root / hw / misc / ivshmem.c @ baefb8bf
History | View | Annotate | Download (22.8 kB)
1 |
/*
|
---|---|
2 |
* Inter-VM Shared Memory PCI device.
|
3 |
*
|
4 |
* Author:
|
5 |
* Cam Macdonell <cam@cs.ualberta.ca>
|
6 |
*
|
7 |
* Based On: cirrus_vga.c
|
8 |
* Copyright (c) 2004 Fabrice Bellard
|
9 |
* Copyright (c) 2004 Makoto Suzuki (suzu)
|
10 |
*
|
11 |
* and rtl8139.c
|
12 |
* Copyright (c) 2006 Igor Kovalenko
|
13 |
*
|
14 |
* This code is licensed under the GNU GPL v2.
|
15 |
*
|
16 |
* Contributions after 2012-01-13 are licensed under the terms of the
|
17 |
* GNU GPL, version 2 or (at your option) any later version.
|
18 |
*/
|
19 |
#include "hw/hw.h" |
20 |
#include "hw/i386/pc.h" |
21 |
#include "hw/pci/pci.h" |
22 |
#include "hw/pci/msix.h" |
23 |
#include "sysemu/kvm.h" |
24 |
#include "migration/migration.h" |
25 |
#include "qapi/qmp/qerror.h" |
26 |
#include "qemu/event_notifier.h" |
27 |
#include "sysemu/char.h" |
28 |
|
29 |
#include <sys/mman.h> |
30 |
#include <sys/types.h> |
31 |
|
32 |
#define PCI_VENDOR_ID_IVSHMEM PCI_VENDOR_ID_REDHAT_QUMRANET
|
33 |
#define PCI_DEVICE_ID_IVSHMEM 0x1110 |
34 |
|
35 |
#define IVSHMEM_IOEVENTFD 0 |
36 |
#define IVSHMEM_MSI 1 |
37 |
|
38 |
#define IVSHMEM_PEER 0 |
39 |
#define IVSHMEM_MASTER 1 |
40 |
|
41 |
#define IVSHMEM_REG_BAR_SIZE 0x100 |
42 |
|
43 |
//#define DEBUG_IVSHMEM
|
44 |
#ifdef DEBUG_IVSHMEM
|
45 |
#define IVSHMEM_DPRINTF(fmt, ...) \
|
46 |
do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0) |
47 |
#else
|
48 |
#define IVSHMEM_DPRINTF(fmt, ...)
|
49 |
#endif
|
50 |
|
51 |
typedef struct Peer { |
52 |
int nb_eventfds;
|
53 |
EventNotifier *eventfds; |
54 |
} Peer; |
55 |
|
56 |
typedef struct EventfdEntry { |
57 |
PCIDevice *pdev; |
58 |
int vector;
|
59 |
} EventfdEntry; |
60 |
|
61 |
typedef struct IVShmemState { |
62 |
PCIDevice dev; |
63 |
uint32_t intrmask; |
64 |
uint32_t intrstatus; |
65 |
uint32_t doorbell; |
66 |
|
67 |
CharDriverState **eventfd_chr; |
68 |
CharDriverState *server_chr; |
69 |
MemoryRegion ivshmem_mmio; |
70 |
|
71 |
/* We might need to register the BAR before we actually have the memory.
|
72 |
* So prepare a container MemoryRegion for the BAR immediately and
|
73 |
* add a subregion when we have the memory.
