]>
Commit | Line | Data |
---|---|---|
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.h" | |
20 | #include "pc.h" | |
21 | #include "pci.h" | |
22 | #include "msix.h" | |
23 | #include "kvm.h" | |
24 | #include "migration.h" | |
25 | #include "qerror.h" | |
26 | ||
27 | #include <sys/mman.h> | |
28 | #include <sys/types.h> | |
29 | ||
30 | #define IVSHMEM_IOEVENTFD 0 | |
31 | #define IVSHMEM_MSI 1 | |
32 | ||
33 | #define IVSHMEM_PEER 0 | |
34 | #define IVSHMEM_MASTER 1 | |
35 | ||
36 | #define IVSHMEM_REG_BAR_SIZE 0x100 | |
37 | ||
38 | //#define DEBUG_IVSHMEM | |
39 | #ifdef DEBUG_IVSHMEM | |
40 | #define IVSHMEM_DPRINTF(fmt, ...) \ | |
41 | do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0) | |
42 | #else | |
43 | #define IVSHMEM_DPRINTF(fmt, ...) | |
44 | #endif | |
45 | ||
46 | typedef struct Peer { | |
47 | int nb_eventfds; | |
48 | int *eventfds; | |
49 | } Peer; | |
50 | ||
51 | typedef struct EventfdEntry { | |
52 | PCIDevice *pdev; | |
53 | int vector; | |
54 | } EventfdEntry; | |
55 | ||
56 | typedef struct IVShmemState { | |
57 | PCIDevice dev; | |
58 | uint32_t intrmask; | |
59 | uint32_t intrstatus; | |
60 | uint32_t doorbell; | |
61 | ||
62 | CharDriverState **eventfd_chr; | |
63 | CharDriverState *server_chr; | |
64 | MemoryRegion ivshmem_mmio; | |
65 | ||
66 | pcibus_t mmio_addr; | |
67 | /* We might need to register the BAR before we actually have the memory. | |
68 | * So prepare a container MemoryRegion for the BAR immediately and | |
69 | * add a subregion when we have the memory. | |
70 | */ | |
71 | MemoryRegion bar; | |
72 | MemoryRegion ivshmem; | |
73 | uint64_t ivshmem_size; /* size of shared memory region */ | |
74 | int shm_fd; /* shared memory file descriptor */ | |
75 | ||
76 | Peer *peers; | |
77 | int nb_peers; /* how many guests we have space for */ | |
78 | int max_peer; /* maximum numbered peer */ | |
79 | ||
80 | int vm_id; | |
81 | uint32_t vectors; | |
82 | uint32_t features; | |
83 | EventfdEntry *eventfd_table; | |
84 | ||
85 | Error *migration_blocker; | |
86 | ||
87 | char * shmobj; | |
88 | char * sizearg; | |
89 | char * role; | |
90 | int role_val; /* scalar to avoid multiple string comparisons */ | |
91 | } IVShmemState; | |
92 | ||
93 | /* registers for the Inter-VM shared memory device */ | |
94 | enum ivshmem_registers { | |
95 | INTRMASK = 0, | |
96 | INTRSTATUS = 4, | |
97 | IVPOSITION = 8, | |
98 | DOORBELL = 12, | |
99 | }; | |
100 | ||
101 | static inline uint32_t ivshmem_has_feature(IVShmemState *ivs, | |
102 | unsigned int feature) { | |
103 | return (ivs->features & (1 << feature)); | |
104 | } | |
105 | ||
106 | static inline bool is_power_of_two(uint64_t x) { | |
107 | return (x & (x - 1)) == 0; | |
108 | } | |
109 | ||
110 | /* accessing registers - based on rtl8139 */ | |
111 | static void ivshmem_update_irq(IVShmemState *s, int val) | |
112 | { | |
113 | int isr; | |
114 | isr = (s->intrstatus & s->intrmask) & 0xffffffff; | |
115 | ||
116 | /* don't print ISR resets */ | |
117 | if (isr) { | |
118 | IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n", | |
119 | isr ? 1 : 0, s->intrstatus, s->intrmask); | |
120 | } | |
121 | ||
122 | qemu_set_irq(s->dev.irq[0], (isr != 0)); | |
123 | } | |
124 | ||
125 | static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val) | |
126 | { | |
127 | IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val); | |
128 | ||
