| Message ID | 20200826174636.23873-5-guennadi.liakhovetski@linux.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | Add a vhost RPMsg API | expand |
Good afternoon, On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > Linux supports running the RPMsg protocol over the VirtIO transport > protocol, but currently there is only support for VirtIO clients and > no support for a VirtIO server. This patch adds a vhost-based RPMsg > server implementation. This changelog is very confusing... At this time the name service in the remoteproc space runs as a server on the application processor. But from the above the remoteproc usecase seems to be considered to be a client configuration. And I don't see a server implementation per se... It is more like a client implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends messages from host to guest. Perhaps it is my lack of familiarity with vhost terminology. > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > --- > drivers/vhost/Kconfig | 7 + > drivers/vhost/Makefile | 3 + > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > 4 files changed, 457 insertions(+) > create mode 100644 drivers/vhost/rpmsg.c > create mode 100644 drivers/vhost/vhost_rpmsg.h > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > index 587fbae06182..046b948fc411 100644 > --- a/drivers/vhost/Kconfig > +++ b/drivers/vhost/Kconfig > @@ -38,6 +38,13 @@ config VHOST_NET > To compile this driver as a module, choose M here: the module will > be called vhost_net. > > +config VHOST_RPMSG > + tristate > + select VHOST > + help > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > + drivers, using the RPMsg over VirtIO protocol. I had to assume vhost drivers are running on the host and virtIO drivers on the guests. This may be common knowledge for people familiar with vhosts but certainly obscur for commoners Having a help section that is clear on what is happening would remove any ambiguity. > + > config VHOST_SCSI > tristate "VHOST_SCSI TCM fabric driver" > depends on TARGET_CORE && EVENTFD > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > index f3e1897cce85..9cf459d59f97 100644 > --- a/drivers/vhost/Makefile > +++ b/drivers/vhost/Makefile > @@ -2,6 +2,9 @@ > obj-$(CONFIG_VHOST_NET) += vhost_net.o > vhost_net-y := net.o > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > +vhost_rpmsg-y := rpmsg.o > + > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > vhost_scsi-y := scsi.o > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > new file mode 100644 > index 000000000000..c26d7a4afc6d > --- /dev/null > +++ b/drivers/vhost/rpmsg.c > @@ -0,0 +1,373 @@ > +// SPDX-License-Identifier: GPL-2.0-only > +/* > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > + * > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > + * > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c Again, very confusing. The changelog refers to a server implementation but to me this refers to a client implementation, especially if rpmsg_recv_single() and rpmsg_ns_cb() are used on the other side of the pipe. > + * These functions handle creation of 2 virtual queues, handling of endpoint > + * addresses, sending a name-space announcement to the guest as well as any > + * user messages. This API can be used by any vhost driver to handle RPMsg > + * specific processing. > + * Specific vhost drivers, using this API will use their own VirtIO device > + * IDs, that should then also be added to the ID table in virtio_rpmsg_bus.c > + */ > + > +#include <linux/compat.h> > +#include <linux/file.h> > +#include <linux/miscdevice.h> > +#include <linux/module.h> > +#include <linux/mutex.h> > +#include <linux/vhost.h> > +#include <linux/virtio_rpmsg.h> > +#include <uapi/linux/rpmsg.h> > + > +#include "vhost.h" > +#include "vhost_rpmsg.h" > + > +/* > + * All virtio-rpmsg virtual queue kicks always come with just one buffer - > + * either input or output, but we can also handle split messages > + */ > +static int vhost_rpmsg_get_msg(struct vhost_virtqueue *vq, unsigned int *cnt) > +{ > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > + unsigned int out, in; > + int head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), &out, &in, > + NULL, NULL); > + if (head < 0) { > + vq_err(vq, "%s(): error %d getting buffer\n", > + __func__, head); > + return head; > + } > + > + /* Nothing new? */ > + if (head == vq->num) > + return head; > + > + if (vq == &vr->vq[VIRTIO_RPMSG_RESPONSE]) { > + if (out) { > + vq_err(vq, "%s(): invalid %d output in response queue\n", > + __func__, out); > + goto return_buf; > + } > + > + *cnt = in; > + } > + > + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST]) { > + if (in) { > + vq_err(vq, "%s(): invalid %d input in request queue\n", > + __func__, in); > + goto return_buf; > + } > + > + *cnt = out; > + } > + > + return head; > + > +return_buf: > + vhost_add_used(vq, head, 0); > + > + return -EINVAL; > +} > + > +static const struct vhost_rpmsg_ept *vhost_rpmsg_ept_find(struct vhost_rpmsg *vr, int addr) > +{ > + unsigned int i; > + > + for (i = 0; i < vr->n_epts; i++) > + if (vr->ept[i].addr == addr) > + return vr->ept + i; > + > + return NULL; > +} > + > +/* > + * if len < 0, then for reading a request, the complete virtual queue buffer > + * size is prepared, for sending a response, the length in the iterator is used > + */ > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > + unsigned int qid, ssize_t len) > + __acquires(vq->mutex) > +{ > + struct vhost_virtqueue *vq = vr->vq + qid; > + unsigned int cnt; > + ssize_t ret; > + size_t tmp; > + > + if (qid >= VIRTIO_RPMSG_NUM_OF_VQS) > + return -EINVAL; > + > + iter->vq = vq; > + > + mutex_lock(&vq->mutex); > + vhost_disable_notify(&vr->dev, vq); > + > + iter->head = vhost_rpmsg_get_msg(vq, &cnt); > + if (iter->head == vq->num) > + iter->head = -EAGAIN; > + > + if (iter->head < 0) { > + ret = iter->head; > + goto unlock; > + } > + > + tmp = iov_length(vq->iov, cnt); > + if (tmp < sizeof(iter->rhdr)) { > + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); > + ret = -ENOBUFS; > + goto return_buf; > + } > + > + switch (qid) { > + case VIRTIO_RPMSG_REQUEST: > + if (len >= 0) { > + if (tmp < sizeof(iter->rhdr) + len) { > + ret = -ENOBUFS; > + goto return_buf; > + } > + > + tmp = len + sizeof(iter->rhdr); > + } > + > + /* len is now the size of the payload */ > + iov_iter_init(&iter->iov_iter, WRITE, vq->iov, cnt, tmp); > + > + /* Read the RPMSG header with endpoint addresses */ > + tmp = copy_from_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > + if (tmp != sizeof(iter->rhdr)) { > + vq_err(vq, "%s(): got %zu instead of %zu\n", __func__, > + tmp, sizeof(iter->rhdr)); > + ret = -EIO; > + goto return_buf; > + } > + > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.dst)); > + if (!iter->ept) { > + vq_err(vq, "%s(): no endpoint with address %d\n", > + __func__, vhost32_to_cpu(vq, iter->rhdr.dst)); > + ret = -ENOENT; > + goto return_buf; > + } > + > + /* Let the endpoint read the payload */ > + if (iter->ept->read) { > + ret = iter->ept->read(vr, iter); > + if (ret < 0) > + goto return_buf; > + > + iter->rhdr.len = cpu_to_vhost16(vq, ret); > + } else { > + iter->rhdr.len = 0; > + } > + > + /* Prepare for the response phase */ > + iter->rhdr.dst = iter->rhdr.src; > + iter->rhdr.src = cpu_to_vhost32(vq, iter->ept->addr); > + > + break; > + case VIRTIO_RPMSG_RESPONSE: > + if (!iter->ept && iter->rhdr.dst != cpu_to_vhost32(vq, RPMSG_NS_ADDR)) { > + /* > + * Usually the iterator is configured when processing a > + * message on the request queue, but it's also possible > + * to send a message on the response queue without a > + * preceding request, in that case the iterator must > + * contain source and destination addresses. > + */ > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.src)); > + if (!iter->ept) { > + ret = -ENOENT; > + goto return_buf; > + } > + } > + > + if (len >= 0) { > + if (tmp < sizeof(iter->rhdr) + len) { > + ret = -ENOBUFS; > + goto return_buf; > + } > + > + iter->rhdr.len = cpu_to_vhost16(vq, len); > + tmp = len + sizeof(iter->rhdr); > + } > + > + /* len is now the size of the payload */ > + iov_iter_init(&iter->iov_iter, READ, vq->iov, cnt, tmp); > + > + /* Write the RPMSG header with endpoint addresses */ > + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > + if (tmp != sizeof(iter->rhdr)) { > + ret = -EIO; > + goto return_buf; > + } > + > + /* Let the endpoint write the payload */ > + if (iter->ept && iter->ept->write) { > + ret = iter->ept->write(vr, iter); > + if (ret < 0) > + goto return_buf; > + } > + > + break; > + } > + > + return 0; > + > +return_buf: > + vhost_add_used(vq, iter->head, 0); > +unlock: > + vhost_enable_notify(&vr->dev, vq); > + mutex_unlock(&vq->mutex); > + > + return ret; > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_start_lock); > + > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > + void *data, size_t size) > +{ > + /* > + * We could check for excess data, but copy_{to,from}_iter() don't do > + * that either > + */ > + if (iter->vq == vr->vq + VIRTIO_RPMSG_RESPONSE) > + return copy_to_iter(data, size, &iter->iov_iter); > + > + return copy_from_iter(data, size, &iter->iov_iter); > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_copy); > + > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > + struct vhost_rpmsg_iter *iter) > + __releases(vq->mutex) > +{ > + if (iter->head >= 0) > + vhost_add_used_and_signal(iter->vq->dev, iter->vq, iter->head, > + vhost16_to_cpu(iter->vq, iter->rhdr.len) + > + sizeof(iter->rhdr)); > + > + vhost_enable_notify(&vr->dev, iter->vq); > + mutex_unlock(&iter->vq->mutex); > + > + return iter->head; > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_finish_unlock); > + > +/* > + * Return false to terminate the external loop only if we fail to obtain either > + * a request or a response buffer > + */ > +static bool handle_rpmsg_req_single(struct vhost_rpmsg *vr, > + struct vhost_virtqueue *vq) > +{ > + struct vhost_rpmsg_iter iter; > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_REQUEST, -EINVAL); > + if (!ret) > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > + if (ret < 0) { > + if (ret != -EAGAIN) > + vq_err(vq, "%s(): RPMSG processing failed %d\n", > + __func__, ret); > + return false; > + } > + > + if (!iter.ept->write) > + return true; > + > + ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, -EINVAL); > + if (!ret) > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > + if (ret < 0) { > + vq_err(vq, "%s(): RPMSG finalising failed %d\n", __func__, ret); > + return false; > + } > + > + return true; > +} > + > +static void handle_rpmsg_req_kick(struct vhost_work *work) > +{ > + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, > + poll.work); > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > + > + while (handle_rpmsg_req_single(vr, vq)) > + ; > +} > + > +/* > + * initialise two virtqueues with an array of endpoints, > + * request and response callbacks > + */ > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > + unsigned int n_epts) > +{ > + unsigned int i; > + > + for (i = 0; i < ARRAY_SIZE(vr->vq); i++) > + vr->vq_p[i] = &vr->vq[i]; > + > + /* vq[0]: host -> guest, vq[1]: host <- guest */ > + vr->vq[VIRTIO_RPMSG_REQUEST].handle_kick = handle_rpmsg_req_kick; > + vr->vq[VIRTIO_RPMSG_RESPONSE].handle_kick = NULL; > + > + vr->ept = ept; > + vr->n_epts = n_epts; > + > + vhost_dev_init(&vr->dev, vr->vq_p, VIRTIO_RPMSG_NUM_OF_VQS, > + UIO_MAXIOV, 0, 0, true, NULL); > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_init); > + > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr) > +{ > + if (vhost_dev_has_owner(&vr->dev)) > + vhost_poll_flush(&vr->vq[VIRTIO_RPMSG_REQUEST].poll); > + > + vhost_dev_cleanup(&vr->dev); > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_destroy); > + > +/* send namespace */ > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, unsigned int src) > +{ > + struct vhost_virtqueue *vq = &vr->vq[VIRTIO_RPMSG_RESPONSE]; > + struct vhost_rpmsg_iter iter = { > + .rhdr = { > + .src = 0, > + .dst = cpu_to_vhost32(vq, RPMSG_NS_ADDR), > + .flags = cpu_to_vhost16(vq, RPMSG_NS_CREATE), /* rpmsg_recv_single() */ Where is the flag used in rpmsg_recv_single()? It is used for the name space message (as you have below) but not in the header when doing a name space announcement. > + }, > + }; > + struct rpmsg_ns_msg ns = { > + .addr = cpu_to_vhost32(vq, src), > + .flags = cpu_to_vhost32(vq, RPMSG_NS_CREATE), /* for rpmsg_ns_cb() */ > + }; > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, sizeof(ns)); > + > + if (ret < 0) > + return ret; > + > + strlcpy(ns.name, name, sizeof(ns.name)); > + > + ret = vhost_rpmsg_copy(vr, &iter, &ns, sizeof(ns)); > + if (ret != sizeof(ns)) > + vq_err(iter.vq, "%s(): added %d instead of %zu bytes\n", > + __func__, ret, sizeof(ns)); > + > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > + if (ret < 0) > + vq_err(iter.vq, "%s(): namespace announcement failed: %d\n", > + __func__, ret); > + > + return ret; > +} > +EXPORT_SYMBOL_GPL(vhost_rpmsg_ns_announce); > + > +MODULE_LICENSE("GPL v2"); > +MODULE_AUTHOR("Intel, Inc."); > +MODULE_DESCRIPTION("Vhost RPMsg API"); > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > new file mode 100644 > index 000000000000..30072cecb8a0 > --- /dev/null > +++ b/drivers/vhost/vhost_rpmsg.h > @@ -0,0 +1,74 @@ > +/* SPDX-License-Identifier: GPL-2.0 */ > +/* > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > + * > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > + */ > + > +#ifndef VHOST_RPMSG_H > +#define VHOST_RPMSG_H > + > +#include <linux/uio.h> > +#include <linux/virtio_rpmsg.h> > + > +#include "vhost.h" > + > +/* RPMsg uses two VirtQueues: one for each direction */ > +enum { > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > + /* Keep last */ > + VIRTIO_RPMSG_NUM_OF_VQS, > +}; > + > +struct vhost_rpmsg_ept; > + > +struct vhost_rpmsg_iter { > + struct iov_iter iov_iter; > + struct rpmsg_hdr rhdr; > + struct vhost_virtqueue *vq; > + const struct vhost_rpmsg_ept *ept; > + int head; > + void *priv; I don't see @priv being used anywhere. > +}; > + > +struct vhost_rpmsg { > + struct vhost_dev dev; > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > + const struct vhost_rpmsg_ept *ept; > + unsigned int n_epts; > +}; > + > +struct vhost_rpmsg_ept { > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > + int addr; > +}; > + > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > +{ > + return iter->rhdr.len; > +} Again, I don't see where this is used. > + > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > + .rhdr = { \ > + .src = cpu_to_vhost32(_vq, _src), \ > + .dst = cpu_to_vhost32(_vq, _dst), \ > + }, \ > + } Same. Thanks, Mathieu > + > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > + unsigned int n_epts); > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > + unsigned int src); > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > + struct vhost_rpmsg_iter *iter, > + unsigned int qid, ssize_t len); > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > + void *data, size_t size); > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > + struct vhost_rpmsg_iter *iter); > + > +#endif > -- > 2.28.0 >
