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 <linux/virtio-uml.h>
31 #include <shared/as-layout.h>
35 #include "vhost_user.h"
37 #define MAX_SUPPORTED_QUEUE_SIZE 256
39 #define to_virtio_uml_device(_vdev) \
40 container_of(_vdev, struct virtio_uml_device, vdev)
42 struct virtio_uml_platform_data {
44 const char *socket_path;
45 struct work_struct conn_broken_wk;
46 struct platform_device *pdev;
49 struct virtio_uml_device {
50 struct virtio_device vdev;
51 struct platform_device *pdev;
54 int sock, req_fd, irq;
56 u64 protocol_features;
62 u8 config_changed_irq:1;
63 uint64_t vq_irq_vq_map;
66 struct virtio_uml_vq_info {
72 extern unsigned long long physmem_size, highmem;
74 #define vu_err(vu_dev, ...) dev_err(&(vu_dev)->pdev->dev, ##__VA_ARGS__)
76 /* Vhost-user protocol */
78 static int full_sendmsg_fds(int fd, const void *buf, unsigned int len,
79 const int *fds, unsigned int fds_num)
84 rc = os_sendmsg_fds(fd, buf, len, fds, fds_num);
91 } while (len && (rc >= 0 || rc == -EINTR));
98 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_read_message(struct virtio_uml_device *vu_dev,
356 struct time_travel_event *ev)
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 vu_dev->config_changed_irq = true;
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 vu_dev->vq_irq_vq_map |= BIT_ULL(vq->index);
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 (ev && !vu_dev->suspended)
397 time_travel_add_irq_event(ev);
399 if (msg.msg.header.flags & VHOST_USER_FLAG_NEED_REPLY)
400 vhost_user_reply(vu_dev, &msg.msg, response);
405 static irqreturn_t vu_req_interrupt(int irq, void *data)
407 struct virtio_uml_device *vu_dev = data;
408 irqreturn_t ret = IRQ_HANDLED;
410 if (!um_irq_timetravel_handler_used())
411 ret = vu_req_read_message(vu_dev, NULL);
413 if (vu_dev->vq_irq_vq_map) {
414 struct virtqueue *vq;
416 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
417 if (vu_dev->vq_irq_vq_map & BIT_ULL(vq->index))
418 vring_interrupt(0 /* ignored */, vq);
420 vu_dev->vq_irq_vq_map = 0;
421 } else if (vu_dev->config_changed_irq) {
422 virtio_config_changed(&vu_dev->vdev);
423 vu_dev->config_changed_irq = false;
429 static void vu_req_interrupt_comm_handler(int irq, int fd, void *data,
430 struct time_travel_event *ev)
432 vu_req_read_message(data, ev);
435 static int vhost_user_init_slave_req(struct virtio_uml_device *vu_dev)
439 /* Use a pipe for slave req fd, SIGIO is not supported for eventfd */
440 rc = os_pipe(req_fds, true, true);
443 vu_dev->req_fd = req_fds[0];
445 rc = um_request_irq_tt(UM_IRQ_ALLOC, vu_dev->req_fd, IRQ_READ,
446 vu_req_interrupt, IRQF_SHARED,
447 vu_dev->pdev->name, vu_dev,
448 vu_req_interrupt_comm_handler);
454 rc = vhost_user_send_no_payload_fd(vu_dev, VHOST_USER_SET_SLAVE_REQ_FD,
462 um_free_irq(vu_dev->irq, vu_dev);
464 os_close_file(req_fds[0]);
466 /* Close unused write end of request fds */
467 os_close_file(req_fds[1]);
471 static int vhost_user_init(struct virtio_uml_device *vu_dev)
473 int rc = vhost_user_set_owner(vu_dev);
477 rc = vhost_user_get_features(vu_dev, &vu_dev->features);
481 if (vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)) {
482 rc = vhost_user_get_protocol_features(vu_dev,
483 &vu_dev->protocol_features);
486 vu_dev->protocol_features &= VHOST_USER_SUPPORTED_PROTOCOL_F;
487 rc = vhost_user_set_protocol_features(vu_dev,
488 vu_dev->protocol_features);
493 if (vu_dev->protocol_features &
494 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