129 | s->intrmask = val; | |
130 | ||
131 | ivshmem_update_irq(s, val); | |
132 | } | |
133 | ||
134 | static uint32_t ivshmem_IntrMask_read(IVShmemState *s) | |
135 | { | |
136 | uint32_t ret = s->intrmask; | |
137 | ||
138 | IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret); | |
139 | ||
140 | return ret; | |
141 | } | |
142 | ||
143 | static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val) | |
144 | { | |
145 | IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val); | |
146 | ||
147 | s->intrstatus = val; | |
148 | ||
149 | ivshmem_update_irq(s, val); | |
150 | return; | |
151 | } | |
152 | ||
153 | static uint32_t ivshmem_IntrStatus_read(IVShmemState *s) | |
154 | { | |
155 | uint32_t ret = s->intrstatus; | |
156 | ||
157 | /* reading ISR clears all interrupts */ | |
158 | s->intrstatus = 0; | |
159 | ||
160 | ivshmem_update_irq(s, 0); | |
161 | ||
162 | return ret; | |
163 | } | |
164 | ||
165 | static void ivshmem_io_write(void *opaque, target_phys_addr_t addr, | |
166 | uint64_t val, unsigned size) | |
167 | { | |
168 | IVShmemState *s = opaque; | |
169 | ||
170 | uint64_t write_one = 1; | |
171 | uint16_t dest = val >> 16; | |
172 | uint16_t vector = val & 0xff; | |
173 | ||
174 | addr &= 0xfc; | |
175 | ||
176 | IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr); | |
177 | switch (addr) | |
178 | { | |
179 | case INTRMASK: | |
180 | ivshmem_IntrMask_write(s, val); | |
181 | break; | |
182 | ||
183 | case INTRSTATUS: | |
184 | ivshmem_IntrStatus_write(s, val); | |
185 | break; | |
186 | ||
187 | case DOORBELL: | |
188 | /* check that dest VM ID is reasonable */ | |
189 | if (dest > s->max_peer) { | |
190 | IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest); | |
191 | break; | |
192 | } | |
193 | ||
194 | /* check doorbell range */ | |
195 | if (vector < s->peers[dest].nb_eventfds) { | |
196 | IVSHMEM_DPRINTF("Writing %" PRId64 " to VM %d on vector %d\n", | |
197 | write_one, dest, vector); | |
198 | if (write(s->peers[dest].eventfds[vector], | |
199 | &(write_one), 8) != 8) { | |
200 | IVSHMEM_DPRINTF("error writing to eventfd\n"); | |
201 | } | |
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, target_phys_addr_t 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, int eventfd, | |
282 | int vector) | |
283 | { | |
284 | /* create a event character device based on the passed eventfd */ | |
285 | IVShmemState *s = opaque; | |
286 | CharDriverState * chr; | |
287 | ||
288 | chr = qemu_chr_open_eventfd(eventfd); | |
289 | ||
290 | if (chr == NULL) { | |
291 | fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd); | |
292 | exit(-1); | |
293 | } | |
294 | ||
295 | /* if MSI is supported we need multiple interrupts */ | |
296 | if (ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
297 | s->eventfd_table[vector].pdev = &s->dev; | |
298 | s->eventfd_table[vector].vector = vector; | |
299 | ||
300 | qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd, | |
301 | ivshmem_event, &s->eventfd_table[vector]); | |
302 | } else { | |
303 | qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive, | |
304 | ivshmem_event, s); | |
305 | } | |
306 | ||
307 | return chr; | |
308 | ||
309 | } | |
310 | ||
311 | static int check_shm_size(IVShmemState *s, int fd) { | |
312 | /* check that the guest isn't going to try and map more memory than the | |
313 | * the object has allocated return -1 to indicate error */ | |
