xref: /openbmc/qemu/hw/pci/pci.c (revision 5accc840)
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_set_irq(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         qemu_set_irq(dev->irq[i], 0);
165     }
166 }
167 
168 /*
169  * This function is called on #RST and FLR.
170  * FLR if PCI_EXP_DEVCTL_BCR_FLR is set
171  */
172 void pci_device_reset(PCIDevice *dev)
173 {
174     int r;
175 
176     qdev_reset_all(&dev->qdev);
177 
178     dev->irq_state = 0;
179     pci_update_irq_status(dev);
180     pci_device_deassert_intx(dev);
181     /* Clear all writable bits */
182     pci_word_test_and_clear_mask(dev->config + PCI_COMMAND,
183                                  pci_get_word(dev->wmask + PCI_COMMAND) |
184                                  pci_get_word(dev->w1cmask + PCI_COMMAND));
185     pci_word_test_and_clear_mask(dev->config + PCI_STATUS,
186                                  pci_get_word(dev->wmask + PCI_STATUS) |
187                                  pci_get_word(dev->w1cmask + PCI_STATUS));
188     dev->config[PCI_CACHE_LINE_SIZE] = 0x0;
189     dev->config[PCI_INTERRUPT_LINE] = 0x0;
190     for (r = 0; r < PCI_NUM_REGIONS; ++r) {
191         PCIIORegion *region = &dev->io_regions[r];
192         if (!region->size) {
193             continue;
194         }
195 
196         if (!(region->type & PCI_BASE_ADDRESS_SPACE_IO) &&
197             region->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
198             pci_set_quad(dev->config + pci_bar(dev, r), region->type);
199         } else {
200             pci_set_long(dev->config + pci_bar(dev, r), region->type);
201         }
202     }
203     pci_update_mappings(dev);
204 
205     msi_reset(dev);
206     msix_reset(dev);
207 }
208 
209 /*
210  * Trigger pci bus reset under a given bus.
211  * To be called on RST# assert.
212  */
213 void pci_bus_reset(PCIBus *bus)
214 {
215     int i;
216 
217     for (i = 0; i < bus->nirq; i++) {
218         bus->irq_count[i] = 0;
219     }
220     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
221         if (bus->devices[i]) {
222             pci_device_reset(bus->devices[i]);
223         }
224     }
225 }
226 
227 static int pcibus_reset(BusState *qbus)
228 {
229     pci_bus_reset(DO_UPCAST(PCIBus, qbus, qbus));
230 
231     /* topology traverse is done by pci_bus_reset().
232        Tell qbus/qdev walker not to traverse the tree */
233     return 1;
234 }
235 
236 static void pci_host_bus_register(PCIBus *bus, DeviceState *parent)
237 {
238     PCIHostState *host_bridge = PCI_HOST_BRIDGE(parent);
239 
240     QLIST_INSERT_HEAD(&pci_host_bridges, host_bridge, next);
241 }
242 
243 PCIBus *pci_find_primary_bus(void)
244 {
245     PCIBus *primary_bus = NULL;
246     PCIHostState *host;
247 
248     QLIST_FOREACH(host, &pci_host_bridges, next) {
249         if (primary_bus) {
250             /* We have multiple root buses, refuse to select a primary */
251             return NULL;
252         }
253         primary_bus = host->bus;
254     }
255 
256     return primary_bus;
257 }
258 
259 PCIBus *pci_device_root_bus(const PCIDevice *d)
260 {
261     PCIBus *bus = d->bus;
262 
263     while ((d = bus->parent_dev) != NULL) {
264         bus = d->bus;
265     }
266 
267     return bus;
268 }
269 
270 const char *pci_root_bus_path(PCIDevice *dev)
271 {
272     PCIBus *rootbus = pci_device_root_bus(dev);
273     PCIHostState *host_bridge = PCI_HOST_BRIDGE(rootbus->qbus.parent);
274     PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_GET_CLASS(host_bridge);
275 
276     assert(!rootbus->parent_dev);
277     assert(host_bridge->bus == rootbus);
278 
279     if (hc->root_bus_path) {
280         return (*hc->root_bus_path)(host_bridge, rootbus);
281     }
282 
283     return rootbus->qbus.name;
284 }
285 
286 static void pci_bus_init(PCIBus *bus, DeviceState *parent,
287                          const char *name,
288                          MemoryRegion *address_space_mem,
289                          MemoryRegion *address_space_io,
290                          uint8_t devfn_min)
291 {
292     assert(PCI_FUNC(devfn_min) == 0);
293     bus->devfn_min = devfn_min;
294     bus->address_space_mem = address_space_mem;
295     bus->address_space_io = address_space_io;
296 
297     /* host bridge */
298     QLIST_INIT(&bus->child);
299 
300     pci_host_bus_register(bus, parent);
301 
302     vmstate_register(NULL, -1, &vmstate_pcibus, bus);
303 }
304 
305 bool pci_bus_is_express(PCIBus *bus)
306 {
307     return object_dynamic_cast(OBJECT(bus), TYPE_PCIE_BUS);
308 }
309 
310 bool pci_bus_is_root(PCIBus *bus)
311 {
312     return !bus->parent_dev;
313 }
314 
315 void pci_bus_new_inplace(PCIBus *bus, size_t bus_size, DeviceState *parent,
316                          const char *name,
317                          MemoryRegion *address_space_mem,
318                          MemoryRegion *address_space_io,
319                          uint8_t devfn_min, const char *typename)
320 {
321     qbus_create_inplace(bus, bus_size, typename, parent, name);
322     pci_bus_init(bus, parent, name, address_space_mem,
323                  address_space_io, devfn_min);
324 }
325 
326 PCIBus *pci_bus_new(DeviceState *parent, const char *name,
327                     MemoryRegion *address_space_mem,
328                     MemoryRegion *address_space_io,
329                     uint8_t devfn_min, const char *typename)
330 {
331     PCIBus *bus;
332 
333     bus = PCI_BUS(qbus_create(typename, parent, name));
334     pci_bus_init(bus, parent, name, address_space_mem,
335                  address_space_io, devfn_min);
336     return bus;
337 }
338 
339 void pci_bus_irqs(PCIBus *bus, pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
340                   void *irq_opaque, int nirq)
341 {
342     bus->set_irq = set_irq;
343     bus->map_irq = map_irq;
344     bus->irq_opaque = irq_opaque;
345     bus->nirq = nirq;
346     bus->irq_count = g_malloc0(nirq * sizeof(bus->irq_count[0]));
347 }
348 
349 void pci_bus_hotplug(PCIBus *bus, pci_hotplug_fn hotplug, DeviceState *qdev)
350 {
351     bus->qbus.allow_hotplug = 1;
352     bus->hotplug = hotplug;
353     bus->hotplug_qdev = qdev;
354 }
355 
356 PCIBus *pci_register_bus(DeviceState *parent, const char *name,
357                          pci_set_irq_fn set_irq, pci_map_irq_fn map_irq,
358                          void *irq_opaque,
359                          MemoryRegion *address_space_mem,
360                          MemoryRegion *address_space_io,
361                          uint8_t devfn_min, int nirq, const char *typename)
362 {
363     PCIBus *bus;
364 
365     bus = pci_bus_new(parent, name, address_space_mem,
366                       address_space_io, devfn_min, typename);
367     pci_bus_irqs(bus, set_irq, map_irq, irq_opaque, nirq);
368     return bus;
369 }
370 
371 int pci_bus_num(PCIBus *s)
372 {
373     if (pci_bus_is_root(s))
374         return 0;       /* pci host bridge */
375     return s->parent_dev->config[PCI_SECONDARY_BUS];
376 }
377 
378 static int get_pci_config_device(QEMUFile *f, void *pv, size_t size)
379 {
380     PCIDevice *s = container_of(pv, PCIDevice, config);
381     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(s);
382     uint8_t *config;
383     int i;
384 
385     assert(size == pci_config_size(s));
386     config = g_malloc(size);
387 
388     qemu_get_buffer(f, config, size);
389     for (i = 0; i < size; ++i) {
390         if ((config[i] ^ s->config[i]) &
391             s->cmask[i] & ~s->wmask[i] & ~s->w1cmask[i]) {
392             g_free(config);
393             return -EINVAL;
394         }
395     }
396     memcpy(s->config, config, size);
397 
398     pci_update_mappings(s);
399     if (pc->is_bridge) {
400         PCIBridge *b = PCI_BRIDGE(s);
401         pci_bridge_update_mappings(b);
402     }
403 
404     memory_region_set_enabled(&s->bus_master_enable_region,
405                               pci_get_word(s->config + PCI_COMMAND)
406                               & PCI_COMMAND_MASTER);
407 
408     g_free(config);
409     return 0;
410 }
411 
412 /* just put buffer */
413 static void put_pci_config_device(QEMUFile *f, void *pv, size_t size)
414 {
415     const uint8_t **v = pv;
416     assert(size == pci_config_size(container_of(pv, PCIDevice, config)));
417     qemu_put_buffer(f, *v, size);
418 }
419 
420 static VMStateInfo vmstate_info_pci_config = {
421     .name = "pci config",
422     .get  = get_pci_config_device,
423     .put  = put_pci_config_device,
424 };
425 
426 static int get_pci_irq_state(QEMUFile *f, void *pv, size_t size)
427 {
428     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
429     uint32_t irq_state[PCI_NUM_PINS];
430     int i;
431     for (i = 0; i < PCI_NUM_PINS; ++i) {
432         irq_state[i] = qemu_get_be32(f);
433         if (irq_state[i] != 0x1 && irq_state[i] != 0) {
434             fprintf(stderr, "irq state %d: must be 0 or 1.\n",
435                     irq_state[i]);
436             return -EINVAL;
437         }
438     }
439 
440     for (i = 0; i < PCI_NUM_PINS; ++i) {
441         pci_set_irq_state(s, i, irq_state[i]);
442     }
443 
444     return 0;
445 }
446 
447 static void put_pci_irq_state(QEMUFile *f, void *pv, size_t size)
448 {
449     int i;
450     PCIDevice *s = container_of(pv, PCIDevice, irq_state);
451 
452     for (i = 0; i < PCI_NUM_PINS; ++i) {
453         qemu_put_be32(f, pci_irq_state(s, i));
454     }
455 }
456 
457 static VMStateInfo vmstate_info_pci_irq_state = {
458     .name = "pci irq state",
459     .get  = get_pci_irq_state,
460     .put  = put_pci_irq_state,
461 };
462 
463 const VMStateDescription vmstate_pci_device = {
464     .name = "PCIDevice",
465     .version_id = 2,
466     .minimum_version_id = 1,
467     .minimum_version_id_old = 1,
468     .fields      = (VMStateField []) {
469         VMSTATE_INT32_LE(version_id, PCIDevice),
470         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
471                                    vmstate_info_pci_config,
472                                    PCI_CONFIG_SPACE_SIZE),
473         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
474 				   vmstate_info_pci_irq_state,
475 				   PCI_NUM_PINS * sizeof(int32_t)),
476         VMSTATE_END_OF_LIST()
477     }
478 };
479 
480 const VMStateDescription vmstate_pcie_device = {
481     .name = "PCIEDevice",
482     .version_id = 2,
483     .minimum_version_id = 1,
484     .minimum_version_id_old = 1,
485     .fields      = (VMStateField []) {
486         VMSTATE_INT32_LE(version_id, PCIDevice),
487         VMSTATE_BUFFER_UNSAFE_INFO(config, PCIDevice, 0,
488                                    vmstate_info_pci_config,
489                                    PCIE_CONFIG_SPACE_SIZE),
490         VMSTATE_BUFFER_UNSAFE_INFO(irq_state, PCIDevice, 2,
491 				   vmstate_info_pci_irq_state,
492 				   PCI_NUM_PINS * sizeof(int32_t)),
493         VMSTATE_END_OF_LIST()
494     }
495 };
496 
497 static inline const VMStateDescription *pci_get_vmstate(PCIDevice *s)
498 {
499     return pci_is_express(s) ? &vmstate_pcie_device : &vmstate_pci_device;
500 }
501 
502 void pci_device_save(PCIDevice *s, QEMUFile *f)
503 {
504     /* Clear interrupt status bit: it is implicit
505      * in irq_state which we are saving.