Hi Mathieu, On Wed, Sep 09, 2020 at 04:39:46PM -0600, Mathieu Poirier wrote: > Good afternoon, > > On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > > Linux supports running the RPMsg protocol over the VirtIO transport > > protocol, but currently there is only support for VirtIO clients and > > no support for a VirtIO server. This patch adds a vhost-based RPMsg > > server implementation. > > This changelog is very confusing... At this time the name service in the > remoteproc space runs as a server on the application processor. But from the > above the remoteproc usecase seems to be considered to be a client > configuration. I agree that this isn't very obvious. But I think it is common to call the host "a server" and guests "clients." E.g. in vhost.c in the top-of-thefile comment: * Generic code for virtio server in host kernel. I think the generic concept behind this notation is, that as guests boot, they send their requests to the host, e.g. VirtIO device drivers on guests send requests over VirtQueues to VirtIO servers on the host, which can run either in the user- or in the kernel-space. And I think you can follow that logic in case of devices or remote processors too: it's the main CPU(s) that boot(s) and start talking to devices and remote processors, so in that sence devices are servers and the CPUs are their clients. And yes, the name-space announcement use-case seems confusing to me too - it reverts the relationship in a way: once a guest has booted and established connections to any rpmsg "devices," those send their namespace announcements back. But I think this can be regarded as server identification: you connect to a server and it replies with its identification and capabilities. > And I don't see a server implementation per se... It is more like a client > implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends > messages from host to guest. > > Perhaps it is my lack of familiarity with vhost terminology. > > > > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > --- > > drivers/vhost/Kconfig | 7 + > > drivers/vhost/Makefile | 3 + > > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > > 4 files changed, 457 insertions(+) > > create mode 100644 drivers/vhost/rpmsg.c > > create mode 100644 drivers/vhost/vhost_rpmsg.h > > > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > > index 587fbae06182..046b948fc411 100644 > > --- a/drivers/vhost/Kconfig > > +++ b/drivers/vhost/Kconfig > > @@ -38,6 +38,13 @@ config VHOST_NET > > To compile this driver as a module, choose M here: the module will > > be called vhost_net. > > > > +config VHOST_RPMSG > > + tristate > > + select VHOST > > + help > > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > > + drivers, using the RPMsg over VirtIO protocol. > > I had to assume vhost drivers are running on the host and virtIO drivers on the > guests. This may be common knowledge for people familiar with vhosts but > certainly obscur for commoners Having a help section that is clear on what is > happening would remove any ambiguity. It is the terminology, yes, but you're right, the wording isn't very clear, will improve. > > + > > config VHOST_SCSI > > tristate "VHOST_SCSI TCM fabric driver" > > depends on TARGET_CORE && EVENTFD > > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > > index f3e1897cce85..9cf459d59f97 100644 > > --- a/drivers/vhost/Makefile > > +++ b/drivers/vhost/Makefile > > @@ -2,6 +2,9 @@ > > obj-$(CONFIG_VHOST_NET) += vhost_net.o > > vhost_net-y := net.o > > > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > > +vhost_rpmsg-y := rpmsg.o > > + > > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > > vhost_scsi-y := scsi.o > > > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > > new file mode 100644 > > index 000000000000..c26d7a4afc6d > > --- /dev/null > > +++ b/drivers/vhost/rpmsg.c > > @@ -0,0 +1,373 @@ > > +// SPDX-License-Identifier: GPL-2.0-only > > +/* > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > + * > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > + * > > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c > > Again, very confusing. The changelog refers to a server implementation but to > me this refers to a client implementation, especially if rpmsg_recv_single() and > rpmsg_ns_cb() are used on the other side of the pipe. I think the above is correct. "Vhost" indicates, that this is running on the host. "match the guest side" means, that you can use this API on the host and it is designed to work together with the RPMsg VirtIO drivers running on guests, as implemented *on guests* by virtio_rpmsg_bus.c. Would "to work together" be a better description than "to match?" > > + * These functions handle creation of 2 virtual queues, handling of endpoint > > + * addresses, sending a name-space announcement to the guest as well as any > > + * user messages. This API can be used by any vhost driver to handle RPMsg > > + * specific processing. > > + * Specific vhost drivers, using this API will use their own VirtIO device > > + * IDs, that should then also be added to the ID table in virtio_rpmsg_bus.c > > + */ > > + > > +#include <linux/compat.h> > > +#include <linux/file.h> > > +#include <linux/miscdevice.h> > > +#include <linux/module.h> > > +#include <linux/mutex.h> > > +#include <linux/vhost.h> > > +#include <linux/virtio_rpmsg.h> > > +#include <uapi/linux/rpmsg.h> > > + > > +#include "vhost.h" > > +#include "vhost_rpmsg.h" > > + > > +/* > > + * All virtio-rpmsg virtual queue kicks always come with just one buffer - > > + * either input or output, but we can also handle split messages > > + */ > > +static int vhost_rpmsg_get_msg(struct vhost_virtqueue *vq, unsigned int *cnt) > > +{ > > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > > + unsigned int out, in; > > + int head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), &out, &in, > > + NULL, NULL); > > + if (head < 0) { > > + vq_err(vq, "%s(): error %d getting buffer\n", > > + __func__, head); > > + return head; > > + } > > + > > + /* Nothing new? */ > > + if (head == vq->num) > > + return head; > > + > > + if (vq == &vr->vq[VIRTIO_RPMSG_RESPONSE]) { > > + if (out) { > > + vq_err(vq, "%s(): invalid %d output in response queue\n", > > + __func__, out); > > + goto return_buf; > > + } > > + > > + *cnt = in; > > + } > > + > > + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST]) { > > + if (in) { > > + vq_err(vq, "%s(): invalid %d input in request queue\n", > > + __func__, in); > > + goto return_buf; > > + } > > + > > + *cnt = out; > > + } > > + > > + return head; > > + > > +return_buf: > > + vhost_add_used(vq, head, 0); > > + > > + return -EINVAL; > > +} > > + > > +static const struct vhost_rpmsg_ept *vhost_rpmsg_ept_find(struct vhost_rpmsg *vr, int addr) > > +{ > > + unsigned int i; > > + > > + for (i = 0; i < vr->n_epts; i++) > > + if (vr->ept[i].addr == addr) > > + return vr->ept + i; > > + > > + return NULL; > > +} > > + > > +/* > > + * if len < 0, then for reading a request, the complete virtual queue buffer > > + * size is prepared, for sending a response, the length in the iterator is used > > + */ > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > + unsigned int qid, ssize_t len) > > + __acquires(vq->mutex) > > +{ > > + struct vhost_virtqueue *vq = vr->vq + qid; > > + unsigned int cnt; > > + ssize_t ret; > > + size_t tmp; > > + > > + if (qid >= VIRTIO_RPMSG_NUM_OF_VQS) > > + return -EINVAL; > > + > > + iter->vq = vq; > > + > > + mutex_lock(&vq->mutex); > > + vhost_disable_notify(&vr->dev, vq); > > + > > + iter->head = vhost_rpmsg_get_msg(vq, &cnt); > > + if (iter->head == vq->num) > > + iter->head = -EAGAIN; > > + > > + if (iter->head < 0) { > > + ret = iter->head; > > + goto unlock; > > + } > > + > > + tmp = iov_length(vq->iov, cnt); > > + if (tmp < sizeof(iter->rhdr)) { > > + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); > > + ret = -ENOBUFS; > > + goto return_buf; > > + } > > + > > + switch (qid) { > > + case VIRTIO_RPMSG_REQUEST: > > + if (len >= 0) { > > + if (tmp < sizeof(iter->rhdr) + len) { > > + ret = -ENOBUFS; > > + goto return_buf; > > + } > > + > > + tmp = len + sizeof(iter->rhdr); > > + } > > + > > + /* len is now the size of the payload */ > > + iov_iter_init(&iter->iov_iter, WRITE, vq->iov, cnt, tmp); > > + > > + /* Read the RPMSG header with endpoint addresses */ > > + tmp = copy_from_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > > + if (tmp != sizeof(iter->rhdr)) { > > + vq_err(vq, "%s(): got %zu instead of %zu\n", __func__, > > + tmp, sizeof(iter->rhdr)); > > + ret = -EIO; > > + goto return_buf; > > + } > > + > > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.dst)); > > + if (!iter->ept) { > > + vq_err(vq, "%s(): no endpoint with address %d\n", > > + __func__, vhost32_to_cpu(vq, iter->rhdr.dst)); > > + ret = -ENOENT; > > + goto return_buf; > > + } > > + > > + /* Let the endpoint read the payload */ > > + if (iter->ept->read) { > > + ret = iter->ept->read(vr, iter); > > + if (ret < 0) > > + goto return_buf; > > + > > + iter->rhdr.len = cpu_to_vhost16(vq, ret); > > + } else { > > + iter->rhdr.len = 0; > > + } > > + > > + /* Prepare for the response phase */ > > + iter->rhdr.dst = iter->rhdr.src; > > + iter->rhdr.src = cpu_to_vhost32(vq, iter->ept->addr); > > + > > + break; > > + case VIRTIO_RPMSG_RESPONSE: > > + if (!iter->ept && iter->rhdr.dst != cpu_to_vhost32(vq, RPMSG_NS_ADDR)) { > > + /* > > + * Usually the iterator is configured when processing a > > + * message on the request queue, but it's also possible > > + * to send a message on the response queue without a > > + * preceding request, in that case the iterator must > > + * contain source and destination addresses. > > + */ > > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.src)); > > + if (!iter->ept) { > > + ret = -ENOENT; > > + goto return_buf; > > + } > > + } > > + > > + if (len >= 0) { > > + if (tmp < sizeof(iter->rhdr) + len) { > > + ret = -ENOBUFS; > > + goto return_buf; > > + } > > + > > + iter->rhdr.len = cpu_to_vhost16(vq, len); > > + tmp = len + sizeof(iter->rhdr); > > + } > > + > > + /* len is now the size of the payload */ > > + iov_iter_init(&iter->iov_iter, READ, vq->iov, cnt, tmp); > > + > > + /* Write the RPMSG header with endpoint addresses */ > > + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > > + if (tmp != sizeof(iter->rhdr)) { > > + ret = -EIO; > > + goto return_buf; > > + } > > + > > + /* Let the endpoint write the payload */ > > + if (iter->ept && iter->ept->write) { > > + ret = iter->ept->write(vr, iter); > > + if (ret < 0) > > + goto return_buf; > > + } > > + > > + break; > > + } > > + > > + return 0; > > + > > +return_buf: > > + vhost_add_used(vq, iter->head, 0); > > +unlock: > > + vhost_enable_notify(&vr->dev, vq); > > + mutex_unlock(&vq->mutex); > > + > > + return ret; > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_start_lock); > > + > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > + void *data, size_t size) > > +{ > > + /* > > + * We could check for excess data, but copy_{to,from}_iter() don't do > > + * that either > > + */ > > + if (iter->vq == vr->vq + VIRTIO_RPMSG_RESPONSE) > > + return copy_to_iter(data, size, &iter->iov_iter); > > + > > + return copy_from_iter(data, size, &iter->iov_iter); > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_copy); > > + > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > + struct vhost_rpmsg_iter *iter) > > + __releases(vq->mutex) > > +{ > > + if (iter->head >= 0) > > + vhost_add_used_and_signal(iter->vq->dev, iter->vq, iter->head, > > + vhost16_to_cpu(iter->vq, iter->rhdr.len) + > > + sizeof(iter->rhdr)); > > + > > + vhost_enable_notify(&vr->dev, iter->vq); > > + mutex_unlock(&iter->vq->mutex); > > + > > + return iter->head; > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_finish_unlock); > > + > > +/* > > + * Return false to terminate the external loop only if we fail to obtain either > > + * a request or a response buffer > > + */ > > +static bool handle_rpmsg_req_single(struct vhost_rpmsg *vr, > > + struct vhost_virtqueue *vq) > > +{ > > + struct vhost_rpmsg_iter iter; > > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_REQUEST, -EINVAL); > > + if (!ret) > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > + if (ret < 0) { > > + if (ret != -EAGAIN) > > + vq_err(vq, "%s(): RPMSG processing failed %d\n", > > + __func__, ret); > > + return false; > > + } > > + > > + if (!iter.ept->write) > > + return true; > > + > > + ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, -EINVAL); > > + if (!ret) > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > + if (ret < 0) { > > + vq_err(vq, "%s(): RPMSG finalising failed %d\n", __func__, ret); > > + return false; > > + } > > + > > + return true; > > +} > > + > > +static void handle_rpmsg_req_kick(struct vhost_work *work) > > +{ > > + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, > > + poll.work); > > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > > + > > + while (handle_rpmsg_req_single(vr, vq)) > > + ; > > +} > > + > > +/* > > + * initialise two virtqueues with an