018 
1019 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1020 {
1021     return pci_dev->io_regions[region_num].addr;
1022 }
1023 
1024 static pcibus_t pci_bar_address(PCIDevice *d,
1025 				int reg, uint8_t type, pcibus_t size)
1026 {
1027     pcibus_t new_addr, last_addr;
1028     int bar = pci_bar(d, reg);
1029     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1030 
1031     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1032         if (!(cmd & PCI_COMMAND_IO)) {
1033             return PCI_BAR_UNMAPPED;
1034         }
1035         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1036         last_addr = new_addr + size - 1;
1037         /* Check if 32 bit BAR wraps around explicitly.
1038          * TODO: make priorities correct and remove this work around.
1039          */
1040         if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
1041             return PCI_BAR_UNMAPPED;
1042         }
1043         return new_addr;
1044     }
1045 
1046     if (!(cmd & PCI_COMMAND_MEMORY)) {
1047         return PCI_BAR_UNMAPPED;
1048     }
1049     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1050         new_addr = pci_get_quad(d->config + bar);
1051     } else {
1052         new_addr = pci_get_long(d->config + bar);
1053     }
1054     /* the ROM slot has a specific enable bit */
1055     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1056         return PCI_BAR_UNMAPPED;
1057     }
1058     new_addr &= ~(size - 1);
1059     last_addr = new_addr + size - 1;
1060     /* NOTE: we do not support wrapping */
1061     /* XXX: as we cannot support really dynamic
1062        mappings, we handle specific values as invalid
1063        mappings. */
1064     if (last_addr <= new_addr || new_addr == 0 ||
1065         last_addr == PCI_BAR_UNMAPPED) {
1066         return PCI_BAR_UNMAPPED;
1067     }
1068 
1069     /* Now pcibus_t is 64bit.
1070      * Check if 32 bit BAR wraps around explicitly.
1071      * Without this, PC ide doesn't work well.
1072      * TODO: remove this work around.
1073      */
1074     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1075         return PCI_BAR_UNMAPPED;
1076     }
1077 
1078     /*
1079      * OS is allowed to set BAR beyond its addressable
1080      * bits. For example, 32 bit OS can set 64bit bar
1081      * to >4G. Check it. TODO: we might need to support
1082      * it in the future for e.g. PAE.
1083      */
1084     if (last_addr >= HWADDR_MAX) {
1085         return PCI_BAR_UNMAPPED;
1086     }
1087 
1088     return new_addr;
1089 }
1090 
1091 static void pci_update_mappings(PCIDevice *d)
1092 {
1093     PCIIORegion *r;
1094     int i;
1095     pcibus_t new_addr;
1096 
1097     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1098         r = &d->io_regions[i];
1099 
1100         /* this region isn't registered */
1101         if (!r->size)
1102             continue;
1103 
1104         new_addr = pci_bar_address(d, i, r->type, r->size);
1105 
1106         /* This bar isn't changed */
1107         if (new_addr == r->addr)
1108             continue;
1109 
1110         /* now do the real mapping */
1111         if (r->addr != PCI_BAR_UNMAPPED) {
1112             memory_region_del_subregion(r->address_space, r->memory);
1113         }
1114         r->addr = new_addr;
1115         if (r->addr != PCI_BAR_UNMAPPED) {
1116             memory_region_add_subregion_overlap(r->address_space,
1117                                                 r->addr, r->memory, 1);
1118         }
1119     }
1120 
1121     pci_update_vga(d);
1122 }
1123 
1124 static inline int pci_irq_disabled(PCIDevice *d)
1125 {
1126     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1127 }
1128 
1129 /* Called after interrupt disabled field update in config space,
1130  * assert/deassert interrupts if necessary.