506      * This makes us compatible with old devices
507      * which never set or clear this bit. */
508     s->config[PCI_STATUS] &= ~PCI_STATUS_INTERRUPT;
509     vmstate_save_state(f, pci_get_vmstate(s), s);
510     /* Restore the interrupt status bit. */
511     pci_update_irq_status(s);
512 }
513 
514 int pci_device_load(PCIDevice *s, QEMUFile *f)
515 {
516     int ret;
517     ret = vmstate_load_state(f, pci_get_vmstate(s), s, s->version_id);
518     /* Restore the interrupt status bit. */
519     pci_update_irq_status(s);
520     return ret;
521 }
522 
523 static void pci_set_default_subsystem_id(PCIDevice *pci_dev)
524 {
525     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
526                  pci_default_sub_vendor_id);
527     pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
528                  pci_default_sub_device_id);
529 }
530 
531 /*
532  * Parse [[<domain>:]<bus>:]<slot>, return -1 on error if funcp == NULL
533  *       [[<domain>:]<bus>:]<slot>.<func>, return -1 on error
534  */
535 int pci_parse_devaddr(const char *addr, int *domp, int *busp,
536                       unsigned int *slotp, unsigned int *funcp)
537 {
538     const char *p;
539     char *e;
540     unsigned long val;
541     unsigned long dom = 0, bus = 0;
542     unsigned int slot = 0;
543     unsigned int func = 0;
544 
545     p = addr;
546     val = strtoul(p, &e, 16);
547     if (e == p)
548 	return -1;
549     if (*e == ':') {
550 	bus = val;
551 	p = e + 1;
552 	val = strtoul(p, &e, 16);
553 	if (e == p)
554 	    return -1;
555 	if (*e == ':') {
556 	    dom = bus;
557 	    bus = val;
558 	    p = e + 1;
559 	    val = strtoul(p, &e, 16);
560 	    if (e == p)
561 		return -1;
562 	}
563     }
564 
565     slot = val;
566 
567     if (funcp != NULL) {
568         if (*e != '.')
569             return -1;
570 
571         p = e + 1;
572         val = strtoul(p, &e, 16);
573         if (e == p)
574             return -1;
575 
576         func = val;
577     }
578 
579     /* if funcp == NULL func is 0 */
580     if (dom > 0xffff || bus > 0xff || slot > 0x1f || func > 7)
581 	return -1;
582 
583     if (*e)
584 	return -1;
585 
586     *domp = dom;
587     *busp = bus;
588     *slotp = slot;
589     if (funcp != NULL)
590         *funcp = func;
591     return 0;
592 }
593 
594 PCIBus *pci_get_bus_devfn(int *devfnp, PCIBus *root, const char *devaddr)
595 {
596     int dom, bus;
597     unsigned slot;
598 
599     assert(!root->parent_dev);
600 
601     if (!root) {
602         fprintf(stderr, "No primary PCI bus\n");
603         return NULL;
604     }
605 
606     if (!devaddr) {
607         *devfnp = -1;
608         return pci_find_bus_nr(root, 0);
609     }
610 
611     if (pci_parse_devaddr(devaddr, &dom, &bus, &slot, NULL) < 0) {
612         return NULL;
613     }
614 
615     if (dom != 0) {
616         fprintf(stderr, "No support for non-zero PCI domains\n");
617         return NULL;
618     }
619 
620     *devfnp = PCI_DEVFN(slot, 0);
621     return pci_find_bus_nr(root, bus);
622 }
623 
624 static void pci_init_cmask(PCIDevice *dev)
625 {
626     pci_set_word(dev->cmask + PCI_VENDOR_ID, 0xffff);
627     pci_set_word(dev->cmask + PCI_DEVICE_ID, 0xffff);
628     dev->cmask[PCI_STATUS] = PCI_STATUS_CAP_LIST;
629     dev->cmask[PCI_REVISION_ID] = 0xff;
630     dev->cmask[PCI_CLASS_PROG] = 0xff;
631     pci_set_word(dev->cmask + PCI_CLASS_DEVICE, 0xffff);
632     dev->cmask[PCI_HEADER_TYPE] = 0xff;
633     dev->cmask[PCI_CAPABILITY_LIST] = 0xff;
634 }
635 
636 static void pci_init_wmask(PCIDevice *dev)
637 {
638     int config_size = pci_config_size(dev);
639 
640     dev->wmask[PCI_CACHE_LINE_SIZE] = 0xff;
641     dev->wmask[PCI_INTERRUPT_LINE] = 0xff;
642     pci_set_word(dev->wmask + PCI_COMMAND,
643                  PCI_COMMAND_IO | PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER |
644                  PCI_COMMAND_INTX_DISABLE);
645     if (dev->cap_present & QEMU_PCI_CAP_SERR) {
646         pci_word_test_and_set_mask(dev->wmask + PCI_COMMAND, PCI_COMMAND_SERR);
647     }
648 
649     memset(dev->wmask + PCI_CONFIG_HEADER_SIZE, 0xff,
650            config_size - PCI_CONFIG_HEADER_SIZE);
651 }
652 
653 static void pci_init_w1cmask(PCIDevice *dev)
654 {
655     /*
656      * Note: It's okay to set w1cmask even for readonly bits as
657      * long as their value is hardwired to 0.
658      */
659     pci_set_word(dev->w1cmask + PCI_STATUS,
660                  PCI_STATUS_PARITY | PCI_STATUS_SIG_TARGET_ABORT |
661                  PCI_STATUS_REC_TARGET_ABORT | PCI_STATUS_REC_MASTER_ABORT |
662                  PCI_STATUS_SIG_SYSTEM_ERROR | PCI_STATUS_DETECTED_PARITY);
663 }
664 
665 static void pci_init_mask_bridge(PCIDevice *d)
666 {
667     /* PCI_PRIMARY_BUS, PCI_SECONDARY_BUS, PCI_SUBORDINATE_BUS and
668        PCI_SEC_LETENCY_TIMER */
669     memset(d->wmask + PCI_PRIMARY_BUS, 0xff, 4);
670 
671     /* base and limit */
672     d->wmask[PCI_IO_BASE] = PCI_IO_RANGE_MASK & 0xff;
673     d->wmask[PCI_IO_LIMIT] = PCI_IO_RANGE_MASK & 0xff;
674     pci_set_word(d->wmask + PCI_MEMORY_BASE,
675                  PCI_MEMORY_RANGE_MASK & 0xffff);
676     pci_set_word(d->wmask + PCI_MEMORY_LIMIT,
677                  PCI_MEMORY_RANGE_MASK & 0xffff);
678     pci_set_word(d->wmask + PCI_PREF_MEMORY_BASE,
679                  PCI_PREF_RANGE_MASK & 0xffff);
680     pci_set_word(d->wmask + PCI_PREF_MEMORY_LIMIT,
681                  PCI_PREF_RANGE_MASK & 0xffff);
682 
683     /* PCI_PREF_BASE_UPPER32 and PCI_PREF_LIMIT_UPPER32 */
684     memset(d->wmask + PCI_PREF_BASE_UPPER32, 0xff, 8);
685 
686     /* Supported memory and i/o types */
687     d->config[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_16;
688     d->config[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_16;
689     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_BASE,
690                                PCI_PREF_RANGE_TYPE_64);
691     pci_word_test_and_set_mask(d->config + PCI_PREF_MEMORY_LIMIT,
692                                PCI_PREF_RANGE_TYPE_64);
693 
694     /*
695      * TODO: Bridges default to 10-bit VGA decoding but we currently only
696      * implement 16-bit decoding (no alias support).
697      */
698     pci_set_word(d->wmask + PCI_BRIDGE_CONTROL,
699                  PCI_BRIDGE_CTL_PARITY |
700                  PCI_BRIDGE_CTL_SERR |
701                  PCI_BRIDGE_CTL_ISA |
702                  PCI_BRIDGE_CTL_VGA |
703                  PCI_BRIDGE_CTL_VGA_16BIT |
704                  PCI_BRIDGE_CTL_MASTER_ABORT |
705                  PCI_BRIDGE_CTL_BUS_RESET |
706                  PCI_BRIDGE_CTL_FAST_BACK |
707                  PCI_BRIDGE_CTL_DISCARD |
708                  PCI_BRIDGE_CTL_SEC_DISCARD |
709                  PCI_BRIDGE_CTL_DISCARD_SERR);
710     /* Below does not do anything as we never set this bit, put here for
711      * completeness. */
712     pci_set_word(d->w1cmask + PCI_BRIDGE_CONTROL,
713                  PCI_BRIDGE_CTL_DISCARD_STATUS);
714     d->cmask[PCI_IO_BASE] |= PCI_IO_RANGE_TYPE_MASK;
715     d->cmask[PCI_IO_LIMIT] |= PCI_IO_RANGE_TYPE_MASK;
716     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_BASE,
717                                PCI_PREF_RANGE_TYPE_MASK);
718     pci_word_test_and_set_mask(d->cmask + PCI_PREF_MEMORY_LIMIT,
719                                PCI_PREF_RANGE_TYPE_MASK);
720 }
721 
722 static int pci_init_multifunction(PCIBus *bus, PCIDevice *dev)
723 {
724     uint8_t slot = PCI_SLOT(dev->devfn);
725     uint8_t func;
726 
727     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
728         dev->config[PCI_HEADER_TYPE] |= PCI_HEADER_TYPE_MULTI_FUNCTION;
729     }
730 
731     /*
732      * multifunction bit is interpreted in two ways as follows.