array of endpoints, > > + * request and response callbacks > > + */ > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > + unsigned int n_epts) > > +{ > > + unsigned int i; > > + > > + for (i = 0; i < ARRAY_SIZE(vr->vq); i++) > > + vr->vq_p[i] = &vr->vq[i]; > > + > > + /* vq[0]: host -> guest, vq[1]: host <- guest */ > > + vr->vq[VIRTIO_RPMSG_REQUEST].handle_kick = handle_rpmsg_req_kick; > > + vr->vq[VIRTIO_RPMSG_RESPONSE].handle_kick = NULL; > > + > > + vr->ept = ept; > > + vr->n_epts = n_epts; > > + > > + vhost_dev_init(&vr->dev, vr->vq_p, VIRTIO_RPMSG_NUM_OF_VQS, > > + UIO_MAXIOV, 0, 0, true, NULL); > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_init); > > + > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr) > > +{ > > + if (vhost_dev_has_owner(&vr->dev)) > > + vhost_poll_flush(&vr->vq[VIRTIO_RPMSG_REQUEST].poll); > > + > > + vhost_dev_cleanup(&vr->dev); > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_destroy); > > + > > +/* send namespace */ > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, unsigned int src) > > +{ > > + struct vhost_virtqueue *vq = &vr->vq[VIRTIO_RPMSG_RESPONSE]; > > + struct vhost_rpmsg_iter iter = { > > + .rhdr = { > > + .src = 0, > > + .dst = cpu_to_vhost32(vq, RPMSG_NS_ADDR), > > + .flags = cpu_to_vhost16(vq, RPMSG_NS_CREATE), /* rpmsg_recv_single() */ > > Where is the flag used in rpmsg_recv_single()? It is used for the name space > message (as you have below) but not in the header when doing a name space > announcement. I think you're right, it isn't needed here, will remove. > > + }, > > + }; > > + struct rpmsg_ns_msg ns = { > > + .addr = cpu_to_vhost32(vq, src), > > + .flags = cpu_to_vhost32(vq, RPMSG_NS_CREATE), /* for rpmsg_ns_cb() */ > > + }; > > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, sizeof(ns)); > > + > > + if (ret < 0) > > + return ret; > > + > > + strlcpy(ns.name, name, sizeof(ns.name)); > > + > > + ret = vhost_rpmsg_copy(vr, &iter, &ns, sizeof(ns)); > > + if (ret != sizeof(ns)) > > + vq_err(iter.vq, "%s(): added %d instead of %zu bytes\n", > > + __func__, ret, sizeof(ns)); > > + > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > + if (ret < 0) > > + vq_err(iter.vq, "%s(): namespace announcement failed: %d\n", > > + __func__, ret); > > + > > + return ret; > > +} > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_ns_announce); > > + > > +MODULE_LICENSE("GPL v2"); > > +MODULE_AUTHOR("Intel, Inc."); > > +MODULE_DESCRIPTION("Vhost RPMsg API"); > > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > > new file mode 100644 > > index 000000000000..30072cecb8a0 > > --- /dev/null > > +++ b/drivers/vhost/vhost_rpmsg.h > > @@ -0,0 +1,74 @@ > > +/* SPDX-License-Identifier: GPL-2.0 */ > > +/* > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > + * > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > + */ > > + > > +#ifndef VHOST_RPMSG_H > > +#define VHOST_RPMSG_H > > + > > +#include <linux/uio.h> > > +#include <linux/virtio_rpmsg.h> > > + > > +#include "vhost.h" > > + > > +/* RPMsg uses two VirtQueues: one for each direction */ > > +enum { > > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > > + /* Keep last */ > > + VIRTIO_RPMSG_NUM_OF_VQS, > > +}; > > + > > +struct vhost_rpmsg_ept; > > + > > +struct vhost_rpmsg_iter { > > + struct iov_iter iov_iter; > > + struct rpmsg_hdr rhdr; > > + struct vhost_virtqueue *vq; > > + const struct vhost_rpmsg_ept *ept; > > + int head; > > + void *priv; > > I don't see @priv being used anywhere. That's logical: this is a field, private to the API users, so the core shouldn't use it :-) It's used in later patches. > > > +}; > > + > > +struct vhost_rpmsg { > > + struct vhost_dev dev; > > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > > + const struct vhost_rpmsg_ept *ept; > > + unsigned int n_epts; > > +}; > > + > > +struct vhost_rpmsg_ept { > > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > + int addr; > > +}; > > + > > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > > +{ > > + return iter->rhdr.len; > > +} > > Again, I don't see where this is used. This is exported API, it's used by users. > > + > > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > > + .rhdr = { \ > > + .src = cpu_to_vhost32(_vq, _src), \ > > + .dst = cpu_to_vhost32(_vq, _dst), \ > > + }, \ > > + } > > Same. ditto. Thanks Guennadi > Thanks, > Mathieu > > > + > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > + unsigned int n_epts); > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > > + unsigned int src); > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > > + struct vhost_rpmsg_iter *iter, > > + unsigned int qid, ssize_t len); > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > + void *data, size_t size); > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > + struct vhost_rpmsg_iter *iter); > > + > > +#endif > > -- > > 2.28.0 > >
Good morning Guennadi, On Thu, Sep 10, 2020 at 10:38:54AM +0200, Guennadi Liakhovetski wrote: > Hi Mathieu, > > On Wed, Sep 09, 2020 at 04:39:46PM -0600, Mathieu Poirier wrote: > > Good afternoon, > > > > On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > > > Linux supports running the RPMsg protocol over the VirtIO transport > > > protocol, but currently there is only support for VirtIO clients and > > > no support for a VirtIO server. This patch adds a vhost-based RPMsg > > > server implementation. > > > > This changelog is very confusing... At this time the name service in the > > remoteproc space runs as a server on the application processor. But from the > > above the remoteproc usecase seems to be considered to be a client > > configuration. > > I agree that this isn't very obvious. But I think it is common to call the > host "a server" and guests "clients." E.g. in vhost.c in the top-of-thefile > comment: Ok - that part we agree on. > > * Generic code for virtio server in host kernel. > > I think the generic concept behind this notation is, that as guests boot, > they send their requests to the host, e.g. VirtIO device drivers on guests > send requests over VirtQueues to VirtIO servers on the host, which can run > either in the user- or in the kernel-space. And I think you can follow that I can see that process taking place. After all virtIO devices on guests are only stubs that need host support for access to HW. > logic in case of devices or remote processors too: it's the main CPU(s) > that boot(s) and start talking to devices and remote processors, so in that > sence devices are servers and the CPUs are their clients. In the remote processor case, the remoteproc core (application processor) sets up the name service but does not initiate the communication with a remote processor. It simply waits there for a name space request to come in from the remote processor. > > And yes, the name-space announcement use-case seems confusing to me too - it > reverts the relationship in a way: once a guest has booted and established > connections to any rpmsg "devices," those send their namespace announcements > back. But I think this can be regarded as server identification: you connect > to a server and it replies with its identification and capabilities. Based on the above can I assume vhost_rpmsg_ns_announce() is sent from the guest? I saw your V7, something I will look into. In the mean time I need to bring your attention to this set [1] from Arnaud. Please have a look as it will impact your work. https://patchwork.kernel.org/project/linux-remoteproc/list/?series=338335 > > > And I don't see a server implementation per se... It is more like a client > > implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends > > messages from host to guest. > > > > Perhaps it is my lack of familiarity with vhost terminology. > > > > > > > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > --- > > > drivers/vhost/Kconfig | 7 + > > > drivers/vhost/Makefile | 3 + > > > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > > > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > > > 4 files changed, 457 insertions(+) > > > create mode 100644 drivers/vhost/rpmsg.c > > > create mode 100644 drivers/vhost/vhost_rpmsg.h > > > > > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > > > index 587fbae06182..046b948fc411 100644 > > > --- a/drivers/vhost/Kconfig > > > +++ b/drivers/vhost/Kconfig > > > @@ -38,6 +38,13 @@ config VHOST_NET > > > To compile this driver as a module, choose M here: the module will > > > be called vhost_net. > > > > > > +config VHOST_RPMSG > > > + tristate > > > + select VHOST > > > + help > > > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > > > + drivers, using the RPMsg over VirtIO protocol. > > > > I had to assume vhost drivers are running on the host and virtIO drivers on the > > guests. This may be common knowledge for people familiar with vhosts but > > certainly obscur for commoners Having a help section that is clear on what is > > happening would remove any ambiguity. > > It is the terminology, yes, but you're right, the wording isn't very clear, will > improve. > > > > + > > > config VHOST_SCSI > > > tristate "VHOST_SCSI TCM fabric driver" > > > depends on TARGET_CORE && EVENTFD > > > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > > > index f3e1897cce85..9cf459d59f97 100644 > > > --- a/drivers/vhost/Makefile > > > +++ b/drivers/vhost/Makefile > > > @@ -2,6 +2,9 @@ > > > obj-$(CONFIG_VHOST_NET) += vhost_net.o > > > vhost_net-y := net.o > > > > > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > > > +vhost_rpmsg-y := rpmsg.o > > > + > > > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > > > vhost_scsi-y := scsi.o > > > > > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > > > new file mode 100644 > > > index 000000000000..c26d7a4afc6d > > > --- /dev/null > > > +++ b/drivers/vhost/rpmsg.c > > > @@ -0,0 +1,373 @@ > > > +// SPDX-License-Identifier: GPL-2.0-only > > > +/* > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > + * > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > + * > > > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > > > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c > > > > Again, very confusing. The changelog refers to a server implementation but to > > me this refers to a client implementation, especially if rpmsg_recv_single() and > > rpmsg_ns_cb() are used on the other side of the pipe. > > I think the above is correct. "Vhost" indicates, that this is running on the host. > "match the guest side" means, that you can use this API on the host and it is > designed to work together with the RPMsg VirtIO drivers running on guests, as > implemented *on guests* by virtio_rpmsg_bus.c. Would "to work together" be a better > description than "to match?" Lets forget about this part now and concentrate on the above conversation. Things will start to make sense at one point. > > > > + * These functions handle creation of 2 virtual queues, handling of endpoint > > > + * addresses, sending a name-space announcement to the guest as well as any > > > + * user messages. This API can be used by any vhost driver to handle RPMsg > > > + * specific processing. > > > + * Specific vhost drivers, using this API will use their own VirtIO device > > > + * IDs, that should then also be added to the ID table in virtio_rpmsg_bus.c > > > + */ > > > + > > > +#include <linux/compat.h> > > > +#include <linux/file.h> > > > +#include <linux/miscdevice.h> > > > +#include <linux/module.h> > > > +#include <linux/mutex.h> > > > +#include <linux/vhost.h> > > > +#include <linux/virtio_rpmsg.h> > > > +#include <uapi/linux/rpmsg.h> > > > + > > > +#include "vhost.h" > > > +#include "vhost_rpmsg.h" > > > + > > > +/* > > > + * All virtio-rpmsg virtual queue kicks always come with just one buffer - > > > + * either input or output, but we can also handle split messages > > > + */ > > > +static int vhost_rpmsg_get_msg(struct vhost_virtqueue *vq, unsigned int *cnt) > > > +{ > > > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > > > + unsigned int out, in; > > > + int head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), &out, &in, > > > + NULL, NULL); > > > + if (head < 0) { > > > + vq_err(vq, "%s(): error %d getting buffer\n", > > > + __func__, head); > > > + return head; > > > + } > > > + > > > + /* Nothing new? */ > > > + if (head == vq->num) > > > + return head; > > > + > > > + if (vq == &vr->vq[VIRTIO_RPMSG_RESPONSE]) { > > > + if (out) { > > > + vq_err(vq, "%s(): invalid %d output in response queue\n", > > > + __func__, out); > > > + goto return_buf; > > > + } > > > + > > > + *cnt = in; > > > + } > > > + > > > + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST]) { > > > + if (in) { > > > + vq_err(vq, "%s(): invalid %d input in request queue\n", > > > + __func__, in); > > > + goto return_buf; > > > + } > > > + > > > + *cnt = out; > > > + } > > > + > > > + return head; > > > + > > > +return_buf: > > > + vhost_add_used(vq, head, 0); > > > + > > > + return -EINVAL; > > > +} > > > + > > > +static const struct vhost_rpmsg_ept *vhost_rpmsg_ept_find(struct vhost_rpmsg *vr, int addr) > > > +{ > > > + unsigned int i; > > > + > > > + for (i = 0; i < vr->n_epts; i++) > > > + if (vr->ept[i].addr == addr) > > > + return vr->ept + i; > > > + > > > + return NULL; > > > +} > > > + > > > +/* > > > + * if len < 0, then for reading a request, the complete virtual queue buffer > > > + * size is prepared, for sending a response, the length in the iterator is used > > > + */ > > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > + unsigned int qid, ssize_t len) > > > + __acquires(vq->mutex) > > > +{ > > > + struct vhost_virtqueue *vq = vr->vq + qid; > > > + unsigned int cnt; > > > + ssize_t ret; > > > + size_t tmp; > > > + > > > + if (qid >= VIRTIO_RPMSG_NUM_OF_VQS) > > > + return -EINVAL; > > > + > > > + iter->vq = vq; > > > + > > > + mutex_lock(&vq->mutex); > > > + vhost_disable_notify(&vr->dev, vq); > > > + > > > + iter->head = vhost_rpmsg_get_msg(vq, &cnt); > > > + if (iter->head == vq->num) > > > + iter->head = -EAGAIN; > > > + > > > + if (iter->head < 0) { > > > + ret = iter->head; > > > + goto unlock; > > > + } > > > + > > > + tmp = iov_length(vq->iov, cnt); > > > + if (tmp < sizeof(iter->rhdr)) { > > > + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); > > > + ret = -ENOBUFS; > > > + goto return_buf; > > > + } > > > + > > > + switch (qid) { > > > + case VIRTIO_RPMSG_REQUEST: > > > + if (len >= 0) { > > > + if (tmp < sizeof(iter->rhdr) + len) { > > > + ret = -ENOBUFS; > > > + goto return_buf; > > > + } > > > + > > > + tmp = len + sizeof(iter->rhdr); > > > + } > > > + > > > + /* len is now the size of the payload */ > > > + iov_iter_init(&iter->iov_iter, WRITE, vq->iov, cnt, tmp); > > > + > > > + /* Read the RPMSG header with endpoint addresses */ > > > + tmp = copy_from_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > > > + if (tmp != sizeof(iter->rhdr)) { > > > + vq_err(vq, "%s(): got %zu instead of %zu\n", __func__, > > > + tmp, sizeof(iter->rhdr)); > > > + ret = -EIO; > > > + goto return_buf; > > > + } > > > + > > > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.dst)); > > > + if (!iter->ept) { > > > + vq_err(vq, "%s(): no endpoint with address %d\n", > > > + __func__, vhost32_to_cpu(vq, iter->rhdr.dst)); > > > + ret = -ENOENT; > > > + goto return_buf; > > > + } > > > + > > > + /* Let the endpoint read the payload */ > > > + if (iter->ept->read) { > > > + ret = iter->ept->read(vr, iter); > > > + if (ret < 0) > > > + goto return_buf; > > > + > > > + iter->rhdr.len = cpu_to_vhost16(vq, ret); > > > + } else { > > > + iter->rhdr.len = 0; > > > + } > > > + > > > + /* Prepare for the response phase */ > > > + iter->rhdr.dst = iter->rhdr.src; > > > + iter->rhdr.src = cpu_to_vhost32(vq, iter->ept->addr); > > > + > > > + break; > > > + case VIRTIO_RPMSG_RESPONSE: > > > + if (!iter->ept && iter->rhdr.dst != cpu_to_vhost32(vq, RPMSG_NS_ADDR)) { > > > + /* > > > + * Usually the iterator is configured when processing a > > > + * message on the request queue, but it's also possible > > > + * to send a message on the response queue without a > > > + * preceding request, in that case the iterator must > > > + * contain source and destination addresses. > > > + */ > > > + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.src)); > > > + if (!iter->ept) { > > > + ret = -ENOENT; > > > + goto return_buf; > > > + } > > > + } > > > + > > > + if (len >= 0) { > > > + if (tmp < sizeof(iter->rhdr) + len) { > > > + ret = -ENOBUFS; > > > + goto return_buf; > > > + } > > > + > > > + iter->rhdr.len = cpu_to_vhost16(vq, len); > > > + tmp = len + sizeof(iter->rhdr); > > > + } > > > + > > > + /* len is now the size of the payload */ > > > + iov_iter_init(&iter->iov_iter, READ, vq->iov, cnt, tmp); > > > + > > > + /* Write the RPMSG header with endpoint addresses */ > > > + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); > > > + if (tmp != sizeof(iter->rhdr)) { > > > + ret = -EIO; > > > + goto return_buf; > > > + } > > > + > > > + /* Let the endpoint write the payload */ > > > + if (iter->ept && iter->ept->write) { > > > + ret = iter->ept->write(vr, iter); > > > + if (ret < 0) > > > + goto return_buf; > > > + } > > > + > > > + break; > > > + } > > > + > > > + return 0; > > > + > > > +return_buf: > > > + vhost_add_used(vq, iter->head, 0); > > > +unlock: > > > + vhost_enable_notify(&vr->dev, vq); > > > + mutex_unlock(&vq->mutex); > > > + > > > + return ret; > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_start_lock); > > > + > > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > + void *data, size_t size) > > > +{ > > > + /* > > > + * We could check for excess data, but copy_{to,from}_iter() don't do > > > + * that either > > > + */ > > > + if (iter->vq == vr->vq + VIRTIO_RPMSG_RESPONSE) > > > + return copy_to_iter(data, size, &iter->iov_iter); > > > + > > > + return copy_from_iter(data, size, &iter->iov_iter); > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_copy); > > > + > > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > > + struct vhost_rpmsg_iter *iter) > > > + __releases(vq->mutex) > > > +{ > > > + if (iter->head >= 0) > > > + vhost_add_used_and_signal(iter->vq->dev, iter->vq, iter->head, > > > + vhost16_to_cpu(iter->vq, iter->rhdr.len) + > > > + sizeof(iter->rhdr)); > > > + > > > + vhost_enable_notify(&vr->dev, iter->vq); > > > + mutex_unlock(&iter->vq->mutex); > > > + > > > + return iter->head; > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_finish_unlock); > > > + > > > +/* > > > + * Return false to terminate the external loop only if we fail to obtain either > > > + * a request or a response buffer > > > + */ > > > +static bool handle_rpmsg_req_single(struct vhost_rpmsg *vr, > > > + struct vhost_virtqueue *vq) > > > +{ > > > + struct vhost_rpmsg_iter iter; > > > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_REQUEST, -EINVAL); > > > + if (!ret) > > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > > + if (ret < 0) { > > > + if (ret != -EAGAIN) > > > + vq_err(vq, "%s(): RPMSG processing failed %d\n", > > > + __func__, ret); > > > + return false; > > > + } > > > + > > > + if (!iter.ept->write) > > > + return true; > > > + > > > + ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, -EINVAL); > > > + if (!ret) > > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > > + if (ret < 0) { > > > + vq_err(vq, "%s(): RPMSG finalising failed %d\n", __func__, ret); > > > + return false; > > > + } > > > + > > > + return true; > > > +} > > > + > > > +static void handle_rpmsg_req_kick(struct vhost_work *work) > > > +{ > > > + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, > > > + poll.work); > > > + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); > > > + > > > + while (handle_rpmsg_req_single(vr, vq)) > > > + ; > > > +} > > > + > > > +/* > > > + * initialise two virtqueues with an array of endpoints, > > > + * request and response callbacks > > > + */ > > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > > + unsigned int n_epts) > > > +{ > > > + unsigned int i; > > > + > > > + for (i = 0; i < ARRAY_SIZE(vr->vq); i++) > > > + vr->vq_p[i] = &vr->vq[i]; > > > + > > > + /* vq[0]: host -> guest, vq[1]: host <- guest */ > > > + vr->vq[VIRTIO_RPMSG_REQUEST].handle_kick = handle_rpmsg_req_kick; > > > + vr->vq[VIRTIO_RPMSG_RESPONSE].handle_kick = NULL; > > > + > > > + vr->ept = ept; > > > + vr->n_epts = n_epts; > > > + > > > + vhost_dev_init(&vr->dev, vr->vq_p, VIRTIO_RPMSG_NUM_OF_VQS, > > > + UIO_MAXIOV, 0, 0, true, NULL); > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_init); > > > + > > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr) > > > +{ > > > + if (vhost_dev_has_owner(&vr->dev)) > > > + vhost_poll_flush(&vr->vq[VIRTIO_RPMSG_REQUEST].poll); > > > + > > > + vhost_dev_cleanup(&vr->dev); > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_destroy); > > > + > > > +/* send namespace */ > > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, unsigned int src) > > > +{ > > > + struct vhost_virtqueue *vq = &vr->vq[VIRTIO_RPMSG_RESPONSE]; > > > + struct vhost_rpmsg_iter iter = { > > > + .rhdr = { > > > + .src = 0, > > > + .dst = cpu_to_vhost32(vq, RPMSG_NS_ADDR), > > > + .flags = cpu_to_vhost16(vq, RPMSG_NS_CREATE), /* rpmsg_recv_single() */ > > > > Where is the flag used in rpmsg_recv_single()? It is used for the name space > > message (as you have below) but not in the header when doing a name space > > announcement. > > I think you're right, it isn't needed here, will remove. > > > > + }, > > > + }; > > > + struct rpmsg_ns_msg ns = { > > > + .addr = cpu_to_vhost32(vq, src), > > > + .flags = cpu_to_vhost32(vq, RPMSG_NS_CREATE), /* for rpmsg_ns_cb() */ > > > + }; > > > + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, sizeof(ns)); > > > + > > > + if (ret < 0) > > > + return ret; > > > + > > > + strlcpy(ns.name, name, sizeof(ns.name)); > > > + > > > + ret = vhost_rpmsg_copy(vr, &iter, &ns, sizeof(ns)); > > > + if (ret != sizeof(ns)) > > > + vq_err(iter.vq, "%s(): added %d instead of %zu bytes\n", > > > + __func__, ret, sizeof(ns)); > > > + > > > + ret = vhost_rpmsg_finish_unlock(vr, &iter); > > > + if (ret < 0) > > > + vq_err(iter.vq, "%s(): namespace announcement failed: %d\n", > > > + __func__, ret); > > > + > > > + return ret; > > > +} > > > +EXPORT_SYMBOL_GPL(vhost_rpmsg_ns_announce); > > > + > > > +MODULE_LICENSE("GPL v2"); > > > +MODULE_AUTHOR("Intel, Inc."); > > > +MODULE_DESCRIPTION("Vhost RPMsg API"); > > > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > > > new file mode 100644 > > > index 000000000000..30072cecb8a0 > > > --- /dev/null > > > +++ b/drivers/vhost/vhost_rpmsg.h > > > @@ -0,0 +1,74 @@ > > > +/* SPDX-License-Identifier: GPL-2.0 */ > > > +/* > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > + * > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > + */ > > > + > > > +#ifndef VHOST_RPMSG_H > > > +#define VHOST_RPMSG_H > > > + > > > +#include <linux/uio.h> > > > +#include <linux/virtio_rpmsg.h> > > > + > > > +#include "vhost.h" > > > + > > > +/* RPMsg uses two VirtQueues: one for each direction */ > > > +enum { > > > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > > > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > > > + /* Keep last */ > > > + VIRTIO_RPMSG_NUM_OF_VQS, > > > +}; > > > + > > > +struct vhost_rpmsg_ept; > > > + > > > +struct vhost_rpmsg_iter { > > > + struct iov_iter iov_iter; > > > + struct rpmsg_hdr rhdr; > > > + struct vhost_virtqueue *vq; > > > + const struct vhost_rpmsg_ept *ept; > > > + int head; > > > + void *priv; > > > > I don't see @priv being used anywhere. > > That's logical: this is a field, private to the API users, so the core shouldn't > use it :-) It's used in later patches. That is where structure documentation is useful. I will let Michael decide what he wants to do. Thanks for the feedback, Mathieu > > > > > > +}; > > > + > > > +struct vhost_rpmsg { > > > + struct vhost_dev dev; > > > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > > > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > > > + const struct vhost_rpmsg_ept *ept; > > > + unsigned int n_epts; > > > +}; > > > + > > > +struct vhost_rpmsg_ept { > > > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > + int addr; > > > +}; > > > + > > > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > > > +{ > > > + return iter->rhdr.len; > > > +} > > > > Again, I don't see where this is used. > > This is exported API, it's used by users. > > > > + > > > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > > > + .rhdr = { \ > > > + .src = cpu_to_vhost32(_vq, _src), \ > > > + .dst = cpu_to_vhost32(_vq, _dst), \ > > > + }, \ > > > + } > > > > Same. > > ditto. > > Thanks > Guennadi > > > Thanks, > > Mathieu > > > > > + > > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > > + unsigned int n_epts); > > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > > > + unsigned int src); > > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > > > + struct vhost_rpmsg_iter *iter, > > > + unsigned int qid, ssize_t len); > > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > + void *data, size_t size); > > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > > + struct vhost_rpmsg_iter *iter); > > > + > > > +#endif > > > -- > > > 2.28.0 > > >
On Fri, Sep 11, 2020 at 09:46:56AM +0200, Guennadi Liakhovetski wrote: > Hi Mathieu, > > On Thu, Sep 10, 2020 at 11:22:11AM -0600, Mathieu Poirier wrote: > > Good morning Guennadi, > > > > On Thu, Sep 10, 2020 at 10:38:54AM +0200, Guennadi Liakhovetski wrote: > > > Hi Mathieu, > > > > > > On Wed, Sep 09, 2020 at 04:39:46PM -0600, Mathieu Poirier wrote: > > > > Good afternoon, > > > > > > > > On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > > > > > Linux supports running the RPMsg protocol over the VirtIO transport > > > > > protocol, but currently there is only support for VirtIO clients and > > > > > no support for a VirtIO server. This patch adds a vhost-based RPMsg > > > > > server implementation. > > > > > > > > This changelog is very confusing... At this time the name service in the > > > > remoteproc space runs as a server on the application processor. But from the > > > > above the remoteproc usecase seems to be considered to be a client > > > > configuration. > > > > > > I agree that this isn't very obvious. But I think it is common to call the > > > host "a server" and guests "clients." E.g. in vhost.c in the top-of-thefile > > > comment: > > > > Ok - that part we agree on. > > > > > > > > * Generic code for virtio server in host kernel. > > > > > > I think the generic concept behind this notation is, that as guests boot, > > > they send their requests to the host, e.g. VirtIO device drivers on guests > > > send requests over VirtQueues to VirtIO servers on the host, which can run > > > either in the user- or in the kernel-space. And I think you can follow that > > > > I can see that process taking place. After all virtIO devices on guests are > > only stubs that need host support for access to HW. > > > > > logic in case of devices or remote processors too: it's the main CPU(s) > > > that boot(s) and start talking to devices and remote processors, so in that > > > sence devices are servers and the CPUs are their clients. > > > > In the remote processor case, the remoteproc core (application processor) sets up > > the name service but does not initiate the communication with a remote > > processor. It simply waits there for a name space request to come in from the > > remote processor. > > Hm, I don't see that in two examples, that I looked at: mtk and virtio. In both > cases the announcement seems to be directly coming from the application processor > maybe after some initialisation. > Can you expand on that part - perhaps point me to the (virtio) code you are referring to? > > > And yes, the name-space announcement use-case seems confusing to me too - it > > > reverts the relationship in a way: once a guest has booted and established > > > connections to any rpmsg "devices," those send their namespace announcements > > > back. But I think this can be regarded as server identification: you connect > > > to a server and it replies with its identification and capabilities. > > > > Based on the above can I assume vhost_rpmsg_ns_announce() is sent from the > > guest? > > No, it's "vhost_..." so it's running on the host. Ok, that's