xref: /qemu/hw/pci/pci.c (revision de4c2dcf7fedf5fa727113e1dec0d0e5dd0462a0)
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 uint64_t master_abort_mem_read(void *opaque, hwaddr addr, unsigned size)
287 {
288    return -1ULL;
289 }
290 
291 static void master_abort_mem_write(void *opaque, hwaddr addr, uint64_t val,
292                                    unsigned size)
293 {
294 }
295 
296 static const MemoryRegionOps master_abort_mem_ops = {
297     .read = master_abort_mem_read,
298     .write = master_abort_mem_write,
299     .endianness = DEVICE_LITTLE_ENDIAN,
300 };
301 
302 #define MASTER_ABORT_MEM_PRIORITY INT_MIN
303 
304 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
305                          const char *name,
306                          MemoryRegion *address_space_mem,
307                          MemoryRegion *address_space_io,
308                          uint8_t devfn_min)
309 {
310     assert(PCI_FUNC(devfn_min) == 0);
311     bus->devfn_min = devfn_min;
312     bus->address_space_mem = address_space_mem;
313     bus->address_space_io = address_space_io;
314 
315 
316     memory_region_init_io(&bus->master_abort_mem, OBJECT(bus),
317                           &master_abort_mem_ops, bus, "pci-master-abort",
318                           memory_region_size(bus->address_space_mem));
319     memory_region_add_subregion_overlap(bus->address_space_mem,
320                                         0, &bus->master_abort_mem,
321                                         MASTER_ABORT_MEM_PRIORITY);
322 
323     /* host bridge */
324     QLIST_INIT(&bus->child);
325 
326     pci_host_bus_register(bus, parent);
327 
328     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
329 }
330 
331 bool pci_bus_is_express(PCIBus *bus)
332 {
333     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
334 }
335 
336 bool pci_bus_is_root(PCIBus *bus)
337 {
338     return !bus->parent_dev;
339 }
340 
341 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
342                          const char *name,
343                          MemoryRegion *address_space_mem,
344                          MemoryRegion *address_space_io,
345                          uint8_t devfn_min, const char *typename)
346 {
347     qbus_create_inplace(bus, bus_size, typename, parent, name);
348     pci_bus_init(bus, parent, name, address_space_mem,
349                  address_space_io, devfn_min);
350 }
351 
352 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
353                     MemoryRegion *address_space_mem,
354                     MemoryRegion *address_space_io,
355                     uint8_t devfn_min, const char *typename)
356 {
357     PCIBus *bus;
358 
359     bus = PCI_BUS(qbus_create(typename, parent, name));
360     pci_bus_init(bus, parent, name, address_space_mem,
361                  address_space_io, devfn_min);
362     return bus;
363 }
364 
365 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
366                   void *irq_opaque, int nirq)
367 {
368     bus->set_irq = set_irq;
369     bus->map_irq = map_irq;
370     bus->irq_opaque = irq_opaque;
371     bus->nirq = nirq;
372     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
373 }
374 
375 void pci_bus_hotplug(PCIBus *bus, pci_hotplug_fn hotplug, DeviceState *qdev)
376 {
377     bus->qbus.allow_hotplug = 1;
378     bus->hotplug = hotplug;
379     bus->hotplug_qdev = qdev;
380 }
381 
382 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
383                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
384                          void *irq_opaque,
385                          MemoryRegion *address_space_mem,
386                          MemoryRegion *address_space_io,
387                          uint8_t devfn_min, int nirq, const char *typename)
388 {
389     PCIBus *bus;
390 
391     bus = pci_bus_new(parent, name, address_space_mem,
392                       address_space_io, devfn_min, typename);
393     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
394     return bus;
395 }
396 
397 int pci_bus_num(PCIBus *s)
398 {
399     if (pci_bus_is_root(s))
400         return 0;       /* pci host bridge */
401     return s->parent_dev->config[PCI_SECONDARY_BUS];
402 }
403 
404 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size)
405 {
406     PCIDevice *s = container_of(pv, PCIDevice, config);
407     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
408     uint8_t *config;
409     int i;
410 
411     assert(size == pci_config_size(s));
412     config = g_malloc(size);
413 
414     qemu_get_buffer(f, config, size);
415     for (i = 0; i < size; ++i) {
416         if ((config[i] ^ s->config[i]) &
417             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
418             g_free(config);
419             return -EINVAL;
420         }
421     }
422     memcpy(s->config, config, size);
423 
424     pci_update_mappings(s);
425     if (pc->is_bridge) {
426         PCIBridge *b = PCI_BRIDGE(s);
427         pci_bridge_update_mappings(b);
428     }
429 
430     memory_region_set_enabled(&s->bus_master_enable_region,
431                               pci_get_word(s->config + PCI_COMMAND)
432                               & PCI_COMMAND_MASTER);
433 
434     g_free(config);
435     return 0;
436 }
437 
438 /* just put buffer */
439 static void put_pci_config_device(QEMUFile *f, void *pv, size_t size)
440 {
441     const uint8_t **v = pv;
442     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
443     qemu_put_buffer(f, *v, size);
444 }
445 
446 static VMStateInfo vmstate_info_pci_config = {
447     .name = "pci config",
448     .get  = get_pci_config_device,
449     .put  = put_pci_config_device,
450 };
451 
452 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size)
453 {
454     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
455     uint32_t irq_state[PCI_NUM_PINS];
456     int i;
457     for (i = 0; i < PCI_NUM_PINS; ++i) {
458         irq_state[i] = qemu_get_be32(f);
459         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
460             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
461                     irq_state[i]);
462             return -EINVAL;
463         }
464     }
465 
466     for (i = 0; i < PCI_NUM_PINS; ++i) {
467         pci_set_irq_state(s, i, irq_state[i]);
468     }
469 
470     return 0;
471 }
472 
473 static void put_pci_irq_state(QEMUFile *f, void *pv, size_t size)
474 {
475     int i;
476     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
477 
478     for (i = 0; i < PCI_NUM_PINS; ++i) {
479         qemu_put_be32(f, pci_irq_state(s, i));
480     }
481 }
482 
483 static VMStateInfo vmstate_info_pci_irq_state = {
484     .name = "pci irq state",
485     .get  = get_pci_irq_state,
486     .put  = put_pci_irq_state,
487 };
488 
489 const VMStateDescription vmstate_pci_device = {
490     .name = "PCIDevice",
491     .version_id = 2,
492     .minimum_version_id = 1,
493     .minimum_version_id_old = 1,
494     .fields      = (VMStateField []) {
495         VMSTATE_INT32_LE(version_id, PCIDevice),
496         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
497                                    vmstate_info_pci_config,
498                                    PCI_CONFIG_SPACE_SIZE),
499         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
500 				   vmstate_info_pci_irq_state,
501 				   PCI_NUM_PINS * sizeof(int32_t)),
502         VMSTATE_END_OF_LIST()
503     }
504 };
505 
506 const VMStateDescription vmstate_pcie_device = {
507     .name = "PCIEDevice",
508     .version_id = 2,
509     .minimum_version_id = 1,
510     .minimum_version_id_old = 1,
511     .fields      = (VMStateField []) {
512         VMSTATE_INT32_LE(version_id, PCIDevice),
513         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
514                                    vmstate_info_pci_config,
515                                    PCIE_CONFIG_SPACE_SIZE),
516         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
517 				   vmstate_info_pci_irq_state,
518 				   PCI_NUM_PINS * sizeof(int32_t)),
519         VMSTATE_END_OF_LIST()
520     }
521 };
522 
523 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
524 {
525     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
526 }
527 
528 void pci_device_save(PCIDevice *s, QEMUFile *f)
529 {
530     /* Clear interrupt status bit: it is implicit
531      * in irq_state which we are saving.