|
74 |
*/
|
75 |
MemoryRegion bar; |
76 |
MemoryRegion ivshmem; |
77 |
uint64_t ivshmem_size; /* size of shared memory region */
|
78 |
uint32_t ivshmem_attr; |
79 |
uint32_t ivshmem_64bit; |
80 |
int shm_fd; /* shared memory file descriptor */ |
81 |
|
82 |
Peer *peers; |
83 |
int nb_peers; /* how many guests we have space for */ |
84 |
int max_peer; /* maximum numbered peer */ |
85 |
|
86 |
int vm_id;
|
87 |
uint32_t vectors; |
88 |
uint32_t features; |
89 |
EventfdEntry *eventfd_table; |
90 |
|
91 |
Error *migration_blocker; |
92 |
|
93 |
char * shmobj;
|
94 |
char * sizearg;
|
95 |
char * role;
|
96 |
int role_val; /* scalar to avoid multiple string comparisons */ |
97 |
} IVShmemState; |
98 |
|
99 |
/* registers for the Inter-VM shared memory device */
|
100 |
enum ivshmem_registers {
|
101 |
INTRMASK = 0,
|
102 |
INTRSTATUS = 4,
|
103 |
IVPOSITION = 8,
|
104 |
DOORBELL = 12,
|
105 |
}; |
106 |
|
107 |
static inline uint32_t ivshmem_has_feature(IVShmemState *ivs, |
108 |
unsigned int feature) { |
109 |
return (ivs->features & (1 << feature)); |
110 |
} |
111 |
|
112 |
static inline bool is_power_of_two(uint64_t x) { |
113 |
return (x & (x - 1)) == 0; |
114 |
} |
115 |
|
116 |
/* accessing registers - based on rtl8139 */
|
117 |
static void ivshmem_update_irq(IVShmemState *s, int val) |
118 |
{ |
119 |
int isr;
|
120 |
isr = (s->intrstatus & s->intrmask) & 0xffffffff;
|
121 |
|
122 |
/* don't print ISR resets */
|
123 |
if (isr) {
|
124 |
IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n",
|
125 |
isr ? 1 : 0, s->intrstatus, s->intrmask); |
126 |
} |
127 |
|
128 |
qemu_set_irq(s->dev.irq[0], (isr != 0)); |
129 |
} |
130 |
|
131 |
static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val) |
132 |
{ |
133 |
IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val);
|
134 |
|
135 |
s->intrmask = val; |
136 |
|
137 |
ivshmem_update_irq(s, val); |
138 |
} |
139 |
|
140 |
static uint32_t ivshmem_IntrMask_read(IVShmemState *s)
|
141 |
{ |
142 |
uint32_t ret = s->intrmask; |
143 |
|
144 |
IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret);
|
145 |
|
146 |
return ret;
|
147 |
} |
148 |
|
149 |
static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val) |
150 |
{ |
151 |
IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val);
|
152 |
|
153 |
s->intrstatus = val; |
154 |
|
155 |
ivshmem_update_irq(s, val); |
156 |
} |
157 |
|
158 |
static uint32_t ivshmem_IntrStatus_read(IVShmemState *s)
|
159 |
{ |
160 |
uint32_t ret = s->intrstatus; |
161 |
|
162 |
/* reading ISR clears all interrupts */
|
163 |
s->intrstatus = 0;
|
164 |
|
165 |
ivshmem_update_irq(s, 0);
|
166 |
|
167 |
return ret;
|
168 |
} |
169 |
|
170 |
static void ivshmem_io_write(void *opaque, hwaddr addr, |
171 |
uint64_t val, unsigned size)
|
172 |
{ |
173 |
IVShmemState *s = opaque; |
174 |
|
175 |
uint16_t dest = val >> 16;
|
176 |
uint16_t vector = val & 0xff;
|
177 |
|
178 |
addr &= 0xfc;
|
179 |
|
180 |
IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr); |
181 |
switch (addr)
|
182 |
{ |
183 |
case INTRMASK:
|
184 |
ivshmem_IntrMask_write(s, val); |
185 |
break;
|
186 |
|
187 |
case INTRSTATUS:
|
188 |
ivshmem_IntrStatus_write(s, val); |
189 |
break;
|
190 |
|
191 |
case DOORBELL:
|
192 |
/* check that dest VM ID is reasonable */
|
193 |
if (dest > s->max_peer) {
|
194 |
IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest);
|
