1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Virtio vhost-user driver
4 *
5 * Copyright(c) 2019 Intel Corporation
6 *
7 * This driver allows virtio devices to be used over a vhost-user socket.
8 *
9 * Guest devices can be instantiated by kernel module or command line
10 * parameters. One device will be created for each parameter. Syntax:
11 *
12 * virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]
13 * where:
14 * <socket> := vhost-user socket path to connect
15 * <virtio_id> := virtio device id (as in virtio_ids.h)
16 * <platform_id> := (optional) platform device id
17 *
18 * example:
19 * virtio_uml.device=/var/uml.socket:1
20 *
21 * Based on Virtio MMIO driver by Pawel Moll, copyright 2011-2014, ARM Ltd.
22 */
23 #include <linux/module.h>
24 #include <linux/of.h>
25 #include <linux/platform_device.h>
26 #include <linux/slab.h>
27 #include <linux/virtio.h>
28 #include <linux/virtio_config.h>
29 #include <linux/virtio_ring.h>
30 #include <linux/time-internal.h>
31 #include <linux/virtio-uml.h>
32 #include <shared/as-layout.h>
33 #include <irq_kern.h>
34 #include <init.h>
35 #include <os.h>
36 #include "vhost_user.h"
37
38 #define MAX_SUPPORTED_QUEUE_SIZE 256
39
40 #define to_virtio_uml_device(_vdev) \
41 container_of(_vdev, struct virtio_uml_device, vdev)
42
43 struct virtio_uml_platform_data {
44 u32 virtio_device_id;
45 const char *socket_path;
46 struct work_struct conn_broken_wk;
47 struct platform_device *pdev;
48 };
49
50 struct virtio_uml_device {
51 struct virtio_device vdev;
52 struct platform_device *pdev;
53 struct virtio_uml_platform_data *pdata;
54
55 spinlock_t sock_lock;
56 int sock, req_fd, irq;
57 u64 features;
58 u64 protocol_features;
59 u64 max_vqs;
60 u8 status;
61 u8 registered:1;
62 u8 suspended:1;
63 u8 no_vq_suspend:1;
64
65 u8 config_changed_irq:1;
66 uint64_t vq_irq_vq_map;
67 int recv_rc;
68 };
69
70 struct virtio_uml_vq_info {
71 int kick_fd, call_fd;
72 char name[32];
73 bool suspended;
74 };
75
76 #define vu_err(vu_dev, ...) dev_err(&(vu_dev)->pdev->dev, ##__VA_ARGS__)
77
78 /* Vhost-user protocol */
79
full_sendmsg_fds(int fd,const void * buf,unsigned int len,const int * fds,unsigned int fds_num)80 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
81 const int *fds, unsigned int fds_num)
82 {
83 int rc;
84
85 do {
86 rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
87 if (rc > 0) {
88 buf += rc;
89 len -= rc;
90 fds = NULL;
91 fds_num = 0;
92 }
93 } while (len && (rc >= 0 || rc == -EINTR));
94
95 if (rc < 0)
96 return rc;
97 return 0;
98 }
99
full_read(int fd,void * buf,int len,bool abortable)100 static int full_read(int fd, void *buf, int len, bool abortable)
101 {
102 int rc;
103
104 if (!len)
105 return 0;
106
107 do {
108 rc = os_read_file(fd, buf, len);
109 if (rc > 0) {
110 buf += rc;
111 len -= rc;
112 }
113 } while (len && (rc > 0 || rc == -EINTR || (!abortable && rc == -EAGAIN)));
114
115 if (rc < 0)
116 return rc;
117 if (rc == 0)
118 return -ECONNRESET;
119 return 0;
120 }
121
vhost_user_recv_header(int fd,struct vhost_user_msg * msg)122 static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
123 {
124 return full_read(fd, msg, sizeof(msg->header), true);
125 }
126
vhost_user_recv(struct virtio_uml_device * vu_dev,int fd,struct vhost_user_msg * msg,size_t max_payload_size,bool wait)127 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
128 int fd, struct vhost_user_msg *msg,
129 size_t max_payload_size, bool wait)
130 {
131 size_t size;
132 int rc;
133
134 /*
135 * In virtio time-travel mode, we're handling all the vhost-user
136 * FDs by polling them whenever appropriate. However, we may get
137 * into a situation where we're sending out an interrupt message
138 * to a device (e.g. a net device) and need to handle a simulation
139 * time message while doing so, e.g. one that tells us to update
140 * our idea of how long we can run without scheduling.
141 *
142 * Thus, we need to not just read() from the given fd, but need
143 * to also handle messages for the simulation time - this function
144 * does that for us while waiting for the given fd to be readable.
145 */
146 if (wait)
147 time_travel_wait_readable(fd);
148
149 rc = vhost_user_recv_header(fd, msg);
150
151 if (rc)
152 return rc;
153 size = msg->header.size;
154 if (size > max_payload_size)
155 return -EPROTO;
156 return full_read(fd, &msg->payload, size, false);
157 }
158
vhost_user_check_reset(struct virtio_uml_device * vu_dev,int rc)159 static void vhost_user_check_reset(struct virtio_uml_device *vu_dev,
160 int rc)
161 {
162 struct virtio_uml_platform_data *pdata = vu_dev->pdata;
163
164 if (rc != -ECONNRESET)
165 return;
166
167 if (!vu_dev->registered)
168 return;
169
170 vu_dev->registered = 0;
171
172 schedule_work(&pdata->conn_broken_wk);
173 }
174
vhost_user_recv_resp(struct virtio_uml_device * vu_dev,struct vhost_user_msg * msg,size_t max_payload_size)175 static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
176 struct vhost_user_msg *msg,
177 size_t max_payload_size)
178 {
179 int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg,
180 max_payload_size, true);
181
182 if (rc) {
183 vhost_user_check_reset(vu_dev, rc);
184 return rc;
185 }
186
187 if (msg->header.flags != (VHOST_USER_FLAG_REPLY | VHOST_USER_VERSION))
188 return -EPROTO;
189
190 return 0;
191 }
192
vhost_user_recv_u64(struct virtio_uml_device * vu_dev,u64 * value)193 static int vhost_user_recv_u64(struct virtio_uml_device *vu_dev,
194 u64 *value)
195 {
196 struct vhost_user_msg msg;
197 int rc = vhost_user_recv_resp(vu_dev, &msg,
198 sizeof(msg.payload.integer));
199
200 if (rc)
201 return rc;
202 if (msg.header.size != sizeof(msg.payload.integer))
203 return -EPROTO;
204 *value = msg.payload.integer;
205 return 0;
206 }
207
vhost_user_recv_req(struct virtio_uml_device * vu_dev,struct vhost_user_msg * msg,size_t max_payload_size)208 static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
209 struct vhost_user_msg *msg,
210 size_t max_payload_size)
211 {
212 int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg,
213 max_payload_size, false);
214
215 if (rc)
216 return rc;
217
218 if ((msg->header.flags & ~VHOST_USER_FLAG_NEED_REPLY) !=
219 VHOST_USER_VERSION)
220 return -EPROTO;
221
222 return 0;
223 }
224
vhost_user_send(struct virtio_uml_device * vu_dev,bool need_response,struct vhost_user_msg * msg,int * fds,size_t num_fds)225 static int vhost_user_send(struct virtio_uml_device *vu_dev,
226 bool need_response, struct vhost_user_msg *msg,
227 int *fds, size_t num_fds)
228 {
229 size_t size = sizeof(msg->header) + msg->header.size;
230 unsigned long flags;
231 bool request_ack;
232 int rc;
233
234 msg->header.flags |= VHOST_USER_VERSION;
235
236 /*
237 * The need_response flag indicates that we already need a response,
238 * e.g. to read the features. In these cases, don't request an ACK as
239 * it is meaningless. Also request an ACK only if supported.