495 rc = vhost_user_init_slave_req(vu_dev);
503 static void vhost_user_get_config(struct virtio_uml_device *vu_dev,
504 u32 offset, void *buf, u32 len)
506 u32 cfg_size = offset + len;
507 struct vhost_user_msg *msg;
508 size_t payload_size = sizeof(msg->payload.config) + cfg_size;
509 size_t msg_size = sizeof(msg->header) + payload_size;
512 if (!(vu_dev->protocol_features &
513 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
516 msg = kzalloc(msg_size, GFP_KERNEL);
519 msg->header.request = VHOST_USER_GET_CONFIG;
520 msg->header.size = payload_size;
521 msg->payload.config.offset = 0;
522 msg->payload.config.size = cfg_size;
524 rc = vhost_user_send(vu_dev, true, msg, NULL, 0);
526 vu_err(vu_dev, "sending VHOST_USER_GET_CONFIG failed: %d\n",
531 rc = vhost_user_recv_resp(vu_dev, msg, msg_size);
534 "receiving VHOST_USER_GET_CONFIG response failed: %d\n",
539 if (msg->header.size != payload_size ||
540 msg->payload.config.size != cfg_size) {
543 "Invalid VHOST_USER_GET_CONFIG sizes (payload %d expected %zu, config %u expected %u)\n",
544 msg->header.size, payload_size,
545 msg->payload.config.size, cfg_size);
548 memcpy(buf, msg->payload.config.payload + offset, len);
554 static void vhost_user_set_config(struct virtio_uml_device *vu_dev,
555 u32 offset, const void *buf, u32 len)
557 struct vhost_user_msg *msg;
558 size_t payload_size = sizeof(msg->payload.config) + len;
559 size_t msg_size = sizeof(msg->header) + payload_size;
562 if (!(vu_dev->protocol_features &
563 BIT_ULL(VHOST_USER_PROTOCOL_F_CONFIG)))
566 msg = kzalloc(msg_size, GFP_KERNEL);
569 msg->header.request = VHOST_USER_SET_CONFIG;
570 msg->header.size = payload_size;
571 msg->payload.config.offset = offset;
572 msg->payload.config.size = len;
573 memcpy(msg->payload.config.payload, buf, len);
575 rc = vhost_user_send(vu_dev, false, msg, NULL, 0);
577 vu_err(vu_dev, "sending VHOST_USER_SET_CONFIG failed: %d\n",
583 static int vhost_user_init_mem_region(u64 addr, u64 size, int *fd_out,
584 struct vhost_user_mem_region *region_out)
586 unsigned long long mem_offset;
587 int rc = phys_mapping(addr, &mem_offset);
589 if (WARN(rc < 0, "phys_mapping of 0x%llx returned %d\n", addr, rc))
592 region_out->guest_addr = addr;
593 region_out->user_addr = addr;
594 region_out->size = size;
595 region_out->mmap_offset = mem_offset;
597 /* Ensure mapping is valid for the entire region */
598 rc = phys_mapping(addr + size - 1, &mem_offset);
599 if (WARN(rc != *fd_out, "phys_mapping of 0x%llx failed: %d != %d\n",
600 addr + size - 1, rc, *fd_out))
605 static int vhost_user_set_mem_table(struct virtio_uml_device *vu_dev)
607 struct vhost_user_msg msg = {
608 .header.request = VHOST_USER_SET_MEM_TABLE,
609 .header.size = sizeof(msg.payload.mem_regions),
610 .payload.mem_regions.num = 1,
612 unsigned long reserved = uml_reserved - uml_physmem;
617 * This is a bit tricky, see also the comment with setup_physmem().
619 * Essentially, setup_physmem() uses a file to mmap() our physmem,
620 * but the code and data we *already* have is omitted. To us, this
621 * is no difference, since they both become part of our address
622 * space and memory consumption. To somebody looking in from the
623 * outside, however, it is different because the part of our memory
624 * consumption that's already part of the binary (code/data) is not
625 * mapped from the file, so it's not visible to another mmap from
626 * the file descriptor.
