1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Virtio vhost-user driver
5 * Copyright(c) 2019 Intel Corporation
7 * This driver allows virtio devices to be used over a vhost-user socket.
9 * Guest devices can be instantiated by kernel module or command line
10 * parameters. One device will be created for each parameter. Syntax:
12 * virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]
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
19 * virtio_uml.device=/var/uml.socket:1
21 * Based on Virtio MMIO driver by Pawel Moll, copyright 2011-2014, ARM Ltd.
23 #include <linux/module.h>
24 #include <linux/platform_device.h>
25 #include <linux/slab.h>
26 #include <linux/virtio.h>
27 #include <linux/virtio_config.h>
28 #include <linux/virtio_ring.h>
29 #include <linux/time-internal.h>
30 #include <shared/as-layout.h>
34 #include "vhost_user.h"
36 /* Workaround due to a conflict between irq_user.h and irqreturn.h */
41 #define MAX_SUPPORTED_QUEUE_SIZE 256
43 #define to_virtio_uml_device(_vdev) \
44 container_of(_vdev, struct virtio_uml_device, vdev)
46 struct virtio_uml_platform_data {
48 const char *socket_path;
49 struct work_struct conn_broken_wk;
50 struct platform_device *pdev;
53 struct virtio_uml_device {
54 struct virtio_device vdev;
55 struct platform_device *pdev;
60 u64 protocol_features;
65 struct virtio_uml_vq_info {
68 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
70 vq_callback_t *callback;
71 struct time_travel_event defer;
75 extern unsigned long long physmem_size, highmem;
77 #define vu_err(vu_dev, ...) dev_err(&(vu_dev)->pdev->dev, ##__VA_ARGS__)
79 /* Vhost-user protocol */
81 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
82 const int *fds, unsigned int fds_num)
87 rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
94 } while (len && (rc >= 0 || rc == -EINTR));
101 static int full_read(int fd, void *buf, int len, bool abortable)
106 rc = os_read_file(fd, buf, len);
111 } while (len && (rc > 0 || rc == -EINTR || (!abortable && rc == -EAGAIN)));
120 static int vhost_user_recv_header(int fd, struct vhost_user_msg *msg)
122 return full_read(fd, msg, sizeof(msg->header), true);
125 static int vhost_user_recv(struct virtio_uml_device *vu_dev,
126 int fd, struct vhost_user_msg *msg,
127 size_t max_payload_size, bool wait)
133 * In virtio time-travel mode, we're handling all the vhost-user
134 * FDs by polling them whenever appropriate. However, we may get
135 * into a situation where we're sending out an interrupt message
136 * to a device (e.g. a net device) and need to handle a simulation
137 * time message while doing so, e.g. one that tells us to update
138 * our idea of how long we can run without scheduling.
140 * Thus, we need to not just read() from the given fd, but need
141 * to also handle messages for the simulation time - this function
142 * does that for us while waiting for the given fd to be readable.
145 time_travel_wait_readable(fd);
147 rc = vhost_user_recv_header(fd, msg);
149 if (rc == -ECONNRESET && vu_dev->registered) {
150 struct virtio_uml_platform_data *pdata;
152 pdata = vu_dev->pdev->dev.platform_data;
154 virtio_break_device(&vu_dev->vdev);
155 schedule_work(&pdata->conn_broken_wk);
159 size = msg->header.size;
160 if (size > max_payload_size)
162 return full_read(fd, &msg->payload, size, false);
165 static int vhost_user_recv_resp(struct virtio_uml_device *vu_dev,
166 struct vhost_user_msg *msg,
167 size_t max_payload_size)
169 int rc = vhost_user_recv(vu_dev, vu_dev->sock, msg,
170 max_payload_size, true);
175 if (msg->header.flags != (VHOST_USER_FLAG_REPLY | VHOST_USER_VERSION))
181 static int vhost_user_recv_u64(struct virtio_uml_device *vu_dev,
184 struct vhost_user_msg msg;
185 int rc = vhost_user_recv_resp(vu_dev, &msg,
186 sizeof(msg.payload.integer));
190 if (msg.header.size != sizeof(msg.payload.integer))
192 *value = msg.payload.integer;
196 static int vhost_user_recv_req(struct virtio_uml_device *vu_dev,
197 struct vhost_user_msg *msg,
198 size_t max_payload_size)
200 int rc = vhost_user_recv(vu_dev, vu_dev->req_fd, msg,
201 max_payload_size, false);
206 if ((msg->header.flags & ~VHOST_USER_FLAG_NEED_REPLY) !=
213 static int vhost_user_send(struct virtio_uml_device *vu_dev,
214 bool need_response, struct vhost_user_msg *msg,
215 int *fds, size_t num_fds)
217 size_t size = sizeof(msg->header) + msg->header.size;
222 msg->header.flags |= VHOST_USER_VERSION;
225 * The need_response flag indicates that we already need a response,
226 * e.g. to read the features. In these cases, don't request an ACK as
227 * it is meaningless. Also request an ACK only if supported.