240 */
241 request_ack = !need_response;
242 if (!(vu_dev->protocol_features &
243 BIT_ULL(VHOST_USER_PROTOCOL_F_REPLY_ACK)))
244 request_ack = false;
245
246 if (request_ack)
247 msg->header.flags |= VHOST_USER_FLAG_NEED_REPLY;
248
249 spin_lock_irqsave(&vu_dev->sock_lock, flags);
250 rc = full_sendmsg_fds(vu_dev->sock, msg, size, fds, num_fds);
251 if (rc < 0)
252 goto out;
253
254 if (request_ack) {
255 uint64_t status;
256
257 rc = vhost_user_recv_u64(vu_dev, &status);
258 if (rc)
259 goto out;
260
261 if (status) {
262 vu_err(vu_dev, "slave reports error: %llu\n", status);
263 rc = -EIO;
264 goto out;
265 }
266 }
267
268 out:
269 spin_unlock_irqrestore(&vu_dev->sock_lock, flags);
270 return rc;
271 }
272
vhost_user_send_no_payload(struct virtio_uml_device * vu_dev,bool need_response,u32 request)273 static int vhost_user_send_no_payload(struct virtio_uml_device *vu_dev,
274 bool need_response, u32 request)
275 {
276 struct vhost_user_msg msg = {
277 .header.request = request,
278 };
279
280 return vhost_user_send(vu_dev, need_response, &msg, NULL, 0);
281 }
282
vhost_user_send_no_payload_fd(struct virtio_uml_device * vu_dev,u32 request,int fd)283 static int vhost_user_send_no_payload_fd(struct virtio_uml_device *vu_dev,
284 u32 request, int fd)
285 {
286 struct vhost_user_msg msg = {
287 .header.request = request,
288 };
289
290 return vhost_user_send(vu_dev, false, &msg, &fd, 1);
291 }
292
vhost_user_send_u64(struct virtio_uml_device * vu_dev,u32 request,u64 value)293 static int vhost_user_send_u64(struct virtio_uml_device *vu_dev,
294 u32 request, u64 value)
295 {
296 struct vhost_user_msg msg = {
297 .header.request = request,
298 .header.size = sizeof(msg.payload.integer),
299 .payload.integer = value,
300 };
301
302 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
303 }
304
vhost_user_set_owner(struct virtio_uml_device * vu_dev)305 static int vhost_user_set_owner(struct virtio_uml_device *vu_dev)
306 {
307 return vhost_user_send_no_payload(vu_dev, false, VHOST_USER_SET_OWNER);
308 }
309
vhost_user_get_features(struct virtio_uml_device * vu_dev,u64 * features)310 static int vhost_user_get_features(struct virtio_uml_device *vu_dev,
311 u64 *features)
312 {
313 int rc = vhost_user_send_no_payload(vu_dev, true,
314 VHOST_USER_GET_FEATURES);
315
316 if (rc)
317 return rc;
318 return vhost_user_recv_u64(vu_dev, features);
319 }
320
vhost_user_set_features(struct virtio_uml_device * vu_dev,u64 features)321 static int vhost_user_set_features(struct virtio_uml_device *vu_dev,
322 u64 features)
323 {
324 return vhost_user_send_u64(vu_dev, VHOST_USER_SET_FEATURES, features);
325 }
326
vhost_user_get_protocol_features(struct virtio_uml_device * vu_dev,u64 * protocol_features)327 static int vhost_user_get_protocol_features(struct virtio_uml_device *vu_dev,
328 u64 *protocol_features)
329 {
330 int rc = vhost_user_send_no_payload(vu_dev, true,
331 VHOST_USER_GET_PROTOCOL_FEATURES);
332
333 if (rc)
334 return rc;
335 return vhost_user_recv_u64(vu_dev, protocol_features);
336 }
337
vhost_user_set_protocol_features(struct virtio_uml_device * vu_dev,u64 protocol_features)338 static int vhost_user_set_protocol_features(struct virtio_uml_device *vu_dev,
339 u64 protocol_features)
340 {
341 return vhost_user_send_u64(vu_dev, VHOST_USER_SET_PROTOCOL_FEATURES,
342 protocol_features);
343 }
344
vhost_user_get_queue_num(struct virtio_uml_device * vu_dev,u64 * queue_num)345 static int vhost_user_get_queue_num(struct virtio_uml_device *vu_dev,
346 u64 *queue_num)
347 {
348 int rc = vhost_user_send_no_payload(vu_dev, true,
349 VHOST_USER_GET_QUEUE_NUM);
350
351 if (rc)
352 return rc;
353 return vhost_user_recv_u64(vu_dev, queue_num);
354 }
355
vhost_user_reply(struct virtio_uml_device * vu_dev,struct vhost_user_msg * msg,int response)356 static void vhost_user_reply(struct virtio_uml_device *vu_dev,
357 struct vhost_user_msg *msg, int response)
358 {
359 struct vhost_user_msg reply = {
360 .payload.integer = response,
361 };
362 size_t size = sizeof(reply.header) + sizeof(reply.payload.integer);
363 int rc;
364
365 reply.header = msg->header;
366 reply.header.flags &= ~VHOST_USER_FLAG_NEED_REPLY;
367 reply.header.flags |= VHOST_USER_FLAG_REPLY;
368 reply.header.size = sizeof(reply.payload.integer);
369
370 rc = full_sendmsg_fds(vu_dev->req_fd, &reply, size, NULL, 0);
371
372 if (rc)
373 vu_err(vu_dev,
374 "sending reply to slave request failed: %d (size %zu)\n",
375 rc, size);
376 }
377
vu_req_read_message(struct virtio_uml_device * vu_dev,struct time_travel_event * ev)378 static irqreturn_t vu_req_read_message(struct virtio_uml_device *vu_dev,
379 struct time_travel_event *ev)
380 {
381 struct virtqueue *vq;
382 int response = 1;
383 struct {
384 struct vhost_user_msg msg;
385 u8 extra_payload[512];
386 } msg;
387 int rc;
388 irqreturn_t irq_rc = IRQ_NONE;
