]>
Commit | Line | Data |
---|---|---|
6cbf4c8c CM |
1 | /* |
2 | * Inter-VM Shared Memory PCI device. | |
3 | * | |
4 | * Author: | |
5 | * Cam Macdonell <[email protected]> | |
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 | #include "hw.h" | |
17 | #include "pc.h" | |
18 | #include "pci.h" | |
19 | #include "msix.h" | |
20 | #include "kvm.h" | |
21 | ||
22 | #include <sys/mman.h> | |
23 | #include <sys/types.h> | |
24 | ||
25 | #define IVSHMEM_IOEVENTFD 0 | |
26 | #define IVSHMEM_MSI 1 | |
27 | ||
28 | #define IVSHMEM_PEER 0 | |
29 | #define IVSHMEM_MASTER 1 | |
30 | ||
31 | #define IVSHMEM_REG_BAR_SIZE 0x100 | |
32 | ||
33 | //#define DEBUG_IVSHMEM | |
34 | #ifdef DEBUG_IVSHMEM | |
35 | #define IVSHMEM_DPRINTF(fmt, ...) \ | |
36 | do {printf("IVSHMEM: " fmt, ## __VA_ARGS__); } while (0) | |
37 | #else | |
38 | #define IVSHMEM_DPRINTF(fmt, ...) | |
39 | #endif | |
40 | ||
41 | typedef struct Peer { | |
42 | int nb_eventfds; | |
43 | int *eventfds; | |
44 | } Peer; | |
45 | ||
46 | typedef struct EventfdEntry { | |
47 | PCIDevice *pdev; | |
48 | int vector; | |
49 | } EventfdEntry; | |
50 | ||
51 | typedef struct IVShmemState { | |
52 | PCIDevice dev; | |
53 | uint32_t intrmask; | |
54 | uint32_t intrstatus; | |
55 | uint32_t doorbell; | |
56 | ||
57 | CharDriverState **eventfd_chr; | |
58 | CharDriverState *server_chr; | |
cb06608e | 59 | MemoryRegion ivshmem_mmio; |
6cbf4c8c CM |
60 | |
61 | pcibus_t mmio_addr; | |
cb06608e AK |
62 | /* We might need to register the BAR before we actually have the memory. |
63 | * So prepare a container MemoryRegion for the BAR immediately and | |
64 | * add a subregion when we have the memory. | |
65 | */ | |
66 | MemoryRegion bar; | |
67 | MemoryRegion ivshmem; | |
95524ae8 | 68 | MemoryRegion msix_bar; |
6cbf4c8c CM |
69 | uint64_t ivshmem_size; /* size of shared memory region */ |
70 | int shm_fd; /* shared memory file descriptor */ | |
71 | ||
72 | Peer *peers; | |
73 | int nb_peers; /* how many guests we have space for */ | |
74 | int max_peer; /* maximum numbered peer */ | |
75 | ||
76 | int vm_id; | |
77 | uint32_t vectors; | |
78 | uint32_t features; | |
79 | EventfdEntry *eventfd_table; | |
80 | ||
81 | char * shmobj; | |
82 | char * sizearg; | |
83 | char * role; | |
84 | int role_val; /* scalar to avoid multiple string comparisons */ | |
85 | } IVShmemState; | |
86 | ||
87 | /* registers for the Inter-VM shared memory device */ | |
88 | enum ivshmem_registers { | |
89 | INTRMASK = 0, | |
90 | INTRSTATUS = 4, | |
91 | IVPOSITION = 8, | |
92 | DOORBELL = 12, | |
93 | }; | |
94 | ||
95 | static inline uint32_t ivshmem_has_feature(IVShmemState *ivs, | |
96 | unsigned int feature) { | |
97 | return (ivs->features & (1 << feature)); | |
98 | } | |
99 | ||
100 | static inline bool is_power_of_two(uint64_t x) { | |
101 | return (x & (x - 1)) == 0; | |
102 | } | |
103 | ||
6cbf4c8c CM |
104 | /* accessing registers - based on rtl8139 */ |
105 | static void ivshmem_update_irq(IVShmemState *s, int val) | |
106 | { | |
107 | int isr; | |
108 | isr = (s->intrstatus & s->intrmask) & 0xffffffff; | |
109 | ||
110 | /* don't print ISR resets */ | |
111 | if (isr) { | |
112 | IVSHMEM_DPRINTF("Set IRQ to %d (%04x %04x)\n", | |