628 * Thus, don't advertise this space to the vhost-user slave. This
629 * means that the slave will likely abort or similar when we give
630 * it an address from the hidden range, since it's not marked as
631 * a valid address, but at least that way we detect the issue and
632 * don't just have the slave read an all-zeroes buffer from the
633 * shared memory file, or write something there that we can never
634 * see (depending on the direction of the virtqueue traffic.)
636 * Since we usually don't want to use .text for virtio buffers,
637 * this effectively means that you cannot use
638 * 1) global variables, which are in the .bss and not in the shm
640 * 2) the stack in some processes, depending on where they have
641 * their stack (or maybe only no interrupt stack?)
643 * The stack is already not typically valid for DMA, so this isn't
644 * much of a restriction, but global variables might be encountered.
646 * It might be possible to fix it by copying around the data that's
647 * between bss_start and where we map the file now, but it's not
648 * something that you typically encounter with virtio drivers, so
649 * it didn't seem worthwhile.
651 rc = vhost_user_init_mem_region(reserved, physmem_size - reserved,
653 &msg.payload.mem_regions.regions[0]);
658 msg.payload.mem_regions.num++;
659 rc = vhost_user_init_mem_region(__pa(end_iomem), highmem,
660 &fds[1], &msg.payload.mem_regions.regions[1]);
665 return vhost_user_send(vu_dev, false, &msg, fds,
666 msg.payload.mem_regions.num);
669 static int vhost_user_set_vring_state(struct virtio_uml_device *vu_dev,
670 u32 request, u32 index, u32 num)
672 struct vhost_user_msg msg = {
673 .header.request = request,
674 .header.size = sizeof(msg.payload.vring_state),
675 .payload.vring_state.index = index,
676 .payload.vring_state.num = num,
679 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
682 static int vhost_user_set_vring_num(struct virtio_uml_device *vu_dev,
685 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_NUM,
689 static int vhost_user_set_vring_base(struct virtio_uml_device *vu_dev,
690 u32 index, u32 offset)
692 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_BASE,
696 static int vhost_user_set_vring_addr(struct virtio_uml_device *vu_dev,
697 u32 index, u64 desc, u64 used, u64 avail,
700 struct vhost_user_msg msg = {
701 .header.request = VHOST_USER_SET_VRING_ADDR,
702 .header.size = sizeof(msg.payload.vring_addr),
703 .payload.vring_addr.index = index,
704 .payload.vring_addr.desc = desc,
705 .payload.vring_addr.used = used,
706 .payload.vring_addr.avail = avail,
707 .payload.vring_addr.log = log,
710 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
713 static int vhost_user_set_vring_fd(struct virtio_uml_device *vu_dev,
714 u32 request, int index, int fd)
716 struct vhost_user_msg msg = {
717 .header.request = request,
718 .header.size = sizeof(msg.payload.integer),
719 .payload.integer = index,
722 if (index & ~VHOST_USER_VRING_INDEX_MASK)
725 msg.payload.integer |= VHOST_USER_VRING_POLL_MASK;
726 return vhost_user_send(vu_dev, false, &msg, NULL, 0);
728 return vhost_user_send(vu_dev, false, &msg, &fd, 1);
731 static int vhost_user_set_vring_call(struct virtio_uml_device *vu_dev,
734 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_CALL,
738 static int vhost_user_set_vring_kick(struct virtio_uml_device *vu_dev,
741 return vhost_user_set_vring_fd(vu_dev, VHOST_USER_SET_VRING_KICK,