195 |
break;
|
196 |
} |
197 |
|
198 |
/* check doorbell range */
|
199 |
if (vector < s->peers[dest].nb_eventfds) {
|
200 |
IVSHMEM_DPRINTF("Notifying VM %d on vector %d\n", dest, vector);
|
201 |
event_notifier_set(&s->peers[dest].eventfds[vector]); |
202 |
} |
203 |
break;
|
204 |
default:
|
205 |
IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest);
|
206 |
} |
207 |
} |
208 |
|
209 |
static uint64_t ivshmem_io_read(void *opaque, hwaddr addr, |
210 |
unsigned size)
|
211 |
{ |
212 |
|
213 |
IVShmemState *s = opaque; |
214 |
uint32_t ret; |
215 |
|
216 |
switch (addr)
|
217 |
{ |
218 |
case INTRMASK:
|
219 |
ret = ivshmem_IntrMask_read(s); |
220 |
break;
|
221 |
|
222 |
case INTRSTATUS:
|
223 |
ret = ivshmem_IntrStatus_read(s); |
224 |
break;
|
225 |
|
226 |
case IVPOSITION:
|
227 |
/* return my VM ID if the memory is mapped */
|
228 |
if (s->shm_fd > 0) { |
229 |
ret = s->vm_id; |
230 |
} else {
|
231 |
ret = -1;
|
232 |
} |
233 |
break;
|
234 |
|
235 |
default:
|
236 |
IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr); |
237 |
ret = 0;
|
238 |
} |
239 |
|
240 |
return ret;
|
241 |
} |
242 |
|
243 |
static const MemoryRegionOps ivshmem_mmio_ops = { |
244 |
.read = ivshmem_io_read, |
245 |
.write = ivshmem_io_write, |
246 |
.endianness = DEVICE_NATIVE_ENDIAN, |
247 |
.impl = { |
248 |
.min_access_size = 4,
|
249 |
.max_access_size = 4,
|
250 |
}, |
251 |
}; |
252 |
|
253 |
static void ivshmem_receive(void *opaque, const uint8_t *buf, int size) |
254 |
{ |
255 |
IVShmemState *s = opaque; |
256 |
|
257 |
ivshmem_IntrStatus_write(s, *buf); |
258 |
|
259 |
IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf);
|
260 |
} |
261 |
|
262 |
static int ivshmem_can_receive(void * opaque) |
263 |
{ |
264 |
return 8; |
265 |
} |
266 |
|
267 |
static void ivshmem_event(void *opaque, int event) |
268 |
{ |
269 |
IVSHMEM_DPRINTF("ivshmem_event %d\n", event);
|
270 |
} |
271 |
|
272 |
static void fake_irqfd(void *opaque, const uint8_t *buf, int size) { |
273 |
|
274 |
EventfdEntry *entry = opaque; |
275 |
PCIDevice *pdev = entry->pdev; |
276 |
|
277 |
IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector);
|
278 |
msix_notify(pdev, entry->vector); |
279 |
} |
280 |
|
281 |
static CharDriverState* create_eventfd_chr_device(void * opaque, EventNotifier *n, |
282 |
int vector)
|
283 |
{ |
284 |
/* create a event character device based on the passed eventfd */
|
285 |
IVShmemState *s = opaque; |
286 |
CharDriverState * chr; |
287 |
int eventfd = event_notifier_get_fd(n);
|
288 |
|
289 |
chr = qemu_chr_open_eventfd(eventfd); |
290 |
|
291 |
if (chr == NULL) { |
292 |
fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd);
|
293 |
exit(-1);
|
294 |
} |
295 |
qemu_chr_fe_claim_no_fail(chr); |
296 |
|
297 |
/* if MSI is supported we need multiple interrupts */
|
298 |
if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
|
299 |
s->eventfd_table[vector].pdev = &s->dev; |
300 |
s->eventfd_table[vector].vector = vector; |
301 |
|
302 |
qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd, |
303 |
ivshmem_event, &s->eventfd_table[vector]); |
304 |
} else {
|
305 |
qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive, |
306 |
ivshmem_event, s); |
307 |
} |
308 |
|
309 |
return chr;
|
310 |
|
311 |
} |
312 |
|
313 |
static int check_shm_size(IVShmemState *s, int fd) { |
314 |
/* check that the guest isn't going to try and map more memory than the