113 | isr ? 1 : 0, s->intrstatus, s->intrmask); | |
114 | } | |
115 | ||
116 | qemu_set_irq(s->dev.irq[0], (isr != 0)); | |
117 | } | |
118 | ||
119 | static void ivshmem_IntrMask_write(IVShmemState *s, uint32_t val) | |
120 | { | |
121 | IVSHMEM_DPRINTF("IntrMask write(w) val = 0x%04x\n", val); | |
122 | ||
123 | s->intrmask = val; | |
124 | ||
125 | ivshmem_update_irq(s, val); | |
126 | } | |
127 | ||
128 | static uint32_t ivshmem_IntrMask_read(IVShmemState *s) | |
129 | { | |
130 | uint32_t ret = s->intrmask; | |
131 | ||
132 | IVSHMEM_DPRINTF("intrmask read(w) val = 0x%04x\n", ret); | |
133 | ||
134 | return ret; | |
135 | } | |
136 | ||
137 | static void ivshmem_IntrStatus_write(IVShmemState *s, uint32_t val) | |
138 | { | |
139 | IVSHMEM_DPRINTF("IntrStatus write(w) val = 0x%04x\n", val); | |
140 | ||
141 | s->intrstatus = val; | |
142 | ||
143 | ivshmem_update_irq(s, val); | |
144 | return; | |
145 | } | |
146 | ||
147 | static uint32_t ivshmem_IntrStatus_read(IVShmemState *s) | |
148 | { | |
149 | uint32_t ret = s->intrstatus; | |
150 | ||
151 | /* reading ISR clears all interrupts */ | |
152 | s->intrstatus = 0; | |
153 | ||
154 | ivshmem_update_irq(s, 0); | |
155 | ||
156 | return ret; | |
157 | } | |
158 | ||
cb06608e AK |
159 | static void ivshmem_io_write(void *opaque, target_phys_addr_t addr, |
160 | uint64_t val, unsigned size) | |
6cbf4c8c CM |
161 | { |
162 | IVShmemState *s = opaque; | |
163 | ||
164 | uint64_t write_one = 1; | |
165 | uint16_t dest = val >> 16; | |
166 | uint16_t vector = val & 0xff; | |
167 | ||
168 | addr &= 0xfc; | |
169 | ||
170 | IVSHMEM_DPRINTF("writing to addr " TARGET_FMT_plx "\n", addr); | |
171 | switch (addr) | |
172 | { | |
173 | case INTRMASK: | |
174 | ivshmem_IntrMask_write(s, val); | |
175 | break; | |
176 | ||
177 | case INTRSTATUS: | |
178 | ivshmem_IntrStatus_write(s, val); | |
179 | break; | |
180 | ||
181 | case DOORBELL: | |
182 | /* check that dest VM ID is reasonable */ | |
1b27d7a1 | 183 | if (dest > s->max_peer) { |
6cbf4c8c CM |
184 | IVSHMEM_DPRINTF("Invalid destination VM ID (%d)\n", dest); |
185 | break; | |
186 | } | |
187 | ||
188 | /* check doorbell range */ | |
1b27d7a1 | 189 | if (vector < s->peers[dest].nb_eventfds) { |
6cbf4c8c CM |
190 | IVSHMEM_DPRINTF("Writing %" PRId64 " to VM %d on vector %d\n", |
191 | write_one, dest, vector); | |
192 | if (write(s->peers[dest].eventfds[vector], | |
193 | &(write_one), 8) != 8) { | |
194 | IVSHMEM_DPRINTF("error writing to eventfd\n"); | |
195 | } | |
196 | } | |
197 | break; | |
198 | default: | |
199 | IVSHMEM_DPRINTF("Invalid VM Doorbell VM %d\n", dest); | |
200 | } | |
201 | } | |
202 | ||
cb06608e AK |
203 | static uint64_t ivshmem_io_read(void *opaque, target_phys_addr_t addr, |
204 | unsigned size) | |
6cbf4c8c CM |
205 | { |
206 | ||
207 | IVShmemState *s = opaque; | |
208 | uint32_t ret; | |
209 | ||
210 | switch (addr) | |
211 | { | |
212 | case INTRMASK: | |
213 | ret = ivshmem_IntrMask_read(s); | |
214 | break; | |
215 | ||
216 | case INTRSTATUS: | |
217 | ret = ivshmem_IntrStatus_read(s); | |
218 | break; | |
219 | ||
220 | case IVPOSITION: | |
221 | /* return my VM ID if the memory is mapped */ | |