314 | ||
315 | struct stat buf; | |
316 | ||
317 | fstat(fd, &buf); | |
318 | ||
319 | if (s->ivshmem_size > buf.st_size) { | |
320 | fprintf(stderr, | |
321 | "IVSHMEM ERROR: Requested memory size greater" | |
322 | " than shared object size (%" PRIu64 " > %" PRIu64")\n", | |
323 | s->ivshmem_size, (uint64_t)buf.st_size); | |
324 | return -1; | |
325 | } else { | |
326 | return 0; | |
327 | } | |
328 | } | |
329 | ||
330 | /* create the shared memory BAR when we are not using the server, so we can | |
331 | * create the BAR and map the memory immediately */ | |
332 | static void create_shared_memory_BAR(IVShmemState *s, int fd) { | |
333 | ||
334 | void * ptr; | |
335 | ||
336 | s->shm_fd = fd; | |
337 | ||
338 | ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); | |
339 | ||
340 | memory_region_init_ram_ptr(&s->ivshmem, "ivshmem.bar2", | |
341 | s->ivshmem_size, ptr); | |
342 | vmstate_register_ram(&s->ivshmem, &s->dev.qdev); | |
343 | memory_region_add_subregion(&s->bar, 0, &s->ivshmem); | |
344 | ||
345 | /* region for shared memory */ | |
346 | pci_register_bar(&s->dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar); | |
347 | } | |
348 | ||
349 | static void close_guest_eventfds(IVShmemState *s, int posn) | |
350 | { | |
351 | int i, guest_curr_max; | |
352 | ||
353 | guest_curr_max = s->peers[posn].nb_eventfds; | |
354 | ||
355 | for (i = 0; i < guest_curr_max; i++) { | |
356 | kvm_set_ioeventfd_mmio(s->peers[posn].eventfds[i], | |
357 | s->mmio_addr + DOORBELL, (posn << 16) | i, 0, 4); | |
358 | close(s->peers[posn].eventfds[i]); | |
359 | } | |
360 | ||
361 | g_free(s->peers[posn].eventfds); | |
362 | s->peers[posn].nb_eventfds = 0; | |
363 | } | |
364 | ||
365 | static void setup_ioeventfds(IVShmemState *s) { | |
366 | ||
367 | int i, j; | |
368 | ||
369 | for (i = 0; i <= s->max_peer; i++) { | |
370 | for (j = 0; j < s->peers[i].nb_eventfds; j++) { | |
371 | memory_region_add_eventfd(&s->ivshmem_mmio, | |
372 | DOORBELL, | |
373 | 4, | |
374 | true, | |
375 | (i << 16) | j, | |
376 | s->peers[i].eventfds[j]); | |
377 | } | |
378 | } | |
379 | } | |
380 | ||
381 | /* this function increase the dynamic storage need to store data about other | |
382 | * guests */ | |
383 | static void increase_dynamic_storage(IVShmemState *s, int new_min_size) { | |
384 | ||
385 | int j, old_nb_alloc; | |
386 | ||
387 | old_nb_alloc = s->nb_peers; | |
388 | ||
389 | while (new_min_size >= s->nb_peers) | |
390 | s->nb_peers = s->nb_peers * 2; | |
391 | ||
392 | IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers); | |
393 | s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer)); | |
394 | ||
395 | /* zero out new pointers */ | |
396 | for (j = old_nb_alloc; j < s->nb_peers; j++) { | |
397 | s->peers[j].eventfds = NULL; | |
398 | s->peers[j].nb_eventfds = 0; | |
399 | } | |
400 | } | |
401 | ||
402 | static void ivshmem_read(void *opaque, const uint8_t * buf, int flags) | |
403 | { | |
404 | IVShmemState *s = opaque; | |
405 | int incoming_fd, tmp_fd; | |
406 | int guest_max_eventfd; | |
407 | long incoming_posn; | |
408 | ||
409 | memcpy(&incoming_posn, buf, sizeof(long)); | |
410 | /* pick off s->server_chr->msgfd and store it, posn should accompany msg */ | |
411 | tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr); | |
412 | IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd); | |
413 | ||
414 | /* make sure we have enough space for this guest */ | |