1131  * Gets original interrupt disable bit value (before update). */
1132 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1133 {
1134     int i, disabled = pci_irq_disabled(d);
1135     if (disabled == was_irq_disabled)
1136         return;
1137     for (i = 0; i < PCI_NUM_PINS; ++i) {
1138         int state = pci_irq_state(d, i);
1139         pci_change_irq_level(d, i, disabled ? -state : state);
1140     }
1141 }
1142 
1143 uint32_t pci_default_read_config(PCIDevice *d,
1144                                  uint32_t address, int len)
1145 {
1146     uint32_t val = 0;
1147 
1148     memcpy(&val, d->config + address, len);
1149     return le32_to_cpu(val);
1150 }
1151 
1152 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
1153 {
1154     int i, was_irq_disabled = pci_irq_disabled(d);
1155 
1156     for (i = 0; i < l; val >>= 8, ++i) {
1157         uint8_t wmask = d->wmask[addr + i];
1158         uint8_t w1cmask = d->w1cmask[addr + i];
1159         assert(!(wmask & w1cmask));
1160         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1161         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1162     }
1163     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1164         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1165         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1166         range_covers_byte(addr, l, PCI_COMMAND))
1167         pci_update_mappings(d);
1168 
1169     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1170         pci_update_irq_disabled(d, was_irq_disabled);
1171         memory_region_set_enabled(&d->bus_master_enable_region,
1172                                   pci_get_word(d->config + PCI_COMMAND)
1173                                     & PCI_COMMAND_MASTER);
1174     }
1175 
1176     msi_write_config(d, addr, val, l);
1177     msix_write_config(d, addr, val, l);
1178 }
1179 
1180 /***********************************************************/
1181 /* generic PCI irq support */
1182 
1183 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1184 static void pci_irq_handler(void *opaque, int irq_num, int level)
1185 {
1186     PCIDevice *pci_dev = opaque;
1187     int change;
1188 
1189     change = level - pci_irq_state(pci_dev, irq_num);
1190     if (!change)
1191         return;
1192 
1193     pci_set_irq_state(pci_dev, irq_num, level);
1194     pci_update_irq_status(pci_dev);
1195     if (pci_irq_disabled(pci_dev))
1196         return;
1197     pci_change_irq_level(pci_dev, irq_num, change);
1198 }
1199 
1200 static inline int pci_intx(PCIDevice *pci_dev)
1201 {
1202     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1203 }
1204 
1205 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1206 {
1207     int intx = pci_intx(pci_dev);
1208 
1209     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1210 }
1211 
1212 void pci_set_irq(PCIDevice *pci_dev, int level)
1213 {
1214     int intx = pci_intx(pci_dev);
1215     pci_irq_handler(pci_dev, intx, level);
1216 }
1217 
1218 /* Special hooks used by device assignment */
1219 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1220 {
1221     assert(pci_bus_is_root(bus));
1222     bus->route_intx_to_irq = route_intx_to_irq;
1223 }
1224 
1225 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1226 {
1227     PCIBus *bus;
1228 
1229     do {
1230          bus = dev->bus;
1231          pin = bus->map_irq(dev, pin);
1232          dev = bus->parent_dev;
1233     } while (dev);
1234 
1235     if (!bus->route_intx_to_irq) {
1236         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1237                      object_get_typename(OBJECT(bus->qbus.parent)));
1238         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1239     }
1240 
1241     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1242 }
1243 
1244 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1245 {
1246     return old->mode != new->mode || old->irq != new->irq;
1247 }
1248 
1249 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1250 {
1251     PCIDevice *dev;
1252     PCIBus *sec;
1253     int i;
1254 
1255     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1256         dev = bus->devices[i];
1257         if (dev && dev->intx_routing_notifier) {
1258             dev->intx_routing_notifier(dev);
1259         }
1260     }
1261 
1262     QLIST_FOREACH(sec, &bus->child, sibling) {
1263         pci_bus_fire_intx_routing_notifier(sec);
1264     }
1265 }
1266 
1267 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1268                                           PCIINTxRoutingNotifier notifier)
1269 {
1270     dev->intx_routing_notifier = notifier;
1271 }
1272 
1273 /*
1274  * PCI-to-PCI bridge specification
1275  * 9.1: Interrupt routing. Table 9-1
1276  *
1277  * the PCI Express Base Specification, Revision 2.1
1278  * 2.2.8.1: INTx interrutp signaling - Rules
1279  *          the Implementation Note
1280  *          Table 2-20
1281  */
1282 /*
1283  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1284  * 0-origin unlike PCI interrupt pin register.
1285  */
1286 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1287 {
1288     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1289 }
1290 
1291 /***********************************************************/
1292 /* monitor info on PCI */
1293 
1294 typedef struct {
1295     uint16_t class;
1296     const char *desc;
1297     const char *fw_name;
1298     uint16_t fw_ign_bits;
1299 } pci_class_desc;
1300 
1301 static const pci_class_desc pci_class_descriptions[] =
1302 {
1303     { 0x0001, "VGA controller", "display"},
1304     { 0x0100, "SCSI controller", "scsi"},
1305     { 0x0101, "IDE controller", "ide"},
1306     { 0x0102, "Floppy controller", "fdc"},
1307     { 0x0103, "IPI controller", "ipi"},
1308     { 0x0104, "RAID controller", "raid"},
1309     { 0x0106, "SATA controller"},
1310     { 0x0107, "SAS controller"},
1311     { 0x0180, "Storage controller"},