222 | if (s->shm_fd > 0) { | |
223 | ret = s->vm_id; | |
224 | } else { | |
225 | ret = -1; | |
226 | } | |
227 | break; | |
228 | ||
229 | default: | |
230 | IVSHMEM_DPRINTF("why are we reading " TARGET_FMT_plx "\n", addr); | |
231 | ret = 0; | |
232 | } | |
233 | ||
234 | return ret; | |
235 | } | |
236 | ||
cb06608e AK |
237 | static const MemoryRegionOps ivshmem_mmio_ops = { |
238 | .read = ivshmem_io_read, | |
239 | .write = ivshmem_io_write, | |
240 | .endianness = DEVICE_NATIVE_ENDIAN, | |
241 | .impl = { | |
242 | .min_access_size = 4, | |
243 | .max_access_size = 4, | |
244 | }, | |
6cbf4c8c CM |
245 | }; |
246 | ||
247 | static void ivshmem_receive(void *opaque, const uint8_t *buf, int size) | |
248 | { | |
249 | IVShmemState *s = opaque; | |
250 | ||
251 | ivshmem_IntrStatus_write(s, *buf); | |
252 | ||
253 | IVSHMEM_DPRINTF("ivshmem_receive 0x%02x\n", *buf); | |
254 | } | |
255 | ||
256 | static int ivshmem_can_receive(void * opaque) | |
257 | { | |
258 | return 8; | |
259 | } | |
260 | ||
261 | static void ivshmem_event(void *opaque, int event) | |
262 | { | |
263 | IVSHMEM_DPRINTF("ivshmem_event %d\n", event); | |
264 | } | |
265 | ||
266 | static void fake_irqfd(void *opaque, const uint8_t *buf, int size) { | |
267 | ||
268 | EventfdEntry *entry = opaque; | |
269 | PCIDevice *pdev = entry->pdev; | |
270 | ||
271 | IVSHMEM_DPRINTF("interrupt on vector %p %d\n", pdev, entry->vector); | |
272 | msix_notify(pdev, entry->vector); | |
273 | } | |
274 | ||
275 | static CharDriverState* create_eventfd_chr_device(void * opaque, int eventfd, | |
276 | int vector) | |
277 | { | |
278 | /* create a event character device based on the passed eventfd */ | |
279 | IVShmemState *s = opaque; | |
280 | CharDriverState * chr; | |
281 | ||
282 | chr = qemu_chr_open_eventfd(eventfd); | |
283 | ||
284 | if (chr == NULL) { | |
285 | fprintf(stderr, "creating eventfd for eventfd %d failed\n", eventfd); | |
286 | exit(-1); | |
287 | } | |
288 | ||
289 | /* if MSI is supported we need multiple interrupts */ | |
290 | if (ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
291 | s->eventfd_table[vector].pdev = &s->dev; | |
292 | s->eventfd_table[vector].vector = vector; | |
293 | ||
294 | qemu_chr_add_handlers(chr, ivshmem_can_receive, fake_irqfd, | |
295 | ivshmem_event, &s->eventfd_table[vector]); | |
296 | } else { | |
297 | qemu_chr_add_handlers(chr, ivshmem_can_receive, ivshmem_receive, | |
298 | ivshmem_event, s); | |
299 | } | |
300 | ||
301 | return chr; | |
302 | ||
303 | } | |
304 | ||
305 | static int check_shm_size(IVShmemState *s, int fd) { | |
306 | /* check that the guest isn't going to try and map more memory than the | |
307 | * the object has allocated return -1 to indicate error */ | |
308 | ||
309 | struct stat buf; | |
310 | ||
311 | fstat(fd, &buf); | |
312 | ||
313 | if (s->ivshmem_size > buf.st_size) { | |
ad0a4ac1 AK |
314 | fprintf(stderr, |
315 | "IVSHMEM ERROR: Requested memory size greater" | |
316 | " than shared object size (%" PRIu64 " > %" PRIu64")\n", | |
317 | s->ivshmem_size, (uint64_t)buf.st_size); | |
6cbf4c8c CM |
318 | return -1; |
319 | } else { | |
320 | return 0; | |
321 | } | |
322 | } | |
323 | ||