745 static int vhost_user_set_vring_enable(struct virtio_uml_device *vu_dev,
746 u32 index, bool enable)
748 if (!(vu_dev->features & BIT_ULL(VHOST_USER_F_PROTOCOL_FEATURES)))
751 return vhost_user_set_vring_state(vu_dev, VHOST_USER_SET_VRING_ENABLE,
756 /* Virtio interface */
758 static bool vu_notify(struct virtqueue *vq)
760 struct virtio_uml_vq_info *info = vq->priv;
761 const uint64_t n = 1;
767 time_travel_propagate_time();
769 if (info->kick_fd < 0) {
770 struct virtio_uml_device *vu_dev;
772 vu_dev = to_virtio_uml_device(vq->vdev);
774 return vhost_user_set_vring_state(vu_dev, VHOST_USER_VRING_KICK,
779 rc = os_write_file(info->kick_fd, &n, sizeof(n));
780 } while (rc == -EINTR);
781 return !WARN(rc != sizeof(n), "write returned %d\n", rc);
784 static irqreturn_t vu_interrupt(int irq, void *opaque)
786 struct virtqueue *vq = opaque;
787 struct virtio_uml_vq_info *info = vq->priv;
790 irqreturn_t ret = IRQ_NONE;
793 rc = os_read_file(info->call_fd, &n, sizeof(n));
795 ret |= vring_interrupt(irq, vq);
796 } while (rc == sizeof(n) || rc == -EINTR);
797 WARN(rc != -EAGAIN, "read returned %d\n", rc);
802 static void vu_get(struct virtio_device *vdev, unsigned offset,
803 void *buf, unsigned len)
805 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
807 vhost_user_get_config(vu_dev, offset, buf, len);
810 static void vu_set(struct virtio_device *vdev, unsigned offset,
811 const void *buf, unsigned len)
813 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
815 vhost_user_set_config(vu_dev, offset, buf, len);
818 static u8 vu_get_status(struct virtio_device *vdev)
820 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
822 return vu_dev->status;
825 static void vu_set_status(struct virtio_device *vdev, u8 status)
827 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
829 vu_dev->status = status;
832 static void vu_reset(struct virtio_device *vdev)
834 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
839 static void vu_del_vq(struct virtqueue *vq)
841 struct virtio_uml_vq_info *info = vq->priv;
843 if (info->call_fd >= 0) {
844 struct virtio_uml_device *vu_dev;
846 vu_dev = to_virtio_uml_device(vq->vdev);
848 um_free_irq(vu_dev->irq, vq);
849 os_close_file(info->call_fd);
852 if (info->kick_fd >= 0)
853 os_close_file(info->kick_fd);
855 vring_del_virtqueue(vq);
859 static void vu_del_vqs(struct virtio_device *vdev)
861 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
862 struct virtqueue *vq, *n;
865 /* Note: reverse order as a workaround to a decoding bug in snabb */
866 list_for_each_entry_reverse(vq, &vdev->vqs, list)
867 WARN_ON(vhost_user_set_vring_enable(vu_dev, vq->index, false));
869 /* Ensure previous messages have been processed */
870 WARN_ON(vhost_user_get_features(vu_dev, &features));
872 list_for_each_entry_safe(vq, n, &vdev->vqs, list)
876 static int vu_setup_vq_call_fd(struct virtio_uml_device *vu_dev,
877 struct virtqueue *vq)
879 struct virtio_uml_vq_info *info = vq->priv;
883 /* no call FD needed/desired in this case */
884 if (vu_dev->protocol_features &
885 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS) &&
886 vu_dev->protocol_features &
887 BIT_ULL(VHOST_USER_PROTOCOL_F_SLAVE_REQ)) {
892 /* Use a pipe for call fd, since SIGIO is not supported for eventfd */
893 rc = os_pipe(call_fds, true, true);
897 info->call_fd = call_fds[0];
898 rc = um_request_irq(vu_dev->irq, info->call_fd, IRQ_READ,
899 vu_interrupt, IRQF_SHARED, info->name, vq);
903 rc = vhost_user_set_vring_call(vu_dev, vq->index, call_fds[1]);