733      *   - all functions must set the bit to 1.
734      *     Example: Intel X53
735      *   - function 0 must set the bit, but the rest function (> 0)
736      *     is allowed to leave the bit to 0.
737      *     Example: PIIX3(also in qemu), PIIX4(also in qemu), ICH10,
738      *
739      * So OS (at least Linux) checks the bit of only function 0,
740      * and doesn't see the bit of function > 0.
741      *
742      * The below check allows both interpretation.
743      */
744     if (PCI_FUNC(dev->devfn)) {
745         PCIDevice *f0 = bus->devices[PCI_DEVFN(slot, 0)];
746         if (f0 && !(f0->cap_present & QEMU_PCI_CAP_MULTIFUNCTION)) {
747             /* function 0 should set multifunction bit */
748             error_report("PCI: single function device can't be populated "
749                          "in function %x.%x", slot, PCI_FUNC(dev->devfn));
750             return -1;
751         }
752         return 0;
753     }
754 
755     if (dev->cap_present & QEMU_PCI_CAP_MULTIFUNCTION) {
756         return 0;
757     }
758     /* function 0 indicates single function, so function > 0 must be NULL */
759     for (func = 1; func < PCI_FUNC_MAX; ++func) {
760         if (bus->devices[PCI_DEVFN(slot, func)]) {
761             error_report("PCI: %x.0 indicates single function, "
762                          "but %x.%x is already populated.",
763                          slot, slot, func);
764             return -1;
765         }
766     }
767     return 0;
768 }
769 
770 static void pci_config_alloc(PCIDevice *pci_dev)
771 {
772     int config_size = pci_config_size(pci_dev);
773 
774     pci_dev->config = g_malloc0(config_size);
775     pci_dev->cmask = g_malloc0(config_size);
776     pci_dev->wmask = g_malloc0(config_size);
777     pci_dev->w1cmask = g_malloc0(config_size);
778     pci_dev->used = g_malloc0(config_size);
779 }
780 
781 static void pci_config_free(PCIDevice *pci_dev)
782 {
783     g_free(pci_dev->config);
784     g_free(pci_dev->cmask);
785     g_free(pci_dev->wmask);
786     g_free(pci_dev->w1cmask);
787     g_free(pci_dev->used);
788 }
789 
790 /* -1 for devfn means auto assign */
791 static PCIDevice *do_pci_register_device(PCIDevice *pci_dev, PCIBus *bus,
792                                          const char *name, int devfn)
793 {
794     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
795     PCIConfigReadFunc *config_read = pc->config_read;
796     PCIConfigWriteFunc *config_write = pc->config_write;
797     AddressSpace *dma_as;
798 
799     if (devfn < 0) {
800         for(devfn = bus->devfn_min ; devfn < ARRAY_SIZE(bus->devices);
801             devfn += PCI_FUNC_MAX) {
802             if (!bus->devices[devfn])
803                 goto found;
804         }
805         error_report("PCI: no slot/function available for %s, all in use", name);
806         return NULL;
807     found: ;
808     } else if (bus->devices[devfn]) {
809         error_report("PCI: slot %d function %d not available for %s, in use by %s",
810                      PCI_SLOT(devfn), PCI_FUNC(devfn), name, bus->devices[devfn]->name);
811         return NULL;
812     }
813 
814     pci_dev->bus = bus;
815     dma_as = pci_device_iommu_address_space(pci_dev);
816 
817     memory_region_init_alias(&pci_dev->bus_master_enable_region,
818                              OBJECT(pci_dev), "bus master",
819                              dma_as->root, 0, memory_region_size(dma_as->root));
820     memory_region_set_enabled(&pci_dev->bus_master_enable_region, false);
821     address_space_init(&pci_dev->bus_master_as, &pci_dev->bus_master_enable_region,
822                        name);
823 
824     pci_dev->devfn = devfn;
825     pstrcpy(pci_dev->name, sizeof(pci_dev->name), name);
826     pci_dev->irq_state = 0;
827     pci_config_alloc(pci_dev);
828 
829     pci_config_set_vendor_id(pci_dev->config, pc->vendor_id);
830     pci_config_set_device_id(pci_dev->config, pc->device_id);
831     pci_config_set_revision(pci_dev->config, pc->revision);
832     pci_config_set_class(pci_dev->config, pc->class_id);
833 
834     if (!pc->is_bridge) {
835         if (pc->subsystem_vendor_id || pc->subsystem_id) {
836             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_VENDOR_ID,
837                          pc->subsystem_vendor_id);
838             pci_set_word(pci_dev->config + PCI_SUBSYSTEM_ID,
839                          pc->subsystem_id);
840         } else {
841             pci_set_default_subsystem_id(pci_dev);
842         }
843     } else {
844         /* subsystem_vendor_id/subsystem_id are only for header type 0 */
845         assert(!pc->subsystem_vendor_id);
846         assert(!pc->subsystem_id);
847     }
848     pci_init_cmask(pci_dev);
849     pci_init_wmask(pci_dev);
850     pci_init_w1cmask(pci_dev);
851     if (pc->is_bridge) {
852         pci_init_mask_bridge(pci_dev);
853     }
854     if (pci_init_multifunction(bus, pci_dev)) {
855         pci_config_free(pci_dev);
856         return NULL;
857     }
858 
859     if (!config_read)
860         config_read = pci_default_read_config;
861     if (!config_write)
862         config_write = pci_default_write_config;
863     pci_dev->config_read = config_read;
864     pci_dev->config_write = config_write;
865     bus->devices[devfn] = pci_dev;
866     pci_dev->irq = qemu_allocate_irqs(pci_set_irq, pci_dev, PCI_NUM_PINS);
867     pci_dev->version_id = 2; /* Current pci device vmstate version */
868     return pci_dev;
869 }
870 
871 static void do_pci_unregister_device(PCIDevice *pci_dev)
872 {
873     qemu_free_irqs(pci_dev->irq);
874     pci_dev->bus->devices[pci_dev->devfn] = NULL;
875     pci_config_free(pci_dev);
876 
877     address_space_destroy(&pci_dev->bus_master_as);
878     memory_region_destroy(&pci_dev->bus_master_enable_region);
879 }
880 
881 static void pci_unregister_io_regions(PCIDevice *pci_dev)
882 {
883     PCIIORegion *r;
884     int i;
885 
886     for(i = 0; i < PCI_NUM_REGIONS; i++) {
887         r = &pci_dev->io_regions[i];
888         if (!r->size || r->addr == PCI_BAR_UNMAPPED)
889             continue;
890         memory_region_del_subregion(r->address_space, r->memory);
891     }
892 
893     pci_unregister_vga(pci_dev);
894 }
895 
896 static int pci_unregister_device(DeviceState *dev)
897 {
898     PCIDevice *pci_dev = PCI_DEVICE(dev);
899     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
900 
901     pci_unregister_io_regions(pci_dev);
902     pci_del_option_rom(pci_dev);
903 
904     if (pc->exit) {
905         pc->exit(pci_dev);
906     }
907 
908     do_pci_unregister_device(pci_dev);
909     return 0;
910 }
911 
912 void pci_register_bar(PCIDevice *pci_dev, int region_num,
913                       uint8_t type, MemoryRegion *memory)
914 {
915     PCIIORegion *r;
916     uint32_t addr;
917     uint64_t wmask;
918     pcibus_t size = memory_region_size(memory);
919 
920     assert(region_num >= 0);
921     assert(region_num < PCI_NUM_REGIONS);
922     if (size & (size-1)) {
923         fprintf(stderr, "ERROR: PCI region size must be pow2 "
924                     "type=0x%x, size=0x%"FMT_PCIBUS"\n", type, size);
925         exit(1);
926     }
927 
928     r = &pci_dev->io_regions[region_num];
929     r->addr = PCI_BAR_UNMAPPED;
930     r->size = size;
931     r->type = type;
932     r->memory = NULL;
933 
934     wmask = ~(size - 1);
935     addr = pci_bar(pci_dev, region_num);
936     if (region_num == PCI_ROM_SLOT) {
937         /* ROM enable bit is writable */
938         wmask |= PCI_ROM_ADDRESS_ENABLE;
939     }
940     pci_set_long(pci_dev->config + addr, type);
941     if (!(r->type & PCI_BASE_ADDRESS_SPACE_IO) &&
942         r->type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
943         pci_set_quad(pci_dev->wmask + addr, wmask);
944         pci_set_quad(pci_dev->cmask + addr, ~0ULL);
945     } else {
946         pci_set_long(pci_dev->wmask + addr, wmask & 0xffffffff);
947         pci_set_long(pci_dev->cmask + addr, 0xffffffff);
948     }
949     pci_dev->io_regions[region_num].memory = memory;
950     pci_dev->io_regions[region_num].address_space
951         = type & PCI_BASE_ADDRESS_SPACE_IO
952         ? pci_dev->bus->address_space_io
953         : pci_dev->bus->address_space_mem;
954 }
955 
956 static void pci_update_vga(PCIDevice *pci_dev)
957 {
958     uint16_t cmd;
959 
960     if (!pci_dev->has_vga) {
961         return;
962     }
963 
964     cmd = pci_get_word(pci_dev->config + PCI_COMMAND);
965 
966     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_MEM],
967                               cmd & PCI_COMMAND_MEMORY);
968     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO],
969                               cmd & PCI_COMMAND_IO);
970     memory_region_set_enabled(pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI],
971                               cmd & PCI_COMMAND_IO);
972 }
973 
974 void pci_register_vga(PCIDevice *pci_dev, MemoryRegion *mem,
975                       MemoryRegion *io_lo, MemoryRegion *io_hi)
976 {
977     assert(!pci_dev->has_vga);
978 
979     assert(memory_region_size(mem) == QEMU_PCI_VGA_MEM_SIZE);
980     pci_dev->vga_regions[QEMU_PCI_VGA_MEM] = mem;
981     memory_region_add_subregion_overlap(pci_dev->bus->address_space_mem,
982                                         QEMU_PCI_VGA_MEM_BASE, mem, 1);
983 
984     assert(memory_region_size(io_lo) == QEMU_PCI_VGA_IO_LO_SIZE);
985     pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO] = io_lo;
986     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