324 | /* create the shared memory BAR when we are not using the server, so we can | |
325 | * create the BAR and map the memory immediately */ | |
326 | static void create_shared_memory_BAR(IVShmemState *s, int fd) { | |
327 | ||
328 | void * ptr; | |
329 | ||
330 | s->shm_fd = fd; | |
331 | ||
332 | ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, fd, 0); | |
333 | ||
cb06608e AK |
334 | memory_region_init_ram_ptr(&s->ivshmem, &s->dev.qdev, "ivshmem.bar2", |
335 | s->ivshmem_size, ptr); | |
336 | memory_region_add_subregion(&s->bar, 0, &s->ivshmem); | |
6cbf4c8c CM |
337 | |
338 | /* region for shared memory */ | |
e824b2cc | 339 | pci_register_bar(&s->dev, 2, PCI_BASE_ADDRESS_SPACE_MEMORY, &s->bar); |
6cbf4c8c CM |
340 | } |
341 | ||
342 | static void close_guest_eventfds(IVShmemState *s, int posn) | |
343 | { | |
344 | int i, guest_curr_max; | |
345 | ||
346 | guest_curr_max = s->peers[posn].nb_eventfds; | |
347 | ||
348 | for (i = 0; i < guest_curr_max; i++) { | |
349 | kvm_set_ioeventfd_mmio_long(s->peers[posn].eventfds[i], | |
350 | s->mmio_addr + DOORBELL, (posn << 16) | i, 0); | |
351 | close(s->peers[posn].eventfds[i]); | |
352 | } | |
353 | ||
7267c094 | 354 | g_free(s->peers[posn].eventfds); |
6cbf4c8c CM |
355 | s->peers[posn].nb_eventfds = 0; |
356 | } | |
357 | ||
358 | static void setup_ioeventfds(IVShmemState *s) { | |
359 | ||
360 | int i, j; | |
361 | ||
362 | for (i = 0; i <= s->max_peer; i++) { | |
363 | for (j = 0; j < s->peers[i].nb_eventfds; j++) { | |
cb06608e AK |
364 | memory_region_add_eventfd(&s->ivshmem_mmio, |
365 | DOORBELL, | |
366 | 4, | |
367 | true, | |
368 | (i << 16) | j, | |
369 | s->peers[i].eventfds[j]); | |
6cbf4c8c CM |
370 | } |
371 | } | |
372 | } | |
373 | ||
374 | /* this function increase the dynamic storage need to store data about other | |
375 | * guests */ | |
376 | static void increase_dynamic_storage(IVShmemState *s, int new_min_size) { | |
377 | ||
378 | int j, old_nb_alloc; | |
379 | ||
380 | old_nb_alloc = s->nb_peers; | |
381 | ||
382 | while (new_min_size >= s->nb_peers) | |
383 | s->nb_peers = s->nb_peers * 2; | |
384 | ||
385 | IVSHMEM_DPRINTF("bumping storage to %d guests\n", s->nb_peers); | |
7267c094 | 386 | s->peers = g_realloc(s->peers, s->nb_peers * sizeof(Peer)); |
6cbf4c8c CM |
387 | |
388 | /* zero out new pointers */ | |
389 | for (j = old_nb_alloc; j < s->nb_peers; j++) { | |
390 | s->peers[j].eventfds = NULL; | |
391 | s->peers[j].nb_eventfds = 0; | |
392 | } | |
393 | } | |
394 | ||
395 | static void ivshmem_read(void *opaque, const uint8_t * buf, int flags) | |
396 | { | |
397 | IVShmemState *s = opaque; | |
398 | int incoming_fd, tmp_fd; | |
399 | int guest_max_eventfd; | |
400 | long incoming_posn; | |
401 | ||
402 | memcpy(&incoming_posn, buf, sizeof(long)); | |
403 | /* pick off s->server_chr->msgfd and store it, posn should accompany msg */ | |
74c0d6f0 | 404 | tmp_fd = qemu_chr_fe_get_msgfd(s->server_chr); |
6cbf4c8c CM |
405 | IVSHMEM_DPRINTF("posn is %ld, fd is %d\n", incoming_posn, tmp_fd); |
406 | ||
407 | /* make sure we have enough space for this guest */ | |
408 | if (incoming_posn >= s->nb_peers) { | |
409 | increase_dynamic_storage(s, incoming_posn); | |
410 | } | |
411 | ||