415 | if (incoming_posn >= s->nb_peers) { | |
416 | increase_dynamic_storage(s, incoming_posn); | |
417 | } | |
418 | ||
419 | if (tmp_fd == -1) { | |
420 | /* if posn is positive and unseen before then this is our posn*/ | |
421 | if ((incoming_posn >= 0) && | |
422 | (s->peers[incoming_posn].eventfds == NULL)) { | |
423 | /* receive our posn */ | |
424 | s->vm_id = incoming_posn; | |
425 | return; | |
426 | } else { | |
427 | /* otherwise an fd == -1 means an existing guest has gone away */ | |
428 | IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn); | |
429 | close_guest_eventfds(s, incoming_posn); | |
430 | return; | |
431 | } | |
432 | } | |
433 | ||
434 | /* because of the implementation of get_msgfd, we need a dup */ | |
435 | incoming_fd = dup(tmp_fd); | |
436 | ||
437 | if (incoming_fd == -1) { | |
438 | fprintf(stderr, "could not allocate file descriptor %s\n", | |
439 | strerror(errno)); | |
440 | return; | |
441 | } | |
442 | ||
443 | /* if the position is -1, then it's shared memory region fd */ | |
444 | if (incoming_posn == -1) { | |
445 | ||
446 | void * map_ptr; | |
447 | ||
448 | s->max_peer = 0; | |
449 | ||
450 | if (check_shm_size(s, incoming_fd) == -1) { | |
451 | exit(-1); | |
452 | } | |
453 | ||
454 | /* mmap the region and map into the BAR2 */ | |
455 | map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, | |
456 | incoming_fd, 0); | |
457 | memory_region_init_ram_ptr(&s->ivshmem, | |
458 | "ivshmem.bar2", s->ivshmem_size, map_ptr); | |
459 | vmstate_register_ram(&s->ivshmem, &s->dev.qdev); | |
460 | ||
461 | IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n", | |
462 | s->ivshmem_offset, s->ivshmem_size); | |
463 | ||
464 | memory_region_add_subregion(&s->bar, 0, &s->ivshmem); | |
465 | ||
466 | /* only store the fd if it is successfully mapped */ | |
467 | s->shm_fd = incoming_fd; | |
468 | ||
469 | return; | |
470 | } | |
471 | ||
472 | /* each guest has an array of eventfds, and we keep track of how many | |
473 | * guests for each VM */ | |
474 | guest_max_eventfd = s->peers[incoming_posn].nb_eventfds; | |
475 | ||
476 | if (guest_max_eventfd == 0) { | |
477 | /* one eventfd per MSI vector */ | |
478 | s->peers[incoming_posn].eventfds = (int *) g_malloc(s->vectors * | |
479 | sizeof(int)); | |
480 | } | |
481 | ||
482 | /* this is an eventfd for a particular guest VM */ | |
483 | IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn, | |
484 | guest_max_eventfd, incoming_fd); | |
485 | s->peers[incoming_posn].eventfds[guest_max_eventfd] = incoming_fd; | |
486 | ||
487 | /* increment count for particular guest */ | |
488 | s->peers[incoming_posn].nb_eventfds++; | |
489 | ||
490 | /* keep track of the maximum VM ID */ | |
491 | if (incoming_posn > s->max_peer) { | |
492 | s->max_peer = incoming_posn; | |
493 | } | |
494 | ||
495 | if (incoming_posn == s->vm_id) { | |
496 | s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s, | |
497 | s->peers[s->vm_id].eventfds[guest_max_eventfd], | |
498 | guest_max_eventfd); | |
499 | } | |
500 | ||
501 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) { | |
502 | if (kvm_set_ioeventfd_mmio(incoming_fd, s->mmio_addr + DOORBELL, | |
503 | (incoming_posn << 16) | guest_max_eventfd, 1, 4) < 0) { | |
504 | fprintf(stderr, "ivshmem: ioeventfd not available\n"); | |
505 | } | |
506 | } | |
507 | ||
508 | return; | |
509 | } | |
510 | ||