229 request_ack = !need_response;
230 if (!(vu_dev->protocol_features &
231 BIT_ULL(VHOST_USER_PROTOCOL_F_REPLY_ACK)))
235 msg->header.flags |= VHOST_USER_FLAG_NEED_REPLY;
237 spin_lock_irqsave(&vu_dev->sock_lock, flags);
238 rc = full_sendmsg_fds(vu_dev->sock, msg, size, fds, num_fds);
245 rc = vhost_user_recv_u64(vu_dev, &status);
250 vu_err(vu_dev, "slave reports error: %llu\n", status);
257 spin_unlock_irqrestore(&vu_dev->sock_lock, flags);
261 static int vhost_user_send_no_payload(struct virtio_uml_device *vu_dev,
262 bool need_response, u32 request)
264 struct vhost_user_msg msg = {
265 .header.request = request,
268 return vhost_user_send(vu_dev, need_response, &msg, NULL, 0);
271 static int vhost_user_send_no_payload_fd(struct virtio_uml_device *vu_dev,
274 struct vhost_user_msg msg = {
275 .header.request = request,
278 return vhost_user_send(vu_dev, false, &msg, &fd, 1);
281 static int vhost_user_send_u64(struct virtio_uml_device *vu_dev,
282 u32 request, u64 value)
284 struct vhost_user_msg msg = {
285 .header.request = request,
286 .header.size = sizeof(msg.payload.integer),
287 .payload.integer = value,
290 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
293 static int vhost_user_set_owner(struct virtio_uml_device *vu_dev)
295 return vhost_user_send_no_payload(vu_dev, false, VHOST_USER_SET_OWNER);
298 static int vhost_user_get_features(struct virtio_uml_device *vu_dev,
301 int rc = vhost_user_send_no_payload(vu_dev, true,
302 VHOST_USER_GET_FEATURES);
306 return vhost_user_recv_u64(vu_dev, features);
309 static int vhost_user_set_features(struct virtio_uml_device *vu_dev,
312 return vhost_user_send_u64(vu_dev, VHOST_USER_SET_FEATURES, features);
315 static int vhost_user_get_protocol_features(struct virtio_uml_device *vu_dev,
316 u64 *protocol_features)
318 int rc = vhost_user_send_no_payload(vu_dev, true,
319 VHOST_USER_GET_PROTOCOL_FEATURES);
323 return vhost_user_recv_u64(vu_dev, protocol_features);
326 static int vhost_user_set_protocol_features(struct virtio_uml_device *vu_dev,
327 u64 protocol_features)
329 return vhost_user_send_u64(vu_dev, VHOST_USER_SET_PROTOCOL_FEATURES,
333 static void vhost_user_reply(struct virtio_uml_device *vu_dev,
334 struct vhost_user_msg *msg, int response)
336 struct vhost_user_msg reply = {
337 .payload.integer = response,
339 size_t size = sizeof(reply.header) + sizeof(reply.payload.integer);
342 reply.header = msg->header;
343 reply.header.flags &= ~VHOST_USER_FLAG_NEED_REPLY;
344 reply.header.flags |= VHOST_USER_FLAG_REPLY;
345 reply.header.size = sizeof(reply.payload.integer);
347 rc = full_sendmsg_fds(vu_dev->req_fd, &reply, size, NULL, 0);
351 "sending reply to slave request failed: %d (size %zu)\n",
355 static irqreturn_t vu_req_interrupt(int irq, void *data)
357 struct virtio_uml_device *vu_dev = data;
358 struct virtqueue *vq;
361 struct vhost_user_msg msg;
362 u8 extra_payload[512];
366 rc = vhost_user_recv_req(vu_dev, &msg.msg,
367 sizeof(msg.msg.payload) +
368 sizeof(msg.extra_payload));
373 switch (msg.msg.header.request) {
374 case VHOST_USER_SLAVE_CONFIG_CHANGE_MSG:
375 virtio_config_changed(&vu_dev->vdev);
378 case VHOST_USER_SLAVE_VRING_CALL:
379 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
380 if (vq->index == msg.msg.payload.vring_state.index) {
382 vring_interrupt(0 /* ignored */, vq);
387 case VHOST_USER_SLAVE_IOTLB_MSG:
388 /* not supported - VIRTIO_F_ACCESS_PLATFORM */
389 case VHOST_USER_SLAVE_VRING_HOST_NOTIFIER_MSG:
390 /* not supported - VHOST_USER_PROTOCOL_F_HOST_NOTIFIER */
392 vu_err(vu_dev, "unexpected slave request %d\n",
393 msg.msg.header.request);
396 if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
397 vhost_user_reply(vu_dev, &msg.msg, response);
402 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