910 um_free_irq(vu_dev->irq, vq);
912 os_close_file(call_fds[0]);
914 /* Close (unused) write end of call fds */
915 os_close_file(call_fds[1]);
920 static struct virtqueue *vu_setup_vq(struct virtio_device *vdev,
921 unsigned index, vq_callback_t *callback,
922 const char *name, bool ctx)
924 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
925 struct platform_device *pdev = vu_dev->pdev;
926 struct virtio_uml_vq_info *info;
927 struct virtqueue *vq;
928 int num = MAX_SUPPORTED_QUEUE_SIZE;
931 info = kzalloc(sizeof(*info), GFP_KERNEL);
936 snprintf(info->name, sizeof(info->name), "%s.%d-%s", pdev->name,
939 vq = vring_create_virtqueue(index, num, PAGE_SIZE, vdev, true, true,
940 ctx, vu_notify, callback, info->name);
946 num = virtqueue_get_vring_size(vq);
948 if (vu_dev->protocol_features &
949 BIT_ULL(VHOST_USER_PROTOCOL_F_INBAND_NOTIFICATIONS)) {
952 rc = os_eventfd(0, 0);
958 rc = vu_setup_vq_call_fd(vu_dev, vq);
962 rc = vhost_user_set_vring_num(vu_dev, index, num);
966 rc = vhost_user_set_vring_base(vu_dev, index, 0);
970 rc = vhost_user_set_vring_addr(vu_dev, index,
971 virtqueue_get_desc_addr(vq),
972 virtqueue_get_used_addr(vq),
973 virtqueue_get_avail_addr(vq),
981 if (info->call_fd >= 0) {
982 um_free_irq(vu_dev->irq, vq);
983 os_close_file(info->call_fd);
986 if (info->kick_fd >= 0)
987 os_close_file(info->kick_fd);
989 vring_del_virtqueue(vq);
996 static int vu_find_vqs(struct virtio_device *vdev, unsigned nvqs,
997 struct virtqueue *vqs[], vq_callback_t *callbacks[],
998 const char * const names[], const bool *ctx,
999 struct irq_affinity *desc)
1001 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1002 int i, queue_idx = 0, rc;
1003 struct virtqueue *vq;
1005 /* not supported for now */
1006 if (WARN_ON(nvqs > 64))
1009 rc = vhost_user_set_mem_table(vu_dev);
1013 for (i = 0; i < nvqs; ++i) {
1019 vqs[i] = vu_setup_vq(vdev, queue_idx++, callbacks[i], names[i],
1020 ctx ? ctx[i] : false);
1021 if (IS_ERR(vqs[i])) {
1022 rc = PTR_ERR(vqs[i]);
1027 list_for_each_entry(vq, &vdev->vqs, list) {
1028 struct virtio_uml_vq_info *info = vq->priv;
1030 if (info->kick_fd >= 0) {
1031 rc = vhost_user_set_vring_kick(vu_dev, vq->index,
1037 rc = vhost_user_set_vring_enable(vu_dev, vq->index, true);
1049 static u64 vu_get_features(struct virtio_device *vdev)
1051 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1053 return vu_dev->features;
1056 static int vu_finalize_features(struct virtio_device *vdev)
1058 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1059 u64 supported = vdev->features & VHOST_USER_SUPPORTED_F;
1061 vring_transport_features(vdev);
1062 vu_dev->features = vdev->features | supported;
1064 return vhost_user_set_features(vu_dev, vu_dev->features);
1067 static const char *vu_bus_name(struct virtio_device *vdev)
1069 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1071 return vu_dev->pdev->name;
1074 static const struct virtio_config_ops virtio_uml_config_ops = {
1077 .get_status = vu_get_status,
1078 .set_status = vu_set_status,
1080 .find_vqs = vu_find_vqs,
1081 .del_vqs = vu_del_vqs,
1082 .get_features = vu_get_features,
1083 .finalize_features = vu_finalize_features,
1084 .bus_name = vu_bus_name,
1087 static void virtio_uml_release_dev(struct device *d)
1089 struct virtio_device *vdev =
1090 container_of(d, struct virtio_device, dev);
1091 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1093 /* might not have been opened due to not negotiating the feature */