389
390 while (1) {
391 rc = vhost_user_recv_req(vu_dev, &msg.msg,
392 sizeof(msg.msg.payload) +
393 sizeof(msg.extra_payload));
394 if (rc)
395 break;
396
397 switch (msg.msg.header.request) {
398 case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG:
399 vu_dev->config_changed_irq = true;
400 response = 0;
401 break;
402 case VHOST_USER_SLAVE_VRING_CALL:
403 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
404 if (vq->index == msg.msg.payload.vring_state.index) {
405 response = 0;
406 vu_dev->vq_irq_vq_map |= BIT_ULL(vq->index);
407 break;
408 }
409 }
410 break;
411 case VHOST_USER_SLAVE_IOTLB_MSG:
412 /* not supported - VIRTIO_F_ACCESS_PLATFORM */
413 case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
414 /* not supported - VHOST_USER_PROTOCOL_F_HOST_NOTIFIER */
415 default:
416 vu_err(vu_dev, "unexpected slave request %d\n",
417 msg.msg.header.request);
418 }
419
420 if (ev && !vu_dev->suspended)
421 time_travel_add_irq_event(ev);
422
423 if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
424 vhost_user_reply(vu_dev, &msg.msg, response);
425 irq_rc = IRQ_HANDLED;
426 }
427 /* mask EAGAIN as we try non-blocking read until socket is empty */
428 vu_dev->recv_rc = (rc == -EAGAIN) ? 0 : rc;
429 return irq_rc;
430 }
431
vu_req_interrupt(int irq,void * data)432 static irqreturn_t vu_req_interrupt(int irq, void *data)
433 {
434 struct virtio_uml_device *vu_dev = data;
435 irqreturn_t ret = IRQ_HANDLED;
436
437 if (!um_irq_timetravel_handler_used())
438 ret = vu_req_read_message(vu_dev, NULL);
439
440 if (vu_dev->recv_rc) {
441 vhost_user_check_reset(vu_dev, vu_dev->recv_rc);
442 } else if (vu_dev->vq_irq_vq_map) {
443 struct virtqueue *vq;
444
445 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
446 if (vu_dev->vq_irq_vq_map & BIT_ULL(vq->index))
447 vring_interrupt(0 /* ignored */, vq);
448 }
449 vu_dev->vq_irq_vq_map = 0;
450 } else if (vu_dev->config_changed_irq) {
451 virtio_config_changed(&vu_dev->vdev);
452 vu_dev->config_changed_irq = false;
453 }
454
455 return ret;
456 }
457
vu_req_interrupt_comm_handler(int irq,int fd,void * data,struct time_travel_event * ev)458 static void vu_req_interrupt_comm_handler(int irq, int fd, void *data,
459 struct time_travel_event *ev)
460 {
461 vu_req_read_message(data, ev);
462 }
463
vhost_user_init_slave_req(struct virtio_uml_device * vu_dev)464 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
465 {
466 int rc, req_fds[2];
467
468 /* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
469 rc = os_pipe(req_fds, true, true);
470 if (rc < 0)
471 return rc;
472 vu_dev->req_fd = req_fds[0];
473
474 rc = um_request_irq_tt(UM_IRQ_ALLOC, vu_dev->req_fd, IRQ_READ,
475 vu_req_interrupt, IRQF_SHARED,
476 vu_dev->pdev->name, vu_dev,
477 vu_req_interrupt_comm_handler);
478 if (rc < 0)
479 goto err_close;
480
481 vu_dev->irq = rc;
482
483 rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
484 req_fds[1]);
485 if (rc)
486 goto err_free_irq;
487
488 goto out;
489
490 err_free_irq:
491 um_free_irq(vu_dev->irq, vu_dev);
492 err_close:
493 os_close_file(req_fds[0]);
494 out:
495 /* Close unused write end of request fds */
496 os_close_file(req_fds[1]);
497 return rc;
498 }
499
vhost_user_init(struct virtio_uml_device * vu_dev)500 static int vhost_user_init(struct virtio_uml_device *vu_dev)
501 {
502 int rc = vhost_user_set_owner(vu_dev);
503
504 if (rc)
505 return rc;
506 rc = vhost_user_get_features(vu_dev, &vu_dev->features);
507 if (rc)
508 return rc;
509
510 if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
511 rc = vhost_user_get_protocol_features(vu_dev,
512 &vu_dev->protocol_features);
513 if (rc)
514 return rc;
515 vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
516 rc = vhost_user_set_protocol_features(vu_dev,
517 vu_dev->protocol_features);
518 if (rc)
519 return rc;
520 }
521
522 if (vu_dev->protocol_features &
523 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
524 rc = vhost_user_init_slave_req(vu_dev);
525 if (rc)
526 return rc;
527 }
528
529 if (vu_dev->protocol_features &
530 BIT_ULL(VHOST_USER_PROTOCOL_F_MQ)) {
531 rc = vhost_user_get_queue_num(vu_dev, &vu_dev->max_vqs);
532 if (rc)
533 return rc;
534 } else {
535 vu_dev->max_vqs = U64_MAX;
536 }
537
538 return 0;
539 }
540
vhost_user_get_config(struct virtio_uml_device * vu_dev,u32 offset,void * buf,u32 len)541 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
542 u32 offset, void *buf, u32 len)
543 {
544 u32 cfg_size = offset + len;
545 struct vhost_user_msg *msg;
546 size_t payload_size = sizeof(msg->payload.config) + cfg_size;
547 size_t msg_size = sizeof(msg->header) + payload_size;
548 int rc;
549
550 if (!(vu_dev->protocol_features &
551 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
552 return;
553
554 msg = kzalloc(msg_size, GFP_KERNEL);
555 if (!msg)
556 return;