1312     { 0x0200, "Ethernet controller", "ethernet"},
1313     { 0x0201, "Token Ring controller", "token-ring"},
1314     { 0x0202, "FDDI controller", "fddi"},
1315     { 0x0203, "ATM controller", "atm"},
1316     { 0x0280, "Network controller"},
1317     { 0x0300, "VGA controller", "display", 0x00ff},
1318     { 0x0301, "XGA controller"},
1319     { 0x0302, "3D controller"},
1320     { 0x0380, "Display controller"},
1321     { 0x0400, "Video controller", "video"},
1322     { 0x0401, "Audio controller", "sound"},
1323     { 0x0402, "Phone"},
1324     { 0x0403, "Audio controller", "sound"},
1325     { 0x0480, "Multimedia controller"},
1326     { 0x0500, "RAM controller", "memory"},
1327     { 0x0501, "Flash controller", "flash"},
1328     { 0x0580, "Memory controller"},
1329     { 0x0600, "Host bridge", "host"},
1330     { 0x0601, "ISA bridge", "isa"},
1331     { 0x0602, "EISA bridge", "eisa"},
1332     { 0x0603, "MC bridge", "mca"},
1333     { 0x0604, "PCI bridge", "pci-bridge"},
1334     { 0x0605, "PCMCIA bridge", "pcmcia"},
1335     { 0x0606, "NUBUS bridge", "nubus"},
1336     { 0x0607, "CARDBUS bridge", "cardbus"},
1337     { 0x0608, "RACEWAY bridge"},
1338     { 0x0680, "Bridge"},
1339     { 0x0700, "Serial port", "serial"},
1340     { 0x0701, "Parallel port", "parallel"},
1341     { 0x0800, "Interrupt controller", "interrupt-controller"},
1342     { 0x0801, "DMA controller", "dma-controller"},
1343     { 0x0802, "Timer", "timer"},
1344     { 0x0803, "RTC", "rtc"},
1345     { 0x0900, "Keyboard", "keyboard"},
1346     { 0x0901, "Pen", "pen"},
1347     { 0x0902, "Mouse", "mouse"},
1348     { 0x0A00, "Dock station", "dock", 0x00ff},
1349     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1350     { 0x0c00, "Fireware contorller", "fireware"},
1351     { 0x0c01, "Access bus controller", "access-bus"},
1352     { 0x0c02, "SSA controller", "ssa"},
1353     { 0x0c03, "USB controller", "usb"},
1354     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1355     { 0x0c05, "SMBus"},
1356     { 0, NULL}
1357 };
1358 
1359 static void pci_for_each_device_under_bus(PCIBus *bus,
1360                                           void (*fn)(PCIBus *b, PCIDevice *d,
1361                                                      void *opaque),
1362                                           void *opaque)
1363 {
1364     PCIDevice *d;
1365     int devfn;
1366 
1367     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1368         d = bus->devices[devfn];
1369         if (d) {
1370             fn(bus, d, opaque);
1371         }
1372     }
1373 }
1374 
1375 void pci_for_each_device(PCIBus *bus, int bus_num,
1376                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1377                          void *opaque)
1378 {
1379     bus = pci_find_bus_nr(bus, bus_num);
1380 
1381     if (bus) {
1382         pci_for_each_device_under_bus(bus, fn, opaque);
1383     }
1384 }
1385 
1386 static const pci_class_desc *get_class_desc(int class)
1387 {
1388     const pci_class_desc *desc;
1389 
1390     desc = pci_class_descriptions;
1391     while (desc->desc && class != desc->class) {
1392         desc++;
1393     }
1394 
1395     return desc;
1396 }
1397 
1398 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1399 
1400 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1401 {
1402     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1403     int i;
1404 
1405     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1406         const PCIIORegion *r = &dev->io_regions[i];
1407         PciMemoryRegionList *region;
1408 
1409         if (!r->size) {
1410             continue;
1411         }
1412 
1413         region = g_malloc0(sizeof(*region));
1414         region->value = g_malloc0(sizeof(*region->value));
1415 
1416         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1417             region->value->type = g_strdup("io");
1418         } else {
1419             region->value->type = g_strdup("memory");
1420             region->value->has_prefetch = true;
1421             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1422             region->value->has_mem_type_64 = true;
1423             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1424         }
1425 
1426         region->value->bar = i;
1427         region->value->address = r->addr;
1428         region->value->size = r->size;
1429 
1430         /* XXX: waiting for the qapi to support GSList */
1431         if (!cur_item) {
1432             head = cur_item = region;
1433         } else {
1434             cur_item->next = region;
1435             cur_item = region;
1436         }
1437     }
1438 
1439     return head;
1440 }
1441 
1442 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1443                                            int bus_num)
1444 {
1445     PciBridgeInfo *info;
1446 
1447     info = g_malloc0(sizeof(*info));
1448 
1449     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1450     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1451     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1452 
1453     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1454     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1455     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1456 
1457     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1458     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1459     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1460 
1461     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1462     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1463     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1464 
1465     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1466         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1467         if (child_bus) {
1468             info->has_devices = true;
1469             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1470         }
1471     }
1472 