1094 if (vu_dev->req_fd >= 0) {
1095 um_free_irq(vu_dev->irq, vu_dev);
1096 os_close_file(vu_dev->req_fd);
1099 os_close_file(vu_dev->sock);
1103 void virtio_uml_set_no_vq_suspend(struct virtio_device *vdev,
1106 struct virtio_uml_device *vu_dev = to_virtio_uml_device(vdev);
1108 if (WARN_ON(vdev->config != &virtio_uml_config_ops))
1111 vu_dev->no_vq_suspend = no_vq_suspend;
1112 dev_info(&vdev->dev, "%sabled VQ suspend\n",
1113 no_vq_suspend ? "dis" : "en");
1116 /* Platform device */
1118 static int virtio_uml_probe(struct platform_device *pdev)
1120 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1121 struct virtio_uml_device *vu_dev;
1127 vu_dev = kzalloc(sizeof(*vu_dev), GFP_KERNEL);
1131 vu_dev->vdev.dev.parent = &pdev->dev;
1132 vu_dev->vdev.dev.release = virtio_uml_release_dev;
1133 vu_dev->vdev.config = &virtio_uml_config_ops;
1134 vu_dev->vdev.id.device = pdata->virtio_device_id;
1135 vu_dev->vdev.id.vendor = VIRTIO_DEV_ANY_ID;
1136 vu_dev->pdev = pdev;
1137 vu_dev->req_fd = -1;
1140 rc = os_connect_socket(pdata->socket_path);
1141 } while (rc == -EINTR);
1146 spin_lock_init(&vu_dev->sock_lock);
1148 rc = vhost_user_init(vu_dev);
1152 platform_set_drvdata(pdev, vu_dev);
1154 device_set_wakeup_capable(&vu_dev->vdev.dev, true);
1156 rc = register_virtio_device(&vu_dev->vdev);
1158 put_device(&vu_dev->vdev.dev);
1159 vu_dev->registered = 1;
1163 os_close_file(vu_dev->sock);
1169 static int virtio_uml_remove(struct platform_device *pdev)
1171 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1173 unregister_virtio_device(&vu_dev->vdev);
1177 /* Command line device list */
1179 static void vu_cmdline_release_dev(struct device *d)
1183 static struct device vu_cmdline_parent = {
1184 .init_name = "virtio-uml-cmdline",
1185 .release = vu_cmdline_release_dev,
1188 static bool vu_cmdline_parent_registered;
1189 static int vu_cmdline_id;
1191 static int vu_unregister_cmdline_device(struct device *dev, void *data)
1193 struct platform_device *pdev = to_platform_device(dev);
1194 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1196 kfree(pdata->socket_path);
1197 platform_device_unregister(pdev);
1201 static void vu_conn_broken(struct work_struct *wk)
1203 struct virtio_uml_platform_data *pdata;
1205 pdata = container_of(wk, struct virtio_uml_platform_data, conn_broken_wk);
1206 vu_unregister_cmdline_device(&pdata->pdev->dev, NULL);
1209 static int vu_cmdline_set(const char *device, const struct kernel_param *kp)
1211 const char *ids = strchr(device, ':');
1212 unsigned int virtio_device_id;
1213 int processed, consumed, err;
1215 struct virtio_uml_platform_data pdata, *ppdata;
1216 struct platform_device *pdev;
1218 if (!ids || ids == device)
1221 processed = sscanf(ids, ":%u%n:%d%n",
1222 &virtio_device_id, &consumed,
1223 &vu_cmdline_id, &consumed);
1225 if (processed < 1 || ids[consumed])
1228 if (!vu_cmdline_parent_registered) {
1229 err = device_register(&vu_cmdline_parent);
1231 pr_err("Failed to register parent device!\n");
1232 put_device(&vu_cmdline_parent);
1235 vu_cmdline_parent_registered = true;
1238 socket_path = kmemdup_nul(device, ids - device, GFP_KERNEL);
1242 pdata.virtio_device_id = (u32) virtio_device_id;
1243 pdata.socket_path = socket_path;
1245 pr_info("Registering device virtio-uml.%d id=%d at %s\n",
1246 vu_cmdline_id, virtio_device_id, socket_path);
1248 pdev = platform_device_register_data(&vu_cmdline_parent, "virtio-uml",
1249 vu_cmdline_id++, &pdata,