|
315 |
* the object has allocated return -1 to indicate error */
|
316 |
|
317 |
struct stat buf;
|
318 |
|
319 |
fstat(fd, &buf); |
320 |
|
321 |
if (s->ivshmem_size > buf.st_size) {
|
322 |
fprintf(stderr, |
323 |
"IVSHMEM ERROR: Requested memory size greater"
|
324 |
" than shared object size (%" PRIu64 " > %" PRIu64")\n", |
325 |
s->ivshmem_size, (uint64_t)buf.st_size); |
326 |
return -1; |
327 |
} else {
|
328 |
return 0; |
329 |
} |
330 |
} |
331 |
|
332 |
/* create the shared memory BAR when we are not using the server, so we can
|
333 |
* create the BAR and map the memory immediately */
|
334 |
static void create_shared_memory_BAR(IVShmemState *s, int fd) { |
335 |
|
336 |
void * ptr;
|
337 |
|
338 |
s->shm_fd = fd; |
339 |
|
340 |
ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); |
341 |
|
342 |
memory_region_init_ram_ptr(&s->ivshmem, "ivshmem.bar2",
|
343 |
s->ivshmem_size, ptr); |
344 |
vmstate_register_ram(&s->ivshmem, &s->dev.qdev); |
345 |
memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
|
346 |
|
347 |
/* region for shared memory */
|
348 |
pci_register_bar(&s->dev, 2, s->ivshmem_attr, &s->bar);
|
349 |
} |
350 |
|
351 |
static void ivshmem_add_eventfd(IVShmemState *s, int posn, int i) |
352 |
{ |
353 |
memory_region_add_eventfd(&s->ivshmem_mmio, |
354 |
DOORBELL, |
355 |
4,
|
356 |
true,
|
357 |
(posn << 16) | i,
|
358 |
&s->peers[posn].eventfds[i]); |
359 |
} |
360 |
|
361 |
static void ivshmem_del_eventfd(IVShmemState *s, int posn, int i) |
362 |
{ |
363 |
memory_region_del_eventfd(&s->ivshmem_mmio, |
364 |
DOORBELL, |
365 |
4,
|
366 |
true,
|
367 |
(posn << 16) | i,
|
368 |
&s->peers[posn].eventfds[i]); |
369 |
} |
370 |
|
371 |
static void close_guest_eventfds(IVShmemState *s, int posn) |
372 |
{ |
373 |
int i, guest_curr_max;
|
374 |
|
375 |
if (!ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
|
376 |
return;
|
377 |
} |
378 |
|
379 |
guest_curr_max = s->peers[posn].nb_eventfds; |
380 |
|
381 |
memory_region_transaction_begin(); |
382 |
for (i = 0; i < guest_curr_max; i++) { |
383 |
ivshmem_del_eventfd(s, posn, i); |
384 |
} |
385 |
memory_region_transaction_commit(); |
386 |
for (i = 0; i < guest_curr_max; i++) { |
387 |
event_notifier_cleanup(&s->peers[posn].eventfds[i]); |
388 |
} |
389 |
|
390 |
g_free(s->peers[posn].eventfds); |
391 |
s->peers[posn].nb_eventfds = 0;
|
392 |
} |
393 |
|
394 |
/* this function increase the dynamic storage need to store data about other
|
395 |
* guests */
|
396 |
static void increase_dynamic_storage(IVShmemState *s, int new_min_size) { |
397 |
|
398 |
int j, old_nb_alloc;
|
399 |
|
400 |
old_nb_alloc = s->nb_peers; |
401 |
|
402 |
while (new_min_size >= s->nb_peers)
|
403 |
s->nb_peers = s->nb_peers * 2;
|
404 |
|
405 |
IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers);
|
406 |
s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer));
|
407 |
|
408 |
/* zero out new pointers */
|
409 |
for (j = old_nb_alloc; j < s->nb_peers; j++) {
|
410 |
s->peers[j].eventfds = NULL;
|
411 |
s->peers[j].nb_eventfds = 0;
|
412 |
} |
413 |
} |
414 |
|
415 |
static void ivshmem_read(void *opaque, const uint8_t * buf, int flags) |
416 |
{ |
417 |
IVShmemState *s = opaque; |
418 |
int incoming_fd, tmp_fd;
|
419 |
int guest_max_eventfd;
|
420 |
long incoming_posn;
|
421 |
|
422 |
memcpy(&incoming_posn, buf, sizeof(long)); |
423 |