557 msg->header.request = VHOST_USER_GET_CONFIG;
558 msg->header.size = payload_size;
559 msg->payload.config.offset = 0;
560 msg->payload.config.size = cfg_size;
561
562 rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
563 if (rc) {
564 vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
565 rc);
566 goto free;
567 }
568
569 rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
570 if (rc) {
571 vu_err(vu_dev,
572 "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
573 rc);
574 goto free;
575 }
576
577 if (msg->header.size != payload_size ||
578 msg->payload.config.size != cfg_size) {
579 rc = -EPROTO;
580 vu_err(vu_dev,
581 "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
582 msg->header.size, payload_size,
583 msg->payload.config.size, cfg_size);
584 goto free;
585 }
586 memcpy(buf, msg->payload.config.payload + offset, len);
587
588 free:
589 kfree(msg);
590 }
591
vhost_user_set_config(struct virtio_uml_device * vu_dev,u32 offset,const void * buf,u32 len)592 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
593 u32 offset, const void *buf, u32 len)
594 {
595 struct vhost_user_msg *msg;
596 size_t payload_size = sizeof(msg->payload.config) + len;
597 size_t msg_size = sizeof(msg->header) + payload_size;
598 int rc;
599
600 if (!(vu_dev->protocol_features &
601 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
602 return;
603
604 msg = kzalloc(msg_size, GFP_KERNEL);
605 if (!msg)
606 return;
607 msg->header.request = VHOST_USER_SET_CONFIG;
608 msg->header.size = payload_size;
609 msg->payload.config.offset = offset;
610 msg->payload.config.size = len;
611 memcpy(msg->payload.config.payload, buf, len);
612
613 rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
614 if (rc)
615 vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
616 rc);
617
618 kfree(msg);
619 }
620
vhost_user_init_mem_region(u64 addr,u64 size,int * fd_out,struct vhost_user_mem_region * region_out)621 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
622 struct vhost_user_mem_region *region_out)
623 {
624 unsigned long long mem_offset;
625 int rc = phys_mapping(addr, &mem_offset);
626
627 if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
628 return -EFAULT;
629 *fd_out = rc;
630 region_out->guest_addr = addr;
631 region_out->user_addr = addr;
632 region_out->size = size;
633 region_out->mmap_offset = mem_offset;
634
635 /* Ensure mapping is valid for the entire region */
636 rc = phys_mapping(addr + size - 1, &mem_offset);
637 if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
638 addr + size - 1, rc, *fd_out))
639 return -EFAULT;
640 return 0;
641 }
642
vhost_user_set_mem_table(struct virtio_uml_device * vu_dev)643 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
644 {
645 struct vhost_user_msg msg = {
646 .header.request = VHOST_USER_SET_MEM_TABLE,
647 .header.size = offsetof(typeof(msg.payload.mem_regions), regions[1]),
648 .payload.mem_regions.num = 1,
649 };
650 unsigned long reserved = uml_reserved - uml_physmem;
651 int fds[2];
652 int rc;
653
654 /*
655 * This is a bit tricky, see also the comment with setup_physmem().
656 *
657 * Essentially, setup_physmem() uses a file to mmap() our physmem,
658 * but the code and data we *already* have is omitted. To us, this
659 * is no difference, since they both become part of our address
660 * space and memory consumption. To somebody looking in from the
661 * outside, however, it is different because the part of our memory
662 * consumption that's already part of the binary (code/data) is not
663 * mapped from the file, so it's not visible to another mmap from
664 * the file descriptor.
665 *
666 * Thus, don't advertise this space to the vhost-user slave. This
667 * means that the slave will likely abort or similar when we give
668 * it an address from the hidden range, since it's not marked as
669 * a valid address, but at least that way we detect the issue and
670 * don't just have the slave read an all-zeroes buffer from the
671 * shared memory file, or write something there that we can never
672 * see (depending on the direction of the virtqueue traffic.)
673 *
674 * Since we usually don't want to use .text for virtio buffers,
675 * this effectively means that you cannot use
676 * 1) global variables, which are in the .bss and not in the shm
677 * file-backed memory
678 * 2) the stack in some processes, depending on where they have
679 * their stack (or maybe only no interrupt stack?)
680 *
681 * The stack is already not typically valid for DMA, so this isn't
682 * much of a restriction, but global variables might be encountered.
683 *
684 * It might be possible to fix it by copying around the data that's
685 * between bss_start and where we map the file now, but it's not
686 * something that you typically encounter with virtio drivers, so
687 * it didn't seem worthwhile.