better and confirms the usage of the VIRTIO_RPMSG_RESPONSE queue. When reading your explanation above, I thought the term "those" referred to the guest. In light of your explanation I now understand that "those" referred to the rpmgs devices on the host. In the above paragraph you write: ... "once a guest has booted and established connections to any rpmsg "devices", those send their namespace announcements back". I'd like to unpack a few things about this sentence: 1) In this context, how is a "connection" established between a guest and a host? 2) How does the guest now about the rpmsg devices it has made a connection to? 3) Why is a namespace announcement needed at all when guests are aware of the rpmsg devices instantiated on the host, and have already connected to them? > The host (the server, an > analogue of the application processor, IIUC) sends NS announcements to guests. I think we have just found the source of the confusion - in the remoteproc world the application processor receives name announcements, it doesn't send them. > > > I saw your V7, something I will look into. In the mean time I need to bring > > your attention to this set [1] from Arnaud. Please have a look as it will > > impact your work. > > > > https://patchwork.kernel.org/project/linux-remoteproc/list/?series=338335 > > Yes, I've had a look at that series, thanks for forwarding it to me. TBH I > don't quite understand some choices there, e.g. creating a separate driver and > then having to register devices just for the namespace announcement. I don't > think creating virtual devices is taken easily in Linux. But either way I > don't think our series conflict a lot, but I do hope that I can merge my > series first, he'd just have to switch to using the header, that I'm adding. > Hardly too many changes otherwise. It is not the conflicts between the series that I wanted to highlight but the fact that name service is in the process of becoming a driver on its own, and with no dependence on the transport mechanism. > > > > > And I don't see a server implementation per se... It is more like a client > > > > implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends > > > > messages from host to guest. > > > > > > > > Perhaps it is my lack of familiarity with vhost terminology. > > > > > > > > > > > > > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > --- > > > > > drivers/vhost/Kconfig | 7 + > > > > > drivers/vhost/Makefile | 3 + > > > > > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > > > > > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > > > > > 4 files changed, 457 insertions(+) > > > > > create mode 100644 drivers/vhost/rpmsg.c > > > > > create mode 100644 drivers/vhost/vhost_rpmsg.h > > > > > > > > > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > > > > > index 587fbae06182..046b948fc411 100644 > > > > > --- a/drivers/vhost/Kconfig > > > > > +++ b/drivers/vhost/Kconfig > > > > > @@ -38,6 +38,13 @@ config VHOST_NET > > > > > To compile this driver as a module, choose M here: the module will > > > > > be called vhost_net. > > > > > > > > > > +config VHOST_RPMSG > > > > > + tristate > > > > > + select VHOST > > > > > + help > > > > > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > > > > > + drivers, using the RPMsg over VirtIO protocol. > > > > > > > > I had to assume vhost drivers are running on the host and virtIO drivers on the > > > > guests. This may be common knowledge for people familiar with vhosts but > > > > certainly obscur for commoners Having a help section that is clear on what is > > > > happening would remove any ambiguity. > > > > > > It is the terminology, yes, but you're right, the wording isn't very clear, will > > > improve. > > > > > > > > + > > > > > config VHOST_SCSI > > > > > tristate "VHOST_SCSI TCM fabric driver" > > > > > depends on TARGET_CORE && EVENTFD > > > > > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > > > > > index f3e1897cce85..9cf459d59f97 100644 > > > > > --- a/drivers/vhost/Makefile > > > > > +++ b/drivers/vhost/Makefile > > > > > @@ -2,6 +2,9 @@ > > > > > obj-$(CONFIG_VHOST_NET) += vhost_net.o > > > > > vhost_net-y := net.o > > > > > > > > > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > > > > > +vhost_rpmsg-y := rpmsg.o > > > > > + > > > > > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > > > > > vhost_scsi-y := scsi.o > > > > > > > > > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > > > > > new file mode 100644 > > > > > index 000000000000..c26d7a4afc6d > > > > > --- /dev/null > > > > > +++ b/drivers/vhost/rpmsg.c > > > > > @@ -0,0 +1,373 @@ > > > > > +// SPDX-License-Identifier: GPL-2.0-only > > > > > +/* > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > + * > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > + * > > > > > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > > > > > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c > > > > > > > > Again, very confusing. The changelog refers to a server implementation but to > > > > me this refers to a client implementation, especially if rpmsg_recv_single() and > > > > rpmsg_ns_cb() are used on the other side of the pipe. > > > > > > I think the above is correct. "Vhost" indicates, that this is running on the host. > > > "match the guest side" means, that you can use this API on the host and it is > > > designed to work together with the RPMsg VirtIO drivers running on guests, as > > > implemented *on guests* by virtio_rpmsg_bus.c. Would "to work together" be a better > > > description than "to match?" > > > > Lets forget about this part now and concentrate on the above conversation. > > Things will start to make sense at one point. > > I've improved that description a bit, it was indeed rather clumsy. Much appreciated - I'll take a look a V7 next week. > > [snip] > > > > > > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > > > > > new file mode 100644 > > > > > index 000000000000..30072cecb8a0 > > > > > --- /dev/null > > > > > +++ b/drivers/vhost/vhost_rpmsg.h > > > > > @@ -0,0 +1,74 @@ > > > > > +/* SPDX-License-Identifier: GPL-2.0 */ > > > > > +/* > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > + * > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > + */ > > > > > + > > > > > +#ifndef VHOST_RPMSG_H > > > > > +#define VHOST_RPMSG_H > > > > > + > > > > > +#include <linux/uio.h> > > > > > +#include <linux/virtio_rpmsg.h> > > > > > + > > > > > +#include "vhost.h" > > > > > + > > > > > +/* RPMsg uses two VirtQueues: one for each direction */ > > > > > +enum { > > > > > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > > > > > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > > > > > + /* Keep last */ > > > > > + VIRTIO_RPMSG_NUM_OF_VQS, > > > > > +}; > > > > > + > > > > > +struct vhost_rpmsg_ept; > > > > > + > > > > > +struct vhost_rpmsg_iter { > > > > > + struct iov_iter iov_iter; > > > > > + struct rpmsg_hdr rhdr; > > > > > + struct vhost_virtqueue *vq; > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > + int head; > > > > > + void *priv; > > > > > > > > I don't see @priv being used anywhere. > > > > > > That's logical: this is a field, private to the API users, so the core shouldn't > > > use it :-) It's used in later patches. > > > > That is where structure documentation is useful. I will let Michael decide what > > he wants to do. > > I can add some kerneldoc documentation there, no problem. > > > Thanks for the feedback, > > Thanks for your reviews! I'd very much like to close all the still open points > and merge the series ASAP. > > Thanks > Guennadi > > > Mathieu > > > > > > > > > > > > > > +}; > > > > > + > > > > > +struct vhost_rpmsg { > > > > > + struct vhost_dev dev; > > > > > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > + unsigned int n_epts; > > > > > +}; > > > > > + > > > > > +struct vhost_rpmsg_ept { > > > > > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > + int addr; > > > > > +}; > > > > > + > > > > > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > > > > > +{ > > > > > + return iter->rhdr.len; > > > > > +} > > > > > > > > Again, I don't see where this is used. > > > > > > This is exported API, it's used by users. > > > > > > > > + > > > > > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > > > > > + .rhdr = { \ > > > > > + .src = cpu_to_vhost32(_vq, _src), \ > > > > > + .dst = cpu_to_vhost32(_vq, _dst), \ > > > > > + }, \ > > > > > + } > > > > > > > > Same. > > > > > > ditto. > > > > > > Thanks > > > Guennadi > > > > > > > Thanks, > > > > Mathieu > > > > > > > > > + > > > > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > > > > + unsigned int n_epts); > > > > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > > > > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > > > > > + unsigned int src); > > > > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > > > > > + struct vhost_rpmsg_iter *iter, > > > > > + unsigned int qid, ssize_t len); > > > > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > > > + void *data, size_t size); > > > > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > > > > + struct vhost_rpmsg_iter *iter); > > > > > + > > > > > +#endif > > > > > -- > > > > > 2.28.0 > > > > >
On Fri, Sep 11, 2020 at 11:33:13AM -0600, Mathieu Poirier wrote: > On Fri, Sep 11, 2020 at 09:46:56AM +0200, Guennadi Liakhovetski wrote: > > Hi Mathieu, > > > > On Thu, Sep 10, 2020 at 11:22:11AM -0600, Mathieu Poirier wrote: > > > Good morning Guennadi, > > > > > > On Thu, Sep 10, 2020 at 10:38:54AM +0200, Guennadi Liakhovetski wrote: > > > > Hi Mathieu, > > > > > > > > On Wed, Sep 09, 2020 at 04:39:46PM -0600, Mathieu Poirier wrote: > > > > > Good afternoon, > > > > > > > > > > On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > > > > > > Linux supports running the RPMsg protocol over the VirtIO transport > > > > > > protocol, but currently there is only support for VirtIO clients and > > > > > > no support for a VirtIO server. This patch adds a vhost-based RPMsg > > > > > > server implementation. > > > > > > > > > > This changelog is very confusing... At this time the name service in the > > > > > remoteproc space runs as a server on the application processor. But from the > > > > > above the remoteproc usecase seems to be considered to be a client > > > > > configuration. > > > > > > > > I agree that this isn't very obvious. But I think it is common to call the > > > > host "a server" and guests "clients." E.g. in vhost.c in the top-of-thefile > > > > comment: > > > > > > Ok - that part we agree on. > > > > > > > > > > > * Generic code for virtio server in host kernel. > > > > > > > > I think the generic concept behind this notation is, that as guests boot, > > > > they send their requests to the host, e.g. VirtIO device drivers on guests > > > > send requests over VirtQueues to VirtIO servers on the host, which can run > > > > either in the user- or in the kernel-space. And I think you can follow that > > > > > > I can see that process taking place. After all virtIO devices on guests are > > > only stubs that need host support for access to HW. > > > > > > > logic in case of devices or remote processors too: it's the main CPU(s) > > > > that boot(s) and start talking to devices and remote processors, so in that > > > > sence devices are servers and the CPUs are their clients. > > > > > > In the remote processor case, the remoteproc core (application processor) sets up > > > the name service but does not initiate the communication with a remote > > > processor. It simply waits there for a name space request to come in from the > > > remote processor. > > > > Hm, I don't see that in two examples, that I looked at: mtk and virtio. In both > > cases the announcement seems to be directly coming from the application processor > > maybe after some initialisation. > > Can you expand on that part - perhaps point me to the (virtio) code you are > referring to? Ok, we're both right: it goes both ways. Here's my understanding of the control flow of virtio_rpmsg_bus.c: 1. The driver registers a VirtIO driver with the VIRTIO_ID_RPMSG ID. 2. When the driver is probed, if the server / the application processor supports the VIRTIO_RPMSG_F_NS feature, the driver calls __rpmsg_create_ept() to create an endpoint with rpmsg_ns_cb() as a callback. 3. When a namespace announcement arrives from the server, the callback is called, which then registers a new channel (in case of CREATE). That then created an rpmsg device. 