532      * This makes us compatible with old devices
533      * which never set or clear this bit. */
534     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
535     vmstate_save_state(f, pci_get_vmstate(s), s);
536     /* Restore the interrupt status bit. */
537     pci_update_irq_status(s);
538 }
539 
540 int pci_device_load(PCIDevice *s, QEMUFile *f)
541 {
542     int ret;
543     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
544     /* Restore the interrupt status bit. */
545     pci_update_irq_status(s);
546     return ret;
547 }
548 
549 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
550 {
551     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
552                  pci_default_sub_vendor_id);
553     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
554                  pci_default_sub_device_id);
555 }
556 
557 /*
558  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
559  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
560  */
561 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
562                       unsigned int *slotp, unsigned int *funcp)
563 {
564     const char *p;
565     char *e;
566     unsigned long val;
567     unsigned long dom = 0, bus = 0;
568     unsigned int slot = 0;
569     unsigned int func = 0;
570 
571     p = addr;
572     val = strtoul(p, &e, 16);
573     if (e == p)
574 	return -1;
575     if (*e == ':') {
576 	bus = val;
577 	p = e + 1;
578 	val = strtoul(p, &e, 16);
579 	if (e == p)
580 	    return -1;
581 	if (*e == ':') {
582 	    dom = bus;
583 	    bus = val;
584 	    p = e + 1;
585 	    val = strtoul(p, &e, 16);
586 	    if (e == p)
587 		return -1;
588 	}
589     }
590 
591     slot = val;
592 
593     if (funcp != NULL) {
594         if (*e != '.')
595             return -1;
596 
597         p = e + 1;
598         val = strtoul(p, &e, 16);
599         if (e == p)
600             return -1;
601 
602         func = val;
603     }
604 
605     /* if funcp == NULL func is 0 */
606     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
607 	return -1;
608 
609     if (*e)
610 	return -1;
611 
612     *domp = dom;
613     *busp = bus;
614     *slotp = slot;
615     if (funcp != NULL)
616         *funcp = func;
617     return 0;
618 }
619 
620 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
621 {
622     int dom, bus;
623     unsigned slot;
624 
625     assert(!root->parent_dev);
626 
627     if (!root) {
628         fprintf(stderr, "No primary PCI bus\n");
629         return NULL;
630     }
631 
632     if (!devaddr) {
633         *devfnp = -1;
634         return pci_find_bus_nr(root, 0);
635     }
636 
637     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
638         return NULL;
639     }
640 
641     if (dom != 0) {
642         fprintf(stderr, "No support for non-zero PCI domains\n");
643         return NULL;
644     }
645 
646     *devfnp = PCI_DEVFN(slot, 0);
647     return pci_find_bus_nr(root, bus);
648 }
649 
650 static void pci_init_cmask(PCIDevice *dev)
651 {
652     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
653     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
654     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
655     dev->cmask[PCI_REVISION_ID] = 0xff;
656     dev->cmask[PCI_CLASS_PROG] = 0xff;
657     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
658     dev->cmask[PCI_HEADER_TYPE] = 0xff;
659     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
660 }
661 
662 static void pci_init_wmask(PCIDevice *dev)
663 {
664     int config_size = pci_config_size(dev);
665 
666     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
667     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
668     pci_set_word(dev->wmask + PCI_COMMAND,
669                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
670                  PCI_COMMAND_INTX_DISABLE);
671     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
672         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
673     }
674 
675     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
676            config_size - PCI_CONFIG_HEADER_SIZE);
677 }
678 
679 static void pci_init_w1cmask(PCIDevice *dev)
680 {
681     /*
682      * Note: It's okay to set w1cmask even for readonly bits as
683      * long as their value is hardwired to 0.
684      */
685     pci_set_word(dev->w1cmask + PCI_STATUS,
686                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
687                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
688                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
689 }
690 
691 static void pci_init_mask_bridge(PCIDevice *d)
692 {
693     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
694        PCI_SEC_LETENCY_TIMER */
695     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
696 
697     /* base and limit */
698     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
699     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
700     pci_set_word(d->wmask + PCI_MEMORY_BASE,
701                  PCI_MEMORY_RANGE_MASK & 0xffff);
702     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
703                  PCI_MEMORY_RANGE_MASK & 0xffff);
704     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
705                  PCI_PREF_RANGE_MASK & 0xffff);
706     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
707                  PCI_PREF_RANGE_MASK & 0xffff);
708 
709     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
710     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
711 
712     /* Supported memory and i/o types */
713     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
714     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
715     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
716                                PCI_PREF_RANGE_TYPE_64);
717     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
718                                PCI_PREF_RANGE_TYPE_64);
719 
720     /*
721      * TODO: Bridges default to 10-bit VGA decoding but we currently only
722      * implement 16-bit decoding (no alias support).
723      */
724     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
725                  PCI_BRIDGE_CTL_PARITY |
726                  PCI_BRIDGE_CTL_SERR |
727                  PCI_BRIDGE_CTL_ISA |
728                  PCI_BRIDGE_CTL_VGA |
729                  PCI_BRIDGE_CTL_VGA_16BIT |
730                  PCI_BRIDGE_CTL_MASTER_ABORT |
731                  PCI_BRIDGE_CTL_BUS_RESET |
732                  PCI_BRIDGE_CTL_FAST_BACK |
733                  PCI_BRIDGE_CTL_DISCARD |
734                  PCI_BRIDGE_CTL_SEC_DISCARD |
735                  PCI_BRIDGE_CTL_DISCARD_SERR);
736     /* Below does not do anything as we never set this bit, put here for
737      * completeness. */
738     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
739                  PCI_BRIDGE_CTL_DISCARD_STATUS);
740     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
741     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
742     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
743                                PCI_PREF_RANGE_TYPE_MASK);
744     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
745                                PCI_PREF_RANGE_TYPE_MASK);
746 }
747 
748 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
749 {
750     uint8_t slot = PCI_SLOT(dev->devfn);
751     uint8_t func;
752 
753     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
754         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
755     }
756 
757     /*
758      * multifunction bit is interpreted in two ways as follows.
759      *   - all functions must set the bit to 1.
760      *     Example: Intel X53
761      *   - function 0 must set the bit, but the rest function (> 0)
762      *     is allowed to leave the bit to 0.
763      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
764      *
765      * So OS (at least Linux) checks the bit of only function 0,
766      * and doesn't see the bit of function > 0.
767      *
768      * The below check allows both interpretation.