406 /* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
407 rc = os_pipe(req_fds, true, true);
410 vu_dev->req_fd = req_fds[0];
412 rc = um_request_irq(VIRTIO_IRQ, vu_dev->req_fd, IRQ_READ,
413 vu_req_interrupt, IRQF_SHARED,
414 vu_dev->pdev->name, vu_dev);
418 rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
426 um_free_irq(VIRTIO_IRQ, vu_dev);
428 os_close_file(req_fds[0]);
430 /* Close unused write end of request fds */
431 os_close_file(req_fds[1]);
435 static int vhost_user_init(struct virtio_uml_device *vu_dev)
437 int rc = vhost_user_set_owner(vu_dev);
441 rc = vhost_user_get_features(vu_dev, &vu_dev->features);
445 if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
446 rc = vhost_user_get_protocol_features(vu_dev,
447 &vu_dev->protocol_features);
450 vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
451 rc = vhost_user_set_protocol_features(vu_dev,
452 vu_dev->protocol_features);
457 if (vu_dev->protocol_features &
458 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
459 rc = vhost_user_init_slave_req(vu_dev);
467 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
468 u32 offset, void *buf, u32 len)
470 u32 cfg_size = offset + len;
471 struct vhost_user_msg *msg;
472 size_t payload_size = sizeof(msg->payload.config) + cfg_size;
473 size_t msg_size = sizeof(msg->header) + payload_size;
476 if (!(vu_dev->protocol_features &
477 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
480 msg = kzalloc(msg_size, GFP_KERNEL);
483 msg->header.request = VHOST_USER_GET_CONFIG;
484 msg->header.size = payload_size;
485 msg->payload.config.offset = 0;
486 msg->payload.config.size = cfg_size;
488 rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
490 vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
495 rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
498 "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
503 if (msg->header.size != payload_size ||
504 msg->payload.config.size != cfg_size) {
507 "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
508 msg->header.size, payload_size,
509 msg->payload.config.size, cfg_size);
512 memcpy(buf, msg->payload.config.payload + offset, len);
518 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
519 u32 offset, const void *buf, u32 len)
521 struct vhost_user_msg *msg;
522 size_t payload_size = sizeof(msg->payload.config) + len;
523 size_t msg_size = sizeof(msg->header) + payload_size;
526 if (!(vu_dev->protocol_features &
527 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
530 msg = kzalloc(msg_size, GFP_KERNEL);
533 msg->header.request = VHOST_USER_SET_CONFIG;
534 msg->header.size = payload_size;
535 msg->payload.config.offset = offset;
536 msg->payload.config.size = len;
537 memcpy(msg->payload.config.payload, buf, len);
539 rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
541 vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
547 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
548 struct vhost_user_mem_region *region_out)
550 unsigned long long mem_offset;
551 int rc = phys_mapping(addr, &mem_offset);
553 if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
556 region_out->guest_addr = addr;
557 region_out->user_addr = addr;
558 region_out->size = size;
559 region_out->mmap_offset = mem_offset;
561 /* Ensure mapping is valid for the entire region */
562 rc = phys_mapping(addr + size - 1, &mem_offset);
563 if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
564 addr + size - 1, rc, *fd_out))
569 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
571 struct vhost_user_msg msg = {
572 .header.request = VHOST_USER_SET_MEM_TABLE,
573 .header.size = sizeof(msg.payload.mem_regions),
574 .payload.mem_regions.num = 1,
576 unsigned long reserved = uml_reserved - uml_physmem;
581 * This is a bit tricky, see also the comment with setup_physmem().