4. If there's a matching rpmsg driver for that device, it's .probe() method is called, so it can then add its own rpmsg endpoints, to be used for its proper communication. Now there was indeed something in virtio_rpmsg_bus.c that I didn't fully understand: virtio_rpmsg_announce_create() and virtio_rpmsg_announce_destroy() functions. Now I understood, that as the client registers its custom channels, it also then can send name service announcements to the application processor, using those functions. This is also described in [1] as: <quote> Name Service sub-component (optional) This subcomponent is a minimum implementation of the name service which is present in the Linux Kernel implementation of RPMsg. It allows the communicating node both to send announcements about "named" endpoint (in other words, channel) creation or deletion and to receive these announcement taking any user-defined action in an application callback. </quote> Also in Documentation/rpmsg.txt <quote> ...the remote processor announces the existence of a remote rpmsg service by sending a name service message (which contains the name and rpmsg addr of the remote service, see struct rpmsg_ns_msg). </quote> in [2]: <quote> In the current protocol, at startup, the master sends notification to remote to let it know that it can receive name service announcement. </quote> > > > > And yes, the name-space announcement use-case seems confusing to me too - it > > > > reverts the relationship in a way: once a guest has booted and established > > > > connections to any rpmsg "devices," those send their namespace announcements > > > > back. But I think this can be regarded as server identification: you connect > > > > to a server and it replies with its identification and capabilities. > > > > > > Based on the above can I assume vhost_rpmsg_ns_announce() is sent from the > > > guest? > > > > No, it's "vhost_..." so it's running on the host. > > Ok, that's better and confirms the usage of the VIRTIO_RPMSG_RESPONSE queue. > When reading your explanation above, I thought the term "those" referred to the > guest. In light of your explanation I now understand that "those" referred to > the rpmgs devices on the host. > > In the above paragraph you write: > > ... "once a guest has booted and established connections to any rpmsg "devices", > those send their namespace announcements back". > > I'd like to unpack a few things about this sentence: > > 1) In this context, how is a "connection" established between a guest and a host? That's handled by the VirtIO / VirtQueues in the case of virtio_rpmsg_bus.c but in general, as mentioned in [2] <quote> However, master does not consider the fact that if the remote is ready to handle notification at this point in time. </quote> > 2) How does the guest now about the rpmsg devices it has made a connection to? Again, that's the same as with all other VirtIO / KVM / Qemu devices: in a common Qemu case, it's the Qemu which emulates the hardware and registers those devices. > 3) Why is a namespace announcement needed at all when guests are aware of the > rpmsg devices instantiated on the host, and have already connected to them? It is indeed optional according to the protocol, but as described above, without it the virtio_rpmsg_bus.c driver won't create rpmsg channels / devices, so, no probing will take place. > > The host (the server, an > > analogue of the application processor, IIUC) sends NS announcements to guests. > > I think we have just found the source of the confusion - in the remoteproc world > the application processor receives name announcements, it doesn't send them. Interesting, well, we know now that both directions are possible, but I still don't know whether all configurations are valid: only down, only up, none or both. Thanks Guennadi [1] https://nxpmicro.github.io/rpmsg-lite/ [2] https://github.com/OpenAMP/open-amp/wiki/RPMsg-Messaging-Protocol > > > I saw your V7, something I will look into. In the mean time I need to bring > > > your attention to this set [1] from Arnaud. Please have a look as it will > > > impact your work. > > > > > > https://patchwork.kernel.org/project/linux-remoteproc/list/?series=338335 > > > > Yes, I've had a look at that series, thanks for forwarding it to me. TBH I > > don't quite understand some choices there, e.g. creating a separate driver and > > then having to register devices just for the namespace announcement. I don't > > think creating virtual devices is taken easily in Linux. But either way I > > don't think our series conflict a lot, but I do hope that I can merge my > > series first, he'd just have to switch to using the header, that I'm adding. > > Hardly too many changes otherwise. > > It is not the conflicts between the series that I wanted to highlight but the > fact that name service is in the process of becoming a driver on its own, and > with no dependence on the transport mechanism. > > > > > > > > And I don't see a server implementation per se... It is more like a client > > > > > implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends > > > > > messages from host to guest. > > > > > > > > > > Perhaps it is my lack of familiarity with vhost terminology. > > > > > > > > > > > > > > > > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > --- > > > > > > drivers/vhost/Kconfig | 7 + > > > > > > drivers/vhost/Makefile | 3 + > > > > > > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > > > > > > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > > > > > > 4 files changed, 457 insertions(+) > > > > > > create mode 100644 drivers/vhost/rpmsg.c > > > > > > create mode 100644 drivers/vhost/vhost_rpmsg.h > > > > > > > > > > > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > > > > > > index 587fbae06182..046b948fc411 100644 > > > > > > --- a/drivers/vhost/Kconfig > > > > > > +++ b/drivers/vhost/Kconfig > > > > > > @@ -38,6 +38,13 @@ config VHOST_NET > > > > > > To compile this driver as a module, choose M here: the module will > > > > > > be called vhost_net. > > > > > > > > > > > > +config VHOST_RPMSG > > > > > > + tristate > > > > > > + select VHOST > > > > > > + help > > > > > > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > > > > > > + drivers, using the RPMsg over VirtIO protocol. > > > > > > > > > > I had to assume vhost drivers are running on the host and virtIO drivers on the > > > > > guests. This may be common knowledge for people familiar with vhosts but > > > > > certainly obscur for commoners Having a help section that is clear on what is > > > > > happening would remove any ambiguity. > > > > > > > > It is the terminology, yes, but you're right, the wording isn't very clear, will > > > > improve. > > > > > > > > > > + > > > > > > config VHOST_SCSI > > > > > > tristate "VHOST_SCSI TCM fabric driver" > > > > > > depends on TARGET_CORE && EVENTFD > > > > > > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > > > > > > index f3e1897cce85..9cf459d59f97 100644 > > > > > > --- a/drivers/vhost/Makefile > > > > > > +++ b/drivers/vhost/Makefile > > > > > > @@ -2,6 +2,9 @@ > > > > > > obj-$(CONFIG_VHOST_NET) += vhost_net.o > > > > > > vhost_net-y := net.o > > > > > > > > > > > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > > > > > > +vhost_rpmsg-y := rpmsg.o > > > > > > + > > > > > > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > > > > > > vhost_scsi-y := scsi.o > > > > > > > > > > > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > > > > > > new file mode 100644 > > > > > > index 000000000000..c26d7a4afc6d > > > > > > --- /dev/null > > > > > > +++ b/drivers/vhost/rpmsg.c > > > > > > @@ -0,0 +1,373 @@ > > > > > > +// SPDX-License-Identifier: GPL-2.0-only > > > > > > +/* > > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > > + * > > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > + * > > > > > > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > > > > > > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c > > > > > > > > > > Again, very confusing. The changelog refers to a server implementation but to > > > > > me this refers to a client implementation, especially if rpmsg_recv_single() and > > > > > rpmsg_ns_cb() are used on the other side of the pipe. > > > > > > > > I think the above is correct. "Vhost" indicates, that this is running on the host. > > > > "match the guest side" means, that you can use this API on the host and it is > > > > designed to work together with the RPMsg VirtIO drivers running on guests, as > > > > implemented *on guests* by virtio_rpmsg_bus.c. Would "to work together" be a better > > > > description than "to match?" > > > > > > Lets forget about this part now and concentrate on the above conversation. > > > Things will start to make sense at one point. > > > > I've improved that description a bit, it was indeed rather clumsy. > > Much appreciated - I'll take a look a V7 next week. > > > > > [snip] > > > > > > > > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > > > > > > new file mode 100644 > > > > > > index 000000000000..30072cecb8a0 > > > > > > --- /dev/null > > > > > > +++ b/drivers/vhost/vhost_rpmsg.h > > > > > > @@ -0,0 +1,74 @@ > > > > > > +/* SPDX-License-Identifier: GPL-2.0 */ > > > > > > +/* > > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > > + * > > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > + */ > > > > > > + > > > > > > +#ifndef VHOST_RPMSG_H > > > > > > +#define VHOST_RPMSG_H > > > > > > + > > > > > > +#include <linux/uio.h> > > > > > > +#include <linux/virtio_rpmsg.h> > > > > > > + > > > > > > +#include "vhost.h" > > > > > > + > > > > > > +/* RPMsg uses two VirtQueues: one for each direction */ > > > > > > +enum { > > > > > > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > > > > > > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > > > > > > + /* Keep last */ > > > > > > + VIRTIO_RPMSG_NUM_OF_VQS, > > > > > > +}; > > > > > > + > > > > > > +struct vhost_rpmsg_ept; > > > > > > + > > > > > > +struct vhost_rpmsg_iter { > > > > > > + struct iov_iter iov_iter; > > > > > > + struct rpmsg_hdr rhdr; > > > > > > + struct vhost_virtqueue *vq; > > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > > + int head; > > > > > > + void *priv; > > > > > > > > > > I don't see @priv being used anywhere. > > > > > > > > That's logical: this is a field, private to the API users, so the core shouldn't > > > > use it :-) It's used in later patches. > > > > > > That is where structure documentation is useful. I will let Michael decide what > > > he wants to do. > > > > I can add some kerneldoc documentation there, no problem. > > > > > Thanks for the feedback, > > > > Thanks for your reviews! I'd very much like to close all the still open points > > and merge the series ASAP. > > > > Thanks > > Guennadi > > > > > Mathieu > > > > > > > > > > > > > > > > > > +}; > > > > > > + > > > > > > +struct vhost_rpmsg { > > > > > > + struct vhost_dev dev; > > > > > > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > > + unsigned int n_epts; > > > > > > +}; > > > > > > + > > > > > > +struct vhost_rpmsg_ept { > > > > > > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > > + int addr; > > > > > > +}; > > > > > > + > > > > > > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > > > > > > +{ > > > > > > + return iter->rhdr.len; > > > > > > +} > > > > > > > > > > Again, I don't see where this is used. > > > > > > > > This is exported API, it's used by users. > > > > > > > > > > + > > > > > > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > > > > > > + .rhdr = { \ > > > > > > + .src = cpu_to_vhost32(_vq, _src), \ > > > > > > + .dst = cpu_to_vhost32(_vq, _dst), \ > > > > > > + }, \ > > > > > > + } > > > > > > > > > > Same. > > > > > > > > ditto. > > > > > > > > Thanks > > > > Guennadi > > > > > > > > > Thanks, > > > > > Mathieu > > > > > > > > > > > + > > > > > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > > > > > + unsigned int n_epts); > > > > > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > > > > > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > > > > > > + unsigned int src); > > > > > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > > > > > > + struct vhost_rpmsg_iter *iter, > > > > > > + unsigned int qid, ssize_t len); > > > > > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > > > > + void *data, size_t size); > > > > > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > > > > > + struct vhost_rpmsg_iter *iter); > > > > > > + > > > > > > +#endif > > > > > > -- > > > > > > 2.28.0 > > > > > >
On Tue, 15 Sep 2020 at 06:49, Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> wrote: > > On Fri, Sep 11, 2020 at 11:33:13AM -0600, Mathieu Poirier wrote: > > On Fri, Sep 11, 2020 at 09:46:56AM +0200, Guennadi Liakhovetski wrote: > > > Hi Mathieu, > > > > > > On Thu, Sep 10, 2020 at 11:22:11AM -0600, Mathieu Poirier wrote: > > > > Good morning Guennadi, > > > > > > > > On Thu, Sep 10, 2020 at 10:38:54AM +0200, Guennadi Liakhovetski wrote: > > > > > Hi Mathieu, > > > > > > > > > > On Wed, Sep 09, 2020 at 04:39:46PM -0600, Mathieu Poirier wrote: > > > > > > Good afternoon, > > > > > > > > > > > > On Wed, Aug 26, 2020 at 07:46:36PM +0200, Guennadi Liakhovetski wrote: > > > > > > > Linux supports running the RPMsg protocol over the VirtIO transport > > > > > > > protocol, but currently there is only support for VirtIO clients and > > > > > > > no support for a VirtIO server. This patch adds a vhost-based RPMsg > > > > > > > server implementation. > > > > > > > > > > > > This changelog is very confusing... At this time the name service in the > > > > > > remoteproc space runs as a server on the application processor. But from the > > > > > > above the remoteproc usecase seems to be considered to be a client > > > > > > configuration. > > > > > > > > > > I agree that this isn't very obvious. But I think it is common to call the > > > > > host "a server" and guests "clients." E.g. in vhost.c in the top-of-thefile > > > > > comment: > > > > > > > > Ok - that part we agree on. > > > > > > > > > > > > > > * Generic code for virtio server in host kernel. > > > > > > > > > > I think the generic concept behind this notation is, that as guests boot, > > > > > they send their requests to the host, e.g. VirtIO device drivers on guests > > > > > send requests over VirtQueues to VirtIO servers on the host, which can run > > > > > either in the user- or in the kernel-space. And I think you can follow that > > > > > > > > I can see that process taking place. After all virtIO devices on guests are > > > > only stubs that need host support for access to HW. > > > > > > > > > logic in case of devices or remote processors too: it's the main CPU(s) > > > > > that boot(s) and start talking to devices and remote processors, so in that > > > > > sence devices are servers and the CPUs are their clients. > > > > > > > > In the remote processor case, the remoteproc core (application processor) sets up > > > > the name service but does not initiate the communication with a remote > > > > processor. It simply waits there for a name space request to come in from the > > > > remote processor. > > > > > > Hm, I don't see that in two examples, that I looked at: mtk and virtio. In both > > > cases the announcement seems to be directly coming from the application processor > > > maybe after some initialisation. > > > > Can you expand on that part - perhaps point me to the (virtio) code you are > > referring to? > > Ok, we're both right: it goes both ways. > > Here's my understanding of the control flow of virtio_rpmsg_bus.c: > > 1. The driver registers a VirtIO driver with the VIRTIO_ID_RPMSG ID. s/virtIO driver/virtIO device > 2. When the driver is probed, if the server / the application processor supports the > VIRTIO_RPMSG_F_NS feature, the driver calls __rpmsg_create_ept() to create an > endpoint with rpmsg_ns_cb() as