987                                         QEMU_PCI_VGA_IO_LO_BASE, io_lo, 1);
988 
989     assert(memory_region_size(io_hi) == QEMU_PCI_VGA_IO_HI_SIZE);
990     pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI] = io_hi;
991     memory_region_add_subregion_overlap(pci_dev->bus->address_space_io,
992                                         QEMU_PCI_VGA_IO_HI_BASE, io_hi, 1);
993     pci_dev->has_vga = true;
994 
995     pci_update_vga(pci_dev);
996 }
997 
998 void pci_unregister_vga(PCIDevice *pci_dev)
999 {
1000     if (!pci_dev->has_vga) {
1001         return;
1002     }
1003 
1004     memory_region_del_subregion(pci_dev->bus->address_space_mem,
1005                                 pci_dev->vga_regions[QEMU_PCI_VGA_MEM]);
1006     memory_region_del_subregion(pci_dev->bus->address_space_io,
1007                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_LO]);
1008     memory_region_del_subregion(pci_dev->bus->address_space_io,
1009                                 pci_dev->vga_regions[QEMU_PCI_VGA_IO_HI]);
1010     pci_dev->has_vga = false;
1011 }
1012 
1013 pcibus_t pci_get_bar_addr(PCIDevice *pci_dev, int region_num)
1014 {
1015     return pci_dev->io_regions[region_num].addr;
1016 }
1017 
1018 static pcibus_t pci_bar_address(PCIDevice *d,
1019 				int reg, uint8_t type, pcibus_t size)
1020 {
1021     pcibus_t new_addr, last_addr;
1022     int bar = pci_bar(d, reg);
1023     uint16_t cmd = pci_get_word(d->config + PCI_COMMAND);
1024 
1025     if (type & PCI_BASE_ADDRESS_SPACE_IO) {
1026         if (!(cmd & PCI_COMMAND_IO)) {
1027             return PCI_BAR_UNMAPPED;
1028         }
1029         new_addr = pci_get_long(d->config + bar) & ~(size - 1);
1030         last_addr = new_addr + size - 1;
1031         /* NOTE: we have only 64K ioports on PC */
1032         if (last_addr <= new_addr || new_addr == 0 || last_addr > UINT16_MAX) {
1033             return PCI_BAR_UNMAPPED;
1034         }
1035         return new_addr;
1036     }
1037 
1038     if (!(cmd & PCI_COMMAND_MEMORY)) {
1039         return PCI_BAR_UNMAPPED;
1040     }
1041     if (type & PCI_BASE_ADDRESS_MEM_TYPE_64) {
1042         new_addr = pci_get_quad(d->config + bar);
1043     } else {
1044         new_addr = pci_get_long(d->config + bar);
1045     }
1046     /* the ROM slot has a specific enable bit */
1047     if (reg == PCI_ROM_SLOT && !(new_addr & PCI_ROM_ADDRESS_ENABLE)) {
1048         return PCI_BAR_UNMAPPED;
1049     }
1050     new_addr &= ~(size - 1);
1051     last_addr = new_addr + size - 1;
1052     /* NOTE: we do not support wrapping */
1053     /* XXX: as we cannot support really dynamic
1054        mappings, we handle specific values as invalid
1055        mappings. */
1056     if (last_addr <= new_addr || new_addr == 0 ||
1057         last_addr == PCI_BAR_UNMAPPED) {
1058         return PCI_BAR_UNMAPPED;
1059     }
1060 
1061     /* Now pcibus_t is 64bit.
1062      * Check if 32 bit BAR wraps around explicitly.
1063      * Without this, PC ide doesn't work well.
1064      * TODO: remove this work around.
1065      */
1066     if  (!(type & PCI_BASE_ADDRESS_MEM_TYPE_64) && last_addr >= UINT32_MAX) {
1067         return PCI_BAR_UNMAPPED;
1068     }
1069 
1070     /*
1071      * OS is allowed to set BAR beyond its addressable
1072      * bits. For example, 32 bit OS can set 64bit bar
1073      * to >4G. Check it. TODO: we might need to support
1074      * it in the future for e.g. PAE.
1075      */
1076     if (last_addr >= HWADDR_MAX) {
1077         return PCI_BAR_UNMAPPED;
1078     }
1079 
1080     return new_addr;
1081 }
1082 
1083 static void pci_update_mappings(PCIDevice *d)
1084 {
1085     PCIIORegion *r;
1086     int i;
1087     pcibus_t new_addr;
1088 
1089     for(i = 0; i < PCI_NUM_REGIONS; i++) {
1090         r = &d->io_regions[i];
1091 
1092         /* this region isn't registered */
1093         if (!r->size)
1094             continue;
1095 
1096         new_addr = pci_bar_address(d, i, r->type, r->size);
1097 
1098         /* This bar isn't changed */
1099         if (new_addr == r->addr)
1100             continue;
1101 
1102         /* now do the real mapping */
1103         if (r->addr != PCI_BAR_UNMAPPED) {
1104             memory_region_del_subregion(r->address_space, r->memory);
1105         }
1106         r->addr = new_addr;
1107         if (r->addr != PCI_BAR_UNMAPPED) {
1108             memory_region_add_subregion_overlap(r->address_space,
1109                                                 r->addr, r->memory, 1);
1110         }
1111     }
1112 
1113     pci_update_vga(d);
1114 }
1115 
1116 static inline int pci_irq_disabled(PCIDevice *d)
1117 {
1118     return pci_get_word(d->config + PCI_COMMAND) & PCI_COMMAND_INTX_DISABLE;
1119 }
1120 
1121 /* Called after interrupt disabled field update in config space,
1122  * assert/deassert interrupts if necessary.
1123  * Gets original interrupt disable bit value (before update). */
1124 static void pci_update_irq_disabled(PCIDevice *d, int was_irq_disabled)
1125 {
1126     int i, disabled = pci_irq_disabled(d);
1127     if (disabled == was_irq_disabled)
1128         return;
1129     for (i = 0; i < PCI_NUM_PINS; ++i) {
1130         int state = pci_irq_state(d, i);
1131         pci_change_irq_level(d, i, disabled ? -state : state);
1132     }
1133 }
1134 
1135 uint32_t pci_default_read_config(PCIDevice *d,
1136                                  uint32_t address, int len)
1137 {
1138     uint32_t val = 0;
1139 
1140     memcpy(&val, d->config + address, len);
1141     return le32_to_cpu(val);
1142 }
1143 
1144 void pci_default_write_config(PCIDevice *d, uint32_t addr, uint32_t val, int l)
1145 {
1146     int i, was_irq_disabled = pci_irq_disabled(d);
1147 
1148     for (i = 0; i < l; val >>= 8, ++i) {
1149         uint8_t wmask = d->wmask[addr + i];
1150         uint8_t w1cmask = d->w1cmask[addr + i];
1151         assert(!(wmask & w1cmask));
1152         d->config[addr + i] = (d->config[addr + i] & ~wmask) | (val & wmask);
1153         d->config[addr + i] &= ~(val & w1cmask); /* W1C: Write 1 to Clear */
1154     }
1155     if (ranges_overlap(addr, l, PCI_BASE_ADDRESS_0, 24) ||
1156         ranges_overlap(addr, l, PCI_ROM_ADDRESS, 4) ||
1157         ranges_overlap(addr, l, PCI_ROM_ADDRESS1, 4) ||
1158         range_covers_byte(addr, l, PCI_COMMAND))
1159         pci_update_mappings(d);
1160 
1161     if (range_covers_byte(addr, l, PCI_COMMAND)) {
1162         pci_update_irq_disabled(d, was_irq_disabled);
1163         memory_region_set_enabled(&d->bus_master_enable_region,
1164                                   pci_get_word(d->config + PCI_COMMAND)
1165                                     & PCI_COMMAND_MASTER);
1166     }
1167 
1168     msi_write_config(d, addr, val, l);
1169     msix_write_config(d, addr, val, l);
1170 }
1171 
1172 /***********************************************************/
1173 /* generic PCI irq support */
1174 
1175 /* 0 <= irq_num <= 3. level must be 0 or 1 */
1176 static void pci_set_irq(void *opaque, int irq_num, int level)
1177 {
1178     PCIDevice *pci_dev = opaque;
1179     int change;
1180 
1181     change = level - pci_irq_state(pci_dev, irq_num);
1182     if (!change)
1183         return;
1184 
1185     pci_set_irq_state(pci_dev, irq_num, level);
1186     pci_update_irq_status(pci_dev);
1187     if (pci_irq_disabled(pci_dev))
1188         return;
1189     pci_change_irq_level(pci_dev, irq_num, change);
1190 }
1191 
1192 /* Special hooks used by device assignment */
1193 void pci_bus_set_route_irq_fn(PCIBus *bus, pci_route_irq_fn route_intx_to_irq)
1194 {
1195     assert(pci_bus_is_root(bus));
1196     bus->route_intx_to_irq = route_intx_to_irq;
1197 }
1198 
1199 PCIINTxRoute pci_device_route_intx_to_irq(PCIDevice *dev, int pin)
1200 {
1201     PCIBus *bus;
1202 
1203     do {
1204          bus = dev->bus;
1205          pin = bus->map_irq(dev, pin);
1206          dev = bus->parent_dev;
1207     } while (dev);
1208 
1209     if (!bus->route_intx_to_irq) {
1210         error_report("PCI: Bug - unimplemented PCI INTx routing (%s)",
1211                      object_get_typename(OBJECT(bus->qbus.parent)));
1212         return (PCIINTxRoute) { PCI_INTX_DISABLED, -1 };
1213     }
1214 
1215     return bus->route_intx_to_irq(bus->irq_opaque, pin);
1216 }
1217 
1218 bool pci_intx_route_changed(PCIINTxRoute *old, PCIINTxRoute *new)
1219 {
1220     return old->mode != new->mode || old->irq != new->irq;
1221 }
1222 
1223 void pci_bus_fire_intx_routing_notifier(PCIBus *bus)
1224 {
1225     PCIDevice *dev;
1226     PCIBus *sec;
1227     int i;
1228 
1229     for (i = 0; i < ARRAY_SIZE(bus->devices); ++i) {
1230         dev = bus->devices[i];
1231         if (dev && dev->intx_routing_notifier) {
1232             dev->intx_routing_notifier(dev);
1233         }
1234     }
1235 
1236     QLIST_FOREACH(sec, &bus->child, sibling) {
1237         pci_bus_fire_intx_routing_notifier(sec);
1238     }
1239 }
1240 
1241 void pci_device_set_intx_routing_notifier(PCIDevice *dev,
1242                                           PCIINTxRoutingNotifier notifier)
1243 {
1244     dev->intx_routing_notifier = notifier;
1245 }
1246 
1247 /*
1248  * PCI-to-PCI bridge specification
1249  * 9.1: Interrupt routing. Table 9-1
1250  *
1251  * the PCI Express Base Specification, Revision 2.1
1252  * 2.2.8.1: INTx interrutp signaling - Rules
1253  *          the Implementation Note
1254  *          Table 2-20
1255  */
1256 /*
1257  * 0 <= pin <= 3 0 = INTA, 1 = INTB, 2 = INTC, 3 = INTD
1258  * 0-origin unlike PCI interrupt pin register.