583 * Essentially, setup_physmem() uses a file to mmap() our physmem,
584 * but the code and data we *already* have is omitted. To us, this
585 * is no difference, since they both become part of our address
586 * space and memory consumption. To somebody looking in from the
587 * outside, however, it is different because the part of our memory
588 * consumption that's already part of the binary (code/data) is not
589 * mapped from the file, so it's not visible to another mmap from
590 * the file descriptor.
592 * Thus, don't advertise this space to the vhost-user slave. This
593 * means that the slave will likely abort or similar when we give
594 * it an address from the hidden range, since it's not marked as
595 * a valid address, but at least that way we detect the issue and
596 * don't just have the slave read an all-zeroes buffer from the
597 * shared memory file, or write something there that we can never
598 * see (depending on the direction of the virtqueue traffic.)
600 * Since we usually don't want to use .text for virtio buffers,
601 * this effectively means that you cannot use
602 * 1) global variables, which are in the .bss and not in the shm
604 * 2) the stack in some processes, depending on where they have
605 * their stack (or maybe only no interrupt stack?)
607 * The stack is already not typically valid for DMA, so this isn't
608 * much of a restriction, but global variables might be encountered.
610 * It might be possible to fix it by copying around the data that's
611 * between bss_start and where we map the file now, but it's not
612 * something that you typically encounter with virtio drivers, so
613 * it didn't seem worthwhile.
615 rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
617 &msg.payload.mem_regions.regions[0]);
622 msg.payload.mem_regions.num++;
623 rc = vhost_user_init_mem_region(__pa(end_iomem), highmem,
624 &fds[1], &msg.payload.mem_regions.regions[1]);
629 return vhost_user_send(vu_dev, false, &msg, fds,
630 msg.payload.mem_regions.num);
633 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
634 u32 request, u32 index, u32 num)
636 struct vhost_user_msg msg = {
637 .header.request = request,
638 .header.size = sizeof(msg.payload.vring_state),
639 .payload.vring_state.index = index,
640 .payload.vring_state.num = num,
643 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
646 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
649 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
653 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
654 u32 index, u32 offset)
656 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
660 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
661 u32 index, u64 desc, u64 used, u64 avail,
664 struct vhost_user_msg msg = {
665 .header.request = VHOST_USER_SET_VRING_ADDR,
666 .header.size = sizeof(msg.payload.vring_addr),
667 .payload.vring_addr.index = index,
668 .payload.vring_addr.desc = desc,
669 .payload.vring_addr.used = used,
670 .payload.vring_addr.avail = avail,
671 .payload.vring_addr.log = log,
674 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
677 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
678 u32 request, int index, int fd)
680 struct vhost_user_msg msg = {
681 .header.request = request,
682 .header.size = sizeof(msg.payload.integer),
683 .payload.integer = index,
686 if (index & ~VHOST_USER_VRING_INDEX_MASK)
689 msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
690 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
692 return vhost_user_send(vu_dev, false, &msg, &fd, 1);
695 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
698 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
702 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
705 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
709 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
710 u32 index, bool enable)
712 if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
715 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
720 /* Virtio interface */
722 static bool vu_notify(struct virtqueue *vq)
724 struct virtio_uml_vq_info *info = vq->priv;
725 const uint64_t n = 1;
728 time_travel_propagate_time();
730 if (info->kick_fd < 0) {