769      */
770     if (PCI_FUNC(dev->devfn)) {
771         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
772         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
773             /* function 0 should set multifunction bit */
774             error_report("PCI: single function device can't be populated "
775                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
776             return -1;
777         }
778         return 0;
779     }
780 
781     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
782         return 0;
783     }
784     /* function 0 indicates single function, so function > 0 must be NULL */
785     for (func = 1; func < PCI_FUNC_MAX; ++func) {
786         if (bus->devices[PCI_DEVFN(slot, func)]) {
787             error_report("PCI: %x.0 indicates single function, "
788                          "but %x.%x is already populated.",
789                          slot, slot, func);
790             return -1;
791         }
792     }
793     return 0;
794 }
795 
796 static void pci_config_alloc(PCIDevice *pci_dev)
797 {
798     int config_size = pci_config_size(pci_dev);
799 
800     pci_dev->config = g_malloc0(config_size);
801     pci_dev->cmask = g_malloc0(config_size);
802     pci_dev->wmask = g_malloc0(config_size);
803     pci_dev->w1cmask = g_malloc0(config_size);
804     pci_dev->used = g_malloc0(config_size);
805 }
806 
807 static void pci_config_free(PCIDevice *pci_dev)
808 {
809     g_free(pci_dev->config);
810     g_free(pci_dev->cmask);
811     g_free(pci_dev->wmask);
812     g_free(pci_dev->w1cmask);
813     g_free(pci_dev->used);
814 }
815 
816 /* -1 for devfn means auto assign */
817 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
818                                          const char *name, int devfn)
819 {
820     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
821     PCIConfigReadFunc *config_read = pc->config_read;
822     PCIConfigWriteFunc *config_write = pc->config_write;
823     AddressSpace *dma_as;
824 
825     if (devfn < 0) {
826         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
827             devfn += PCI_FUNC_MAX) {
828             if (!bus->devices[devfn])
829                 goto found;
830         }
831         error_report("PCI: no slot/function available for %s, all in use", name);
832         return NULL;
833     found: ;
834     } else if (bus->devices[devfn]) {
835         error_report("PCI: slot %d function %d not available for %s, in use by %s",
836                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
837         return NULL;
838     }
839 
840     pci_dev->bus = bus;
841     dma_as = pci_device_iommu_address_space(pci_dev);
842 
843     memory_region_init_alias(&pci_dev->bus_master_enable_region,
844                              OBJECT(pci_dev), "bus master",
845                              dma_as->root, 0, memory_region_size(dma_as->root));
846     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
847     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
848                        name);
849 
850     pci_dev->devfn = devfn;
851     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
852     pci_dev->irq_state = 0;
853     pci_config_alloc(pci_dev);
854 
855     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
856     pci_config_set_device_id(pci_dev->config, pc->device_id);
857     pci_config_set_revision(pci_dev->config, pc->revision);
858     pci_config_set_class(pci_dev->config, pc->class_id);
859 
860     if (!pc->is_bridge) {
861         if (pc->subsystem_vendor_id || pc->subsystem_id) {
862             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
863                          pc->subsystem_vendor_id);
864             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
865                          pc->subsystem_id);
866         } else {
867             pci_set_default_subsystem_id(pci_dev);
868         }
869     } else {
870         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
871         assert(!pc->subsystem_vendor_id);
872         assert(!pc->subsystem_id);
873     }
874     pci_init_cmask(pci_dev);
875     pci_init_wmask(pci_dev);
876     pci_init_w1cmask(pci_dev);
877     if (pc->is_bridge) {
878         pci_init_mask_bridge(pci_dev);
879     }
880     if (pci_init_multifunction(bus, pci_dev)) {
881         pci_config_free(pci_dev);
882         return NULL;
883     }
884 
885     if (!config_read)
886         config_read = pci_default_read_config;
887     if (!config_write)
888         config_write = pci_default_write_config;
889     pci_dev->config_read = config_read;
890     pci_dev->config_write = config_write;
891     bus->devices[devfn] = pci_dev;
892     pci_dev->version_id = 2; /* Current pci device vmstate version */
893     return pci_dev;
894 }
895 
896 static void do_pci_unregister_device(PCIDevice *pci_dev)
897 {
898     pci_dev->bus->devices[pci_dev->devfn] = NULL;
899     pci_config_free(pci_dev);
900 
901     address_space_destroy(&pci_dev->bus_master_as);
902     memory_region_destroy(&pci_dev->bus_master_enable_region);
903 }
904 
905 static void pci_unregister_io_regions(PCIDevice *pci_dev)
906 {
907     PCIIORegion *r;
908     int i;
909 
910     for(i = 0; i < PCI_NUM_REGIONS; i++) {
911         r = &pci_dev->io_regions[i];
912         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
913             continue;
914         memory_region_del_subregion(r->address_space, r->memory);
915     }
916 
917     pci_unregister_vga(pci_dev);
918 }
919 
920 static int pci_unregister_device(DeviceState *dev)
921 {
922     PCIDevice *pci_dev = PCI_DEVICE(dev);
923     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
924 
925     pci_unregister_io_regions(pci_dev);
926     pci_del_option_rom(pci_dev);
927 
928     if (pc->exit) {
929         pc->exit(pci_dev);
930     }
931 
932     do_pci_unregister_device(pci_dev);
933     return 0;
934 }
935 
936 void pci_register_bar(PCIDevice *pci_dev, int region_num,
937                       uint8_t type, MemoryRegion *memory)
938 {
939     PCIIORegion *r;
940     uint32_t addr;
941     uint64_t wmask;
942     pcibus_t size = memory_region_size(memory);
943 
944     assert(region_num >= 0);
945     assert(region_num < PCI_NUM_REGIONS);
946     if (size & (size-1)) {
947         fprintf(stderr, "ERROR: PCI region size must be pow2 "
948                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
949         exit(1);
950     }
951 
952     r = &pci_dev->io_regions[region_num];
953     r->addr = PCI_BAR_UNMAPPED;
954     r->size = size;
955     r->type = type;
956     r->memory = NULL;
957 
958     wmask = ~(size - 1);
959     addr = pci_bar(pci_dev, region_num);
960     if (region_num == PCI_ROM_SLOT) {
961         /* ROM enable bit is writable */
962         wmask |= PCI_ROM_ADDRESS_ENABLE;
963     }
964     pci_set_long(pci_dev->config + addr, type);
965     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
966         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
967         pci_set_quad(pci_dev->wmask + addr, wmask);
968         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
969     } else {
970         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
971         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
972     }
973     pci_dev->io_regions[region_num].memory = memory;
974     pci_dev->io_regions[region_num].address_space
975         = type & PCI_BASE_ADDRESS_SPACE_IO
976         ? pci_dev->bus->address_space_io
977         : pci_dev->bus->address_space_mem;
978 }
979 
980 static void pci_update_vga(PCIDevice *pci_dev)
981 {
982     uint16_t cmd;
983 
984     if (!pci_dev->has_vga) {
985         return;
986     }
987 
988     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
989 
990     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
991                               cmd & PCI_COMMAND_MEMORY);
992     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
993                               cmd & PCI_COMMAND_IO);
994     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
995                               cmd & PCI_COMMAND_IO);
996 }
997 
998 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
999                       MemoryRegion *io_lo, MemoryRegion *io_hi)
1000 {
1001     assert(!pci_dev->has_vga);
1002 
1003     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
1004     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
1005     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
1006                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
1007 
1008     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
1009     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
1010     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1011                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
1012 
1013     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
1014     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
1015     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
1016                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
1017     pci_dev->has_vga = true;
1018 
1019     pci_update_vga(pci_dev);
1020 }
1021 
1022 void pci_unregister_vga(PCIDevice *pci_dev)
1023 {
1024     if (!pci_dev->has_vga) {
1025         return;
1026     }
1027 
1028     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1029                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1030     memory_region_del_subregion(pci_dev->bus->address_space_io,
1031                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1032     memory_region_del_subregion(pci_dev->bus->address_space_io,
1033                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1034     pci_dev->has_vga = false;
1035 }
1036 
1037 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1038 {
1039     return pci_dev->io_regions[region_num].addr;
1040 }
1041 
1042 static pcibus_t pci_bar_address(PCIDevice *d,
1043 				int reg, uint8_t type, pcibus_t size)
1044 {
1045     pcibus_t new_addr, last_addr;
1046     int bar = pci_bar(d, reg);
1047     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1048 
1049     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1050         if (!(cmd & PCI_COMMAND_IO)) {
1051             return PCI_BAR_UNMAPPED;
1052         }
1053         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1054         last_addr = new_addr + size - 1;
1055         /* Check if 32 bit BAR wraps around explicitly.
1056          * TODO: make priorities correct and remove this work around.