412 | if (tmp_fd == -1) { | |
413 | /* if posn is positive and unseen before then this is our posn*/ | |
414 | if ((incoming_posn >= 0) && | |
415 | (s->peers[incoming_posn].eventfds == NULL)) { | |
416 | /* receive our posn */ | |
417 | s->vm_id = incoming_posn; | |
418 | return; | |
419 | } else { | |
420 | /* otherwise an fd == -1 means an existing guest has gone away */ | |
421 | IVSHMEM_DPRINTF("posn %ld has gone away\n", incoming_posn); | |
422 | close_guest_eventfds(s, incoming_posn); | |
423 | return; | |
424 | } | |
425 | } | |
426 | ||
427 | /* because of the implementation of get_msgfd, we need a dup */ | |
428 | incoming_fd = dup(tmp_fd); | |
429 | ||
430 | if (incoming_fd == -1) { | |
431 | fprintf(stderr, "could not allocate file descriptor %s\n", | |
432 | strerror(errno)); | |
433 | return; | |
434 | } | |
435 | ||
436 | /* if the position is -1, then it's shared memory region fd */ | |
437 | if (incoming_posn == -1) { | |
438 | ||
439 | void * map_ptr; | |
440 | ||
441 | s->max_peer = 0; | |
442 | ||
443 | if (check_shm_size(s, incoming_fd) == -1) { | |
444 | exit(-1); | |
445 | } | |
446 | ||
447 | /* mmap the region and map into the BAR2 */ | |
448 | map_ptr = mmap(0, s->ivshmem_size, PROT_READ|PROT_WRITE, MAP_SHARED, | |
449 | incoming_fd, 0); | |
cb06608e AK |
450 | memory_region_init_ram_ptr(&s->ivshmem, &s->dev.qdev, |
451 | "ivshmem.bar2", s->ivshmem_size, map_ptr); | |
6cbf4c8c | 452 | |
cb06608e | 453 | IVSHMEM_DPRINTF("guest h/w addr = %" PRIu64 ", size = %" PRIu64 "\n", |
6cbf4c8c CM |
454 | s->ivshmem_offset, s->ivshmem_size); |
455 | ||
cb06608e | 456 | memory_region_add_subregion(&s->bar, 0, &s->ivshmem); |
6cbf4c8c CM |
457 | |
458 | /* only store the fd if it is successfully mapped */ | |
459 | s->shm_fd = incoming_fd; | |
460 | ||
461 | return; | |
462 | } | |
463 | ||
464 | /* each guest has an array of eventfds, and we keep track of how many | |
465 | * guests for each VM */ | |
466 | guest_max_eventfd = s->peers[incoming_posn].nb_eventfds; | |
467 | ||
468 | if (guest_max_eventfd == 0) { | |
469 | /* one eventfd per MSI vector */ | |
7267c094 | 470 | s->peers[incoming_posn].eventfds = (int *) g_malloc(s->vectors * |
6cbf4c8c CM |
471 | sizeof(int)); |
472 | } | |
473 | ||
474 | /* this is an eventfd for a particular guest VM */ | |
475 | IVSHMEM_DPRINTF("eventfds[%ld][%d] = %d\n", incoming_posn, | |
476 | guest_max_eventfd, incoming_fd); | |
477 | s->peers[incoming_posn].eventfds[guest_max_eventfd] = incoming_fd; | |
478 | ||
479 | /* increment count for particular guest */ | |
480 | s->peers[incoming_posn].nb_eventfds++; | |
481 | ||
482 | /* keep track of the maximum VM ID */ | |
483 | if (incoming_posn > s->max_peer) { | |
484 | s->max_peer = incoming_posn; | |
485 | } | |
486 | ||
487 | if (incoming_posn == s->vm_id) { | |
488 | s->eventfd_chr[guest_max_eventfd] = create_eventfd_chr_device(s, | |
489 | s->peers[s->vm_id].eventfds[guest_max_eventfd], | |
490 | guest_max_eventfd); | |
491 | } | |
492 | ||
493 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) { | |
494 | if (kvm_set_ioeventfd_mmio_long(incoming_fd, s->mmio_addr + DOORBELL, | |
495 | (incoming_posn << 16) | guest_max_eventfd, 1) < 0) { | |