688 */
689 rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
690 &fds[0],
691 &msg.payload.mem_regions.regions[0]);
692
693 if (rc < 0)
694 return rc;
695
696 return vhost_user_send(vu_dev, false, &msg, fds,
697 msg.payload.mem_regions.num);
698 }
699
vhost_user_set_vring_state(struct virtio_uml_device * vu_dev,u32 request,u32 index,u32 num)700 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
701 u32 request, u32 index, u32 num)
702 {
703 struct vhost_user_msg msg = {
704 .header.request = request,
705 .header.size = sizeof(msg.payload.vring_state),
706 .payload.vring_state.index = index,
707 .payload.vring_state.num = num,
708 };
709
710 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
711 }
712
vhost_user_set_vring_num(struct virtio_uml_device * vu_dev,u32 index,u32 num)713 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
714 u32 index, u32 num)
715 {
716 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
717 index, num);
718 }
719
vhost_user_set_vring_base(struct virtio_uml_device * vu_dev,u32 index,u32 offset)720 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
721 u32 index, u32 offset)
722 {
723 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
724 index, offset);
725 }
726
vhost_user_set_vring_addr(struct virtio_uml_device * vu_dev,u32 index,u64 desc,u64 used,u64 avail,u64 log)727 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
728 u32 index, u64 desc, u64 used, u64 avail,
729 u64 log)
730 {
731 struct vhost_user_msg msg = {
732 .header.request = VHOST_USER_SET_VRING_ADDR,
733 .header.size = sizeof(msg.payload.vring_addr),
734 .payload.vring_addr.index = index,
735 .payload.vring_addr.desc = desc,
736 .payload.vring_addr.used = used,
737 .payload.vring_addr.avail = avail,
738 .payload.vring_addr.log = log,
739 };
740
741 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
742 }
743
vhost_user_set_vring_fd(struct virtio_uml_device * vu_dev,u32 request,int index,int fd)744 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
745 u32 request, int index, int fd)
746 {
747 struct vhost_user_msg msg = {
748 .header.request = request,
749 .header.size = sizeof(msg.payload.integer),
750 .payload.integer = index,
751 };
752
753 if (index & ~VHOST_USER_VRING_INDEX_MASK)
754 return -EINVAL;
755 if (fd < 0) {
756 msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
757 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
758 }
759 return vhost_user_send(vu_dev, false, &msg, &fd, 1);
760 }
761
vhost_user_set_vring_call(struct virtio_uml_device * vu_dev,int index,int fd)762 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
763 int index, int fd)
764 {
765 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
766 index, fd);
767 }
768
vhost_user_set_vring_kick(struct virtio_uml_device * vu_dev,int index,int fd)769 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
770 int index, int fd)
771 {
772 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
773 index, fd);
774 }
775
vhost_user_set_vring_enable(struct virtio_uml_device * vu_dev,u32 index,bool enable)776 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
777 u32 index, bool enable)
778 {
779 if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
780 return 0;
781
782 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
783 index, enable);
784 }
785
786
787 /* Virtio interface */
788
vu_notify(struct virtqueue * vq)789 static bool vu_notify(struct virtqueue *vq)
790 {
791 struct virtio_uml_vq_info *info = vq->priv;
792 const uint64_t n = 1;
793 int rc;
794
795 if (info->suspended)
796 return true;
797
798 time_travel_propagate_time();
799
800 if (info->kick_fd < 0) {
801 struct virtio_uml_device *vu_dev;
802
803 vu_dev = to_virtio_uml_device(vq->vdev);
804
805 return vhost_user_set_vring_state(vu_dev, VHOST_USER_VRING_KICK,
806 vq->index, 0) == 0;
807 }
808
809 do {
810 rc = os_write_file(info->kick_fd, &n, sizeof(n));
811 } while (rc == -EINTR);
812 return !WARN(rc != sizeof(n), "write returned %d\n", rc);
813 }
814
vu_interrupt(int irq,void * opaque)815 static irqreturn_t vu_interrupt(int irq, void *opaque)
816 {
817 struct virtqueue *vq = opaque;
818 struct virtio_uml_vq_info *info = vq->priv;
819 uint64_t n;
820 int rc;
821 irqreturn_t ret = IRQ_NONE;
822
823 do {
824 rc = os_read_file(info->call_fd, &n, sizeof(n));
825 if (rc == sizeof(n))
826 ret |= vring_interrupt(irq, vq);
827 } while (rc == sizeof(n) || rc == -EINTR);
828 WARN(rc != -EAGAIN, "read returned %d\n", rc);
829 return ret;
830 }
831
832
vu_get(struct virtio_device * vdev,unsigned offset,void * buf,unsigned len)833 static void vu_get(struct virtio_device *vdev, unsigned offset,
834 void *buf, unsigned len)
835 {
836 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
837
838 vhost_user_get_config(vu_dev, offset, buf, len);
839 }
840
vu_set(struct virtio_device * vdev,unsigned offset,const void * buf,unsigned len)841 static void vu_set(struct virtio_device *vdev, unsigned offset,
842 const void *buf, unsigned len)
843 {
844 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
845
846 vhost_user_set_config(vu_dev, offset, buf, len);
847 }
848
vu_get_status(struct virtio_device * vdev)849 static u8 vu_get_status(struct virtio_device *vdev)
850 {
851 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
852
853 return vu_dev->status;
854 }
855
vu_set_status(struct virtio_device * vdev,u8 status)856 static void vu_set_status(struct virtio_device *vdev, u8 status)
857 {
858 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
859
860 vu_dev->status = status;
861 }
862
vu_reset(struct virtio_device * vdev)863 static void vu_reset(struct virtio_device *vdev)
864 {
865 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
866
867 vu_dev->status = 0;
868 }
869
vu_del_vq(struct virtqueue * vq)870 static void vu_del_vq(struct virtqueue *vq)
871 {
872 struct virtio_uml_vq_info *info = vq->priv;
873
874 if (info->call_fd >= 0) {
875 struct virtio_uml_device *vu_dev;
876
877 vu_dev = to_virtio_uml_device(vq->vdev);
878
879 um_free_irq(vu_dev->irq, vq);
880 os_close_file(info->call_fd);
881 }
882
883 if (info->kick_fd >= 0)
884 os_close_file(info->kick_fd);
885
886 vring_del_virtqueue(vq);
887 kfree(info);
888 }
889
vu_del_vqs(struct virtio_device * vdev)890 static void vu_del_vqs(struct virtio_device *vdev)
891 {
892 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
893 struct virtqueue *vq, *n;
894 u64 features;
895
896 /* Note: reverse order as a workaround to a decoding bug in snabb */
897 list_for_each_entry_reverse(vq, &vdev->vqs, list)
898 WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
899
900 /* Ensure previous messages have been processed */
901 WARN_ON(vhost_user_get_features(vu_dev, &features));
902
903 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
904 vu_del_vq(vq);
905 }
906
vu_setup_vq_call_fd(struct virtio_uml_device * vu_dev,struct virtqueue * vq)907 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
908 struct virtqueue *vq)
909 {
910 struct virtio_uml_vq_info *info = vq->priv;
911 int call_fds[2];
912 int rc, irq;
913
914 /* no call FD needed/desired in this case */
915 if (vu_dev->protocol_features &