731 struct virtio_uml_device *vu_dev;
733 vu_dev = to_virtio_uml_device(vq->vdev);
735 return vhost_user_set_vring_state(vu_dev, VHOST_USER_VRING_KICK,
740 rc = os_write_file(info->kick_fd, &n, sizeof(n));
741 } while (rc == -EINTR);
742 return !WARN(rc != sizeof(n), "write returned %d\n", rc);
745 static irqreturn_t vu_interrupt(int irq, void *opaque)
747 struct virtqueue *vq = opaque;
748 struct virtio_uml_vq_info *info = vq->priv;
751 irqreturn_t ret = IRQ_NONE;
754 rc = os_read_file(info->call_fd, &n, sizeof(n));
756 ret |= vring_interrupt(irq, vq);
757 } while (rc == sizeof(n) || rc == -EINTR);
758 WARN(rc != -EAGAIN, "read returned %d\n", rc);
763 static void vu_get(struct virtio_device *vdev, unsigned offset,
764 void *buf, unsigned len)
766 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
768 vhost_user_get_config(vu_dev, offset, buf, len);
771 static void vu_set(struct virtio_device *vdev, unsigned offset,
772 const void *buf, unsigned len)
774 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
776 vhost_user_set_config(vu_dev, offset, buf, len);
779 static u8 vu_get_status(struct virtio_device *vdev)
781 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
783 return vu_dev->status;
786 static void vu_set_status(struct virtio_device *vdev, u8 status)
788 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
790 vu_dev->status = status;
793 static void vu_reset(struct virtio_device *vdev)
795 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
800 static void vu_del_vq(struct virtqueue *vq)
802 struct virtio_uml_vq_info *info = vq->priv;
804 if (info->call_fd >= 0) {
805 um_free_irq(VIRTIO_IRQ, vq);
806 os_close_file(info->call_fd);
809 if (info->kick_fd >= 0)
810 os_close_file(info->kick_fd);
812 vring_del_virtqueue(vq);
816 static void vu_del_vqs(struct virtio_device *vdev)
818 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
819 struct virtqueue *vq, *n;
822 /* Note: reverse order as a workaround to a decoding bug in snabb */
823 list_for_each_entry_reverse(vq, &vdev->vqs, list)
824 WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
826 /* Ensure previous messages have been processed */
827 WARN_ON(vhost_user_get_features(vu_dev, &features));
829 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
833 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
834 struct virtqueue *vq)
836 struct virtio_uml_vq_info *info = vq->priv;
840 /* no call FD needed/desired in this case */
841 if (vu_dev->protocol_features &
842 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
843 vu_dev->protocol_features &
844 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
849 /* Use a pipe for call fd, since SIGIO is not supported for eventfd */
850 rc = os_pipe(call_fds, true, true);
854 info->call_fd = call_fds[0];
855 rc = um_request_irq(VIRTIO_IRQ, info->call_fd, IRQ_READ,
856 vu_interrupt, IRQF_SHARED, info->name, vq);
860 rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
867 um_free_irq(VIRTIO_IRQ, vq);
869 os_close_file(call_fds[0]);
871 /* Close (unused) write end of call fds */
872 os_close_file(call_fds[1]);
877 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
878 static void vu_defer_irq_handle(struct time_travel_event *d)
880 struct virtio_uml_vq_info *info;
882 info = container_of(d, struct virtio_uml_vq_info, defer);
883 info->callback(info->vq);
886 static void vu_defer_irq_callback(struct virtqueue *vq)
888 struct virtio_uml_vq_info *info = vq->priv;
890 time_travel_add_irq_event(&info->defer);
894 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
895 unsigned index, vq_callback_t *callback,
896 const char *name, bool ctx)
898 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
899 struct platform_device *pdev = vu_dev->pdev;
900 struct virtio_uml_vq_info *info;
901 struct virtqueue *vq;
902 int num = MAX_SUPPORTED_QUEUE_SIZE;
905 info = kzalloc(sizeof(*info), GFP_KERNEL);
910 snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
913 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
915 * When we get an interrupt, we must bounce it through the simulation
916 * calendar (the simtime device), except for the simtime device itself
917 * since that's part of the simulation control.