1057          */
1058         if (last_addr <= new_addr || new_addr == 0 || last_addr >= UINT32_MAX) {
1059             return PCI_BAR_UNMAPPED;
1060         }
1061         return new_addr;
1062     }
1063 
1064     if (!(cmd & PCI_COMMAND_MEMORY)) {
1065         return PCI_BAR_UNMAPPED;
1066     }
1067     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1068         new_addr = pci_get_quad(d->config + bar);
1069     } else {
1070         new_addr = pci_get_long(d->config + bar);
1071     }
1072     /* the ROM slot has a specific enable bit */
1073     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1074         return PCI_BAR_UNMAPPED;
1075     }
1076     new_addr &= ~(size - 1);
1077     last_addr = new_addr + size - 1;
1078     /* NOTE: we do not support wrapping */
1079     /* XXX: as we cannot support really dynamic
1080        mappings, we handle specific values as invalid
1081        mappings. */
1082     if (last_addr <= new_addr || new_addr == 0 ||
1083         last_addr == PCI_BAR_UNMAPPED) {
1084         return PCI_BAR_UNMAPPED;
1085     }
1086 
1087     /* Now pcibus_t is 64bit.
1088      * Check if 32 bit BAR wraps around explicitly.
1089      * Without this, PC ide doesn't work well.
1090      * TODO: remove this work around.
1091      */
1092     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1093         return PCI_BAR_UNMAPPED;
1094     }
1095 
1096     /*
1097      * OS is allowed to set BAR beyond its addressable
1098      * bits. For example, 32 bit OS can set 64bit bar
1099      * to >4G. Check it. TODO: we might need to support
1100      * it in the future for e.g. PAE.
1101      */
1102     if (last_addr >= HWADDR_MAX) {
1103         return PCI_BAR_UNMAPPED;
1104     }
1105 
1106     return new_addr;
1107 }
1108 
1109 static void pci_update_mappings(PCIDevice *d)
1110 {
1111     PCIIORegion *r;
1112     int i;
1113     pcibus_t new_addr;
1114 
1115     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1116         r = &d->io_regions[i];
1117 
1118         /* this region isn't registered */
1119         if (!r->size)
1120             continue;
1121 
1122         new_addr = pci_bar_address(d, i, r->type, r->size);
1123 
1124         /* This bar isn't changed */
1125         if (new_addr == r->addr)
1126             continue;
1127 
1128         /* now do the real mapping */
1129         if (r->addr != PCI_BAR_UNMAPPED) {
1130             memory_region_del_subregion(r->address_space, r->memory);
1131         }
1132         r->addr = new_addr;
1133         if (r->addr != PCI_BAR_UNMAPPED) {
1134             memory_region_add_subregion_overlap(r->address_space,
1135                                                 r->addr, r->memory, 1);
1136         }
1137     }
1138 
1139     pci_update_vga(d);
1140 }
1141 
1142 static inline int pci_irq_disabled(PCIDevice *d)
1143 {
1144     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1145 }
1146 
1147 /* Called after interrupt disabled field update in config space,
1148  * assert/deassert interrupts if necessary.
1149  * Gets original interrupt disable bit value (before update). */
1150 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1151 {
1152     int i, disabled = pci_irq_disabled(d);
1153     if (disabled == was_irq_disabled)
1154         return;
1155     for (i = 0; i < PCI_NUM_PINS; ++i) {
1156         int state = pci_irq_state(d, i);
1157         pci_change_irq_level(d, i, disabled ? -state : state);
1158     }
1159 }
1160 
1161 uint32_t pci_default_read_config(PCIDevice *d,
1162                                  uint32_t address, int len)
1163 {
1164     uint32_t val = 0;
1165 
1166     memcpy(&val, d->config + address, len);
1167     return le32_to_cpu(val);
1168 }
1169 
1170 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
1171 {
1172     int i, was_irq_disabled = pci_irq_disabled(d);
1173 
1174     for (i = 0; i < l; val >>= 8, ++i) {
1175         uint8_t wmask = d->wmask[addr + i];
1176         uint8_t w1cmask = d->w1cmask[addr + i];
1177         assert(!(wmask & w1cmask));
1178         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1179         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1180     }
1181     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1182         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1183         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1184         range_covers_byte(addr, l, PCI_COMMAND))
1185         pci_update_mappings(d);
1186 
1187     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1188         pci_update_irq_disabled(d, was_irq_disabled);
1189         memory_region_set_enabled(&d->bus_master_enable_region,
1190                                   pci_get_word(d->config + PCI_COMMAND)
1191                                     & PCI_COMMAND_MASTER);
1192     }
1193 
1194     msi_write_config(d, addr, val, l);
1195     msix_write_config(d, addr, val, l);
1196 }
1197 
1198 /***********************************************************/
1199 /* generic PCI irq support */
1200 
1201 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1202 static void pci_irq_handler(void *opaque, int irq_num, int level)
1203 {
1204     PCIDevice *pci_dev = opaque;
1205     int change;
1206 
1207     change = level - pci_irq_state(pci_dev, irq_num);
1208     if (!change)
1209         return;
1210 
1211     pci_set_irq_state(pci_dev, irq_num, level);
1212     pci_update_irq_status(pci_dev);
1213     if (pci_irq_disabled(pci_dev))
1214         return;
1215     pci_change_irq_level(pci_dev, irq_num, change);
1216 }
1217 
1218 static inline int pci_intx(PCIDevice *pci_dev)
1219 {
1220     return pci_get_byte(pci_dev->config + PCI_INTERRUPT_PIN) - 1;
1221 }
1222 
1223 qemu_irq pci_allocate_irq(PCIDevice *pci_dev)
1224 {
1225     int intx = pci_intx(pci_dev);
1226 
1227     return qemu_allocate_irq(pci_irq_handler, pci_dev, intx);
1228 }
1229 
1230 void pci_set_irq(PCIDevice *pci_dev, int level)
1231 {
1232     int intx = pci_intx(pci_dev);
1233     pci_irq_handler(pci_dev, intx, level);
1234 }
1235 
1236 /* Special hooks used by device assignment */
1237 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1238 {
1239     assert(pci_bus_is_root(bus));
1240     bus->route_intx_to_irq = route_intx_to_irq;
1241 }
1242 
1243 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1244 {
1245     PCIBus *bus;
1246 
1247     do {
1248          bus = dev->bus;
1249          pin = bus->map_irq(dev, pin);
1250          dev = bus->parent_dev;
1251     } while (dev);
1252 
1253     if (!bus->route_intx_to_irq) {
1254         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1255                      object_get_typename(OBJECT(bus->qbus.parent)));
1256         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1257     }
1258 
1259     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1260 }
1261 
1262 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1263 {
1264     return old->mode != new->mode || old->irq != new->irq;
1265 }
1266 
1267 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1268 {
1269     PCIDevice *dev;
1270     PCIBus *sec;
1271     int i;
1272 
1273     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1274         dev = bus->devices[i];
1275         if (dev && dev->intx_routing_notifier) {
1276             dev->intx_routing_notifier(dev);
1277         }
1278     }
1279 
1280     QLIST_FOREACH(sec, &bus->child, sibling) {
1281         pci_bus_fire_intx_routing_notifier(sec);
1282     }
1283 }
1284 
1285 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1286                                           PCIINTxRoutingNotifier notifier)
1287 {
1288     dev->intx_routing_notifier = notifier;
1289 }
1290 
1291 /*
1292  * PCI-to-PCI bridge specification
1293  * 9.1: Interrupt routing. Table 9-1
1294  *
1295  * the PCI Express Base Specification, Revision 2.1
1296  * 2.2.8.1: INTx interrutp signaling - Rules
1297  *          the Implementation Note
1298  *          Table 2-20
1299  */
1300 /*
1301  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1302  * 0-origin unlike PCI interrupt pin register.