1473     return info;
1474 }
1475 
1476 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1477                                            int bus_num)
1478 {
1479     const pci_class_desc *desc;
1480     PciDeviceInfo *info;
1481     uint8_t type;
1482     int class;
1483 
1484     info = g_malloc0(sizeof(*info));
1485     info->bus = bus_num;
1486     info->slot = PCI_SLOT(dev->devfn);
1487     info->function = PCI_FUNC(dev->devfn);
1488 
1489     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1490     info->class_info.q_class = class;
1491     desc = get_class_desc(class);
1492     if (desc->desc) {
1493         info->class_info.has_desc = true;
1494         info->class_info.desc = g_strdup(desc->desc);
1495     }
1496 
1497     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1498     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1499     info->regions = qmp_query_pci_regions(dev);
1500     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1501 
1502     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1503         info->has_irq = true;
1504         info->irq = dev->config[PCI_INTERRUPT_LINE];
1505     }
1506 
1507     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1508     if (type == PCI_HEADER_TYPE_BRIDGE) {
1509         info->has_pci_bridge = true;
1510         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1511     }
1512 
1513     return info;
1514 }
1515 
1516 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1517 {
1518     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1519     PCIDevice *dev;
1520     int devfn;
1521 
1522     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1523         dev = bus->devices[devfn];
1524         if (dev) {
1525             info = g_malloc0(sizeof(*info));
1526             info->value = qmp_query_pci_device(dev, bus, bus_num);
1527 
1528             /* XXX: waiting for the qapi to support GSList */
1529             if (!cur_item) {
1530                 head = cur_item = info;
1531             } else {
1532                 cur_item->next = info;
1533                 cur_item = info;
1534             }
1535         }
1536     }
1537 
1538     return head;
1539 }
1540 
1541 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1542 {
1543     PciInfo *info = NULL;
1544 
1545     bus = pci_find_bus_nr(bus, bus_num);
1546     if (bus) {
1547         info = g_malloc0(sizeof(*info));
1548         info->bus = bus_num;
1549         info->devices = qmp_query_pci_devices(bus, bus_num);
1550     }
1551 
1552     return info;
1553 }
1554 
1555 PciInfoList *qmp_query_pci(Error **errp)
1556 {
1557     PciInfoList *info, *head = NULL, *cur_item = NULL;
1558     PCIHostState *host_bridge;
1559 
1560     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1561         info = g_malloc0(sizeof(*info));
1562         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1563 
1564         /* XXX: waiting for the qapi to support GSList */
1565         if (!cur_item) {
1566             head = cur_item = info;
1567         } else {
1568             cur_item->next = info;
1569             cur_item = info;
1570         }
1571     }
1572 
1573     return head;
1574 }
1575 
1576 static const char * const pci_nic_models[] = {
1577     "ne2k_pci",
1578     "i82551",
1579     "i82557b",
1580     "i82559er",
1581     "rtl8139",
1582     "e1000",
1583     "pcnet",
1584     "virtio",
1585     NULL
1586 };
1587 
1588 static const char * const pci_nic_names[] = {
1589     "ne2k_pci",
1590     "i82551",
1591     "i82557b",
1592     "i82559er",
1593     "rtl8139",
1594     "e1000",
1595     "pcnet",
1596     "virtio-net-pci",
1597     NULL
1598 };
1599 
1600 /* Initialize a PCI NIC.  */
1601 /* FIXME callers should check for failure, but don't */
1602 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1603                         const char *default_model,
1604                         const char *default_devaddr)
1605 {
1606     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1607     PCIBus *bus;
1608     int devfn;
1609     PCIDevice *pci_dev;
1610     DeviceState *dev;
1611     int i;
1612 
1613     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1614     if (i < 0)
1615         return NULL;
1616 
1617     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1618     if (!bus) {
1619         error_report("Invalid PCI device address %s for device %s",
1620                      devaddr, pci_nic_names[i]);
1621         return NULL;
1622     }
1623 
1624     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1625     dev = &pci_dev->qdev;
1626     qdev_set_nic_properties(dev, nd);
1627     if (qdev_init(dev) < 0)
1628         return NULL;
1629     return pci_dev;
1630 }
1631 
1632 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1633                                const char *default_model,
1634                                const char *default_devaddr)
1635 {
1636     PCIDevice *res;
1637 
1638     if (qemu_show_nic_models(nd->model, pci_nic_models))
1639         exit(0);
1640 
1641     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1642     if (!res)
1643         exit(1);
1644     return res;
1645 }
1646 
1647 PCIDevice *pci_vga_init(PCIBus *bus)
1648 {
1649     switch (vga_interface_type) {
1650     case VGA_CIRRUS:
1651         return pci_create_simple(bus, -1, "cirrus-vga");
1652     case VGA_QXL:
1653         return pci_create_simple(bus, -1, "qxl-vga");
1654     case VGA_STD:
1655         return pci_create_simple(bus, -1, "VGA");
1656     case VGA_VMWARE:
1657         return pci_create_simple(bus, -1, "vmware-svga");
1658     case VGA_NONE:
1659     default: /* Other non-PCI types. Checking for unsupported types is already
1660                 done in vl.c. */
1661         return NULL;
1662     }
1663 }
1664 
1665 /* Whether a given bus number is in range of the secondary
1666  * bus of the given bridge device. */
1667 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1668 {