/* pick off s->server_chr->msgfd and store it, posn should accompany msg */
|
424 |
tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr); |
425 |
IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd);
|
426 |
|
427 |
/* make sure we have enough space for this guest */
|
428 |
if (incoming_posn >= s->nb_peers) {
|
429 |
increase_dynamic_storage(s, incoming_posn); |
430 |
} |
431 |
|
432 |
if (tmp_fd == -1) { |
433 |
/* if posn is positive and unseen before then this is our posn*/
|
434 |
if ((incoming_posn >= 0) && |
435 |
(s->peers[incoming_posn].eventfds == NULL)) {
|
436 |
/* receive our posn */
|
437 |
s->vm_id = incoming_posn; |
438 |
return;
|
439 |
} else {
|
440 |
/* otherwise an fd == -1 means an existing guest has gone away */
|
441 |
IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn);
|
442 |
close_guest_eventfds(s, incoming_posn); |
443 |
return;
|
444 |
} |
445 |
} |
446 |
|
447 |
/* because of the implementation of get_msgfd, we need a dup */
|
448 |
incoming_fd = dup(tmp_fd); |
449 |
|
450 |
if (incoming_fd == -1) { |
451 |
fprintf(stderr, "could not allocate file descriptor %s\n",
|
452 |
strerror(errno)); |
453 |
return;
|
454 |
} |
455 |
|
456 |
/* if the position is -1, then it's shared memory region fd */
|
457 |
if (incoming_posn == -1) { |
458 |
|
459 |
void * map_ptr;
|
460 |
|
461 |
s->max_peer = 0;
|
462 |
|
463 |
if (check_shm_size(s, incoming_fd) == -1) { |
464 |
exit(-1);
|
465 |
} |
466 |
|
467 |
/* mmap the region and map into the BAR2 */
|
468 |
map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED,
|
469 |
incoming_fd, 0);
|
470 |
memory_region_init_ram_ptr(&s->ivshmem, |
471 |
"ivshmem.bar2", s->ivshmem_size, map_ptr);
|
472 |
vmstate_register_ram(&s->ivshmem, &s->dev.qdev); |
473 |
|
474 |
IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n", |
475 |
s->ivshmem_offset, s->ivshmem_size); |
476 |
|
477 |
memory_region_add_subregion(&s->bar, 0, &s->ivshmem);
|
478 |
|
479 |
/* only store the fd if it is successfully mapped */
|
480 |
s->shm_fd = incoming_fd; |
481 |
|
482 |
return;
|
483 |
} |
484 |
|
485 |
/* each guest has an array of eventfds, and we keep track of how many
|
486 |
* guests for each VM */
|
487 |
guest_max_eventfd = s->peers[incoming_posn].nb_eventfds; |
488 |
|
489 |
if (guest_max_eventfd == 0) { |
490 |
/* one eventfd per MSI vector */
|
491 |
s->peers[incoming_posn].eventfds = g_new(EventNotifier, s->vectors); |
492 |
} |
493 |
|
494 |
/* this is an eventfd for a particular guest VM */
|
495 |
IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn,
|
496 |
guest_max_eventfd, incoming_fd); |
497 |
event_notifier_init_fd(&s->peers[incoming_posn].eventfds[guest_max_eventfd], |
498 |
incoming_fd); |
499 |
|
500 |
/* increment count for particular guest */
|
501 |
s->peers[incoming_posn].nb_eventfds++; |
502 |
|
503 |
/* keep track of the maximum VM ID */
|
504 |
if (incoming_posn > s->max_peer) {
|
505 |
s->max_peer = incoming_posn; |
506 |
} |
507 |
|
508 |
if (incoming_posn == s->vm_id) {
|
509 |
s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s, |
510 |
&s->peers[s->vm_id].eventfds[guest_max_eventfd], |
511 |
guest_max_eventfd); |
512 |
} |
513 |
|
514 |
if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) {
|
515 |
ivshmem_add_eventfd(s, incoming_posn, guest_max_eventfd); |
516 |
} |
517 |
} |
518 |
|
519 |
/* Select the MSI-X vectors used by device.