1259  */
1260 int pci_swizzle_map_irq_fn(PCIDevice *pci_dev, int pin)
1261 {
1262     return (pin + PCI_SLOT(pci_dev->devfn)) % PCI_NUM_PINS;
1263 }
1264 
1265 /***********************************************************/
1266 /* monitor info on PCI */
1267 
1268 typedef struct {
1269     uint16_t class;
1270     const char *desc;
1271     const char *fw_name;
1272     uint16_t fw_ign_bits;
1273 } pci_class_desc;
1274 
1275 static const pci_class_desc pci_class_descriptions[] =
1276 {
1277     { 0x0001, "VGA controller", "display"},
1278     { 0x0100, "SCSI controller", "scsi"},
1279     { 0x0101, "IDE controller", "ide"},
1280     { 0x0102, "Floppy controller", "fdc"},
1281     { 0x0103, "IPI controller", "ipi"},
1282     { 0x0104, "RAID controller", "raid"},
1283     { 0x0106, "SATA controller"},
1284     { 0x0107, "SAS controller"},
1285     { 0x0180, "Storage controller"},
1286     { 0x0200, "Ethernet controller", "ethernet"},
1287     { 0x0201, "Token Ring controller", "token-ring"},
1288     { 0x0202, "FDDI controller", "fddi"},
1289     { 0x0203, "ATM controller", "atm"},
1290     { 0x0280, "Network controller"},
1291     { 0x0300, "VGA controller", "display", 0x00ff},
1292     { 0x0301, "XGA controller"},
1293     { 0x0302, "3D controller"},
1294     { 0x0380, "Display controller"},
1295     { 0x0400, "Video controller", "video"},
1296     { 0x0401, "Audio controller", "sound"},
1297     { 0x0402, "Phone"},
1298     { 0x0403, "Audio controller", "sound"},
1299     { 0x0480, "Multimedia controller"},
1300     { 0x0500, "RAM controller", "memory"},
1301     { 0x0501, "Flash controller", "flash"},
1302     { 0x0580, "Memory controller"},
1303     { 0x0600, "Host bridge", "host"},
1304     { 0x0601, "ISA bridge", "isa"},
1305     { 0x0602, "EISA bridge", "eisa"},
1306     { 0x0603, "MC bridge", "mca"},
1307     { 0x0604, "PCI bridge", "pci"},
1308     { 0x0605, "PCMCIA bridge", "pcmcia"},
1309     { 0x0606, "NUBUS bridge", "nubus"},
1310     { 0x0607, "CARDBUS bridge", "cardbus"},
1311     { 0x0608, "RACEWAY bridge"},
1312     { 0x0680, "Bridge"},
1313     { 0x0700, "Serial port", "serial"},
1314     { 0x0701, "Parallel port", "parallel"},
1315     { 0x0800, "Interrupt controller", "interrupt-controller"},
1316     { 0x0801, "DMA controller", "dma-controller"},
1317     { 0x0802, "Timer", "timer"},
1318     { 0x0803, "RTC", "rtc"},
1319     { 0x0900, "Keyboard", "keyboard"},
1320     { 0x0901, "Pen", "pen"},
1321     { 0x0902, "Mouse", "mouse"},
1322     { 0x0A00, "Dock station", "dock", 0x00ff},
1323     { 0x0B00, "i386 cpu", "cpu", 0x00ff},
1324     { 0x0c00, "Fireware contorller", "fireware"},
1325     { 0x0c01, "Access bus controller", "access-bus"},
1326     { 0x0c02, "SSA controller", "ssa"},
1327     { 0x0c03, "USB controller", "usb"},
1328     { 0x0c04, "Fibre channel controller", "fibre-channel"},
1329     { 0x0c05, "SMBus"},
1330     { 0, NULL}
1331 };
1332 
1333 static void pci_for_each_device_under_bus(PCIBus *bus,
1334                                           void (*fn)(PCIBus *b, PCIDevice *d,
1335                                                      void *opaque),
1336                                           void *opaque)
1337 {
1338     PCIDevice *d;
1339     int devfn;
1340 
1341     for(devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1342         d = bus->devices[devfn];
1343         if (d) {
1344             fn(bus, d, opaque);
1345         }
1346     }
1347 }
1348 
1349 void pci_for_each_device(PCIBus *bus, int bus_num,
1350                          void (*fn)(PCIBus *b, PCIDevice *d, void *opaque),
1351                          void *opaque)
1352 {
1353     bus = pci_find_bus_nr(bus, bus_num);
1354 
1355     if (bus) {
1356         pci_for_each_device_under_bus(bus, fn, opaque);
1357     }
1358 }
1359 
1360 static const pci_class_desc *get_class_desc(int class)
1361 {
1362     const pci_class_desc *desc;
1363 
1364     desc = pci_class_descriptions;
1365     while (desc->desc && class != desc->class) {
1366         desc++;
1367     }
1368 
1369     return desc;
1370 }
1371 
1372 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num);
1373 
1374 static PciMemoryRegionList *qmp_query_pci_regions(const PCIDevice *dev)
1375 {
1376     PciMemoryRegionList *head = NULL, *cur_item = NULL;
1377     int i;
1378 
1379     for (i = 0; i < PCI_NUM_REGIONS; i++) {
1380         const PCIIORegion *r = &dev->io_regions[i];
1381         PciMemoryRegionList *region;
1382 
1383         if (!r->size) {
1384             continue;
1385         }
1386 
1387         region = g_malloc0(sizeof(*region));
1388         region->value = g_malloc0(sizeof(*region->value));
1389 
1390         if (r->type & PCI_BASE_ADDRESS_SPACE_IO) {
1391             region->value->type = g_strdup("io");
1392         } else {
1393             region->value->type = g_strdup("memory");
1394             region->value->has_prefetch = true;
1395             region->value->prefetch = !!(r->type & PCI_BASE_ADDRESS_MEM_PREFETCH);
1396             region->value->has_mem_type_64 = true;
1397             region->value->mem_type_64 = !!(r->type & PCI_BASE_ADDRESS_MEM_TYPE_64);
1398         }
1399 
1400         region->value->bar = i;
1401         region->value->address = r->addr;
1402         region->value->size = r->size;
1403 
1404         /* XXX: waiting for the qapi to support GSList */
1405         if (!cur_item) {
1406             head = cur_item = region;
1407         } else {
1408             cur_item->next = region;
1409             cur_item = region;
1410         }
1411     }
1412 
1413     return head;
1414 }
1415 
1416 static PciBridgeInfo *qmp_query_pci_bridge(PCIDevice *dev, PCIBus *bus,
1417                                            int bus_num)
1418 {
1419     PciBridgeInfo *info;
1420 
1421     info = g_malloc0(sizeof(*info));
1422 
1423     info->bus.number = dev->config[PCI_PRIMARY_BUS];
1424     info->bus.secondary = dev->config[PCI_SECONDARY_BUS];
1425     info->bus.subordinate = dev->config[PCI_SUBORDINATE_BUS];
1426 
1427     info->bus.io_range = g_malloc0(sizeof(*info->bus.io_range));
1428     info->bus.io_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_IO);
1429     info->bus.io_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_IO);
1430 
1431     info->bus.memory_range = g_malloc0(sizeof(*info->bus.memory_range));
1432     info->bus.memory_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1433     info->bus.memory_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_SPACE_MEMORY);
1434 
1435     info->bus.prefetchable_range = g_malloc0(sizeof(*info->bus.prefetchable_range));
1436     info->bus.prefetchable_range->base = pci_bridge_get_base(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1437     info->bus.prefetchable_range->limit = pci_bridge_get_limit(dev, PCI_BASE_ADDRESS_MEM_PREFETCH);
1438 
1439     if (dev->config[PCI_SECONDARY_BUS] != 0) {
1440         PCIBus *child_bus = pci_find_bus_nr(bus, dev->config[PCI_SECONDARY_BUS]);
1441         if (child_bus) {
1442             info->has_devices = true;
1443             info->devices = qmp_query_pci_devices(child_bus, dev->config[PCI_SECONDARY_BUS]);
1444         }
1445     }
1446 
1447     return info;
1448 }
1449 
1450 static PciDeviceInfo *qmp_query_pci_device(PCIDevice *dev, PCIBus *bus,
1451                                            int bus_num)
1452 {
1453     const pci_class_desc *desc;
1454     PciDeviceInfo *info;
1455     uint8_t type;
1456     int class;
1457 
1458     info = g_malloc0(sizeof(*info));
1459     info->bus = bus_num;
1460     info->slot = PCI_SLOT(dev->devfn);
1461     info->function = PCI_FUNC(dev->devfn);
1462 
1463     class = pci_get_word(dev->config + PCI_CLASS_DEVICE);
1464     info->class_info.q_class = class;
1465     desc = get_class_desc(class);
1466     if (desc->desc) {
1467         info->class_info.has_desc = true;
1468         info->class_info.desc = g_strdup(desc->desc);
1469     }
1470 
1471     info->id.vendor = pci_get_word(dev->config + PCI_VENDOR_ID);
1472     info->id.device = pci_get_word(dev->config + PCI_DEVICE_ID);
1473     info->regions = qmp_query_pci_regions(dev);
1474     info->qdev_id = g_strdup(dev->qdev.id ? dev->qdev.id : "");
1475 
1476     if (dev->config[PCI_INTERRUPT_PIN] != 0) {
1477         info->has_irq = true;
1478         info->irq = dev->config[PCI_INTERRUPT_LINE];
1479     }
1480 
1481     type = dev->config[PCI_HEADER_TYPE] & ~PCI_HEADER_TYPE_MULTI_FUNCTION;
1482     if (type == PCI_HEADER_TYPE_BRIDGE) {
1483         info->has_pci_bridge = true;
1484         info->pci_bridge = qmp_query_pci_bridge(dev, bus, bus_num);
1485     }
1486 
1487     return info;
1488 }
1489 
1490 static PciDeviceInfoList *qmp_query_pci_devices(PCIBus *bus, int bus_num)
1491 {
1492     PciDeviceInfoList *info, *head = NULL, *cur_item = NULL;
1493     PCIDevice *dev;
1494     int devfn;
1495 
1496     for (devfn = 0; devfn < ARRAY_SIZE(bus->devices); devfn++) {
1497         dev = bus->devices[devfn];
1498         if (dev) {
1499             info = g_malloc0(sizeof(*info));
1500             info->value = qmp_query_pci_device(dev, bus, bus_num);
1501 
1502             /* XXX: waiting for the qapi to support GSList */
1503             if (!cur_item) {
1504                 head = cur_item = info;
1505             } else {
1506                 cur_item->next = info;
1507                 cur_item = info;
1508             }
1509         }
1510     }
1511 
1512     return head;
1513 }
1514 
1515 static PciInfo *qmp_query_pci_bus(PCIBus *bus, int bus_num)
1516 {
1517     PciInfo *info = NULL;
1518 
1519     bus = pci_find_bus_nr(bus, bus_num);
1520     if (bus) {
1521         info = g_malloc0(sizeof(*info));
1522         info->bus = bus_num;
1523         info->devices = qmp_query_pci_devices(bus, bus_num);
1524     }
1525 
1526     return info;
1527 }
1528 
1529 PciInfoList *qmp_query_pci(Error **errp)
1530 {
1531     PciInfoList *info, *head = NULL, *cur_item = NULL;