496 | fprintf(stderr, "ivshmem: ioeventfd not available\n"); | |
497 | } | |
498 | } | |
499 | ||
500 | return; | |
501 | } | |
502 | ||
503 | static void ivshmem_reset(DeviceState *d) | |
504 | { | |
505 | IVShmemState *s = DO_UPCAST(IVShmemState, dev.qdev, d); | |
506 | ||
507 | s->intrstatus = 0; | |
508 | return; | |
509 | } | |
510 | ||
6cbf4c8c CM |
511 | static uint64_t ivshmem_get_size(IVShmemState * s) { |
512 | ||
513 | uint64_t value; | |
514 | char *ptr; | |
515 | ||
516 | value = strtoull(s->sizearg, &ptr, 10); | |
517 | switch (*ptr) { | |
518 | case 0: case 'M': case 'm': | |
519 | value <<= 20; | |
520 | break; | |
521 | case 'G': case 'g': | |
522 | value <<= 30; | |
523 | break; | |
524 | default: | |
525 | fprintf(stderr, "qemu: invalid ram size: %s\n", s->sizearg); | |
526 | exit(1); | |
527 | } | |
528 | ||
529 | /* BARs must be a power of 2 */ | |
530 | if (!is_power_of_two(value)) { | |
531 | fprintf(stderr, "ivshmem: size must be power of 2\n"); | |
532 | exit(1); | |
533 | } | |
534 | ||
535 | return value; | |
536 | } | |
537 | ||
538 | static void ivshmem_setup_msi(IVShmemState * s) { | |
539 | ||
540 | int i; | |
541 | ||
542 | /* allocate the MSI-X vectors */ | |
543 | ||
95524ae8 AK |
544 | memory_region_init(&s->msix_bar, "ivshmem-msix", 4096); |
545 | if (!msix_init(&s->dev, s->vectors, &s->msix_bar, 1, 0)) { | |
e824b2cc AK |
546 | pci_register_bar(&s->dev, 1, PCI_BASE_ADDRESS_SPACE_MEMORY, |
547 | &s->msix_bar); | |
6cbf4c8c CM |
548 | IVSHMEM_DPRINTF("msix initialized (%d vectors)\n", s->vectors); |
549 | } else { | |
550 | IVSHMEM_DPRINTF("msix initialization failed\n"); | |
551 | exit(1); | |
552 | } | |
553 | ||
554 | /* 'activate' the vectors */ | |
555 | for (i = 0; i < s->vectors; i++) { | |
556 | msix_vector_use(&s->dev, i); | |
557 | } | |
558 | ||
559 | /* allocate Qemu char devices for receiving interrupts */ | |
7267c094 | 560 | s->eventfd_table = g_malloc0(s->vectors * sizeof(EventfdEntry)); |
6cbf4c8c CM |
561 | } |
562 | ||
563 | static void ivshmem_save(QEMUFile* f, void *opaque) | |
564 | { | |
565 | IVShmemState *proxy = opaque; | |
566 | ||
567 | IVSHMEM_DPRINTF("ivshmem_save\n"); | |
568 | pci_device_save(&proxy->dev, f); | |
569 | ||
570 | if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) { | |
571 | msix_save(&proxy->dev, f); | |
572 | } else { | |
573 | qemu_put_be32(f, proxy->intrstatus); | |
574 | qemu_put_be32(f, proxy->intrmask); | |
575 | } | |
576 | ||
577 | } | |
578 | ||
579 | static int ivshmem_load(QEMUFile* f, void *opaque, int version_id) | |
580 | { | |
581 | IVSHMEM_DPRINTF("ivshmem_load\n"); | |
582 | ||
583 | IVShmemState *proxy = opaque; | |
584 | int ret, i; | |
585 | ||
586 | if (version_id > 0) { | |
587 | return -EINVAL; | |
588 | } | |
589 | ||
590 | if (proxy->role_val == IVSHMEM_PEER) { | |
591 | fprintf(stderr, "ivshmem: 'peer' devices are not migratable\n"); | |
592 | return -EINVAL; | |
593 | } | |
594 | ||
595 | ret = pci_device_load(&proxy->dev, f); | |
596 | if (ret) { | |
597 | return ret; | |
598 | } | |
599 | ||
600 | if (ivshmem_has_feature(proxy, IVSHMEM_MSI)) { | |
601 | msix_load(&proxy->dev, f); | |
602 | for (i = 0; i < proxy->vectors; i++) { | |