1669     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1670              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1671         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1672         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1673 }
1674 
1675 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1676 {
1677     PCIBus *sec;
1678 
1679     if (!bus) {
1680         return NULL;
1681     }
1682 
1683     if (pci_bus_num(bus) == bus_num) {
1684         return bus;
1685     }
1686 
1687     /* Consider all bus numbers in range for the host pci bridge. */
1688     if (!pci_bus_is_root(bus) &&
1689         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1690         return NULL;
1691     }
1692 
1693     /* try child bus */
1694     for (; bus; bus = sec) {
1695         QLIST_FOREACH(sec, &bus->child, sibling) {
1696             assert(!pci_bus_is_root(sec));
1697             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1698                 return sec;
1699             }
1700             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1701                 break;
1702             }
1703         }
1704     }
1705 
1706     return NULL;
1707 }
1708 
1709 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1710 {
1711     bus = pci_find_bus_nr(bus, bus_num);
1712 
1713     if (!bus)
1714         return NULL;
1715 
1716     return bus->devices[devfn];
1717 }
1718 
1719 static int pci_qdev_init(DeviceState *qdev)
1720 {
1721     PCIDevice *pci_dev = (PCIDevice *)qdev;
1722     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1723     PCIBus *bus;
1724     int rc;
1725     bool is_default_rom;
1726 
1727     /* initialize cap_present for pci_is_express() and pci_config_size() */
1728     if (pc->is_express) {
1729         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1730     }
1731 
1732     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1733     pci_dev = do_pci_register_device(pci_dev, bus,
1734                                      object_get_typename(OBJECT(qdev)),
1735                                      pci_dev->devfn);
1736     if (pci_dev == NULL)
1737         return -1;
1738     if (qdev->hotplugged && pc->no_hotplug) {
1739         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(pci_dev)));
1740         do_pci_unregister_device(pci_dev);
1741         return -1;
1742     }
1743     if (pc->init) {
1744         rc = pc->init(pci_dev);
1745         if (rc != 0) {
1746             do_pci_unregister_device(pci_dev);
1747             return rc;
1748         }
1749     }
1750 
1751     /* rom loading */
1752     is_default_rom = false;
1753     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1754         pci_dev->romfile = g_strdup(pc->romfile);
1755         is_default_rom = true;
1756     }
1757     pci_add_option_rom(pci_dev, is_default_rom);
1758 
1759     if (bus->hotplug) {
1760         /* Let buses differentiate between hotplug and when device is
1761          * enabled during qemu machine creation. */
1762         rc = bus->hotplug(bus->hotplug_qdev, pci_dev,
1763                           qdev->hotplugged ? PCI_HOTPLUG_ENABLED:
1764                           PCI_COLDPLUG_ENABLED);
1765         if (rc != 0) {
1766             int r = pci_unregister_device(&pci_dev->qdev);
1767             assert(!r);
1768             return rc;
1769         }
1770     }
1771     return 0;
1772 }
1773 
1774 static int pci_unplug_device(DeviceState *qdev)
1775 {
1776     PCIDevice *dev = PCI_DEVICE(qdev);
1777     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1778 
1779     if (pc->no_hotplug) {
1780         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(dev)));
1781         return -1;
1782     }
1783     return dev->bus->hotplug(dev->bus->hotplug_qdev, dev,
1784                              PCI_HOTPLUG_DISABLED);
1785 }
1786 
1787 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1788                                     const char *name)
1789 {
1790     DeviceState *dev;
1791 
1792     dev = qdev_create(&bus->qbus, name);
1793     qdev_prop_set_int32(dev, "addr", devfn);
1794     qdev_prop_set_bit(dev, "multifunction", multifunction);
1795     return PCI_DEVICE(dev);
1796 }
1797 
1798 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1799                                            bool multifunction,
1800                                            const char *name)
1801 {
1802     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1803     qdev_init_nofail(&dev->qdev);
1804     return dev;
1805 }
1806 
1807 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1808 {
1809     return pci_create_multifunction(bus, devfn, false, name);
1810 }
1811 
1812 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1813 {
1814     return pci_create_simple_multifunction(bus, devfn, false, name);
1815 }
1816 
1817 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1818 {
1819     int offset = PCI_CONFIG_HEADER_SIZE;
1820     int i;
1821     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1822         if (pdev->used[i])
1823             offset = i + 1;
1824         else if (i - offset + 1 == size)
1825             return offset;
1826     }
1827     return 0;
1828 }
1829 
1830 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1831                                         uint8_t *prev_p)
1832 {
1833     uint8_t next, prev;
1834 
1835     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1836         return 0;
1837 
1838     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1839          prev = next + PCI_CAP_LIST_NEXT)
1840         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1841             break;
1842 
1843     if (prev_p)
1844         *prev_p = prev;
1845     return next;
1846 }
1847 
1848 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1849 {
1850     uint8_t next, prev, found = 0;
1851 
1852     if (!(pdev->used[offset])) {