a callback. > 3. When a namespace announcement arrives from the server, the callback is called, > which then registers a new channel (in case of CREATE). That then created an > rpmsg device. > 4. If there's a matching rpmsg driver for that device, it's .probe() method is > called, so it can then add its own rpmsg endpoints, to be used for its proper > communication. The above depiction is correct. > > Now there was indeed something in virtio_rpmsg_bus.c that I didn't fully understand: > virtio_rpmsg_announce_create() and virtio_rpmsg_announce_destroy() functions. Now I > understood, that as the client registers its custom channels, it also then can > send name service announcements to the application processor, using those functions. Function virtio_rpmsg_announce_create/destroy() are not used when channels are created on the application processor after the reception of a name service announcement from a remote processor. Indeed, the address of the remote service is already conveyed in rpmsg_ns_msg::addr and need not be announced because it is already known by the remote processor. Create/destroy() functions are useful when the application processor registers a channel on its own, i.e not prodded by the remote processor (rpmsg_ns_cb()). It is important to note that at this point we do not have an rpmsg device (that is virtIO based) that uses this functionality. > This is also described in [1] as: > > <quote> > Name Service sub-component (optional) > > This subcomponent is a minimum implementation of the name service which is present > in the Linux Kernel implementation of RPMsg. It allows the communicating node both > to send announcements about "named" endpoint (in other words, channel) creation or > deletion and to receive these announcement taking any user-defined action in an > application callback. Right, we are talking about the same thing. > </quote> > > Also in Documentation/rpmsg.txt > > <quote> > ...the remote processor announces the existence of a remote rpmsg service by > sending a name service message (which contains the name and rpmsg addr of the > remote service, see struct rpmsg_ns_msg). > </quote> > > in [2]: > > <quote> > In the current protocol, at startup, the master sends notification to remote to let > it know that it can receive name service announcement. The RPMSG protocol is not involved in that path. The only notification send by the application processor goes through the virtIO framework with virtio_device_ready() and virtqueue_notify(): https://elixir.bootlin.com/linux/v5.9-rc4/source/drivers/rpmsg/virtio_rpmsg_bus.c#L973 > </quote> > > > > > > And yes, the name-space announcement use-case seems confusing to me too - it > > > > > reverts the relationship in a way: once a guest has booted and established > > > > > connections to any rpmsg "devices," those send their namespace announcements > > > > > back. But I think this can be regarded as server identification: you connect > > > > > to a server and it replies with its identification and capabilities. > > > > > > > > Based on the above can I assume vhost_rpmsg_ns_announce() is sent from the > > > > guest? > > > > > > No, it's "vhost_..." so it's running on the host. > > > > Ok, that's better and confirms the usage of the VIRTIO_RPMSG_RESPONSE queue. > > When reading your explanation above, I thought the term "those" referred to the > > guest. In light of your explanation I now understand that "those" referred to > > the rpmgs devices on the host. > > > > In the above paragraph you write: > > > > ... "once a guest has booted and established connections to any rpmsg "devices", > > those send their namespace announcements back". > > > > I'd like to unpack a few things about this sentence: > > > > 1) In this context, how is a "connection" established between a guest and a host? > > That's handled by the VirtIO / VirtQueues in the case of virtio_rpmsg_bus.c but in > general, as mentioned in [2] > > <quote> > However, master does not consider the fact that if the remote is ready to handle > notification at this point in time. > </quote> > > > 2) How does the guest now about the rpmsg devices it has made a connection to? > > Again, that's the same as with all other VirtIO / KVM / Qemu devices: in a common > Qemu case, it's the Qemu which emulates the hardware and registers those devices. > > > 3) Why is a namespace announcement needed at all when guests are aware of the > > rpmsg devices instantiated on the host, and have already connected to them? > > It is indeed optional according to the protocol, but as described above, without > it the virtio_rpmsg_bus.c driver won't create rpmsg channels / devices, so, no > probing will take place. I now have a better understanding of what you are trying to do. On the flip side this thread is too long to continue with it - I will review V7 and we can pickup from there. Mathieu > > > > The host (the server, an > > > analogue of the application processor, IIUC) sends NS announcements to guests. > > > > I think we have just found the source of the confusion - in the remoteproc world > > the application processor receives name announcements, it doesn't send them. > > Interesting, well, we know now that both directions are possible, but I still > don't know whether all configurations are valid: only down, only up, none or both. > > Thanks > Guennadi > > [1] https://nxpmicro.github.io/rpmsg-lite/ > [2] https://github.com/OpenAMP/open-amp/wiki/RPMsg-Messaging-Protocol > > > > > I saw your V7, something I will look into. In the mean time I need to bring > > > > your attention to this set [1] from Arnaud. Please have a look as it will > > > > impact your work. > > > > > > > > https://patchwork.kernel.org/project/linux-remoteproc/list/?series=338335 > > > > > > Yes, I've had a look at that series, thanks for forwarding it to me. TBH I > > > don't quite understand some choices there, e.g. creating a separate driver and > > > then having to register devices just for the namespace announcement. I don't > > > think creating virtual devices is taken easily in Linux. But either way I > > > don't think our series conflict a lot, but I do hope that I can merge my > > > series first, he'd just have to switch to using the header, that I'm adding. > > > Hardly too many changes otherwise. > > > > It is not the conflicts between the series that I wanted to highlight but the > > fact that name service is in the process of becoming a driver on its own, and > > with no dependence on the transport mechanism. > > > > > > > > > > > And I don't see a server implementation per se... It is more like a client > > > > > > implementation since vhost_rpmsg_announce() uses the RESPONSE queue, which sends > > > > > > messages from host to guest. > > > > > > > > > > > > Perhaps it is my lack of familiarity with vhost terminology. > > > > > > > > > > > > > > > > > > > > Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > > --- > > > > > > > drivers/vhost/Kconfig | 7 + > > > > > > > drivers/vhost/Makefile | 3 + > > > > > > > drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ > > > > > > > drivers/vhost/vhost_rpmsg.h | 74 +++++++ > > > > > > > 4 files changed, 457 insertions(+) > > > > > > > create mode 100644 drivers/vhost/rpmsg.c > > > > > > > create mode 100644 drivers/vhost/vhost_rpmsg.h > > > > > > > > > > > > > > diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig > > > > > > > index 587fbae06182..046b948fc411 100644 > > > > > > > --- a/drivers/vhost/Kconfig > > > > > > > +++ b/drivers/vhost/Kconfig > > > > > > > @@ -38,6 +38,13 @@ config VHOST_NET > > > > > > > To compile this driver as a module, choose M here: the module will > > > > > > > be called vhost_net. > > > > > > > > > > > > > > +config VHOST_RPMSG > > > > > > > + tristate > > > > > > > + select VHOST > > > > > > > + help > > > > > > > + Vhost RPMsg API allows vhost drivers to communicate with VirtIO > > > > > > > + drivers, using the RPMsg over VirtIO protocol. > > > > > > > > > > > > I had to assume vhost drivers are running on the host and virtIO drivers on the > > > > > > guests. This may be common knowledge for people familiar with vhosts but > > > > > > certainly obscur for commoners Having a help section that is clear on what is > > > > > > happening would remove any ambiguity. > > > > > > > > > > It is the terminology, yes, but you're right, the wording isn't very clear, will > > > > > improve. > > > > > > > > > > > > + > > > > > > > config VHOST_SCSI > > > > > > > tristate "VHOST_SCSI TCM fabric driver" > > > > > > > depends on TARGET_CORE && EVENTFD > > > > > > > diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile > > > > > > > index f3e1897cce85..9cf459d59f97 100644 > > > > > > > --- a/drivers/vhost/Makefile > > > > > > > +++ b/drivers/vhost/Makefile > > > > > > > @@ -2,6 +2,9 @@ > > > > > > > obj-$(CONFIG_VHOST_NET) += vhost_net.o > > > > > > > vhost_net-y := net.o > > > > > > > > > > > > > > +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o > > > > > > > +vhost_rpmsg-y := rpmsg.o > > > > > > > + > > > > > > > obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o > > > > > > > vhost_scsi-y := scsi.o > > > > > > > > > > > > > > diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c > > > > > > > new file mode 100644 > > > > > > > index 000000000000..c26d7a4afc6d > > > > > > > --- /dev/null > > > > > > > +++ b/drivers/vhost/rpmsg.c > > > > > > > @@ -0,0 +1,373 @@ > > > > > > > +// SPDX-License-Identifier: GPL-2.0-only > > > > > > > +/* > > > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > > > + * > > > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > > + * > > > > > > > + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the > > > > > > > + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c > > > > > > > > > > > > Again, very confusing. The changelog refers to a server implementation but to > > > > > > me this refers to a client implementation, especially if rpmsg_recv_single() and > > > > > > rpmsg_ns_cb() are used on the other side of the pipe. > > > > > > > > > > I think the above is correct. "Vhost" indicates, that this is running on the host. > > > > > "match the guest side" means, that you can use this API on the host and it is > > > > > designed to work together with the RPMsg VirtIO drivers running on guests, as > > > > > implemented *on guests* by virtio_rpmsg_bus.c. Would "to work together" be a better > > > > > description than "to match?" > > > > > > > > Lets forget about this part now and concentrate on the above conversation. > > > > Things will start to make sense at one point. > > > > > > I've improved that description a bit, it was indeed rather clumsy. > > > > Much appreciated - I'll take a look a V7 next week. > > > > > > > > [snip] > > > > > > > > > > diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h > > > > > > > new file mode 100644 > > > > > > > index 000000000000..30072cecb8a0 > > > > > > > --- /dev/null > > > > > > > +++ b/drivers/vhost/vhost_rpmsg.h > > > > > > > @@ -0,0 +1,74 @@ > > > > > > > +/* SPDX-License-Identifier: GPL-2.0 */ > > > > > > > +/* > > > > > > > + * Copyright(c) 2020 Intel Corporation. All rights reserved. > > > > > > > + * > > > > > > > + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> > > > > > > > + */ > > > > > > > + > > > > > > > +#ifndef VHOST_RPMSG_H > > > > > > > +#define VHOST_RPMSG_H > > > > > > > + > > > > > > > +#include <linux/uio.h> > > > > > > > +#include <linux/virtio_rpmsg.h> > > > > > > > + > > > > > > > +#include "vhost.h" > > > > > > > + > > > > > > > +/* RPMsg uses two VirtQueues: one for each direction */ > > > > > > > +enum { > > > > > > > + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ > > > > > > > + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ > > > > > > > + /* Keep last */ > > > > > > > + VIRTIO_RPMSG_NUM_OF_VQS, > > > > > > > +}; > > > > > > > + > > > > > > > +struct vhost_rpmsg_ept; > > > > > > > + > > > > > > > +struct vhost_rpmsg_iter { > > > > > > > + struct iov_iter iov_iter; > > > > > > > + struct rpmsg_hdr rhdr; > > > > > > > + struct vhost_virtqueue *vq; > > > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > > > + int head; > > > > > > > + void *priv; > > > > > > > > > > > > I don't see @priv being used anywhere. > > > > > > > > > > That's logical: this is a field, private to the API users, so the core shouldn't > > > > > use it :-) It's used in later patches. > > > > > > > > That is where structure documentation is useful. I will let Michael decide what > > > > he wants to do. > > > > > > I can add some kerneldoc documentation there, no problem. > > > > > > > Thanks for the feedback, > > > > > > Thanks for your reviews! I'd very much like to close all the still open points > > > and merge the series ASAP. > > > > > > Thanks > > > Guennadi > > > > > > > Mathieu > > > > > > > > > > > > > > > > > > > > > > +}; > > > > > > > + > > > > > > > +struct vhost_rpmsg { > > > > > > > + struct vhost_dev dev; > > > > > > > + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > > > + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; > > > > > > > + const struct vhost_rpmsg_ept *ept; > > > > > > > + unsigned int n_epts; > > > > > > > +}; > > > > > > > + > > > > > > > +struct vhost_rpmsg_ept { > > > > > > > + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > > > + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); > > > > > > > + int addr; > > > > > > > +}; > > > > > > > + > > > > > > > +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) > > > > > > > +{ > > > > > > > + return iter->rhdr.len; > > > > > > > +} > > > > > > > > > > > > Again, I don't see where this is used. > > > > > > > > > > This is exported API, it's used by users. > > > > > > > > > > > > + > > > > > > > +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ > > > > > > > + .rhdr = { \ > > > > > > > + .src = cpu_to_vhost32(_vq, _src), \ > > > > > > > + .dst = cpu_to_vhost32(_vq, _dst), \ > > > > > > > + }, \ > > > > > > > + } > > > > > > > > > > > > Same. > > > > > > > > > > ditto. > > > > > > > > > > Thanks > > > > > Guennadi > > > > > > > > > > > Thanks, > > > > > > Mathieu > > > > > > > > > > > > > + > > > > > > > +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, > > > > > > > + unsigned int n_epts); > > > > > > > +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); > > > > > > > +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, > > > > > > > + unsigned int src); > > > > > > > +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, > > > > > > > + struct vhost_rpmsg_iter *iter, > > > > > > > + unsigned int qid, ssize_t len); > > > > > > > +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, > > > > > > > + void *data, size_t size); > > > > > > > +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, > > > > > > > + struct vhost_rpmsg_iter *iter); > > > > > > > + > > > > > > > +#endif > > > > > > > -- > > > > > > > 2.28.0 > > > > > > >