|
520 |
* ivshmem maps events to vectors statically, so
|
521 |
* we just enable all vectors on init and after reset. */
|
522 |
static void ivshmem_use_msix(IVShmemState * s) |
523 |
{ |
524 |
int i;
|
525 |
|
526 |
if (!msix_present(&s->dev)) {
|
527 |
return;
|
528 |
} |
529 |
|
530 |
for (i = 0; i < s->vectors; i++) { |
531 |
msix_vector_use(&s->dev, i); |
532 |
} |
533 |
} |
534 |
|
535 |
static void ivshmem_reset(DeviceState *d) |
536 |
{ |
537 |
IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d); |
538 |
|
539 |
s->intrstatus = 0;
|
540 |
ivshmem_use_msix(s); |
541 |
} |
542 |
|
543 |
static uint64_t ivshmem_get_size(IVShmemState * s) {
|
544 |
|
545 |
uint64_t value; |
546 |
char *ptr;
|
547 |
|
548 |
value = strtoull(s->sizearg, &ptr, 10);
|
549 |
switch (*ptr) {
|
550 |
case 0: case 'M': case 'm': |
551 |
value <<= 20;
|
552 |
break;
|
553 |
case 'G': case 'g': |
554 |
value <<= 30;
|
555 |
break;
|
556 |
default:
|
557 |
fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg);
|
558 |
exit(1);
|
559 |
} |
560 |
|
561 |
/* BARs must be a power of 2 */
|
562 |
if (!is_power_of_two(value)) {
|
563 |
fprintf(stderr, "ivshmem: size must be power of 2\n");
|
564 |
exit(1);
|
565 |
} |
566 |
|
567 |
return value;
|
568 |
} |
569 |
|
570 |
static void ivshmem_setup_msi(IVShmemState * s) |
571 |
{ |
572 |
if (msix_init_exclusive_bar(&s->dev, s->vectors, 1)) { |
573 |
IVSHMEM_DPRINTF("msix initialization failed\n");
|
574 |
exit(1);
|
575 |
} |
576 |
|
577 |
IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors);
|
578 |
|
579 |
/* allocate QEMU char devices for receiving interrupts */
|
580 |
s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry));
|
581 |
|
582 |
ivshmem_use_msix(s); |
583 |
} |
584 |
|
585 |
static void ivshmem_save(QEMUFile* f, void *opaque) |
586 |
{ |
587 |
IVShmemState *proxy = opaque; |
588 |
|
589 |
IVSHMEM_DPRINTF("ivshmem_save\n");
|
590 |
pci_device_save(&proxy->dev, f); |
591 |
|
592 |
if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
|
593 |
msix_save(&proxy->dev, f); |
594 |
} else {
|
595 |
qemu_put_be32(f, proxy->intrstatus); |
596 |
qemu_put_be32(f, proxy->intrmask); |
597 |
} |
598 |
|
599 |
} |
600 |
|
601 |
static int ivshmem_load(QEMUFile* f, void *opaque, int version_id) |
602 |
{ |
603 |
IVSHMEM_DPRINTF("ivshmem_load\n");
|
604 |
|
605 |
IVShmemState *proxy = opaque; |
606 |
int ret;
|
607 |
|
608 |
if (version_id > 0) { |
609 |
return -EINVAL;
|
610 |
} |
611 |
|
612 |
if (proxy->role_val == IVSHMEM_PEER) {
|
613 |
fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n");
|
614 |
return -EINVAL;
|
615 |
} |
616 |
|
617 |
ret = pci_device_load(&proxy->dev, f); |
618 |
if (ret) {
|
619 |
return ret;
|
620 |
} |
621 |
|
622 |
if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) {
|
623 |
msix_load(&proxy->dev, f); |
624 |
ivshmem_use_msix(proxy); |
625 |
} else {
|
626 |
proxy->intrstatus = qemu_get_be32(f); |
627 |
proxy->intrmask = qemu_get_be32(f); |
628 |
} |
629 |
|
630 |
return 0; |
631 |
} |
632 |
|
633 |
static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address, |
634 |
uint32_t val, int len)
|
635 |
{ |
636 |
pci_default_write_config(pci_dev, address, val, len); |
637 |
msix_write_config(pci_dev, address, val, len); |