603 | msix_vector_use(&proxy->dev, i); | |
604 | } | |
605 | } else { | |
606 | proxy->intrstatus = qemu_get_be32(f); | |
607 | proxy->intrmask = qemu_get_be32(f); | |
608 | } | |
609 | ||
610 | return 0; | |
611 | } | |
612 | ||
613 | static int pci_ivshmem_init(PCIDevice *dev) | |
614 | { | |
615 | IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); | |
616 | uint8_t *pci_conf; | |
617 | ||
618 | if (s->sizearg == NULL) | |
619 | s->ivshmem_size = 4 << 20; /* 4 MB default */ | |
620 | else { | |
621 | s->ivshmem_size = ivshmem_get_size(s); | |
622 | } | |
623 | ||
624 | register_savevm(&s->dev.qdev, "ivshmem", 0, 0, ivshmem_save, ivshmem_load, | |
625 | dev); | |
626 | ||
627 | /* IRQFD requires MSI */ | |
628 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD) && | |
629 | !ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
630 | fprintf(stderr, "ivshmem: ioeventfd/irqfd requires MSI\n"); | |
631 | exit(1); | |
632 | } | |
633 | ||
634 | /* check that role is reasonable */ | |
635 | if (s->role) { | |
636 | if (strncmp(s->role, "peer", 5) == 0) { | |
637 | s->role_val = IVSHMEM_PEER; | |
638 | } else if (strncmp(s->role, "master", 7) == 0) { | |
639 | s->role_val = IVSHMEM_MASTER; | |
640 | } else { | |
641 | fprintf(stderr, "ivshmem: 'role' must be 'peer' or 'master'\n"); | |
642 | exit(1); | |
643 | } | |
644 | } else { | |
645 | s->role_val = IVSHMEM_MASTER; /* default */ | |
646 | } | |
647 | ||
648 | if (s->role_val == IVSHMEM_PEER) { | |
649 | register_device_unmigratable(&s->dev.qdev, "ivshmem", s); | |
650 | } | |
651 | ||
652 | pci_conf = s->dev.config; | |
6cbf4c8c | 653 | pci_conf[PCI_COMMAND] = PCI_COMMAND_IO | PCI_COMMAND_MEMORY; |
6cbf4c8c CM |
654 | |
655 | pci_config_set_interrupt_pin(pci_conf, 1); | |
656 | ||
6cbf4c8c CM |
657 | s->shm_fd = 0; |
658 | ||
cb06608e AK |
659 | memory_region_init_io(&s->ivshmem_mmio, &ivshmem_mmio_ops, s, |
660 | "ivshmem-mmio", IVSHMEM_REG_BAR_SIZE); | |
661 | ||
662 | if (ivshmem_has_feature(s, IVSHMEM_IOEVENTFD)) { | |
663 | setup_ioeventfds(s); | |
664 | } | |
665 | ||
6cbf4c8c | 666 | /* region for registers*/ |
e824b2cc AK |
667 | pci_register_bar(&s->dev, 0, PCI_BASE_ADDRESS_SPACE_MEMORY, |
668 | &s->ivshmem_mmio); | |
cb06608e AK |
669 | |
670 | memory_region_init(&s->bar, "ivshmem-bar2-container", s->ivshmem_size); | |
6cbf4c8c CM |
671 | |
672 | if ((s->server_chr != NULL) && | |
673 | (strncmp(s->server_chr->filename, "unix:", 5) == 0)) { | |
674 | /* if we get a UNIX socket as the parameter we will talk | |
675 | * to the ivshmem server to receive the memory region */ | |
676 | ||
677 | if (s->shmobj != NULL) { | |
678 | fprintf(stderr, "WARNING: do not specify both 'chardev' " | |
679 | "and 'shm' with ivshmem\n"); | |
680 | } | |
681 | ||
682 | IVSHMEM_DPRINTF("using shared memory server (socket = %s)\n", | |
683 | s->server_chr->filename); | |
684 | ||
685 | if (ivshmem_has_feature(s, IVSHMEM_MSI)) { | |
686 | ivshmem_setup_msi(s); | |
687 | } | |
688 | ||
689 | /* we allocate enough space for 16 guests and grow as needed */ | |
690 | s->nb_peers = 16; | |
691 | s->vm_id = -1; | |
692 | ||
693 | /* allocate/initialize space for interrupt handling */ | |
7267c094 | 694 | s->peers = g_malloc0(s->nb_peers * sizeof(Peer)); |