diff --git a/drivers/vhost/Kconfig b/drivers/vhost/Kconfig index 587fbae06182..046b948fc411 100644 --- a/drivers/vhost/Kconfig +++ b/drivers/vhost/Kconfig @@ -38,6 +38,13 @@ config VHOST_NET To compile this driver as a module, choose M here: the module will be called vhost_net. +config VHOST_RPMSG + tristate + select VHOST + help + Vhost RPMsg API allows vhost drivers to communicate with VirtIO + drivers, using the RPMsg over VirtIO protocol. + config VHOST_SCSI tristate "VHOST_SCSI TCM fabric driver" depends on TARGET_CORE && EVENTFD diff --git a/drivers/vhost/Makefile b/drivers/vhost/Makefile index f3e1897cce85..9cf459d59f97 100644 --- a/drivers/vhost/Makefile +++ b/drivers/vhost/Makefile @@ -2,6 +2,9 @@ obj-$(CONFIG_VHOST_NET) += vhost_net.o vhost_net-y := net.o +obj-$(CONFIG_VHOST_RPMSG) += vhost_rpmsg.o +vhost_rpmsg-y := rpmsg.o + obj-$(CONFIG_VHOST_SCSI) += vhost_scsi.o vhost_scsi-y := scsi.o diff --git a/drivers/vhost/rpmsg.c b/drivers/vhost/rpmsg.c new file mode 100644 index 000000000000..c26d7a4afc6d --- /dev/null +++ b/drivers/vhost/rpmsg.c @@ -0,0 +1,373 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright(c) 2020 Intel Corporation. All rights reserved. + * + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> + * + * Vhost RPMsg VirtIO interface. It provides a set of functions to match the + * guest side RPMsg VirtIO API, provided by drivers/rpmsg/virtio_rpmsg_bus.c + * These functions handle creation of 2 virtual queues, handling of endpoint + * addresses, sending a name-space announcement to the guest as well as any + * user messages. This API can be used by any vhost driver to handle RPMsg + * specific processing. + * Specific vhost drivers, using this API will use their own VirtIO device + * IDs, that should then also be added to the ID table in virtio_rpmsg_bus.c + */ + +#include <linux/compat.h> +#include <linux/file.h> +#include <linux/miscdevice.h> +#include <linux/module.h> +#include <linux/mutex.h> +#include <linux/vhost.h> +#include <linux/virtio_rpmsg.h> +#include <uapi/linux/rpmsg.h> + +#include "vhost.h" +#include "vhost_rpmsg.h" + +/* + * All virtio-rpmsg virtual queue kicks always come with just one buffer - + * either input or output, but we can also handle split messages + */ +static int vhost_rpmsg_get_msg(struct vhost_virtqueue *vq, unsigned int *cnt) +{ + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); + unsigned int out, in; + int head = vhost_get_vq_desc(vq, vq->iov, ARRAY_SIZE(vq->iov), &out, &in, + NULL, NULL); + if (head < 0) { + vq_err(vq, "%s(): error %d getting buffer\n", + __func__, head); + return head; + } + + /* Nothing new? */ + if (head == vq->num) + return head; + + if (vq == &vr->vq[VIRTIO_RPMSG_RESPONSE]) { + if (out) { + vq_err(vq, "%s(): invalid %d output in response queue\n", + __func__, out); + goto return_buf; + } + + *cnt = in; + } + + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST]) { + if (in) { + vq_err(vq, "%s(): invalid %d input in request queue\n", + __func__, in); + goto return_buf; + } + + *cnt = out; + } + + return head; + +return_buf: + vhost_add_used(vq, head, 0); + + return -EINVAL; +} + +static const struct vhost_rpmsg_ept *vhost_rpmsg_ept_find(struct vhost_rpmsg *vr, int addr) +{ + unsigned int i; + + for (i = 0; i < vr->n_epts; i++) + if (vr->ept[i].addr == addr) + return vr->ept + i; + + return NULL; +} + +/* + * if len < 0, then for reading a request, the complete virtual queue buffer + * size is prepared, for sending a response, the length in the iterator is used + */ +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, + unsigned int qid, ssize_t len) + __acquires(vq->mutex) +{ + struct vhost_virtqueue *vq = vr->vq + qid; + unsigned int cnt; + ssize_t ret; + size_t tmp; + + if (qid >= VIRTIO_RPMSG_NUM_OF_VQS) + return -EINVAL; + + iter->vq = vq; + + mutex_lock(&vq->mutex); + vhost_disable_notify(&vr->dev, vq); + + iter->head = vhost_rpmsg_get_msg(vq, &cnt); + if (iter->head == vq->num) + iter->head = -EAGAIN; + + if (iter->head < 0) { + ret = iter->head; + goto unlock; + } + + tmp = iov_length(vq->iov, cnt); + if (tmp < sizeof(iter->rhdr)) { + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); + ret = -ENOBUFS; + goto return_buf; + } + + switch (qid) { + case VIRTIO_RPMSG_REQUEST: + if (len >= 0) { + if (tmp < sizeof(iter->rhdr) + len) { + ret = -ENOBUFS; + goto return_buf; + } + + tmp = len + sizeof(iter->rhdr); + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, WRITE, vq->iov, cnt, tmp); + + /* Read the RPMSG header with endpoint addresses */ + tmp = copy_from_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); + if (tmp != sizeof(iter->rhdr)) { + vq_err(vq, "%s(): got %zu instead of %zu\n", __func__, + tmp, sizeof(iter->rhdr)); + ret = -EIO; + goto return_buf; + } + + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.dst)); + if (!iter->ept) { + vq_err(vq, "%s(): no endpoint with address %d\n", + __func__, vhost32_to_cpu(vq, iter->rhdr.dst)); + ret = -ENOENT; + goto return_buf; + } + + /* Let the endpoint read the payload */ + if (iter->ept->read) { + ret = iter->ept->read(vr, iter); + if (ret < 0) + goto return_buf; + + iter->rhdr.len = cpu_to_vhost16(vq, ret); + } else { + iter->rhdr.len = 0; + } + + /* Prepare for the response phase */ + iter->rhdr.dst = iter->rhdr.src; + iter->rhdr.src = cpu_to_vhost32(vq, iter->ept->addr); + + break; + case VIRTIO_RPMSG_RESPONSE: + if (!iter->ept && iter->rhdr.dst != cpu_to_vhost32(vq, RPMSG_NS_ADDR)) { + /* + * Usually the iterator is configured when processing a + * message on the request queue, but it's also possible + * to send a message on the response queue without a + * preceding request, in that case the iterator must + * contain source and destination addresses. + */ + iter->ept = vhost_rpmsg_ept_find(vr, vhost32_to_cpu(vq, iter->rhdr.src)); + if (!iter->ept) { + ret = -ENOENT; + goto return_buf; + } + } + + if (len >= 0) { + if (tmp < sizeof(iter->rhdr) + len) { + ret = -ENOBUFS; + goto return_buf; + } + + iter->rhdr.len = cpu_to_vhost16(vq, len); + tmp = len + sizeof(iter->rhdr); + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, READ, vq->iov, cnt, tmp); + + /* Write the RPMSG header with endpoint addresses */ + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), &iter->iov_iter); + if (tmp != sizeof(iter->rhdr)) { + ret = -EIO; + goto return_buf; + } + + /* Let the endpoint write the payload */ + if (iter->ept && iter->ept->write) { + ret = iter->ept->write(vr, iter); + if (ret < 0) + goto return_buf; + } + + break; + } + + return 0; + +return_buf: + vhost_add_used(vq, iter->head, 0); +unlock: + vhost_enable_notify(&vr->dev, vq); + mutex_unlock(&vq->mutex); + + return ret; +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_start_lock); + +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, + void *data, size_t size) +{ + /* + * We could check for excess data, but copy_{to,from}_iter() don't do + * that either + */ + if (iter->vq == vr->vq + VIRTIO_RPMSG_RESPONSE) + return copy_to_iter(data, size, &iter->iov_iter); + + return copy_from_iter(data, size, &iter->iov_iter); +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_copy); + +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, + struct vhost_rpmsg_iter *iter) + __releases(vq->mutex) +{ + if (iter->head >= 0) + vhost_add_used_and_signal(iter->vq->dev, iter->vq, iter->head, + vhost16_to_cpu(iter->vq, iter->rhdr.len) + + sizeof(iter->rhdr)); + + vhost_enable_notify(&vr->dev, iter->vq); + mutex_unlock(&iter->vq->mutex); + + return iter->head; +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_finish_unlock); + +/* + * Return false to terminate the external loop only if we fail to obtain either + * a request or a response buffer + */ +static bool handle_rpmsg_req_single(struct vhost_rpmsg *vr, + struct vhost_virtqueue *vq) +{ + struct vhost_rpmsg_iter iter; + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_REQUEST, -EINVAL); + if (!ret) + ret = vhost_rpmsg_finish_unlock(vr, &iter); + if (ret < 0) { + if (ret != -EAGAIN) + vq_err(vq, "%s(): RPMSG processing failed %d\n", + __func__, ret); + return false; + } + + if (!iter.ept->write) + return true; + + ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, -EINVAL); + if (!ret) + ret = vhost_rpmsg_finish_unlock(vr, &iter); + if (ret < 0) { + vq_err(vq, "%s(): RPMSG finalising failed %d\n", __func__, ret); + return false; + } + + return true; +} + +static void handle_rpmsg_req_kick(struct vhost_work *work) +{ + struct vhost_virtqueue *vq = container_of(work, struct vhost_virtqueue, + poll.work); + struct vhost_rpmsg *vr = container_of(vq->dev, struct vhost_rpmsg, dev); + + while (handle_rpmsg_req_single(vr, vq)) + ; +} + +/* + * initialise two virtqueues with an array of endpoints, + * request and response callbacks + */ +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, + unsigned int n_epts) +{ + unsigned int i; + + for (i = 0; i < ARRAY_SIZE(vr->vq); i++) + vr->vq_p[i] = &vr->vq[i]; + + /* vq[0]: host -> guest, vq[1]: host <- guest */ + vr->vq[VIRTIO_RPMSG_REQUEST].handle_kick = handle_rpmsg_req_kick; + vr->vq[VIRTIO_RPMSG_RESPONSE].handle_kick = NULL; + + vr->ept = ept; + vr->n_epts = n_epts; + + vhost_dev_init(&vr->dev, vr->vq_p, VIRTIO_RPMSG_NUM_OF_VQS, + UIO_MAXIOV, 0, 0, true, NULL); +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_init); + +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr) +{ + if (vhost_dev_has_owner(&vr->dev)) + vhost_poll_flush(&vr->vq[VIRTIO_RPMSG_REQUEST].poll); + + vhost_dev_cleanup(&vr->dev); +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_destroy); + +/* send namespace */ +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, unsigned int src) +{ + struct vhost_virtqueue *vq = &vr->vq[VIRTIO_RPMSG_RESPONSE]; + struct vhost_rpmsg_iter iter = { + .rhdr = { + .src = 0, + .dst = cpu_to_vhost32(vq, RPMSG_NS_ADDR), + .flags = cpu_to_vhost16(vq, RPMSG_NS_CREATE), /* rpmsg_recv_single() */ + }, + }; + struct rpmsg_ns_msg ns = { + .addr = cpu_to_vhost32(vq, src), + .flags = cpu_to_vhost32(vq, RPMSG_NS_CREATE), /* for rpmsg_ns_cb() */ + }; + int ret = vhost_rpmsg_start_lock(vr, &iter, VIRTIO_RPMSG_RESPONSE, sizeof(ns)); + + if (ret < 0) + return ret; + + strlcpy(ns.name, name, sizeof(ns.name)); + + ret = vhost_rpmsg_copy(vr, &iter, &ns, sizeof(ns)); + if (ret != sizeof(ns)) + vq_err(iter.vq, "%s(): added %d instead of %zu bytes\n", + __func__, ret, sizeof(ns)); + + ret = vhost_rpmsg_finish_unlock(vr, &iter); + if (ret < 0) + vq_err(iter.vq, "%s(): namespace announcement failed: %d\n", + __func__, ret); + + return ret; +} +EXPORT_SYMBOL_GPL(vhost_rpmsg_ns_announce); + +MODULE_LICENSE("GPL v2"); +MODULE_AUTHOR("Intel, Inc."); +MODULE_DESCRIPTION("Vhost RPMsg API"); diff --git a/drivers/vhost/vhost_rpmsg.h b/drivers/vhost/vhost_rpmsg.h new file mode 100644 index 000000000000..30072cecb8a0 --- /dev/null +++ b/drivers/vhost/vhost_rpmsg.h @@ -0,0 +1,74 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +/* + * Copyright(c) 2020 Intel Corporation. All rights reserved. + * + * Author: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> + */ + +#ifndef VHOST_RPMSG_H +#define VHOST_RPMSG_H + +#include <linux/uio.h> +#include <linux/virtio_rpmsg.h> + +#include "vhost.h" + +/* RPMsg uses two VirtQueues: one for each direction */ +enum { + VIRTIO_RPMSG_RESPONSE, /* RPMsg response (host->guest) buffers */ + VIRTIO_RPMSG_REQUEST, /* RPMsg request (guest->host) buffers */ + /* Keep last */ + VIRTIO_RPMSG_NUM_OF_VQS, +}; + +struct vhost_rpmsg_ept; + +struct vhost_rpmsg_iter { + struct iov_iter iov_iter; + struct rpmsg_hdr rhdr; + struct vhost_virtqueue *vq; + const struct vhost_rpmsg_ept *ept; + int head; + void *priv; +}; + +struct vhost_rpmsg { + struct vhost_dev dev; + struct vhost_virtqueue vq[VIRTIO_RPMSG_NUM_OF_VQS]; + struct vhost_virtqueue *vq_p[VIRTIO_RPMSG_NUM_OF_VQS]; + const struct vhost_rpmsg_ept *ept; + unsigned int n_epts; +}; + +struct vhost_rpmsg_ept { + ssize_t (*read)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); + ssize_t (*write)(struct vhost_rpmsg *, struct vhost_rpmsg_iter *); + int addr; +}; + +static inline size_t vhost_rpmsg_iter_len(const struct vhost_rpmsg_iter *iter) +{ + return iter->rhdr.len; +} + +#define VHOST_RPMSG_ITER(_vq, _src, _dst) { \ + .rhdr = { \ + .src = cpu_to_vhost32(_vq, _src), \ + .dst = cpu_to_vhost32(_vq, _dst), \ + }, \ + } + +void vhost_rpmsg_init(struct vhost_rpmsg *vr, const struct vhost_rpmsg_ept *ept, + unsigned int n_epts); +void vhost_rpmsg_destroy(struct vhost_rpmsg *vr); +int vhost_rpmsg_ns_announce(struct vhost_rpmsg *vr, const char *name, + unsigned int src); +int vhost_rpmsg_start_lock(struct vhost_rpmsg *vr, + struct vhost_rpmsg_iter *iter, + unsigned int qid, ssize_t len); +size_t vhost_rpmsg_copy(struct vhost_rpmsg *vr, struct vhost_rpmsg_iter *iter, + void *data, size_t size); +int vhost_rpmsg_finish_unlock(struct vhost_rpmsg *vr, + struct vhost_rpmsg_iter *iter); + +#endif
Linux supports running the RPMsg protocol over the VirtIO transport protocol, but currently there is only support for VirtIO clients and no support for a VirtIO server. This patch adds a vhost-based RPMsg server implementation. Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 373 ++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost_rpmsg.h | 74 +++++++ 4 files changed, 457 insertions(+) create mode 100644 drivers/vhost/rpmsg.c create mode 100644 drivers/vhost/vhost_rpmsg.h