1303  */
1304 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1305 {
1306     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1307 }
1308 
1309 /***********************************************************/
1310 /* monitor info on PCI */
1311 
1312 typedef struct {
1313     uint16_t class;
1314     const char *desc;
1315     const char *fw_name;
1316     uint16_t fw_ign_bits;
1317 } pci_class_desc;
1318 
1319 static const pci_class_desc pci_class_descriptions[] =
1320 {
1321     { 0x0001, "VGA controller", "display"},
1322     { 0x0100, "SCSI controller", "scsi"},
1323     { 0x0101, "IDE controller", "ide"},
1324     { 0x0102, "Floppy controller", "fdc"},
1325     { 0x0103, "IPI controller", "ipi"},
1326     { 0x0104, "RAID controller", "raid"},
1327     { 0x0106, "SATA controller"},
1328     { 0x0107, "SAS controller"},
1329     { 0x0180, "Storage controller"},
1330     { 0x0200, "Ethernet controller", "ethernet"},
1331     { 0x0201, "Token Ring controller", "token-ring"},
1332     { 0x0202, "FDDI controller", "fddi"},
1333     { 0x0203, "ATM controller", "atm"},
1334     { 0x0280, "Network controller"},
1335     { 0x0300, "VGA controller", "display", 0x00ff},
1336     { 0x0301, "XGA controller"},
1337     { 0x0302, "3D controller"},
1338     { 0x0380, "Display controller"},
1339     { 0x0400, "Video controller", "video"},
1340     { 0x0401, "Audio controller", "sound"},
1341     { 0x0402, "Phone"},
1342     { 0x0403, "Audio controller", "sound"},
1343     { 0x0480, "Multimedia controller"},
1344     { 0x0500, "RAM controller", "memory"},
1345     { 0x0501, "Flash controller", "flash"},
1346     { 0x0580, "Memory controller"},
1347     { 0x0600, "Host bridge", "host"},
1348     { 0x0601, "ISA bridge", "isa"},
1349     { 0x0602, "EISA bridge", "eisa"},
1350     { 0x0603, "MC bridge", "mca"},
1351     { 0x0604, "PCI bridge", "pci"},
1352     { 0x0605, "PCMCIA bridge", "pcmcia"},
1353     { 0x0606, "NUBUS bridge", "nubus"},
1354     { 0x0607, "CARDBUS bridge", "cardbus"},
1355     { 0x0608, "RACEWAY bridge"},
1356     { 0x0680, "Bridge"},
1357     { 0x0700, "Serial port", "serial"},
1358     { 0x0701, "Parallel port", "parallel"},
1359     { 0x0800, "Interrupt controller", "interrupt-controller"},
1360     { 0x0801, "DMA controller", "dma-controller"},
1361     { 0x0802, "Timer", "timer"},
1362     { 0x0803, "RTC", "rtc"},
1363     { 0x0900, "Keyboard", "keyboard"},
1364     { 0x0901, "Pen", "pen"},
1365     { 0x0902, "Mouse", "mouse"},
1366     { 0x0A00, "Dock station", "dock", 0x00ff},
1367     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1368     { 0x0c00, "Fireware contorller", "fireware"},
1369     { 0x0c01, "Access bus controller", "access-bus"},
1370     { 0x0c02, "SSA controller", "ssa"},
1371     { 0x0c03, "USB controller", "usb"},
1372     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1373     { 0x0c05, "SMBus"},
1374     { 0, NULL}
1375 };
1376 
1377 static void pci_for_each_device_under_bus(PCIBus *bus,
1378                                           void (*fn)(PCIBus *b, PCIDevice *d,
1379                                                      void *opaque),
1380                                           void *opaque)
1381 {
1382     PCIDevice *d;
1383     int devfn;
1384 
1385     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1386         d = bus->devices[devfn];
1387         if (d) {
1388             fn(bus, d, opaque);
1389         }
1390     }
1391 }
1392 
1393 void pci_for_each_device(PCIBus *bus, int bus_num,
1394                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1395                          void *opaque)
1396 {
1397     bus = pci_find_bus_nr(bus, bus_num);
1398 
1399     if (bus) {
1400         pci_for_each_device_under_bus(bus, fn, opaque);
1401     }
1402 }
1403 
1404 static const pci_class_desc *get_class_desc(int class)
1405 {
1406     const pci_class_desc *desc;
1407 
1408     desc = pci_class_descriptions;
1409     while (desc->desc && class != desc->class) {
1410         desc++;
1411     }
1412 
1413     return desc;
1414 }
1415 
1416 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1417 
1418 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1419 {
1420     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1421     int i;
1422 
1423     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1424         const PCIIORegion *r = &dev->io_regions[i];
1425         PciMemoryRegionList *region;
1426 
1427         if (!r->size) {
1428             continue;
1429         }
1430 
1431         region = g_malloc0(sizeof(*region));
1432         region->value = g_malloc0(sizeof(*region->value));
1433 
1434         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1435             region->value->type = g_strdup("io");
1436         } else {
1437             region->value->type = g_strdup("memory");
1438             region->value->has_prefetch = true;
1439             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1440             region->value->has_mem_type_64 = true;
1441             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1442         }
1443 
1444         region->value->bar = i;
1445         region->value->address = r->addr;
1446         region->value->size = r->size;
1447 
1448         /* XXX: waiting for the qapi to support GSList */
1449         if (!cur_item) {
1450             head = cur_item = region;
1451         } else {
1452             cur_item->next = region;
1453             cur_item = region;
1454         }
1455     }
1456 
1457     return head;
1458 }
1459 
1460 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1461                                            int bus_num)
1462 {
1463     PciBridgeInfo *info;
1464 
1465     info = g_malloc0(sizeof(*info));
1466 
1467     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1468     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1469     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1470 
1471     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1472     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1473     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1474 
1475     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1476     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1477     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1478 
1479     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1480     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1481     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1482 
1483     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1484         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1485         if (child_bus) {
1486             info->has_devices = true;
1487             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1488         }
1489     }
1490 
1491     return info;
1492 }
1493 
1494 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1495                                            int bus_num)
1496 {
1497     const pci_class_desc *desc;
1498     PciDeviceInfo *info;
1499     uint8_t type;
1500     int class;
1501 
1502     info = g_malloc0(sizeof(*info));
1503     info->bus = bus_num;
1504     info->slot = PCI_SLOT(dev->devfn);
1505     info->function = PCI_FUNC(dev->devfn);
1506 
1507     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1508     info->class_info.q_class = class;
1509     desc = get_class_desc(class);
1510     if (desc->desc) {
1511         info->class_info.has_desc = true;
1512         info->class_info.desc = g_strdup(desc->desc);
1513     }
1514 
1515     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1516     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1517     info->regions = qmp_query_pci_regions(dev);
1518     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1519 
1520     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1521         info->has_irq = true;
1522         info->irq = dev->config[PCI_INTERRUPT_LINE];
1523     }
1524 
1525     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1526     if (type == PCI_HEADER_TYPE_BRIDGE) {
1527         info->has_pci_bridge = true;
1528         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1529     }
1530 
1531     return info;
1532 }
1533 
1534 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1535 {
1536     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1537     PCIDevice *dev;
1538     int devfn;
1539 
1540     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1541         dev = bus->devices[devfn];
1542         if (dev) {
1543             info = g_malloc0(sizeof(*info));
1544             info->value = qmp_query_pci_device(dev, bus, bus_num);
1545 
1546             /* XXX: waiting for the qapi to support GSList */
1547             if (!cur_item) {
1548                 head = cur_item = info;
1549             } else {
1550                 cur_item->next = info;
1551                 cur_item = info;
1552             }
1553         }
1554     }
1555 
1556     return head;
1557 }
1558 