511 | /* Select the MSI-X vectors used by device. | |
512 | * ivshmem maps events to vectors statically, so | |
513 | * we just enable all vectors on init and after reset. */ | |
514 | static void ivshmem_use_msix(IVShmemState * s) | |
515 | { | |
516 | int i; | |
517 | ||
518 | if (!msix_present(&s->dev)) { | |
519 | return; | |
520 | } | |
521 | ||
522 | for (i = 0; i < s->vectors; i++) { | |
523 | msix_vector_use(&s->dev, i); | |
524 | } | |
525 | } | |
526 | ||
527 | static void ivshmem_reset(DeviceState *d) | |
528 | { | |
529 | IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d); | |
530 | ||
531 | s->intrstatus = 0; | |
532 | ivshmem_use_msix(s); | |
533 | return; | |
534 | } | |
535 | ||
536 | static uint64_t ivshmem_get_size(IVShmemState * s) { | |
537 | ||
538 | uint64_t value; | |
539 | char *ptr; | |
540 | ||
541 | value = strtoull(s->sizearg, &ptr, 10); | |
542 | switch (*ptr) { | |
543 | case 0: case 'M': case 'm': | |
544 | value <<= 20; | |
545 | break; | |
546 | case 'G': case 'g': | |
547 | value <<= 30; | |
548 | break; | |
549 | default: | |
550 | fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg); | |
551 | exit(1); | |
552 | } | |
553 | ||
554 | /* BARs must be a power of 2 */ | |
555 | if (!is_power_of_two(value)) { | |
556 | fprintf(stderr, "ivshmem: size must be power of 2\n"); | |
557 | exit(1); | |
558 | } | |
559 | ||
560 | return value; | |
561 | } | |
562 | ||
563 | static void ivshmem_setup_msi(IVShmemState * s) | |
564 | { | |
565 | if (msix_init_exclusive_bar(&s->dev, s->vectors, 1)) { | |
566 | IVSHMEM_DPRINTF("msix initialization failed\n"); | |
567 | exit(1); | |
568 | } | |
569 | ||
570 | IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors); | |
571 | ||
572 | /* allocate QEMU char devices for receiving interrupts */ | |
573 | s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry)); | |
574 | ||
575 | ivshmem_use_msix(s); | |
576 | } | |
577 | ||
578 | static void ivshmem_save(QEMUFile* f, void *opaque) | |
579 | { | |
580 | IVShmemState *proxy = opaque; | |
581 | ||
582 | IVSHMEM_DPRINTF("ivshmem_save\n"); | |
583 | pci_device_save(&proxy->dev, f); | |
584 | ||
585 | if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) { | |
586 | msix_save(&proxy->dev, f); | |
587 | } else { | |
588 | qemu_put_be32(f, proxy->intrstatus); | |
589 | qemu_put_be32(f, proxy->intrmask); | |
590 | } | |
591 | ||
592 | } | |
593 | ||
594 | static int ivshmem_load(QEMUFile* f, void *opaque, int version_id) | |
595 | { | |
596 | IVSHMEM_DPRINTF("ivshmem_load\n"); | |
597 | ||
598 | IVShmemState *proxy = opaque; | |
599 | int ret; | |
600 | ||
601 | if (version_id > 0) { | |
602 | return -EINVAL; | |
603 | } | |
604 | ||
605 | if (proxy->role_val == IVSHMEM_PEER) { | |
606 | fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n"); | |
607 | return -EINVAL; | |
608 | } | |
609 | ||
610 | ret = pci_device_load(&proxy->dev, f); | |
611 | if (ret) { | |
612 | return ret; | |
613 | } | |
614 | ||
615 | if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) { | |
616 | msix_load(&proxy->dev, f); | |
617 | ivshmem_use_msix(proxy); | |
618 | } else { | |
619 | proxy->intrstatus = qemu_get_be32(f); | |
620 | proxy->intrmask = qemu_get_be32(f); | |
621 | } | |
622 | ||
623 | return 0; | |
624 | } | |
625 | ||
626 | static void ivshmem_write_config(PCIDevice *pci_dev, uint32_t address, | |
627 | uint32_t val, int len) | |