919 if (time_travel_mode == TT_MODE_EXTERNAL && callback) {
920 info->callback = callback;
921 callback = vu_defer_irq_callback;
922 time_travel_set_event_fn(&info->defer, vu_defer_irq_handle);
926 vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
927 ctx, vu_notify, callback, info->name);
933 num = virtqueue_get_vring_size(vq);
934 #ifdef CONFIG_UML_TIME_TRAVEL_SUPPORT
938 if (vu_dev->protocol_features &
939 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS)) {
942 rc = os_eventfd(0, 0);
948 rc = vu_setup_vq_call_fd(vu_dev, vq);
952 rc = vhost_user_set_vring_num(vu_dev, index, num);
956 rc = vhost_user_set_vring_base(vu_dev, index, 0);
960 rc = vhost_user_set_vring_addr(vu_dev, index,
961 virtqueue_get_desc_addr(vq),
962 virtqueue_get_used_addr(vq),
963 virtqueue_get_avail_addr(vq),
971 if (info->call_fd >= 0) {
972 um_free_irq(VIRTIO_IRQ, vq);
973 os_close_file(info->call_fd);
976 if (info->kick_fd >= 0)
977 os_close_file(info->kick_fd);
979 vring_del_virtqueue(vq);
986 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
987 struct virtqueue *vqs[], vq_callback_t *callbacks[],
988 const char * const names[], const bool *ctx,
989 struct irq_affinity *desc)
991 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
992 int i, queue_idx = 0, rc;
993 struct virtqueue *vq;
995 rc = vhost_user_set_mem_table(vu_dev);
999 for (i = 0; i < nvqs; ++i) {
1005 vqs[i] = vu_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
1006 ctx ? ctx[i] : false);
1007 if (IS_ERR(vqs[i])) {
1008 rc = PTR_ERR(vqs[i]);
1013 list_for_each_entry(vq, &vdev->vqs, list) {
1014 struct virtio_uml_vq_info *info = vq->priv;
1016 if (info->kick_fd >= 0) {
1017 rc = vhost_user_set_vring_kick(vu_dev, vq->index,
1023 rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
1035 static u64 vu_get_features(struct virtio_device *vdev)
1037 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1039 return vu_dev->features;
1042 static int vu_finalize_features(struct virtio_device *vdev)
1044 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1045 u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
1047 vring_transport_features(vdev);
1048 vu_dev->features = vdev->features | supported;
1050 return vhost_user_set_features(vu_dev, vu_dev->features);
1053 static const char *vu_bus_name(struct virtio_device *vdev)
1055 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1057 return vu_dev->pdev->name;
1060 static const struct virtio_config_ops virtio_uml_config_ops = {
1063 .get_status = vu_get_status,
1064 .set_status = vu_set_status,
1066 .find_vqs = vu_find_vqs,
1067 .del_vqs = vu_del_vqs,
1068 .get_features = vu_get_features,
1069 .finalize_features = vu_finalize_features,
1070 .bus_name = vu_bus_name,
1073 static void virtio_uml_release_dev(struct device *d)
1075 struct virtio_device *vdev =
1076 container_of(d, struct virtio_device, dev);
1077 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1079 time_travel_propagate_time();
1081 /* might not have been opened due to not negotiating the feature */
1082 if (vu_dev->req_fd >= 0) {
1083 um_free_irq(VIRTIO_IRQ, vu_dev);
1084 os_close_file(vu_dev->req_fd);
1087 os_close_file(vu_dev->sock);
1091 /* Platform device */
1093 static int virtio_uml_probe(struct platform_device *pdev)
1095 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1096 struct virtio_uml_device *vu_dev;
1102 vu_dev = kzalloc(sizeof(*vu_dev), GFP_KERNEL);
1106 vu_dev->vdev.dev.parent = &pdev->dev;
1107 vu_dev->vdev.dev.release = virtio_uml_release_dev;
1108 vu_dev->vdev.config = &virtio_uml_config_ops;
1109 vu_dev->vdev.id.device = pdata->virtio_device_id;
1110 vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1111 vu_dev->pdev = pdev;
1112 vu_dev->req_fd = -1;
1114 time_travel_propagate_time();
1117 rc = os_connect_socket(pdata->socket_path);
1118 } while (rc == -EINTR);
1123 spin_lock_init(&vu_dev->sock_lock);
1125 rc = vhost_user_init(vu_dev);
1129 platform_set_drvdata(pdev, vu_dev);
1131 rc = register_virtio_device(&vu_dev->vdev);
1133 put_device(&vu_dev->vdev.dev);
1134 vu_dev->registered = 1;
1138 os_close_file(vu_dev->sock);
1144 static int virtio_uml_remove(struct platform_device *pdev)