6cbf4c8c | 695 | |
e824b2cc AK |
696 | pci_register_bar(&s->dev, 2, |
697 | PCI_BASE_ADDRESS_SPACE_MEMORY, &s->ivshmem); | |
6cbf4c8c | 698 | |
7267c094 | 699 | s->eventfd_chr = g_malloc0(s->vectors * sizeof(CharDriverState *)); |
6cbf4c8c CM |
700 | |
701 | qemu_chr_add_handlers(s->server_chr, ivshmem_can_receive, ivshmem_read, | |
702 | ivshmem_event, s); | |
703 | } else { | |
704 | /* just map the file immediately, we're not using a server */ | |
705 | int fd; | |
706 | ||
707 | if (s->shmobj == NULL) { | |
708 | fprintf(stderr, "Must specify 'chardev' or 'shm' to ivshmem\n"); | |
709 | } | |
710 | ||
711 | IVSHMEM_DPRINTF("using shm_open (shm object = %s)\n", s->shmobj); | |
712 | ||
713 | /* try opening with O_EXCL and if it succeeds zero the memory | |
714 | * by truncating to 0 */ | |
715 | if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR|O_EXCL, | |
716 | S_IRWXU|S_IRWXG|S_IRWXO)) > 0) { | |
717 | /* truncate file to length PCI device's memory */ | |
718 | if (ftruncate(fd, s->ivshmem_size) != 0) { | |
719 | fprintf(stderr, "ivshmem: could not truncate shared file\n"); | |
720 | } | |
721 | ||
722 | } else if ((fd = shm_open(s->shmobj, O_CREAT|O_RDWR, | |
723 | S_IRWXU|S_IRWXG|S_IRWXO)) < 0) { | |
724 | fprintf(stderr, "ivshmem: could not open shared file\n"); | |
725 | exit(-1); | |
726 | ||
727 | } | |
728 | ||
729 | if (check_shm_size(s, fd) == -1) { | |
730 | exit(-1); | |
731 | } | |
732 | ||
733 | create_shared_memory_BAR(s, fd); | |
734 | ||
735 | } | |
736 | ||
737 | return 0; | |
738 | } | |
739 | ||
740 | static int pci_ivshmem_uninit(PCIDevice *dev) | |
741 | { | |
742 | IVShmemState *s = DO_UPCAST(IVShmemState, dev, dev); | |
743 | ||
cb06608e AK |
744 | memory_region_destroy(&s->ivshmem_mmio); |
745 | memory_region_del_subregion(&s->bar, &s->ivshmem); | |
746 | memory_region_destroy(&s->ivshmem); | |
747 | memory_region_destroy(&s->bar); | |
6cbf4c8c CM |
748 | unregister_savevm(&dev->qdev, "ivshmem", s); |
749 | ||
750 | return 0; | |
751 | } | |
752 | ||
753 | static PCIDeviceInfo ivshmem_info = { | |
754 | .qdev.name = "ivshmem", | |
755 | .qdev.size = sizeof(IVShmemState), | |
756 | .qdev.reset = ivshmem_reset, | |
757 | .init = pci_ivshmem_init, | |
758 | .exit = pci_ivshmem_uninit, | |
7aff0f21 IY |
759 | .vendor_id = PCI_VENDOR_ID_REDHAT_QUMRANET, |
760 | .device_id = 0x1110, | |
761 | .class_id = PCI_CLASS_MEMORY_RAM, | |
6cbf4c8c CM |
762 | .qdev.props = (Property[]) { |
763 | DEFINE_PROP_CHR("chardev", IVShmemState, server_chr), | |
764 | DEFINE_PROP_STRING("size", IVShmemState, sizearg), | |
765 | DEFINE_PROP_UINT32("vectors", IVShmemState, vectors, 1), | |
766 | DEFINE_PROP_BIT("ioeventfd", IVShmemState, features, IVSHMEM_IOEVENTFD, false), | |
767 | DEFINE_PROP_BIT("msi", IVShmemState, features, IVSHMEM_MSI, true), | |
768 | DEFINE_PROP_STRING("shm", IVShmemState, shmobj), | |
769 | DEFINE_PROP_STRING("role", IVShmemState, role), | |
770 | DEFINE_PROP_END_OF_LIST(), | |
771 | } | |
772 | }; | |
773 | ||
774 | static void ivshmem_register_devices(void) | |
775 | { | |
776 | pci_qdev_register(&ivshmem_info); | |
777 | } | |
778 | ||
779 | device_init(ivshmem_register_devices) |