1559 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1560 {
1561     PciInfo *info = NULL;
1562 
1563     bus = pci_find_bus_nr(bus, bus_num);
1564     if (bus) {
1565         info = g_malloc0(sizeof(*info));
1566         info->bus = bus_num;
1567         info->devices = qmp_query_pci_devices(bus, bus_num);
1568     }
1569 
1570     return info;
1571 }
1572 
1573 PciInfoList *qmp_query_pci(Error **errp)
1574 {
1575     PciInfoList *info, *head = NULL, *cur_item = NULL;
1576     PCIHostState *host_bridge;
1577 
1578     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1579         info = g_malloc0(sizeof(*info));
1580         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1581 
1582         /* XXX: waiting for the qapi to support GSList */
1583         if (!cur_item) {
1584             head = cur_item = info;
1585         } else {
1586             cur_item->next = info;
1587             cur_item = info;
1588         }
1589     }
1590 
1591     return head;
1592 }
1593 
1594 static const char * const pci_nic_models[] = {
1595     "ne2k_pci",
1596     "i82551",
1597     "i82557b",
1598     "i82559er",
1599     "rtl8139",
1600     "e1000",
1601     "pcnet",
1602     "virtio",
1603     NULL
1604 };
1605 
1606 static const char * const pci_nic_names[] = {
1607     "ne2k_pci",
1608     "i82551",
1609     "i82557b",
1610     "i82559er",
1611     "rtl8139",
1612     "e1000",
1613     "pcnet",
1614     "virtio-net-pci",
1615     NULL
1616 };
1617 
1618 /* Initialize a PCI NIC.  */
1619 /* FIXME callers should check for failure, but don't */
1620 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1621                         const char *default_model,
1622                         const char *default_devaddr)
1623 {
1624     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1625     PCIBus *bus;
1626     int devfn;
1627     PCIDevice *pci_dev;
1628     DeviceState *dev;
1629     int i;
1630 
1631     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1632     if (i < 0)
1633         return NULL;
1634 
1635     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1636     if (!bus) {
1637         error_report("Invalid PCI device address %s for device %s",
1638                      devaddr, pci_nic_names[i]);
1639         return NULL;
1640     }
1641 
1642     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1643     dev = &pci_dev->qdev;
1644     qdev_set_nic_properties(dev, nd);
1645     if (qdev_init(dev) < 0)
1646         return NULL;
1647     return pci_dev;
1648 }
1649 
1650 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1651                                const char *default_model,
1652                                const char *default_devaddr)
1653 {
1654     PCIDevice *res;
1655 
1656     if (qemu_show_nic_models(nd->model, pci_nic_models))
1657         exit(0);
1658 
1659     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1660     if (!res)
1661         exit(1);
1662     return res;
1663 }
1664 
1665 PCIDevice *pci_vga_init(PCIBus *bus)
1666 {
1667     switch (vga_interface_type) {
1668     case VGA_CIRRUS:
1669         return pci_create_simple(bus, -1, "cirrus-vga");
1670     case VGA_QXL:
1671         return pci_create_simple(bus, -1, "qxl-vga");
1672     case VGA_STD:
1673         return pci_create_simple(bus, -1, "VGA");
1674     case VGA_VMWARE:
1675         return pci_create_simple(bus, -1, "vmware-svga");
1676     case VGA_NONE:
1677     default: /* Other non-PCI types. Checking for unsupported types is already
1678                 done in vl.c. */
1679         return NULL;
1680     }
1681 }
1682 
1683 /* Whether a given bus number is in range of the secondary
1684  * bus of the given bridge device. */
1685 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1686 {
1687     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1688              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1689         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1690         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1691 }
1692 
1693 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1694 {
1695     PCIBus *sec;
1696 
1697     if (!bus) {
1698         return NULL;
1699     }
1700 
1701     if (pci_bus_num(bus) == bus_num) {
1702         return bus;
1703     }
1704 
1705     /* Consider all bus numbers in range for the host pci bridge. */
1706     if (!pci_bus_is_root(bus) &&
1707         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1708         return NULL;
1709     }
1710 
1711     /* try child bus */
1712     for (; bus; bus = sec) {
1713         QLIST_FOREACH(sec, &bus->child, sibling) {
1714             assert(!pci_bus_is_root(sec));
1715             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1716                 return sec;
1717             }
1718             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1719                 break;
1720             }
1721         }
1722     }
1723 
1724     return NULL;
1725 }
1726 
1727 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1728 {
1729     bus = pci_find_bus_nr(bus, bus_num);
1730 
1731     if (!bus)
1732         return NULL;
1733 
1734     return bus->devices[devfn];
1735 }
1736 
1737 static int pci_qdev_init(DeviceState *qdev)
1738 {
1739     PCIDevice *pci_dev = (PCIDevice *)qdev;
1740     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1741     PCIBus *bus;
1742     int rc;
1743     bool is_default_rom;
1744 
1745     /* initialize cap_present for pci_is_express() and pci_config_size() */
1746     if (pc->is_express) {
1747         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1748     }
1749 
1750     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1751     pci_dev = do_pci_register_device(pci_dev, bus,
1752                                      object_get_typename(OBJECT(qdev)),
1753                                      pci_dev->devfn);
1754     if (pci_dev == NULL)
1755         return -1;
1756     if (qdev->hotplugged && pc->no_hotplug) {
1757         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(pci_dev)));
1758         do_pci_unregister_device(pci_dev);
1759         return -1;
1760     }
1761     if (pc->init) {
1762         rc = pc->init(pci_dev);
1763         if (rc != 0) {
1764             do_pci_unregister_device(pci_dev);
1765             return rc;
1766         }
1767     }
1768 
1769     /* rom loading */
1770     is_default_rom = false;
1771     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1772         pci_dev->romfile = g_strdup(pc->romfile);
1773         is_default_rom = true;
1774     }
1775     pci_add_option_rom(pci_dev, is_default_rom);
1776 
1777     if (bus->hotplug) {
1778         /* Let buses differentiate between hotplug and when device is
1779          * enabled during qemu machine creation. */
1780         rc = bus->hotplug(bus->hotplug_qdev, pci_dev,
1781                           qdev->hotplugged ? PCI_HOTPLUG_ENABLED:
1782                           PCI_COLDPLUG_ENABLED);
1783         if (rc != 0) {
1784             int r = pci_unregister_device(&pci_dev->qdev);
1785             assert(!r);
1786             return rc;
1787         }
1788     }
1789     return 0;
1790 }
1791 
1792 static int pci_unplug_device(DeviceState *qdev)
1793 {
1794     PCIDevice *dev = PCI_DEVICE(qdev);
1795     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1796 
1797     if (pc->no_hotplug) {
1798         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(dev)));
1799         return -1;
1800     }
1801     return dev->bus->hotplug(dev->bus->hotplug_qdev, dev,
1802                              PCI_HOTPLUG_DISABLED);
1803 }
1804 
1805 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1806                                     const char *name)
1807 {
1808     DeviceState *dev;
1809 
1810     dev = qdev_create(&bus->qbus, name);
1811     qdev_prop_set_int32(dev, "addr", devfn);
1812     qdev_prop_set_bit(dev, "multifunction", multifunction);
1813     return PCI_DEVICE(dev);
1814 }
1815 
1816 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1817                                            bool multifunction,
1818                                            const char *name)
1819 {
1820     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1821     qdev_init_nofail(&dev->qdev);
1822     return dev;
1823 }
1824 
1825 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1826 {
1827     return pci_create_multifunction(bus, devfn, false, name);
1828 }
1829 
1830 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1831 {
1832     return pci_create_simple_multifunction(bus, devfn, false, name);
1833 }
1834 
1835 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1836 {
1837     int offset = PCI_CONFIG_HEADER_SIZE;
1838     int i;
1839     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1840         if (pdev->used[i])
1841             offset = i + 1;
1842         else if (i - offset + 1 == size)
1843             return offset;
1844     }
1845     return 0;
1846 }
1847 
1848 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1849                                         uint8_t *prev_p)