638 |
} |
639 |
|
640 |
static int pci_ivshmem_init(PCIDevice *dev) |
641 |
{ |
642 |
IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); |
643 |
uint8_t *pci_conf; |
644 |
|
645 |
if (s->sizearg == NULL) |
646 |
s->ivshmem_size = 4 << 20; /* 4 MB default */ |
647 |
else {
|
648 |
s->ivshmem_size = ivshmem_get_size(s); |
649 |
} |
650 |
|
651 |
register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load, |
652 |
dev); |
653 |
|
654 |
/* IRQFD requires MSI */
|
655 |
if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) &&
|
656 |
!ivshmem_has_feature(s, IVSHMEM_MSI)) { |
657 |
fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n");
|
658 |
exit(1);
|
659 |
} |
660 |
|
661 |
/* check that role is reasonable */
|
662 |
if (s->role) {
|
663 |
if (strncmp(s->role, "peer", 5) == 0) { |
664 |
s->role_val = IVSHMEM_PEER; |
665 |
} else if (strncmp(s->role, "master", 7) == 0) { |
666 |
s->role_val = IVSHMEM_MASTER; |
667 |
} else {
|
668 |
fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n");
|
669 |
exit(1);
|
670 |
} |
671 |
} else {
|
672 |
s->role_val = IVSHMEM_MASTER; /* default */
|
673 |
} |
674 |
|
675 |
if (s->role_val == IVSHMEM_PEER) {
|
676 |
error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION, |
677 |
"peer mode", "ivshmem"); |
678 |
migrate_add_blocker(s->migration_blocker); |
679 |
} |
680 |
|
681 |
pci_conf = s->dev.config; |
682 |
pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY; |
683 |
|
684 |
pci_config_set_interrupt_pin(pci_conf, 1);
|
685 |
|
686 |
s->shm_fd = 0;
|
687 |
|
688 |
memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s, |
689 |
"ivshmem-mmio", IVSHMEM_REG_BAR_SIZE);
|
690 |
|
691 |
/* region for registers*/
|
692 |
pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY,
|
693 |
&s->ivshmem_mmio); |
694 |
|
695 |
memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size);
|
696 |
s->ivshmem_attr = PCI_BASE_ADDRESS_SPACE_MEMORY | |
697 |
PCI_BASE_ADDRESS_MEM_PREFETCH; |
698 |
if (s->ivshmem_64bit) {
|
699 |
s->ivshmem_attr |= PCI_BASE_ADDRESS_MEM_TYPE_64; |
700 |
} |
701 |
|
702 |
if ((s->server_chr != NULL) && |
703 |
(strncmp(s->server_chr->filename, "unix:", 5) == 0)) { |
704 |
/* if we get a UNIX socket as the parameter we will talk
|
705 |
* to the ivshmem server to receive the memory region */
|
706 |
|
707 |
if (s->shmobj != NULL) { |
708 |
fprintf(stderr, "WARNING: do not specify both 'chardev' "
|
709 |
"and 'shm' with ivshmem\n");
|
710 |
} |
711 |
|
712 |
IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n",
|
713 |
s->server_chr->filename); |
714 |
|
715 |
if (ivshmem_has_feature(s, IVSHMEM_MSI)) {
|
716 |
ivshmem_setup_msi(s); |
717 |
} |
718 |
|
719 |
/* we allocate enough space for 16 guests and grow as needed */
|
720 |
s->nb_peers = 16;
|
721 |
s->vm_id = -1;
|
722 |
|
723 |
/* allocate/initialize space for interrupt handling */
|
724 |
s->peers = g_malloc0(s->nb_peers * sizeof(Peer));
|
725 |
|
726 |
pci_register_bar(&s->dev, 2, s->ivshmem_attr, &s->bar);
|
727 |
|
728 |
s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *));
|
729 |
|
730 |
qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read, |
731 |
ivshmem_event, s); |
732 |
} else {
|
733 |