1532     PCIHostState *host_bridge;
1533 
1534     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
1535         info = g_malloc0(sizeof(*info));
1536         info->value = qmp_query_pci_bus(host_bridge->bus, 0);
1537 
1538         /* XXX: waiting for the qapi to support GSList */
1539         if (!cur_item) {
1540             head = cur_item = info;
1541         } else {
1542             cur_item->next = info;
1543             cur_item = info;
1544         }
1545     }
1546 
1547     return head;
1548 }
1549 
1550 static const char * const pci_nic_models[] = {
1551     "ne2k_pci",
1552     "i82551",
1553     "i82557b",
1554     "i82559er",
1555     "rtl8139",
1556     "e1000",
1557     "pcnet",
1558     "virtio",
1559     NULL
1560 };
1561 
1562 static const char * const pci_nic_names[] = {
1563     "ne2k_pci",
1564     "i82551",
1565     "i82557b",
1566     "i82559er",
1567     "rtl8139",
1568     "e1000",
1569     "pcnet",
1570     "virtio-net-pci",
1571     NULL
1572 };
1573 
1574 /* Initialize a PCI NIC.  */
1575 /* FIXME callers should check for failure, but don't */
1576 PCIDevice *pci_nic_init(NICInfo *nd, PCIBus *rootbus,
1577                         const char *default_model,
1578                         const char *default_devaddr)
1579 {
1580     const char *devaddr = nd->devaddr ? nd->devaddr : default_devaddr;
1581     PCIBus *bus;
1582     int devfn;
1583     PCIDevice *pci_dev;
1584     DeviceState *dev;
1585     int i;
1586 
1587     i = qemu_find_nic_model(nd, pci_nic_models, default_model);
1588     if (i < 0)
1589         return NULL;
1590 
1591     bus = pci_get_bus_devfn(&devfn, rootbus, devaddr);
1592     if (!bus) {
1593         error_report("Invalid PCI device address %s for device %s",
1594                      devaddr, pci_nic_names[i]);
1595         return NULL;
1596     }
1597 
1598     pci_dev = pci_create(bus, devfn, pci_nic_names[i]);
1599     dev = &pci_dev->qdev;
1600     qdev_set_nic_properties(dev, nd);
1601     if (qdev_init(dev) < 0)
1602         return NULL;
1603     return pci_dev;
1604 }
1605 
1606 PCIDevice *pci_nic_init_nofail(NICInfo *nd, PCIBus *rootbus,
1607                                const char *default_model,
1608                                const char *default_devaddr)
1609 {
1610     PCIDevice *res;
1611 
1612     if (qemu_show_nic_models(nd->model, pci_nic_models))
1613         exit(0);
1614 
1615     res = pci_nic_init(nd, rootbus, default_model, default_devaddr);
1616     if (!res)
1617         exit(1);
1618     return res;
1619 }
1620 
1621 PCIDevice *pci_vga_init(PCIBus *bus)
1622 {
1623     switch (vga_interface_type) {
1624     case VGA_CIRRUS:
1625         return pci_create_simple(bus, -1, "cirrus-vga");
1626     case VGA_QXL:
1627         return pci_create_simple(bus, -1, "qxl-vga");
1628     case VGA_STD:
1629         return pci_create_simple(bus, -1, "VGA");
1630     case VGA_VMWARE:
1631         return pci_create_simple(bus, -1, "vmware-svga");
1632     case VGA_NONE:
1633     default: /* Other non-PCI types. Checking for unsupported types is already
1634                 done in vl.c. */
1635         return NULL;
1636     }
1637 }
1638 
1639 /* Whether a given bus number is in range of the secondary
1640  * bus of the given bridge device. */
1641 static bool pci_secondary_bus_in_range(PCIDevice *dev, int bus_num)
1642 {
1643     return !(pci_get_word(dev->config + PCI_BRIDGE_CONTROL) &
1644              PCI_BRIDGE_CTL_BUS_RESET) /* Don't walk the bus if it's reset. */ &&
1645         dev->config[PCI_SECONDARY_BUS] < bus_num &&
1646         bus_num <= dev->config[PCI_SUBORDINATE_BUS];
1647 }
1648 
1649 static PCIBus *pci_find_bus_nr(PCIBus *bus, int bus_num)
1650 {
1651     PCIBus *sec;
1652 
1653     if (!bus) {
1654         return NULL;
1655     }
1656 
1657     if (pci_bus_num(bus) == bus_num) {
1658         return bus;
1659     }
1660 
1661     /* Consider all bus numbers in range for the host pci bridge. */
1662     if (!pci_bus_is_root(bus) &&
1663         !pci_secondary_bus_in_range(bus->parent_dev, bus_num)) {
1664         return NULL;
1665     }
1666 
1667     /* try child bus */
1668     for (; bus; bus = sec) {
1669         QLIST_FOREACH(sec, &bus->child, sibling) {
1670             assert(!pci_bus_is_root(sec));
1671             if (sec->parent_dev->config[PCI_SECONDARY_BUS] == bus_num) {
1672                 return sec;
1673             }
1674             if (pci_secondary_bus_in_range(sec->parent_dev, bus_num)) {
1675                 break;
1676             }
1677         }
1678     }
1679 
1680     return NULL;
1681 }
1682 
1683 PCIDevice *pci_find_device(PCIBus *bus, int bus_num, uint8_t devfn)
1684 {
1685     bus = pci_find_bus_nr(bus, bus_num);
1686 
1687     if (!bus)
1688         return NULL;
1689 
1690     return bus->devices[devfn];
1691 }
1692 
1693 static int pci_qdev_init(DeviceState *qdev)
1694 {
1695     PCIDevice *pci_dev = (PCIDevice *)qdev;
1696     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(pci_dev);
1697     PCIBus *bus;
1698     int rc;
1699     bool is_default_rom;
1700 
1701     /* initialize cap_present for pci_is_express() and pci_config_size() */
1702     if (pc->is_express) {
1703         pci_dev->cap_present |= QEMU_PCI_CAP_EXPRESS;
1704     }
1705 
1706     bus = PCI_BUS(qdev_get_parent_bus(qdev));
1707     pci_dev = do_pci_register_device(pci_dev, bus,
1708                                      object_get_typename(OBJECT(qdev)),
1709                                      pci_dev->devfn);
1710     if (pci_dev == NULL)
1711         return -1;
1712     if (qdev->hotplugged && pc->no_hotplug) {
1713         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(pci_dev)));
1714         do_pci_unregister_device(pci_dev);
1715         return -1;
1716     }
1717     if (pc->init) {
1718         rc = pc->init(pci_dev);
1719         if (rc != 0) {
1720             do_pci_unregister_device(pci_dev);
1721             return rc;
1722         }
1723     }
1724 
1725     /* rom loading */
1726     is_default_rom = false;
1727     if (pci_dev->romfile == NULL && pc->romfile != NULL) {
1728         pci_dev->romfile = g_strdup(pc->romfile);
1729         is_default_rom = true;
1730     }
1731     pci_add_option_rom(pci_dev, is_default_rom);
1732 
1733     if (bus->hotplug) {
1734         /* Let buses differentiate between hotplug and when device is
1735          * enabled during qemu machine creation. */
1736         rc = bus->hotplug(bus->hotplug_qdev, pci_dev,
1737                           qdev->hotplugged ? PCI_HOTPLUG_ENABLED:
1738                           PCI_COLDPLUG_ENABLED);
1739         if (rc != 0) {
1740             int r = pci_unregister_device(&pci_dev->qdev);
1741             assert(!r);
1742             return rc;
1743         }
1744     }
1745     return 0;
1746 }
1747 
1748 static int pci_unplug_device(DeviceState *qdev)
1749 {
1750     PCIDevice *dev = PCI_DEVICE(qdev);
1751     PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev);
1752 
1753     if (pc->no_hotplug) {
1754         qerror_report(QERR_DEVICE_NO_HOTPLUG, object_get_typename(OBJECT(dev)));
1755         return -1;
1756     }
1757     return dev->bus->hotplug(dev->bus->hotplug_qdev, dev,
1758                              PCI_HOTPLUG_DISABLED);
1759 }
1760 
1761 PCIDevice *pci_create_multifunction(PCIBus *bus, int devfn, bool multifunction,
1762                                     const char *name)
1763 {
1764     DeviceState *dev;
1765 
1766     dev = qdev_create(&bus->qbus, name);
1767     qdev_prop_set_int32(dev, "addr", devfn);
1768     qdev_prop_set_bit(dev, "multifunction", multifunction);
1769     return PCI_DEVICE(dev);
1770 }
1771 
1772 PCIDevice *pci_create_simple_multifunction(PCIBus *bus, int devfn,
1773                                            bool multifunction,
1774                                            const char *name)
1775 {
1776     PCIDevice *dev = pci_create_multifunction(bus, devfn, multifunction, name);
1777     qdev_init_nofail(&dev->qdev);
1778     return dev;
1779 }
1780 
1781 PCIDevice *pci_create(PCIBus *bus, int devfn, const char *name)
1782 {
1783     return pci_create_multifunction(bus, devfn, false, name);
1784 }
1785 
1786 PCIDevice *pci_create_simple(PCIBus *bus, int devfn, const char *name)
1787 {
1788     return pci_create_simple_multifunction(bus, devfn, false, name);
1789 }
1790 
1791 static uint8_t pci_find_space(PCIDevice *pdev, uint8_t size)
1792 {
1793     int offset = PCI_CONFIG_HEADER_SIZE;
1794     int i;
1795     for (i = PCI_CONFIG_HEADER_SIZE; i < PCI_CONFIG_SPACE_SIZE; ++i) {
1796         if (pdev->used[i])
1797             offset = i + 1;
1798         else if (i - offset + 1 == size)
1799             return offset;
1800     }
1801     return 0;
1802 }
1803 
1804 static uint8_t pci_find_capability_list(PCIDevice *pdev, uint8_t cap_id,
1805                                         uint8_t *prev_p)
1806 {
1807     uint8_t next, prev;
1808 
1809     if (!(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST))
1810         return 0;
1811 
1812     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1813          prev = next + PCI_CAP_LIST_NEXT)
1814         if (pdev->config[next + PCI_CAP_LIST_ID] == cap_id)
1815             break;
1816 
1817     if (prev_p)
1818         *prev_p = prev;
1819     return next;
1820 }
1821 
1822 static uint8_t pci_find_capability_at_offset(PCIDevice *pdev, uint8_t offset)
1823 {
1824     uint8_t next, prev, found = 0;
1825 
1826     if (!(pdev->used[offset])) {
1827         return 0;
1828     }
1829 
1830     assert(pdev->config[PCI_STATUS] & PCI_STATUS_CAP_LIST);
1831 
1832     for (prev = PCI_CAPABILITY_LIST; (next = pdev->config[prev]);
1833          prev = next + PCI_CAP_LIST_NEXT) {
1834         if (next <= offset && next > found) {
1835             found = next;
1836         }
1837     }
1838     return found;
1839 }
1840 
1841 /* Patch the PCI vendor and device ids in a PCI rom image if necessary.