1850 {
1851     uint8_t next, prev;
1852 
1853     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1854         return 0;
1855 
1856     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1857          prev = next + PCI_CAP_LIST_NEXT)
1858         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1859             break;
1860 
1861     if (prev_p)
1862         *prev_p = prev;
1863     return next;
1864 }
1865 
1866 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1867 {
1868     uint8_t next, prev, found = 0;
1869 
1870     if (!(pdev->used[offset])) {
1871         return 0;
1872     }
1873 
1874     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1875 
1876     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1877          prev = next + PCI_CAP_LIST_NEXT) {
1878         if (next <= offset && next > found) {
1879             found = next;
1880         }
1881     }
1882     return found;
1883 }
1884 
1885 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1886    This is needed for an option rom which is used for more than one device. */
1887 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1888 {
1889     uint16_t vendor_id;
1890     uint16_t device_id;
1891     uint16_t rom_vendor_id;
1892     uint16_t rom_device_id;
1893     uint16_t rom_magic;
1894     uint16_t pcir_offset;
1895     uint8_t checksum;
1896 
1897     /* Words in rom data are little endian (like in PCI configuration),
1898        so they can be read / written with pci_get_word / pci_set_word. */
1899 
1900     /* Only a valid rom will be patched. */
1901     rom_magic = pci_get_word(ptr);
1902     if (rom_magic != 0xaa55) {
1903         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1904         return;
1905     }
1906     pcir_offset = pci_get_word(ptr + 0x18);
1907     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1908         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1909         return;
1910     }
1911 
1912     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1913     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1914     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1915     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1916 
1917     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1918                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1919 
1920     checksum = ptr[6];
1921 
1922     if (vendor_id != rom_vendor_id) {
1923         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1924         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1925         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1926         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1927         ptr[6] = checksum;
1928         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1929     }
1930 
1931     if (device_id != rom_device_id) {
1932         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1933         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1934         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1935         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1936         ptr[6] = checksum;
1937         pci_set_word(ptr + pcir_offset + 6, device_id);
1938     }
1939 }
1940 
1941 /* Add an option rom for the device */
1942 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1943 {
1944     int size;
1945     char *path;
1946     void *ptr;
1947     char name[32];
1948     const VMStateDescription *vmsd;
1949 
1950     if (!pdev->romfile)
1951         return 0;
1952     if (strlen(pdev->romfile) == 0)
1953         return 0;
1954 
1955     if (!pdev->rom_bar) {
1956         /*
1957          * Load rom via fw_cfg instead of creating a rom bar,
1958          * for 0.11 compatibility.
1959          */
1960         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1961         if (class == 0x0300) {
1962             rom_add_vga(pdev->romfile);
1963         } else {
1964             rom_add_option(pdev->romfile, -1);
1965         }
1966         return 0;
1967     }
1968 
1969     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1970     if (path == NULL) {
1971         path = g_strdup(pdev->romfile);
1972     }
1973 
1974     size = get_image_size(path);
1975     if (size < 0) {
1976         error_report("%s: failed to find romfile \"%s\"",
1977                      __func__, pdev->romfile);
1978         g_free(path);
1979         return -1;
1980     } else if (size == 0) {
1981         error_report("%s: ignoring empty romfile \"%s\"",
1982                      __func__, pdev->romfile);
1983         g_free(path);
1984         return -1;
1985     }
1986     if (size & (size - 1)) {
1987         size = 1 << qemu_fls(size);
1988     }
1989 
1990     vmsd = qdev_get_vmsd(DEVICE(pdev));
1991 
1992     if (vmsd) {
1993         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1994     } else {
1995         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1996     }
1997     pdev->has_rom = true;
1998     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size);
1999     vmstate_register_ram(&pdev->rom, &pdev->qdev);
2000     ptr = memory_region_get_ram_ptr(&pdev->rom);
2001     load_image(path, ptr);
2002     g_free(path);
2003 
2004     if (is_default_rom) {
2005         /* Only the default rom images will be patched (if needed). */
2006         pci_patch_ids(pdev, ptr, size);
2007     }
2008 
2009     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
2010 
2011     return 0;
2012 }
2013 
2014 static void pci_del_option_rom(PCIDevice *pdev)
2015 {
2016     if (!pdev->has_rom)
2017         return;
2018 
2019     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
2020     memory_region_destroy(&pdev->rom);
2021     pdev->has_rom = false;
2022 }
2023 
2024 /*
2025  * if !offset
2026  * Reserve space and add capability to the linked list in pci config space
2027  *
2028  * if offset = 0,
2029  * Find and reserve space and add capability to the linked list
2030  * in pci config space */
2031 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
2032                        uint8_t offset, uint8_t size)
2033 {
2034     uint8_t *config;
2035     int i, overlapping_cap;
2036 
2037     if (!offset) {
2038         offset = pci_find_space(pdev, size);
2039         if (!offset) {
2040             return -ENOSPC;
2041         }
2042     } else {
2043         /* Verify that capabilities don't overlap.  Note: device assignment
2044          * depends on this check to verify that the device is not broken.
2045          * Should never trigger for emulated devices, but it's helpful
2046          * for debugging these. */
2047         for (i = offset; i < offset + size; i++) {
2048             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2049             if (overlapping_cap) {
2050                 fprintf(stderr, "ERROR: %s:%02x:%02x.%x "
2051                         "Attempt to add PCI capability %x at offset "
2052                         "%x overlaps existing capability %x at offset %x\n",
2053                         pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2054                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2055                         cap_id, offset, overlapping_cap, i);
2056                 return -EINVAL;
2057             }
2058         }
2059     }
2060 
2061     config = pdev->config + offset;
2062     config[PCI_CAP_LIST_ID] = cap_id;
2063     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2064     pdev->config[PCI_CAPABILITY_LIST] = offset;
2065     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2066     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2067     /* Make capability read-only by default */
2068     memset(pdev->wmask + offset, 0, size);
2069     /* Check capability by default */
2070     memset(pdev->cmask + offset, 0xFF, size);
2071     return offset;
2072 }
2073 
2074 /* Unlink capability from the pci config space. */
2075 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2076 {
2077     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2078     if (!offset)
2079         return;
2080     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2081     /* Make capability writable again */
2082     memset(pdev->wmask + offset, 0xff, size);
2083     memset(pdev->w1cmask + offset, 0, size);
2084     /* Clear cmask as device-specific registers can't be checked */
2085     memset(pdev->cmask + offset, 0, size);
2086     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2087 
2088     if (!pdev->config[PCI_CAPABILITY_LIST])
2089         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2090 }
2091 
2092 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2093 {
2094     return pci_find_capability_list(pdev, cap_id, NULL);
2095 }
2096 
2097 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2098 {
2099     PCIDevice *d = (PCIDevice *)dev;
2100     const pci_class_desc *desc;
2101     char ctxt[64];
2102     PCIIORegion *r;
2103     int i, class;
2104 