916 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
917 vu_dev->protocol_features &
918 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
919 info->call_fd = -1;
920 return 0;
921 }
922
923 /* Use a pipe for call fd, since SIGIO is not supported for eventfd */
924 rc = os_pipe(call_fds, true, true);
925 if (rc < 0)
926 return rc;
927
928 info->call_fd = call_fds[0];
929 irq = um_request_irq(vu_dev->irq, info->call_fd, IRQ_READ,
930 vu_interrupt, IRQF_SHARED, info->name, vq);
931 if (irq < 0) {
932 rc = irq;
933 goto close_both;
934 }
935
936 rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
937 if (rc)
938 goto release_irq;
939
940 vu_dev->irq = irq;
941
942 goto out;
943
944 release_irq:
945 um_free_irq(irq, vq);
946 close_both:
947 os_close_file(call_fds[0]);
948 out:
949 /* Close (unused) write end of call fds */
950 os_close_file(call_fds[1]);
951
952 return rc;
953 }
954
vu_setup_vq(struct virtio_device * vdev,unsigned index,vq_callback_t * callback,const char * name,bool ctx)955 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
956 unsigned index, vq_callback_t *callback,
957 const char *name, bool ctx)
958 {
959 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
960 struct platform_device *pdev = vu_dev->pdev;
961 struct virtio_uml_vq_info *info;
962 struct virtqueue *vq;
963 int num = MAX_SUPPORTED_QUEUE_SIZE;
964 int rc;
965
966 info = kzalloc(sizeof(*info), GFP_KERNEL);
967 if (!info) {
968 rc = -ENOMEM;
969 goto error_kzalloc;
970 }
971 snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
972 pdev->id, name);
973
974 vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
975 ctx, vu_notify, callback, info->name);
976 if (!vq) {
977 rc = -ENOMEM;
978 goto error_create;
979 }
980 vq->priv = info;
981 vq->num_max = num;
982 num = virtqueue_get_vring_size(vq);
983
984 if (vu_dev->protocol_features &
985 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS)) {
986 info->kick_fd = -1;
987 } else {
988 rc = os_eventfd(0, 0);
989 if (rc < 0)
990 goto error_kick;
991 info->kick_fd = rc;
992 }
993
994 rc = vu_setup_vq_call_fd(vu_dev, vq);
995 if (rc)
996 goto error_call;
997
998 rc = vhost_user_set_vring_num(vu_dev, index, num);
999 if (rc)
1000 goto error_setup;
1001
1002 rc = vhost_user_set_vring_base(vu_dev, index, 0);
1003 if (rc)
1004 goto error_setup;
1005
1006 rc = vhost_user_set_vring_addr(vu_dev, index,
1007 virtqueue_get_desc_addr(vq),
1008 virtqueue_get_used_addr(vq),
1009 virtqueue_get_avail_addr(vq),
1010 (u64) -1);
1011 if (rc)
1012 goto error_setup;
1013
1014 return vq;
1015
1016 error_setup:
1017 if (info->call_fd >= 0) {
1018 um_free_irq(vu_dev->irq, vq);
1019 os_close_file(info->call_fd);
1020 }
1021 error_call:
1022 if (info->kick_fd >= 0)
1023 os_close_file(info->kick_fd);
1024 error_kick:
1025 vring_del_virtqueue(vq);
1026 error_create:
1027 kfree(info);
1028 error_kzalloc:
1029 return ERR_PTR(rc);
1030 }
1031
vu_find_vqs(struct virtio_device * vdev,unsigned nvqs,struct virtqueue * vqs[],struct virtqueue_info vqs_info[],struct irq_affinity * desc)1032 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
1033 struct virtqueue *vqs[],
1034 struct virtqueue_info vqs_info[],
1035 struct irq_affinity *desc)
1036 {
1037 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1038 int i, queue_idx = 0, rc;
1039 struct virtqueue *vq;
1040
1041 /* not supported for now */
1042 if (WARN(nvqs > 64 || nvqs > vu_dev->max_vqs,
1043 "%d VQs requested, only up to 64 or %lld supported\n",
1044 nvqs, vu_dev->max_vqs))
1045 return -EINVAL;
1046
1047 rc = vhost_user_set_mem_table(vu_dev);
1048 if (rc)
1049 return rc;
1050
1051 for (i = 0; i < nvqs; ++i) {
1052 struct virtqueue_info *vqi = &vqs_info[i];
1053
1054 if (!vqi->name) {
1055 vqs[i] = NULL;
1056 continue;
1057 }
1058
1059 vqs[i] = vu_setup_vq(vdev, queue_idx++, vqi->callback,
1060 vqi->name, vqi->ctx);
1061 if (IS_ERR(vqs[i])) {
1062 rc = PTR_ERR(vqs[i]);
1063 goto error_setup;
1064 }
1065 }
1066
1067 list_for_each_entry(vq, &vdev->vqs, list) {
1068 struct virtio_uml_vq_info *info = vq->priv;
1069
1070 if (info->kick_fd >= 0) {
1071 rc = vhost_user_set_vring_kick(vu_dev, vq->index,
1072 info->kick_fd);
1073 if (rc)
1074 goto error_setup;
1075 }
1076
1077 rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
1078 if (rc)
1079 goto error_setup;
1080 }
1081
1082 return 0;
1083
1084 error_setup:
1085 vu_del_vqs(vdev);
1086 return rc;
1087 }
1088
vu_get_features(struct virtio_device * vdev)1089 static u64 vu_get_features(struct virtio_device *vdev)
1090 {
1091 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1092
1093 return vu_dev->features;
1094 }
1095
vu_finalize_features(struct virtio_device * vdev)1096 static int vu_finalize_features(struct virtio_device *vdev)
1097 {
1098 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1099 u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
1100
1101 vring_transport_features(vdev);
1102 vu_dev->features = vdev->features | supported;
1103
1104 return vhost_user_set_features(vu_dev, vu_dev->features);
1105 }
1106
vu_bus_name(struct virtio_device * vdev)1107 static const char *vu_bus_name(struct virtio_device *vdev)
1108 {
1109 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1110
1111 return vu_dev->pdev->name;
1112 }
1113
1114 static const struct virtio_config_ops virtio_uml_config_ops = {
1115 .get = vu_get,
1116 .set = vu_set,
1117 .get_status = vu_get_status,
1118 .set_status = vu_set_status,
1119 .reset = vu_reset,
1120 .find_vqs = vu_find_vqs,
1121 .del_vqs = vu_del_vqs,
1122 .get_features = vu_get_features,
1123 .finalize_features = vu_finalize_features,
1124 .bus_name = vu_bus_name,
1125 };
1126
virtio_uml_release_dev(struct device * d)1127 static void virtio_uml_release_dev(struct device *d)
1128 {
1129 struct virtio_device *vdev =
1130 container_of(d, struct virtio_device, dev);
1131 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1132
1133 time_travel_propagate_time();
1134
1135 /* might not have been opened due to not negotiating the feature */
1136 if (vu_dev->req_fd >= 0) {
1137 um_free_irq(vu_dev->irq, vu_dev);
1138 os_close_file(vu_dev->req_fd);
1139 }
1140
1141 os_close_file(vu_dev->sock);
1142 kfree(vu_dev);
1143 }
1144
virtio_uml_set_no_vq_suspend(struct virtio_device * vdev,bool no_vq_suspend)1145 void virtio_uml_set_no_vq_suspend(struct virtio_device *vdev,
1146 bool no_vq_suspend)
1147 {
1148 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1149
1150 if (WARN_ON(vdev->config != &virtio_uml_config_ops))
1151 return;
1152
1153 vu_dev->no_vq_suspend = no_vq_suspend;
1154 dev_info(&vdev->dev, "%sabled VQ suspend\n",
1155 no_vq_suspend ? "dis" : "en");
1156 }
1157
vu_of_conn_broken(struct work_struct * wk)1158 static void vu_of_conn_broken(struct work_struct *wk)
1159 {
1160 struct virtio_uml_platform_data *pdata;
1161 struct virtio_uml_device *vu_dev;
1162
1163 pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1164
1165 vu_dev = platform_get_drvdata(pdata->pdev);
1166
1167 virtio_break_device(&vu_dev->vdev);
1168
1169 /*
1170 * We can't remove the device from the devicetree so the only thing we
1171 * can do is warn.