1842    This is needed for an option rom which is used for more than one device. */
1843 static void pci_patch_ids(PCIDevice *pdev, uint8_t *ptr, int size)
1844 {
1845     uint16_t vendor_id;
1846     uint16_t device_id;
1847     uint16_t rom_vendor_id;
1848     uint16_t rom_device_id;
1849     uint16_t rom_magic;
1850     uint16_t pcir_offset;
1851     uint8_t checksum;
1852 
1853     /* Words in rom data are little endian (like in PCI configuration),
1854        so they can be read / written with pci_get_word / pci_set_word. */
1855 
1856     /* Only a valid rom will be patched. */
1857     rom_magic = pci_get_word(ptr);
1858     if (rom_magic != 0xaa55) {
1859         PCI_DPRINTF("Bad ROM magic %04x\n", rom_magic);
1860         return;
1861     }
1862     pcir_offset = pci_get_word(ptr + 0x18);
1863     if (pcir_offset + 8 >= size || memcmp(ptr + pcir_offset, "PCIR", 4)) {
1864         PCI_DPRINTF("Bad PCIR offset 0x%x or signature\n", pcir_offset);
1865         return;
1866     }
1867 
1868     vendor_id = pci_get_word(pdev->config + PCI_VENDOR_ID);
1869     device_id = pci_get_word(pdev->config + PCI_DEVICE_ID);
1870     rom_vendor_id = pci_get_word(ptr + pcir_offset + 4);
1871     rom_device_id = pci_get_word(ptr + pcir_offset + 6);
1872 
1873     PCI_DPRINTF("%s: ROM id %04x%04x / PCI id %04x%04x\n", pdev->romfile,
1874                 vendor_id, device_id, rom_vendor_id, rom_device_id);
1875 
1876     checksum = ptr[6];
1877 
1878     if (vendor_id != rom_vendor_id) {
1879         /* Patch vendor id and checksum (at offset 6 for etherboot roms). */
1880         checksum += (uint8_t)rom_vendor_id + (uint8_t)(rom_vendor_id >> 8);
1881         checksum -= (uint8_t)vendor_id + (uint8_t)(vendor_id >> 8);
1882         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1883         ptr[6] = checksum;
1884         pci_set_word(ptr + pcir_offset + 4, vendor_id);
1885     }
1886 
1887     if (device_id != rom_device_id) {
1888         /* Patch device id and checksum (at offset 6 for etherboot roms). */
1889         checksum += (uint8_t)rom_device_id + (uint8_t)(rom_device_id >> 8);
1890         checksum -= (uint8_t)device_id + (uint8_t)(device_id >> 8);
1891         PCI_DPRINTF("ROM checksum %02x / %02x\n", ptr[6], checksum);
1892         ptr[6] = checksum;
1893         pci_set_word(ptr + pcir_offset + 6, device_id);
1894     }
1895 }
1896 
1897 /* Add an option rom for the device */
1898 static int pci_add_option_rom(PCIDevice *pdev, bool is_default_rom)
1899 {
1900     int size;
1901     char *path;
1902     void *ptr;
1903     char name[32];
1904     const VMStateDescription *vmsd;
1905 
1906     if (!pdev->romfile)
1907         return 0;
1908     if (strlen(pdev->romfile) == 0)
1909         return 0;
1910 
1911     if (!pdev->rom_bar) {
1912         /*
1913          * Load rom via fw_cfg instead of creating a rom bar,
1914          * for 0.11 compatibility.
1915          */
1916         int class = pci_get_word(pdev->config + PCI_CLASS_DEVICE);
1917         if (class == 0x0300) {
1918             rom_add_vga(pdev->romfile);
1919         } else {
1920             rom_add_option(pdev->romfile, -1);
1921         }
1922         return 0;
1923     }
1924 
1925     path = qemu_find_file(QEMU_FILE_TYPE_BIOS, pdev->romfile);
1926     if (path == NULL) {
1927         path = g_strdup(pdev->romfile);
1928     }
1929 
1930     size = get_image_size(path);
1931     if (size < 0) {
1932         error_report("%s: failed to find romfile \"%s\"",
1933                      __func__, pdev->romfile);
1934         g_free(path);
1935         return -1;
1936     } else if (size == 0) {
1937         error_report("%s: ignoring empty romfile \"%s\"",
1938                      __func__, pdev->romfile);
1939         g_free(path);
1940         return -1;
1941     }
1942     if (size & (size - 1)) {
1943         size = 1 << qemu_fls(size);
1944     }
1945 
1946     vmsd = qdev_get_vmsd(DEVICE(pdev));
1947 
1948     if (vmsd) {
1949         snprintf(name, sizeof(name), "%s.rom", vmsd->name);
1950     } else {
1951         snprintf(name, sizeof(name), "%s.rom", object_get_typename(OBJECT(pdev)));
1952     }
1953     pdev->has_rom = true;
1954     memory_region_init_ram(&pdev->rom, OBJECT(pdev), name, size);
1955     vmstate_register_ram(&pdev->rom, &pdev->qdev);
1956     ptr = memory_region_get_ram_ptr(&pdev->rom);
1957     load_image(path, ptr);
1958     g_free(path);
1959 
1960     if (is_default_rom) {
1961         /* Only the default rom images will be patched (if needed). */
1962         pci_patch_ids(pdev, ptr, size);
1963     }
1964 
1965     pci_register_bar(pdev, PCI_ROM_SLOT, 0, &pdev->rom);
1966 
1967     return 0;
1968 }
1969 
1970 static void pci_del_option_rom(PCIDevice *pdev)
1971 {
1972     if (!pdev->has_rom)
1973         return;
1974 
1975     vmstate_unregister_ram(&pdev->rom, &pdev->qdev);
1976     memory_region_destroy(&pdev->rom);
1977     pdev->has_rom = false;
1978 }
1979 
1980 /*
1981  * if !offset
1982  * Reserve space and add capability to the linked list in pci config space
1983  *
1984  * if offset = 0,
1985  * Find and reserve space and add capability to the linked list
1986  * in pci config space */
1987 int pci_add_capability(PCIDevice *pdev, uint8_t cap_id,
1988                        uint8_t offset, uint8_t size)
1989 {
1990     uint8_t *config;
1991     int i, overlapping_cap;
1992 
1993     if (!offset) {
1994         offset = pci_find_space(pdev, size);
1995         if (!offset) {
1996             return -ENOSPC;
1997         }
1998     } else {
1999         /* Verify that capabilities don't overlap.  Note: device assignment
2000          * depends on this check to verify that the device is not broken.