1853         return 0;
1854     }
1855 
1856     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1857 
1858     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1859          prev = next + PCI_CAP_LIST_NEXT) {
1860         if (next <= offset && next > found) {
1861             found = next;
1862         }
1863     }
1864     return found;
1865 }
1866 
1867 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1868    This is needed for an option rom which is used for more than one device. */
1869 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1870 {
1871     uint16_t vendor_id;
1872     uint16_t device_id;
1873     uint16_t rom_vendor_id;
1874     uint16_t rom_device_id;
1875     uint16_t rom_magic;
1876     uint16_t pcir_offset;
1877     uint8_t checksum;
1878 
1879     /* Words in rom data are little endian (like in PCI configuration),
1880        so they can be read / written with pci_get_word / pci_set_word. */
1881 
1882     /* Only a valid rom will be patched. */
1883     rom_magic = pci_get_word(ptr);
1884     if (rom_magic != 0xaa55) {
1885         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1886         return;
1887     }
1888     pcir_offset = pci_get_word(ptr + 0x18);
1889     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1890         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1891         return;
1892     }
1893 
1894     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1895     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1896     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1897     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1898 
1899     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1900                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1901 
1902     checksum = ptr[6];
1903 
1904     if (vendor_id != rom_vendor_id) {
1905         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1906         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1907         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1908         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1909         ptr[6] = checksum;
1910         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1911     }
1912 
1913     if (device_id != rom_device_id) {
1914         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1915         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1916         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1917         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1918         ptr[6] = checksum;
1919         pci_set_word(ptr + pcir_offset + 6, device_id);
1920     }
1921 }
1922 
1923 /* Add an option rom for the device */
1924 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1925 {
1926     int size;
1927     char *path;
1928     void *ptr;
1929     char name[32];
1930     const VMStateDescription *vmsd;
1931 
1932     if (!pdev->romfile)
1933         return 0;
1934     if (strlen(pdev->romfile) == 0)
1935         return 0;
1936 
1937     if (!pdev->rom_bar) {
1938         /*
1939          * Load rom via fw_cfg instead of creating a rom bar,
1940          * for 0.11 compatibility.
1941          */
1942         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1943         if (class == 0x0300) {
1944             rom_add_vga(pdev->romfile);
1945         } else {
1946             rom_add_option(pdev->romfile, -1);
1947         }
1948         return 0;
1949     }
1950 
1951     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1952     if (path == NULL) {
1953         path = g_strdup(pdev->romfile);
1954     }
1955 
1956     size = get_image_size(path);
1957     if (size < 0) {
1958         error_report("%s: failed to find romfile \"%s\"",
1959                      __func__, pdev->romfile);
1960         g_free(path);
1961         return -1;
1962     } else if (size == 0) {
1963         error_report("%s: ignoring empty romfile \"%s\"",
1964                      __func__, pdev->romfile);
1965         g_free(path);
1966         return -1;
1967     }
1968     if (size & (size - 1)) {
1969         size = 1 << qemu_fls(size);
1970     }
1971 
1972     vmsd = qdev_get_vmsd(DEVICE(pdev));
1973 
1974     if (vmsd) {
1975         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1976     } else {
1977         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1978     }
1979     pdev->has_rom = true;
1980     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size);
1981     vmstate_register_ram(&pdev->rom, &pdev->qdev);
1982     ptr = memory_region_get_ram_ptr(&pdev->rom);
1983     load_image(path, ptr);
1984     g_free(path);
1985 
1986     if (is_default_rom) {
1987         /* Only the default rom images will be patched (if needed). */
1988         pci_patch_ids(pdev, ptr, size);
1989     }
1990 
1991     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
1992 
1993     return 0;
1994 }
1995 
1996 static void pci_del_option_rom(PCIDevice *pdev)
1997 {
1998     if (!pdev->has_rom)
1999         return;
2000 
2001     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2002     memory_region_destroy(&pdev->rom);
2003     pdev->has_rom = false;
2004 }
2005 
2006 /*
2007  * if !offset
2008  * Reserve space and add capability to the linked list in pci config space
2009  *
2010  * if offset = 0,
2011  * Find and reserve space and add capability to the linked list
2012  * in pci config space */
2013 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2014                        uint8_t offset, uint8_t size)
2015 {
2016     uint8_t *config;
2017     int i, overlapping_cap;
2018 
2019     if (!offset) {
2020         offset = pci_find_space(pdev, size);
2021         if (!offset) {
2022             return -ENOSPC;
2023         }
2024     } else {
2025         /* Verify that capabilities don't overlap.  Note: device assignment
2026          * depends on this check to verify that the device is not broken.