628 | { | |
629 | pci_default_write_config(pci_dev, address, val, len); | |
630 | msix_write_config(pci_dev, address, val, len); | |
631 | } | |
632 | ||
633 | static int pci_ivshmem_init(PCIDevice *dev) | |
634 | { | |
635 | IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); | |
636 | uint8_t *pci_conf; | |
637 | ||
638 | if (s->sizearg == NULL) | |
639 | s->ivshmem_size = 4 << 20; /* 4 MB default */ | |
640 | else { | |
641 | s->ivshmem_size = ivshmem_get_size(s); | |
642 | } | |
643 | ||
644 | register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load, | |
645 | dev); | |
646 | ||
647 | /* IRQFD requires MSI */ | |
648 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) && | |
649 | !ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
650 | fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n"); | |
651 | exit(1); | |
652 | } | |
653 | ||
654 | /* check that role is reasonable */ | |
655 | if (s->role) { | |
656 | if (strncmp(s->role, "peer", 5) == 0) { | |
657 | s->role_val = IVSHMEM_PEER; | |
658 | } else if (strncmp(s->role, "master", 7) == 0) { | |
659 | s->role_val = IVSHMEM_MASTER; | |
660 | } else { | |
661 | fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n"); | |
662 | exit(1); | |
663 | } | |
664 | } else { | |
665 | s->role_val = IVSHMEM_MASTER; /* default */ | |
666 | } | |
667 | ||
668 | if (s->role_val == IVSHMEM_PEER) { | |
669 | error_set(&s->migration_blocker, QERR_DEVICE_FEATURE_BLOCKS_MIGRATION, "ivshmem", "peer mode"); | |
670 | migrate_add_blocker(s->migration_blocker); | |
671 | } | |
672 | ||
673 | pci_conf = s->dev.config; | |
674 | pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY; | |
675 | ||
676 | pci_config_set_interrupt_pin(pci_conf, 1); | |
677 | ||
678 | s->shm_fd = 0; | |
679 | ||
680 | memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s, | |
681 | "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE); | |
682 | ||
683 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) { | |
684 | setup_ioeventfds(s); | |
685 | } | |
686 | ||
687 | /* region for registers*/ | |
688 | pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, | |
689 | &s->ivshmem_mmio); | |
690 | ||
691 | memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size); | |
692 | ||
693 | if ((s->server_chr != NULL) && | |
694 | (strncmp(s->server_chr->filename, "unix:", 5) == 0)) { | |
695 | /* if we get a UNIX socket as the parameter we will talk | |
696 | * to the ivshmem server to receive the memory region */ | |
697 | ||
698 | if (s->shmobj != NULL) { | |
699 | fprintf(stderr, "WARNING: do not specify both 'chardev' " | |
700 | "and 'shm' with ivshmem\n"); | |
701 | } | |
702 | ||
703 | IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n", | |
704 | s->server_chr->filename); | |
705 | ||
706 | if (ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
707 | ivshmem_setup_msi(s); | |
708 | } | |
709 | ||
710 | /* we allocate enough space for 16 guests and grow as needed */ | |
711 | s->nb_peers = 16; | |
712 | s->vm_id = -1; | |
713 | ||
714 | /* allocate/initialize space for interrupt handling */ | |
715 | s->peers = g_malloc0(s->nb_peers * sizeof(Peer)); | |
716 | ||
717 | pci_register_bar(&s->dev, 2, | |
718 | PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar); | |
719 | ||
720 | s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *)); | |
721 | ||
722 | qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read, | |