1172 */
1173 WARN_ON(1);
1174 }
1175
1176 /* Platform device */
1177
1178 static struct virtio_uml_platform_data *
virtio_uml_create_pdata(struct platform_device * pdev)1179 virtio_uml_create_pdata(struct platform_device *pdev)
1180 {
1181 struct device_node *np = pdev->dev.of_node;
1182 struct virtio_uml_platform_data *pdata;
1183 int ret;
1184
1185 if (!np)
1186 return ERR_PTR(-EINVAL);
1187
1188 pdata = devm_kzalloc(&pdev->dev, sizeof(*pdata), GFP_KERNEL);
1189 if (!pdata)
1190 return ERR_PTR(-ENOMEM);
1191
1192 INIT_WORK(&pdata->conn_broken_wk, vu_of_conn_broken);
1193 pdata->pdev = pdev;
1194
1195 ret = of_property_read_string(np, "socket-path", &pdata->socket_path);
1196 if (ret)
1197 return ERR_PTR(ret);
1198
1199 ret = of_property_read_u32(np, "virtio-device-id",
1200 &pdata->virtio_device_id);
1201 if (ret)
1202 return ERR_PTR(ret);
1203
1204 return pdata;
1205 }
1206
virtio_uml_probe(struct platform_device * pdev)1207 static int virtio_uml_probe(struct platform_device *pdev)
1208 {
1209 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1210 struct virtio_uml_device *vu_dev;
1211 int rc;
1212
1213 if (!pdata) {
1214 pdata = virtio_uml_create_pdata(pdev);
1215 if (IS_ERR(pdata))
1216 return PTR_ERR(pdata);
1217 }
1218
1219 vu_dev = kzalloc(sizeof(*vu_dev), GFP_KERNEL);
1220 if (!vu_dev)
1221 return -ENOMEM;
1222
1223 vu_dev->pdata = pdata;
1224 vu_dev->vdev.dev.parent = &pdev->dev;
1225 vu_dev->vdev.dev.release = virtio_uml_release_dev;
1226 vu_dev->vdev.config = &virtio_uml_config_ops;
1227 vu_dev->vdev.id.device = pdata->virtio_device_id;
1228 vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1229 vu_dev->pdev = pdev;
1230 vu_dev->req_fd = -1;
1231 vu_dev->irq = UM_IRQ_ALLOC;
1232
1233 time_travel_propagate_time();
1234
1235 do {
1236 rc = os_connect_socket(pdata->socket_path);
1237 } while (rc == -EINTR);
1238 if (rc < 0)
1239 goto error_free;
1240 vu_dev->sock = rc;
1241
1242 spin_lock_init(&vu_dev->sock_lock);
1243
1244 rc = vhost_user_init(vu_dev);
1245 if (rc)
1246 goto error_init;
1247
1248 platform_set_drvdata(pdev, vu_dev);
1249
1250 device_set_wakeup_capable(&vu_dev->vdev.dev, true);
1251
1252 rc = register_virtio_device(&vu_dev->vdev);
1253 if (rc)
1254 put_device(&vu_dev->vdev.dev);
1255 vu_dev->registered = 1;
1256 return rc;
1257
1258 error_init:
1259 os_close_file(vu_dev->sock);
1260 error_free:
1261 kfree(vu_dev);
1262 return rc;
1263 }
1264
virtio_uml_remove(struct platform_device * pdev)1265 static void virtio_uml_remove(struct platform_device *pdev)
1266 {
1267 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1268
1269 unregister_virtio_device(&vu_dev->vdev);
1270 }
1271
1272 /* Command line device list */
1273
vu_cmdline_release_dev(struct device * d)1274 static void vu_cmdline_release_dev(struct device *d)
1275 {
1276 }
1277
1278 static struct device vu_cmdline_parent = {
1279 .init_name = "virtio-uml-cmdline",
1280 .release = vu_cmdline_release_dev,
1281 };
1282
1283 static bool vu_cmdline_parent_registered;
1284 static int vu_cmdline_id;
1285
vu_unregister_cmdline_device(struct device * dev,void * data)1286 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1287 {
1288 struct platform_device *pdev = to_platform_device(dev);
1289 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1290
1291 kfree(pdata->socket_path);
1292 platform_device_unregister(pdev);
1293 return 0;
1294 }
1295
vu_conn_broken(struct work_struct * wk)1296 static void vu_conn_broken(struct work_struct *wk)
1297 {
1298 struct virtio_uml_platform_data *pdata;
1299 struct virtio_uml_device *vu_dev;
1300
1301 pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1302
1303 vu_dev = platform_get_drvdata(pdata->pdev);
1304
1305 virtio_break_device(&vu_dev->vdev);
1306
1307 vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1308 }
1309
vu_cmdline_set(const char * device,const struct kernel_param * kp)1310 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1311 {
1312 const char *ids = strchr(device, ':');
1313 unsigned int virtio_device_id;
1314 int processed, consumed, err;
1315 char *socket_path;
1316 struct virtio_uml_platform_data pdata, *ppdata;
1317 struct platform_device *pdev;
1318
1319 if (!ids || ids == device)
1320 return -EINVAL;
1321
1322 processed = sscanf(ids, ":%u%n:%d%n",
1323 &virtio_device_id, &consumed,
1324 &vu_cmdline_id, &consumed);
1325
1326 if (processed < 1 || ids[consumed])
1327 return -EINVAL;
1328
1329 if (!vu_cmdline_parent_registered) {
1330 err = device_register(&vu_cmdline_parent);
1331 if (err) {
1332 pr_err("Failed to register parent device!\n");
1333 put_device(&vu_cmdline_parent);
1334 return err;
1335 }
1336 vu_cmdline_parent_registered = true;
1337 }