1146 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1148 unregister_virtio_device(&vu_dev->vdev);
1152 /* Command line device list */
1154 static void vu_cmdline_release_dev(struct device *d)
1158 static struct device vu_cmdline_parent = {
1159 .init_name = "virtio-uml-cmdline",
1160 .release = vu_cmdline_release_dev,
1163 static bool vu_cmdline_parent_registered;
1164 static int vu_cmdline_id;
1166 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1168 struct platform_device *pdev = to_platform_device(dev);
1169 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1171 kfree(pdata->socket_path);
1172 platform_device_unregister(pdev);
1176 static void vu_conn_broken(struct work_struct *wk)
1178 struct virtio_uml_platform_data *pdata;
1180 pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1181 vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1184 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1186 const char *ids = strchr(device, ':');
1187 unsigned int virtio_device_id;
1188 int processed, consumed, err;
1190 struct virtio_uml_platform_data pdata, *ppdata;
1191 struct platform_device *pdev;
1193 if (!ids || ids == device)
1196 processed = sscanf(ids, ":%u%n:%d%n",
1197 &virtio_device_id, &consumed,
1198 &vu_cmdline_id, &consumed);
1200 if (processed < 1 || ids[consumed])
1203 if (!vu_cmdline_parent_registered) {
1204 err = device_register(&vu_cmdline_parent);
1206 pr_err("Failed to register parent device!\n");
1207 put_device(&vu_cmdline_parent);
1210 vu_cmdline_parent_registered = true;
1213 socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1217 pdata.virtio_device_id = (u32) virtio_device_id;
1218 pdata.socket_path = socket_path;
1220 pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1221 vu_cmdline_id, virtio_device_id, socket_path);
1223 pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1224 vu_cmdline_id++, &pdata,
1226 err = PTR_ERR_OR_ZERO(pdev);
1230 ppdata = pdev->dev.platform_data;
1231 ppdata->pdev = pdev;
1232 INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1241 static int vu_cmdline_get_device(struct device *dev, void *data)
1243 struct platform_device *pdev = to_platform_device(dev);
1244 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1245 char *buffer = data;
1246 unsigned int len = strlen(buffer);
1248 snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1249 pdata->socket_path, pdata->virtio_device_id, pdev->id);
1253 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1256 if (vu_cmdline_parent_registered)
1257 device_for_each_child(&vu_cmdline_parent, buffer,
1258 vu_cmdline_get_device);
1259 return strlen(buffer) + 1;
1262 static const struct kernel_param_ops vu_cmdline_param_ops = {
1263 .set = vu_cmdline_set,
1264 .get = vu_cmdline_get,
1267 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1268 __uml_help(vu_cmdline_param_ops,
1269 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1270 " Configure a virtio device over a vhost-user socket.\n"
1271 " See virtio_ids.h for a list of possible virtio device id values.\n"
1272 " Optionally use a specific platform_device id.\n\n"
1276 static void vu_unregister_cmdline_devices(void)
1278 if (vu_cmdline_parent_registered) {
1279 device_for_each_child(&vu_cmdline_parent, NULL,
1280 vu_unregister_cmdline_device);
1281 device_unregister(&vu_cmdline_parent);
1282 vu_cmdline_parent_registered = false;
1286 /* Platform driver */
1288 static const struct of_device_id virtio_uml_match[] = {
1289 { .compatible = "virtio,uml", },
1292 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1294 static struct platform_driver virtio_uml_driver = {
1295 .probe = virtio_uml_probe,
1296 .remove = virtio_uml_remove,
1298 .name = "virtio-uml",
1299 .of_match_table = virtio_uml_match,
1303 static int __init virtio_uml_init(void)
1305 return platform_driver_register(&virtio_uml_driver);
1308 static void __exit virtio_uml_exit(void)
1310 platform_driver_unregister(&virtio_uml_driver);
1311 vu_unregister_cmdline_devices();
1314 module_init(virtio_uml_init);
1315 module_exit(virtio_uml_exit);
1316 __uml_exitcall(virtio_uml_exit);
1318 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1319 MODULE_LICENSE("GPL");