2001          * Should never trigger for emulated devices, but it's helpful
2002          * for debugging these. */
2003         for (i = offset; i < offset + size; i++) {
2004             overlapping_cap = pci_find_capability_at_offset(pdev, i);
2005             if (overlapping_cap) {
2006                 fprintf(stderr, "ERROR: %s:%02x:%02x.%x "
2007                         "Attempt to add PCI capability %x at offset "
2008                         "%x overlaps existing capability %x at offset %x\n",
2009                         pci_root_bus_path(pdev), pci_bus_num(pdev->bus),
2010                         PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn),
2011                         cap_id, offset, overlapping_cap, i);
2012                 return -EINVAL;
2013             }
2014         }
2015     }
2016 
2017     config = pdev->config + offset;
2018     config[PCI_CAP_LIST_ID] = cap_id;
2019     config[PCI_CAP_LIST_NEXT] = pdev->config[PCI_CAPABILITY_LIST];
2020     pdev->config[PCI_CAPABILITY_LIST] = offset;
2021     pdev->config[PCI_STATUS] |= PCI_STATUS_CAP_LIST;
2022     memset(pdev->used + offset, 0xFF, QEMU_ALIGN_UP(size, 4));
2023     /* Make capability read-only by default */
2024     memset(pdev->wmask + offset, 0, size);
2025     /* Check capability by default */
2026     memset(pdev->cmask + offset, 0xFF, size);
2027     return offset;
2028 }
2029 
2030 /* Unlink capability from the pci config space. */
2031 void pci_del_capability(PCIDevice *pdev, uint8_t cap_id, uint8_t size)
2032 {
2033     uint8_t prev, offset = pci_find_capability_list(pdev, cap_id, &prev);
2034     if (!offset)
2035         return;
2036     pdev->config[prev] = pdev->config[offset + PCI_CAP_LIST_NEXT];
2037     /* Make capability writable again */
2038     memset(pdev->wmask + offset, 0xff, size);
2039     memset(pdev->w1cmask + offset, 0, size);
2040     /* Clear cmask as device-specific registers can't be checked */
2041     memset(pdev->cmask + offset, 0, size);
2042     memset(pdev->used + offset, 0, QEMU_ALIGN_UP(size, 4));
2043 
2044     if (!pdev->config[PCI_CAPABILITY_LIST])
2045         pdev->config[PCI_STATUS] &= ~PCI_STATUS_CAP_LIST;
2046 }
2047 
2048 uint8_t pci_find_capability(PCIDevice *pdev, uint8_t cap_id)
2049 {
2050     return pci_find_capability_list(pdev, cap_id, NULL);
2051 }
2052 
2053 static void pcibus_dev_print(Monitor *mon, DeviceState *dev, int indent)
2054 {
2055     PCIDevice *d = (PCIDevice *)dev;
2056     const pci_class_desc *desc;
2057     char ctxt[64];
2058     PCIIORegion *r;
2059     int i, class;
2060 
2061     class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2062     desc = pci_class_descriptions;
2063     while (desc->desc && class != desc->class)
2064         desc++;
2065     if (desc->desc) {
2066         snprintf(ctxt, sizeof(ctxt), "%s", desc->desc);
2067     } else {
2068         snprintf(ctxt, sizeof(ctxt), "Class %04x", class);
2069     }
2070 
2071     monitor_printf(mon, "%*sclass %s, addr %02x:%02x.%x, "
2072                    "pci id %04x:%04x (sub %04x:%04x)\n",
2073                    indent, "", ctxt, pci_bus_num(d->bus),
2074                    PCI_SLOT(d->devfn), PCI_FUNC(d->devfn),
2075                    pci_get_word(d->config + PCI_VENDOR_ID),
2076                    pci_get_word(d->config + PCI_DEVICE_ID),
2077                    pci_get_word(d->config + PCI_SUBSYSTEM_VENDOR_ID),
2078                    pci_get_word(d->config + PCI_SUBSYSTEM_ID));
2079     for (i = 0; i < PCI_NUM_REGIONS; i++) {
2080         r = &d->io_regions[i];
2081         if (!r->size)
2082             continue;
2083         monitor_printf(mon, "%*sbar %d: %s at 0x%"FMT_PCIBUS
2084                        " [0x%"FMT_PCIBUS"]\n",
2085                        indent, "",
2086                        i, r->type & PCI_BASE_ADDRESS_SPACE_IO ? "i/o" : "mem",
2087                        r->addr, r->addr + r->size - 1);
2088     }
2089 }
2090 
2091 static char *pci_dev_fw_name(DeviceState *dev, char *buf, int len)
2092 {
2093     PCIDevice *d = (PCIDevice *)dev;
2094     const char *name = NULL;
2095     const pci_class_desc *desc =  pci_class_descriptions;
2096     int class = pci_get_word(d->config + PCI_CLASS_DEVICE);
2097 
2098     while (desc->desc &&
2099           (class & ~desc->fw_ign_bits) !=
2100           (desc->class & ~desc->fw_ign_bits)) {
2101         desc++;
2102     }
2103 
2104     if (desc->desc) {
2105         name = desc->fw_name;
2106     }
2107 
2108     if (name) {
2109         pstrcpy(buf, len, name);
2110     } else {
2111         snprintf(buf, len, "pci%04x,%04x",
2112                  pci_get_word(d->config + PCI_VENDOR_ID),
2113                  pci_get_word(d->config + PCI_DEVICE_ID));
2114     }
2115 
2116     return buf;
2117 }
2118 
2119 static char *pcibus_get_fw_dev_path(DeviceState *dev)
2120 {
2121     PCIDevice *d = (PCIDevice *)dev;
2122     char path[50], name[33];
2123     int off;
2124 
2125     off = snprintf(path, sizeof(path), "%s@%x",
2126                    pci_dev_fw_name(dev, name, sizeof name),
2127                    PCI_SLOT(d->devfn));
2128     if (PCI_FUNC(d->devfn))
2129         snprintf(path + off, sizeof(path) + off, ",%x", PCI_FUNC(d->devfn));
2130     return g_strdup(path);
2131 }
2132 
2133 static char *pcibus_get_dev_path(DeviceState *dev)
2134 {
2135     PCIDevice *d = container_of(dev, PCIDevice, qdev);
2136     PCIDevice *t;
2137     int slot_depth;
2138     /* Path format: Domain:00:Slot.Function:Slot.Function....:Slot.Function.
2139      * 00 is added here to make this format compatible with
2140      * domain:Bus:Slot.Func for systems without nested PCI bridges.
2141      * Slot.Function list specifies the slot and function numbers for all
2142      * devices on the path from root to the specific device. */
2143     const char *root_bus_path;
2144     int root_bus_len;
2145     char slot[] = ":SS.F";
2146     int slot_len = sizeof slot - 1 /* For '\0' */;
2147     int path_len;
2148     char *path, *p;
2149     int s;
2150 
2151     root_bus_path = pci_root_bus_path(d);
2152     root_bus_len = strlen(root_bus_path);
2153 
2154     /* Calculate # of slots on path between device and root. */;
2155     slot_depth = 0;
2156     for (t = d; t; t = t->bus->parent_dev) {
2157         ++slot_depth;
2158     }
2159 
2160     path_len = root_bus_len + slot_len * slot_depth;
2161 
2162     /* Allocate memory, fill in the terminating null byte. */
2163     path = g_malloc(path_len + 1 /* For '\0' */);
2164     path[path_len] = '\0';
2165 
2166     memcpy(path, root_bus_path, root_bus_len);
2167 
2168     /* Fill in slot numbers. We walk up from device to root, so need to print
2169      * them in the reverse order, last to first. */
2170     p = path + path_len;
2171     for (t = d; t; t = t->bus->parent_dev) {
2172         p -= slot_len;
2173         s = snprintf(slot, sizeof slot, ":%02x.%x",
2174                      PCI_SLOT(t->devfn), PCI_FUNC(t->devfn));
2175         assert(s == slot_len);
2176         memcpy(p, slot, slot_len);
2177     }
2178 
2179     return path;
2180 }
2181 
2182 static int pci_qdev_find_recursive(PCIBus *bus,
2183                                    const char *id, PCIDevice **pdev)
2184 {
2185     DeviceState *qdev = qdev_find_recursive(&bus->qbus, id);
2186     if (!qdev) {
2187         return -ENODEV;
2188     }
2189 
2190     /* roughly check if given qdev is pci device */
2191     if (object_dynamic_cast(OBJECT(qdev), TYPE_PCI_DEVICE)) {
2192         *pdev = PCI_DEVICE(qdev);
2193         return 0;
2194     }
2195     return -EINVAL;
2196 }
2197 
2198 int pci_qdev_find_device(const char *id, PCIDevice **pdev)
2199 {
2200     PCIHostState *host_bridge;
2201     int rc = -ENODEV;
2202 
2203     QLIST_FOREACH(host_bridge, &pci_host_bridges, next) {
2204         int tmp = pci_qdev_find_recursive(host_bridge->bus, id, pdev);
2205         if (!tmp) {
2206             rc = 0;
2207             break;
2208         }
2209         if (tmp != -ENODEV) {
2210             rc = tmp;
2211         }
2212     }
2213 
2214     return rc;
2215 }
2216 
2217 MemoryRegion *pci_address_space(PCIDevice *dev)
2218 {
2219     return dev->bus->address_space_mem;
2220 }
2221 
2222 MemoryRegion *pci_address_space_io(PCIDevice *dev)
2223 {
2224     return dev->bus->address_space_io;
2225 }
2226 
2227 static void pci_device_class_init(ObjectClass *klass, void *data)
2228 {
2229     DeviceClass *k = DEVICE_CLASS(klass);
2230     k->init = pci_qdev_init;
2231     k->unplug = pci_unplug_device;
2232     k->exit = pci_unregister_device;
2233     k->bus_type = TYPE_PCI_BUS;
2234     k->props = pci_props;
2235 }
2236 
2237 AddressSpace *pci_device_iommu_address_space(PCIDevice *dev)
2238 {
2239     PCIBus *bus = PCI_BUS(dev->bus);
2240 
2241     if (bus->iommu_fn) {
2242         return bus->iommu_fn(bus, bus->iommu_opaque, dev->devfn);
2243     }
2244 
2245     if (bus->parent_dev) {
2246         /** We are ignoring the bus master DMA bit of the bridge
2247          *  as it would complicate things such as VFIO for no good reason */
2248         return pci_device_iommu_address_space(bus->parent_dev);
2249     }
2250 
2251     return &address_space_memory;
2252 }
2253 
2254 void pci_setup_iommu(PCIBus *bus, PCIIOMMUFunc fn, void *opaque)
2255 {
2256     bus->iommu_fn = fn;
2257     bus->iommu_opaque = opaque;
2258 }
2259 
2260 static const TypeInfo pci_device_type_info = {
2261     .name = TYPE_PCI_DEVICE,
2262     .parent = TYPE_DEVICE,
2263     .instance_size = sizeof(PCIDevice),
2264     .abstract = true,
2265     .class_size = sizeof(PCIDeviceClass),
2266     .class_init = pci_device_class_init,
2267 };
2268 
2269 static void pci_register_types(void)
2270 {
2271     type_register_static(&pci_bus_info);
2272     type_register_static(&pcie_bus_info);
2273     type_register_static(&pci_device_type_info);
2274 }
2275 
2276 type_init(pci_register_types)
2277