723 | ivshmem_event, s); | |
724 | } else { | |
725 | /* just map the file immediately, we're not using a server */ | |
726 | int fd; | |
727 | ||
728 | if (s->shmobj == NULL) { | |
729 | fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n"); | |
730 | } | |
731 | ||
732 | IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj); | |
733 | ||
734 | /* try opening with O_EXCL and if it succeeds zero the memory | |
735 | * by truncating to 0 */ | |
736 | if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL, | |
737 | S_IRWXU|S_IRWXG|S_IRWXO)) > 0) { | |
738 | /* truncate file to length PCI device's memory */ | |
739 | if (ftruncate(fd, s->ivshmem_size) != 0) { | |
740 | fprintf(stderr, "ivshmem: could not truncate shared file\n"); | |
741 | } | |
742 | ||
743 | } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR, | |
744 | S_IRWXU|S_IRWXG|S_IRWXO)) < 0) { | |
745 | fprintf(stderr, "ivshmem: could not open shared file\n"); | |
746 | exit(-1); | |
747 | ||
748 | } | |
749 | ||
750 | if (check_shm_size(s, fd) == -1) { | |
751 | exit(-1); | |
752 | } | |
753 | ||
754 | create_shared_memory_BAR(s, fd); | |
755 | ||
756 | } | |
757 | ||
758 | s->dev.config_write = ivshmem_write_config; | |
759 | ||
760 | return 0; | |
761 | } | |
762 | ||
763 | static void pci_ivshmem_uninit(PCIDevice *dev) | |
764 | { | |
765 | IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); | |
766 | ||
767 | if (s->migration_blocker) { | |
768 | migrate_del_blocker(s->migration_blocker); | |
769 | error_free(s->migration_blocker); | |
770 | } | |
771 | ||
772 | memory_region_destroy(&s->ivshmem_mmio); | |
773 | memory_region_del_subregion(&s->bar, &s->ivshmem); | |
774 | vmstate_unregister_ram(&s->ivshmem, &s->dev.qdev); | |
775 | memory_region_destroy(&s->ivshmem); | |
776 | memory_region_destroy(&s->bar); | |
777 | unregister_savevm(&dev->qdev, "ivshmem", s); | |
778 | } | |
779 | ||
780 | static Property ivshmem_properties[] = { | |
781 | DEFINE_PROP_CHR("chardev", IVShmemState, server_chr), | |
782 | DEFINE_PROP_STRING("size", IVShmemState, sizearg), | |
783 | DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1), | |
784 | DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false), | |
785 | DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true), | |
786 | DEFINE_PROP_STRING("shm", IVShmemState, shmobj), | |
787 | DEFINE_PROP_STRING("role", IVShmemState, role), | |
788 | DEFINE_PROP_END_OF_LIST(), | |
789 | }; | |
790 | ||
791 | static void ivshmem_class_init(ObjectClass *klass, void *data) | |
792 | { | |
793 | DeviceClass *dc = DEVICE_CLASS(klass); | |
794 | PCIDeviceClass *k = PCI_DEVICE_CLASS(klass); | |
795 | ||
796 | k->init = pci_ivshmem_init; | |
797 | k->exit = pci_ivshmem_uninit; | |
798 | k->vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET; | |
799 | k->device_id = 0x1110; | |
800 | k->class_id = PCI_CLASS_MEMORY_RAM; | |
801 | dc->reset = ivshmem_reset; | |
802 | dc->props = ivshmem_properties; | |
803 | } | |
804 | ||
805 | static TypeInfo ivshmem_info = { | |
806 | .name = "ivshmem", | |
807 | .parent = TYPE_PCI_DEVICE, | |
808 | .instance_size = sizeof(IVShmemState), | |
809 | .class_init = ivshmem_class_init, | |
810 | }; | |
811 | ||
812 | static void ivshmem_register_types(void) | |
813 | { | |
814 | type_register_static(&ivshmem_info); | |
815 | } | |
816 | ||
817 | type_init(ivshmem_register_types) |