/* just map the file immediately, we're not using a server */
|
734 |
int fd;
|
735 |
|
736 |
if (s->shmobj == NULL) { |
737 |
fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n");
|
738 |
exit(1);
|
739 |
} |
740 |
|
741 |
IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj);
|
742 |
|
743 |
/* try opening with O_EXCL and if it succeeds zero the memory
|
744 |
* by truncating to 0 */
|
745 |
if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL,
|
746 |
S_IRWXU|S_IRWXG|S_IRWXO)) > 0) {
|
747 |
/* truncate file to length PCI device's memory */
|
748 |
if (ftruncate(fd, s->ivshmem_size) != 0) { |
749 |
fprintf(stderr, "ivshmem: could not truncate shared file\n");
|
750 |
} |
751 |
|
752 |
} else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR, |
753 |
S_IRWXU|S_IRWXG|S_IRWXO)) < 0) {
|
754 |
fprintf(stderr, "ivshmem: could not open shared file\n");
|
755 |
exit(-1);
|
756 |
|
757 |
} |
758 |
|
759 |
if (check_shm_size(s, fd) == -1) { |
760 |
exit(-1);
|
761 |
} |
762 |
|
763 |
create_shared_memory_BAR(s, fd); |
764 |
|
765 |
} |
766 |
|
767 |
s->dev.config_write = ivshmem_write_config; |
768 |
|
769 |
return 0; |
770 |
} |
771 |
|
772 |
static void pci_ivshmem_uninit(PCIDevice *dev) |
773 |
{ |
774 |
IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); |
775 |
|
776 |
if (s->migration_blocker) {
|
777 |
migrate_del_blocker(s->migration_blocker); |
778 |
error_free(s->migration_blocker); |
779 |
} |
780 |
|
781 |
memory_region_destroy(&s->ivshmem_mmio); |
782 |
memory_region_del_subregion(&s->bar, &s->ivshmem); |
783 |
vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev); |
784 |
memory_region_destroy(&s->ivshmem); |
785 |
memory_region_destroy(&s->bar); |
786 |
unregister_savevm(&dev->qdev, "ivshmem", s);
|
787 |
} |
788 |
|
789 |
static Property ivshmem_properties[] = {
|
790 |
DEFINE_PROP_CHR("chardev", IVShmemState, server_chr),
|
791 |
DEFINE_PROP_STRING("size", IVShmemState, sizearg),
|
792 |
DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1), |
793 |
DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false), |
794 |
DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true), |
795 |
DEFINE_PROP_STRING("shm", IVShmemState, shmobj),
|
796 |
DEFINE_PROP_STRING("role", IVShmemState, role),
|
797 |
DEFINE_PROP_UINT32("use64", IVShmemState, ivshmem_64bit, 1), |
798 |
DEFINE_PROP_END_OF_LIST(), |
799 |
}; |
800 |
|
801 |
static void ivshmem_class_init(ObjectClass *klass, void *data) |
802 |
{ |
803 |
DeviceClass *dc = DEVICE_CLASS(klass); |
804 |
PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); |
805 |
|
806 |
k->init = pci_ivshmem_init; |
807 |
k->exit = pci_ivshmem_uninit; |
808 |
k->vendor_id = PCI_VENDOR_ID_IVSHMEM; |
809 |
k->device_id = PCI_DEVICE_ID_IVSHMEM; |
810 |
k->class_id = PCI_CLASS_MEMORY_RAM; |
811 |
dc->reset = ivshmem_reset; |
812 |
dc->props = ivshmem_properties; |
813 |
} |
814 |
|
815 |
static const TypeInfo ivshmem_info = { |
816 |
.name = "ivshmem",
|
817 |
.parent = TYPE_PCI_DEVICE, |
818 |
.instance_size = sizeof(IVShmemState),
|
819 |
.class_init = ivshmem_class_init, |
820 |
}; |
821 |
|
822 |
static void ivshmem_register_types(void) |
823 |
{ |
824 |
type_register_static(&ivshmem_info); |
825 |
} |
826 |
|
827 |
type_init(ivshmem_register_types) |