1338
1339 socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1340 if (!socket_path)
1341 return -ENOMEM;
1342
1343 pdata.virtio_device_id = (u32) virtio_device_id;
1344 pdata.socket_path = socket_path;
1345
1346 pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1347 vu_cmdline_id, virtio_device_id, socket_path);
1348
1349 pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1350 vu_cmdline_id++, &pdata,
1351 sizeof(pdata));
1352 err = PTR_ERR_OR_ZERO(pdev);
1353 if (err)
1354 goto free;
1355
1356 ppdata = pdev->dev.platform_data;
1357 ppdata->pdev = pdev;
1358 INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1359
1360 return 0;
1361
1362 free:
1363 kfree(socket_path);
1364 return err;
1365 }
1366
vu_cmdline_get_device(struct device * dev,void * data)1367 static int vu_cmdline_get_device(struct device *dev, void *data)
1368 {
1369 struct platform_device *pdev = to_platform_device(dev);
1370 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1371 char *buffer = data;
1372 unsigned int len = strlen(buffer);
1373
1374 snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1375 pdata->socket_path, pdata->virtio_device_id, pdev->id);
1376 return 0;
1377 }
1378
vu_cmdline_get(char * buffer,const struct kernel_param * kp)1379 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1380 {
1381 buffer[0] = '\0';
1382 if (vu_cmdline_parent_registered)
1383 device_for_each_child(&vu_cmdline_parent, buffer,
1384 vu_cmdline_get_device);
1385 return strlen(buffer) + 1;
1386 }
1387
1388 static const struct kernel_param_ops vu_cmdline_param_ops = {
1389 .set = vu_cmdline_set,
1390 .get = vu_cmdline_get,
1391 };
1392
1393 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1394 __uml_help(vu_cmdline_param_ops,
1395 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1396 " Configure a virtio device over a vhost-user socket.\n"
1397 " See virtio_ids.h for a list of possible virtio device id values.\n"
1398 " Optionally use a specific platform_device id.\n\n"
1399 );
1400
1401
vu_unregister_cmdline_devices(void)1402 static void vu_unregister_cmdline_devices(void)
1403 {
1404 if (vu_cmdline_parent_registered) {
1405 device_for_each_child(&vu_cmdline_parent, NULL,
1406 vu_unregister_cmdline_device);
1407 device_unregister(&vu_cmdline_parent);
1408 vu_cmdline_parent_registered = false;
1409 }
1410 }
1411
1412 /* Platform driver */
1413
1414 static const struct of_device_id virtio_uml_match[] = {
1415 { .compatible = "virtio,uml", },
1416 { }
1417 };
1418 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1419
virtio_uml_suspend(struct platform_device * pdev,pm_message_t state)1420 static int virtio_uml_suspend(struct platform_device *pdev, pm_message_t state)
1421 {
1422 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1423
1424 if (!vu_dev->no_vq_suspend) {
1425 struct virtqueue *vq;
1426
1427 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1428 struct virtio_uml_vq_info *info = vq->priv;
1429
1430 info->suspended = true;
1431 vhost_user_set_vring_enable(vu_dev, vq->index, false);
1432 }
1433 }
1434
1435 if (!device_may_wakeup(&vu_dev->vdev.dev)) {
1436 vu_dev->suspended = true;
1437 return 0;
1438 }
1439
1440 return irq_set_irq_wake(vu_dev->irq, 1);
1441 }
1442
virtio_uml_resume(struct platform_device * pdev)1443 static int virtio_uml_resume(struct platform_device *pdev)
1444 {
1445 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1446
1447 if (!vu_dev->no_vq_suspend) {
1448 struct virtqueue *vq;
1449
1450 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1451 struct virtio_uml_vq_info *info = vq->priv;
1452
1453 info->suspended = false;
1454 vhost_user_set_vring_enable(vu_dev, vq->index, true);
1455 }
1456 }
1457
1458 vu_dev->suspended = false;
1459
1460 if (!device_may_wakeup(&vu_dev->vdev.dev))
1461 return 0;
1462
1463 return irq_set_irq_wake(vu_dev->irq, 0);
1464 }
1465
1466 static struct platform_driver virtio_uml_driver = {
1467 .probe = virtio_uml_probe,
1468 .remove = virtio_uml_remove,
1469 .driver = {
1470 .name = "virtio-uml",
1471 .of_match_table = virtio_uml_match,
1472 },
1473 .suspend = virtio_uml_suspend,
1474 .resume = virtio_uml_resume,
1475 };
1476
virtio_uml_init(void)1477 static int __init virtio_uml_init(void)
1478 {
1479 return platform_driver_register(&virtio_uml_driver);
1480 }
1481
virtio_uml_exit(void)1482 static void __exit virtio_uml_exit(void)
1483 {
1484 platform_driver_unregister(&virtio_uml_driver);
1485 vu_unregister_cmdline_devices();
1486 }
1487
1488 module_init(virtio_uml_init);
1489 module_exit(virtio_uml_exit);
1490 __uml_exitcall(virtio_uml_exit);
1491
1492 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1493 MODULE_LICENSE("GPL");
1494