2027          * Should never trigger for emulated devices, but it's helpful
2028          * for debugging these. */
2029         for (i = offset; i < offset + size; i++) {
2030             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2031             if (overlapping_cap) {
2032                 fprintf(stderr, "ERROR: %s:%02x:%02x.%x "
2033                         "Attempt to add PCI capability %x at offset "
2034                         "%x overlaps existing capability %x at offset %x\n",
2035                         pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2036                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2037                         cap_id, offset, overlapping_cap, i);
2038                 return -EINVAL;
2039             }
2040         }
2041     }
2042 
2043     config = pdev->config + offset;
2044     config[PCI_CAP_LIST_ID] = cap_id;
2045     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2046     pdev->config[PCI_CAPABILITY_LIST] = offset;
2047     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2048     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2049     /* Make capability read-only by default */
2050     memset(pdev->wmask + offset, 0, size);
2051     /* Check capability by default */
2052     memset(pdev->cmask + offset, 0xFF, size);
2053     return offset;
2054 }
2055 
2056 /* Unlink capability from the pci config space. */
2057 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2058 {
2059     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2060     if (!offset)
2061         return;
2062     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2063     /* Make capability writable again */
2064     memset(pdev->wmask + offset, 0xff, size);
2065     memset(pdev->w1cmask + offset, 0, size);
2066     /* Clear cmask as device-specific registers can't be checked */
2067     memset(pdev->cmask + offset, 0, size);
2068     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2069 
2070     if (!pdev->config[PCI_CAPABILITY_LIST])
2071         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2072 }
2073 
2074 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2075 {
2076     return pci_find_capability_list(pdev, cap_id, NULL);
2077 }
2078 
2079 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2080 {
2081     PCIDevice *d = (PCIDevice *)dev;
2082     const pci_class_desc *desc;
2083     char ctxt[64];
2084     PCIIORegion *r;
2085     int i, class;
2086 
2087     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2088     desc = pci_class_descriptions;
2089     while (desc->desc && class != desc->class)
2090         desc++;
2091     if (desc->desc) {
2092         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2093     } else {
2094         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2095     }
2096 
2097     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2098                    "pci id %04x:%04x (sub %04x:%04x)\n",
2099                    indent, "", ctxt, pci_bus_num(d->bus),
2100                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2101                    pci_get_word(d->config + PCI_VENDOR_ID),
2102                    pci_get_word(d->config + PCI_DEVICE_ID),
2103                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2104                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2105     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2106         r = &d->io_regions[i];
2107         if (!r->size)
2108             continue;
2109         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2110                        " [0x%"FMT_PCIBUS"]\n",
2111                        indent, "",
2112                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2113                        r->addr, r->addr + r->size - 1);
2114     }
2115 }
2116 
2117 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2118 {
2119     PCIDevice *d = (PCIDevice *)dev;
2120     const char *name = NULL;
2121     const pci_class_desc *desc =  pci_class_descriptions;
2122     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2123 
2124     while (desc->desc &&
2125           (class & ~desc->fw_ign_bits) !=
2126           (desc->class & ~desc->fw_ign_bits)) {
2127         desc++;
2128     }
2129 
2130     if (desc->desc) {
2131         name = desc->fw_name;
2132     }
2133 
2134     if (name) {
2135         pstrcpy(buf, len, name);
2136     } else {
2137         snprintf(buf, len, "pci%04x,%04x",
2138                  pci_get_word(d->config + PCI_VENDOR_ID),
2139                  pci_get_word(d->config + PCI_DEVICE_ID));
2140     }
2141 
2142     return buf;
2143 }
2144 
2145 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2146 {
2147     PCIDevice *d = (PCIDevice *)dev;
2148     char path[50], name[33];
2149     int off;
2150 
2151     off = snprintf(path, sizeof(path), "%s@%x",
2152                    pci_dev_fw_name(dev, name, sizeof name),
2153                    PCI_SLOT(d->devfn));
2154     if (PCI_FUNC(d->devfn))
2155         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2156     return g_strdup(path);
2157 }
2158 
2159 static char *pcibus_get_dev_path(DeviceState *dev)
2160 {
2161     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2162     PCIDevice *t;
2163     int slot_depth;
2164     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2165      * 00 is added here to make this format compatible with
2166      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2167      * Slot.Function list specifies the slot and function numbers for all
2168      * devices on the path from root to the specific device. */
2169     const char *root_bus_path;
2170     int root_bus_len;
2171     char slot[] = ":SS.F";