1251 err = PTR_ERR_OR_ZERO(pdev);
1255 ppdata = pdev->dev.platform_data;
1256 ppdata->pdev = pdev;
1257 INIT_WORK(&ppdata->conn_broken_wk, vu_conn_broken);
1266 static int vu_cmdline_get_device(struct device *dev, void *data)
1268 struct platform_device *pdev = to_platform_device(dev);
1269 struct virtio_uml_platform_data *pdata = pdev->dev.platform_data;
1270 char *buffer = data;
1271 unsigned int len = strlen(buffer);
1273 snprintf(buffer + len, PAGE_SIZE - len, "%s:%d:%d\n",
1274 pdata->socket_path, pdata->virtio_device_id, pdev->id);
1278 static int vu_cmdline_get(char *buffer, const struct kernel_param *kp)
1281 if (vu_cmdline_parent_registered)
1282 device_for_each_child(&vu_cmdline_parent, buffer,
1283 vu_cmdline_get_device);
1284 return strlen(buffer) + 1;
1287 static const struct kernel_param_ops vu_cmdline_param_ops = {
1288 .set = vu_cmdline_set,
1289 .get = vu_cmdline_get,
1292 device_param_cb(device, &vu_cmdline_param_ops, NULL, S_IRUSR);
1293 __uml_help(vu_cmdline_param_ops,
1294 "virtio_uml.device=<socket>:<virtio_id>[:<platform_id>]\n"
1295 " Configure a virtio device over a vhost-user socket.\n"
1296 " See virtio_ids.h for a list of possible virtio device id values.\n"
1297 " Optionally use a specific platform_device id.\n\n"
1301 static void vu_unregister_cmdline_devices(void)
1303 if (vu_cmdline_parent_registered) {
1304 device_for_each_child(&vu_cmdline_parent, NULL,
1305 vu_unregister_cmdline_device);
1306 device_unregister(&vu_cmdline_parent);
1307 vu_cmdline_parent_registered = false;
1311 /* Platform driver */
1313 static const struct of_device_id virtio_uml_match[] = {
1314 { .compatible = "virtio,uml", },
1317 MODULE_DEVICE_TABLE(of, virtio_uml_match);
1319 static int virtio_uml_suspend(struct platform_device *pdev, pm_message_t state)
1321 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1323 if (!vu_dev->no_vq_suspend) {
1324 struct virtqueue *vq;
1326 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1327 struct virtio_uml_vq_info *info = vq->priv;
1329 info->suspended = true;
1330 vhost_user_set_vring_enable(vu_dev, vq->index, false);
1334 if (!device_may_wakeup(&vu_dev->vdev.dev)) {
1335 vu_dev->suspended = true;
1339 return irq_set_irq_wake(vu_dev->irq, 1);
1342 static int virtio_uml_resume(struct platform_device *pdev)
1344 struct virtio_uml_device *vu_dev = platform_get_drvdata(pdev);
1346 if (!vu_dev->no_vq_suspend) {
1347 struct virtqueue *vq;
1349 virtio_device_for_each_vq((&vu_dev->vdev), vq) {
1350 struct virtio_uml_vq_info *info = vq->priv;
1352 info->suspended = false;
1353 vhost_user_set_vring_enable(vu_dev, vq->index, true);
1357 vu_dev->suspended = false;
1359 if (!device_may_wakeup(&vu_dev->vdev.dev))
1362 return irq_set_irq_wake(vu_dev->irq, 0);
1365 static struct platform_driver virtio_uml_driver = {
1366 .probe = virtio_uml_probe,
1367 .remove = virtio_uml_remove,
1369 .name = "virtio-uml",
1370 .of_match_table = virtio_uml_match,
1372 .suspend = virtio_uml_suspend,
1373 .resume = virtio_uml_resume,
1376 static int __init virtio_uml_init(void)
1378 return platform_driver_register(&virtio_uml_driver);
1381 static void __exit virtio_uml_exit(void)
1383 platform_driver_unregister(&virtio_uml_driver);
1384 vu_unregister_cmdline_devices();
1387 module_init(virtio_uml_init);
1388 module_exit(virtio_uml_exit);
1389 __uml_exitcall(virtio_uml_exit);
1391 MODULE_DESCRIPTION("UML driver for vhost-user virtio devices");
1392 MODULE_LICENSE("GPL");