2105     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2106     desc = pci_class_descriptions;
2107     while (desc->desc && class != desc->class)
2108         desc++;
2109     if (desc->desc) {
2110         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2111     } else {
2112         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2113     }
2114 
2115     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2116                    "pci id %04x:%04x (sub %04x:%04x)\n",
2117                    indent, "", ctxt, pci_bus_num(d->bus),
2118                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2119                    pci_get_word(d->config + PCI_VENDOR_ID),
2120                    pci_get_word(d->config + PCI_DEVICE_ID),
2121                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2122                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2123     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2124         r = &d->io_regions[i];
2125         if (!r->size)
2126             continue;
2127         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2128                        " [0x%"FMT_PCIBUS"]\n",
2129                        indent, "",
2130                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2131                        r->addr, r->addr + r->size - 1);
2132     }
2133 }
2134 
2135 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2136 {
2137     PCIDevice *d = (PCIDevice *)dev;
2138     const char *name = NULL;
2139     const pci_class_desc *desc =  pci_class_descriptions;
2140     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2141 
2142     while (desc->desc &&
2143           (class & ~desc->fw_ign_bits) !=
2144           (desc->class & ~desc->fw_ign_bits)) {
2145         desc++;
2146     }
2147 
2148     if (desc->desc) {
2149         name = desc->fw_name;
2150     }
2151 
2152     if (name) {
2153         pstrcpy(buf, len, name);
2154     } else {
2155         snprintf(buf, len, "pci%04x,%04x",
2156                  pci_get_word(d->config + PCI_VENDOR_ID),
2157                  pci_get_word(d->config + PCI_DEVICE_ID));
2158     }
2159 
2160     return buf;
2161 }
2162 
2163 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2164 {
2165     PCIDevice *d = (PCIDevice *)dev;
2166     char path[50], name[33];
2167     int off;
2168 
2169     off = snprintf(path, sizeof(path), "%s@%x",
2170                    pci_dev_fw_name(dev, name, sizeof name),
2171                    PCI_SLOT(d->devfn));
2172     if (PCI_FUNC(d->devfn))
2173         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2174     return g_strdup(path);
2175 }
2176 
2177 static char *pcibus_get_dev_path(DeviceState *dev)
2178 {
2179     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2180     PCIDevice *t;
2181     int slot_depth;
2182     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2183      * 00 is added here to make this format compatible with
2184      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2185      * Slot.Function list specifies the slot and function numbers for all
2186      * devices on the path from root to the specific device. */
2187     const char *root_bus_path;
2188     int root_bus_len;
2189     char slot[] = ":SS.F";
2190     int slot_len = sizeof slot - 1 /* For '\0' */;
2191     int path_len;
2192     char *path, *p;
2193     int s;
2194 
2195     root_bus_path = pci_root_bus_path(d);
2196     root_bus_len = strlen(root_bus_path);
2197 
2198     /* Calculate # of slots on path between device and root. */;
2199     slot_depth = 0;
2200     for (t = d; t; t = t->bus->parent_dev) {
2201         ++slot_depth;
2202     }
2203 
2204     path_len = root_bus_len + slot_len * slot_depth;
2205 
2206     /* Allocate memory, fill in the terminating null byte. */
2207     path = g_malloc(path_len + 1 /* For '\0' */);
2208     path[path_len] = '\0';
2209 
2210     memcpy(path, root_bus_path, root_bus_len);
2211 
2212     /* Fill in slot numbers. We walk up from device to root, so need to print
2213      * them in the reverse order, last to first. */
2214     p = path + path_len;
2215     for (t = d; t; t = t->bus->parent_dev) {
2216         p -= slot_len;
2217         s = snprintf(slot, sizeof slot, ":%02x.%x",
2218                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2219         assert(s == slot_len);
2220         memcpy(p, slot, slot_len);
2221     }
2222 
2223     return path;
2224 }
2225 
2226 static int pci_qdev_find_recursive(PCIBus *bus,
2227                                    const char *id, PCIDevice **pdev)
2228 {
2229     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2230     if (!qdev) {
2231         return -ENODEV;
2232     }
2233 
2234     /* roughly check if given qdev is pci device */
2235     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2236         *pdev = PCI_DEVICE(qdev);
2237         return 0;
2238     }
2239     return -EINVAL;
2240 }
2241 
2242 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2243 {
2244     PCIHostState *host_bridge;
2245     int rc = -ENODEV;
2246 
2247     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2248         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2249         if (!tmp) {
2250             rc = 0;
2251             break;
2252         }
2253         if (tmp != -ENODEV) {
2254             rc = tmp;
2255         }
2256     }
2257 
2258     return rc;
2259 }
2260 
2261 MemoryRegion *pci_address_space(PCIDevice *dev)
2262 {
2263     return dev->bus->address_space_mem;
2264 }
2265 
2266 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2267 {
2268     return dev->bus->address_space_io;
2269 }
2270 
2271 static void pci_device_class_init(ObjectClass *klass, void *data)
2272 {
2273     DeviceClass *k = DEVICE_CLASS(klass);
2274     k->init = pci_qdev_init;
2275     k->unplug = pci_unplug_device;
2276     k->exit = pci_unregister_device;
2277     k->bus_type = TYPE_PCI_BUS;
2278     k->props = pci_props;
2279 }
2280 
2281 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2282 {
2283     PCIBus *bus = PCI_BUS(dev->bus);
2284 
2285     if (bus->iommu_fn) {
2286         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2287     }
2288 
2289     if (bus->parent_dev) {
2290         /** We are ignoring the bus master DMA bit of the bridge
2291          *  as it would complicate things such as VFIO for no good reason */
2292         return pci_device_iommu_address_space(bus->parent_dev);
2293     }
2294 
2295     return &address_space_memory;
2296 }
2297 
2298 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2299 {
2300     bus->iommu_fn = fn;
2301     bus->iommu_opaque = opaque;
2302 }
2303 
2304 static void pci_dev_get_w64(PCIBus *b, PCIDevice *dev, void *opaque)
2305 {
2306     Range *range = opaque;
2307     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
2308     uint16_t cmd = pci_get_word(dev->config + PCI_COMMAND);
2309     int i;
2310 
2311     if (!(cmd & PCI_COMMAND_MEMORY)) {
2312         return;
2313     }
2314 
2315     if (pc->is_bridge) {
2316         pcibus_t base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2317         pcibus_t limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
2318 
2319         base = MAX(base, 0x1ULL << 32);
2320 
2321         if (limit >= base) {
2322             Range pref_range;
2323             pref_range.begin = base;
2324             pref_range.end = limit + 1;
2325             range_extend(range, &pref_range);
2326         }
2327     }
2328     for (i = 0; i < PCI_NUM_REGIONS; ++i) {
2329         PCIIORegion *r = &dev->io_regions[i];
2330         Range region_range;
2331 
2332         if (!r->size ||
2333             (r->type & PCI_BASE_ADDRESS_SPACE_IO) ||
2334             !(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64)) {
2335             continue;
2336         }
2337         region_range.begin = pci_bar_address(dev, i, r->type, r->size);
2338         region_range.end = region_range.begin + r->size;
2339 
2340         if (region_range.begin == PCI_BAR_UNMAPPED) {
2341             continue;
2342         }
2343 
2344         region_range.begin = MAX(region_range.begin, 0x1ULL << 32);
2345 
2346         if (region_range.end - 1 >= region_range.begin) {
2347             range_extend(range, &region_range);
2348         }
2349     }
2350 }
2351 
2352 void pci_bus_get_w64_range(PCIBus *bus, Range *range)
2353 {
2354     range->begin = range->end = 0;
2355     pci_for_each_device_under_bus(bus, pci_dev_get_w64, range);
2356 }
2357 
2358 static const TypeInfo pci_device_type_info = {
2359     .name = TYPE_PCI_DEVICE,
2360     .parent = TYPE_DEVICE,
2361     .instance_size = sizeof(PCIDevice),
2362     .abstract = true,
2363     .class_size = sizeof(PCIDeviceClass),
2364     .class_init = pci_device_class_init,
2365 };
2366 
2367 static void pci_register_types(void)
2368 {
2369     type_register_static(&pci_bus_info);
2370     type_register_static(&pcie_bus_info);
2371     type_register_static(&pci_device_type_info);
2372 }
2373 
2374 type_init(pci_register_types)
2375