2172     int slot_len = sizeof slot - 1 /* For '\0' */;
2173     int path_len;
2174     char *path, *p;
2175     int s;
2176 
2177     root_bus_path = pci_root_bus_path(d);
2178     root_bus_len = strlen(root_bus_path);
2179 
2180     /* Calculate # of slots on path between device and root. */;
2181     slot_depth = 0;
2182     for (t = d; t; t = t->bus->parent_dev) {
2183         ++slot_depth;
2184     }
2185 
2186     path_len = root_bus_len + slot_len * slot_depth;
2187 
2188     /* Allocate memory, fill in the terminating null byte. */
2189     path = g_malloc(path_len + 1 /* For '\0' */);
2190     path[path_len] = '\0';
2191 
2192     memcpy(path, root_bus_path, root_bus_len);
2193 
2194     /* Fill in slot numbers. We walk up from device to root, so need to print
2195      * them in the reverse order, last to first. */
2196     p = path + path_len;
2197     for (t = d; t; t = t->bus->parent_dev) {
2198         p -= slot_len;
2199         s = snprintf(slot, sizeof slot, ":%02x.%x",
2200                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2201         assert(s == slot_len);
2202         memcpy(p, slot, slot_len);
2203     }
2204 
2205     return path;
2206 }
2207 
2208 static int pci_qdev_find_recursive(PCIBus *bus,
2209                                    const char *id, PCIDevice **pdev)
2210 {
2211     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2212     if (!qdev) {
2213         return -ENODEV;
2214     }
2215 
2216     /* roughly check if given qdev is pci device */
2217     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2218         *pdev = PCI_DEVICE(qdev);
2219         return 0;
2220     }
2221     return -EINVAL;
2222 }
2223 
2224 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2225 {
2226     PCIHostState *host_bridge;
2227     int rc = -ENODEV;
2228 
2229     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2230         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2231         if (!tmp) {
2232             rc = 0;
2233             break;
2234         }
2235         if (tmp != -ENODEV) {
2236             rc = tmp;
2237         }
2238     }
2239 
2240     return rc;
2241 }
2242 
2243 MemoryRegion *pci_address_space(PCIDevice *dev)
2244 {
2245     return dev->bus->address_space_mem;
2246 }
2247 
2248 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2249 {
2250     return dev->bus->address_space_io;
2251 }
2252 
2253 static void pci_device_class_init(ObjectClass *klass, void *data)
2254 {
2255     DeviceClass *k = DEVICE_CLASS(klass);
2256     k->init = pci_qdev_init;
2257     k->unplug = pci_unplug_device;
2258     k->exit = pci_unregister_device;
2259     k->bus_type = TYPE_PCI_BUS;
2260     k->props = pci_props;
2261 }
2262 
2263 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2264 {
2265     PCIBus *bus = PCI_BUS(dev->bus);
2266 
2267     if (bus->iommu_fn) {
2268         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2269     }
2270 
2271     if (bus->parent_dev) {
2272         /** We are ignoring the bus master DMA bit of the bridge
2273          *  as it would complicate things such as VFIO for no good reason */
2274         return pci_device_iommu_address_space(bus->parent_dev);
2275     }
2276 
2277     return &address_space_memory;
2278 }
2279 
2280 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2281 {
2282     bus->iommu_fn = fn;
2283     bus->iommu_opaque = opaque;
2284 }
2285 
2286 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2287 {
2288     Range *range = opaque;
2289     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2290     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2291     int i;
2292 
2293     if (!(cmd & PCI_COMMAND_MEMORY)) {
2294         return;
2295     }
2296 
2297     if (pc->is_bridge) {
2298         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2299         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2300 
2301         base = MAX(base, 0x1ULL << 32);
2302 
2303         if (limit >= base) {
2304             Range pref_range;
2305             pref_range.begin = base;
2306             pref_range.end = limit + 1;
2307             range_extend(range, &pref_range);
2308         }
2309     }
2310     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2311         PCIIORegion *r = &dev->io_regions[i];
2312         Range region_range;
2313 
2314         if (!r->size ||
2315             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2316             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2317             continue;
2318         }
2319         region_range.begin = pci_bar_address(dev, i, r->type, r->size);
2320         region_range.end = region_range.begin + r->size;
2321 
2322         if (region_range.begin == PCI_BAR_UNMAPPED) {
2323             continue;
2324         }
2325 
2326         region_range.begin = MAX(region_range.begin, 0x1ULL << 32);
2327 
2328         if (region_range.end - 1 >= region_range.begin) {
2329             range_extend(range, &region_range);
2330         }
2331     }
2332 }
2333 
2334 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2335 {
2336     range->begin = range->end = 0;
2337     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2338 }
2339 
2340 static const TypeInfo pci_device_type_info = {
2341     .name = TYPE_PCI_DEVICE,
2342     .parent = TYPE_DEVICE,
2343     .instance_size = sizeof(PCIDevice),
2344     .abstract = true,
2345     .class_size = sizeof(PCIDeviceClass),
2346     .class_init = pci_device_class_init,
2347 };
2348 
2349 static void pci_register_types(void)
2350 {
2351     type_register_static(&pci_bus_info);
2352     type_register_static(&pcie_bus_info);
2353     type_register_static(&pci_device_type_info);
2354 }
2355 
2356 type_init(pci_register_types)
2357