[Sound-open-firmware] [PATCH v3 0/5] Add a vhost RPMsg API
v3: - address several checkpatch warnings - address comments from Mathieu Poirier v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first. Thanks Guennadi Guennadi Liakhovetski (5): vhost: convert VHOST_VSOCK_SET_RUNNING to a generic ioctl vhost: (cosmetic) remove a superfluous variable initialisation rpmsg: move common structures and defines to headers rpmsg: update documentation vhost: add an RPMsg API Documentation/rpmsg.txt | 6 +- drivers/rpmsg/virtio_rpmsg_bus.c | 78 +------- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 382 +++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.c | 2 +- drivers/vhost/vhost_rpmsg.h | 74 ++++++++ include/linux/virtio_rpmsg.h | 81 +++++++++ include/uapi/linux/rpmsg.h | 3 + include/uapi/linux/vhost.h | 4 +- 10 files changed, 559 insertions(+), 81 deletions(-) create mode 100644 drivers/vhost/rpmsg.c create mode 100644 drivers/vhost/vhost_rpmsg.h create mode 100644 include/linux/virtio_rpmsg.h -- 1.9.3
VHOST_VSOCK_SET_RUNNING is used by the vhost vsock driver to perform crucial VirtQueue initialisation, like assigning .private fields and calling vhost_vq_init_access(), and clean up. However, this ioctl is actually extremely useful for any vhost driver, that doesn't have a side channel to inform it of a status change, e.g. upon a guest reboot. This patch makes that ioctl generic, while preserving its numeric value and also keeping the original alias. Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- include/uapi/linux/vhost.h | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/include/uapi/linux/vhost.h b/include/uapi/linux/vhost.h index 9fe72e4..b54af9d 100644 --- a/include/uapi/linux/vhost.h +++ b/include/uapi/linux/vhost.h @@ -93,6 +93,8 @@ #define VHOST_SET_BACKEND_FEATURES _IOW(VHOST_VIRTIO, 0x25, __u64) #define VHOST_GET_BACKEND_FEATURES _IOR(VHOST_VIRTIO, 0x26, __u64) +#define VHOST_SET_RUNNING _IOW(VHOST_VIRTIO, 0x61, int) + /* VHOST_NET specific defines */ /* Attach virtio net ring to a raw socket, or tap device. @@ -114,7 +116,7 @@ /* VHOST_VSOCK specific defines */ #define VHOST_VSOCK_SET_GUEST_CID _IOW(VHOST_VIRTIO, 0x60, __u64) -#define VHOST_VSOCK_SET_RUNNING _IOW(VHOST_VIRTIO, 0x61, int) +#define VHOST_VSOCK_SET_RUNNING VHOST_SET_RUNNING /* VHOST_VDPA specific defines */ -- 1.9.3
Even the compiler is able to figure out that in this case the initialisation is superfluous. Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- drivers/vhost/vhost.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c index 2f4383bb..2b9ad8a 100644 --- a/drivers/vhost/vhost.c +++ b/drivers/vhost/vhost.c @@ -895,7 +895,7 @@ static inline void __user *__vhost_get_user(struct vhost_virtqueue *vq, #define vhost_put_user(vq, x, ptr) \ ({ \ - int ret = -EFAULT; \ + int ret; \ if (!vq->iotlb) { \ ret = __put_user(x, ptr); \ } else { \ -- 1.9.3
virtio_rpmsg_bus.c keeps RPMsg protocol structure declarations and common defines like the ones, needed for name-space announcements, internal. Move them to common headers instead. Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org> Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- drivers/rpmsg/virtio_rpmsg_bus.c | 78 +------------------------------------- include/linux/virtio_rpmsg.h | 81 ++++++++++++++++++++++++++++++++++++++++ include/uapi/linux/rpmsg.h | 3 ++ 3 files changed, 86 insertions(+), 76 deletions(-) create mode 100644 include/linux/virtio_rpmsg.h diff --git a/drivers/rpmsg/virtio_rpmsg_bus.c b/drivers/rpmsg/virtio_rpmsg_bus.c index 07d4f33..f3bd050 100644 --- a/drivers/rpmsg/virtio_rpmsg_bus.c +++ b/drivers/rpmsg/virtio_rpmsg_bus.c @@ -25,7 +25,9 @@ #include <linux/virtio.h> #include <linux/virtio_ids.h> #include <linux/virtio_config.h> +#include <linux/virtio_rpmsg.h> #include <linux/wait.h> +#include <uapi/linux/rpmsg.h> #include "rpmsg_internal.h" @@ -69,58 +71,6 @@ struct virtproc_info { struct rpmsg_endpoint *ns_ept; }; -/* The feature bitmap for virtio rpmsg */ -#define VIRTIO_RPMSG_F_NS 0 /* RP supports name service notifications */ - -/** - * struct rpmsg_hdr - common header for all rpmsg messages - * @src: source address - * @dst: destination address - * @reserved: reserved for future use - * @len: length of payload (in bytes) - * @flags: message flags - * @data: @len bytes of message payload data - * - * Every message sent(/received) on the rpmsg bus begins with this header. - */ -struct rpmsg_hdr { - u32 src; - u32 dst; - u32 reserved; - u16 len; - u16 flags; - u8 data[]; -} __packed; - -/** - * struct rpmsg_ns_msg - dynamic name service announcement message - * @name: name of remote service that is published - * @addr: address of remote service that is published - * @flags: indicates whether service is created or destroyed - * - * This message is sent across to publish a new service, or announce - * about its removal. When we receive these messages, an appropriate - * rpmsg channel (i.e device) is created/destroyed. In turn, the ->probe() - * or ->remove() handler of the appropriate rpmsg driver will be invoked - * (if/as-soon-as one is registered). - */ -struct rpmsg_ns_msg { - char name[RPMSG_NAME_SIZE]; - u32 addr; - u32 flags; -} __packed; - -/** - * enum rpmsg_ns_flags - dynamic name service announcement flags - * - * @RPMSG_NS_CREATE: a new remote service was just created - * @RPMSG_NS_DESTROY: a known remote service was just destroyed - */ -enum rpmsg_ns_flags { - RPMSG_NS_CREATE = 0, - RPMSG_NS_DESTROY = 1, -}; - /** * @vrp: the remote processor this channel belongs to */ @@ -134,36 +84,12 @@ struct virtio_rpmsg_channel { container_of(_rpdev, struct virtio_rpmsg_channel, rpdev) /* - * We're allocating buffers of 512 bytes each for communications. The - * number of buffers will be computed from the number of buffers supported - * by the vring, upto a maximum of 512 buffers (256 in each direction). - * - * Each buffer will have 16 bytes for the msg header and 496 bytes for - * the payload. - * - * This will utilize a maximum total space of 256KB for the buffers. - * - * We might also want to add support for user-provided buffers in time. - * This will allow bigger buffer size flexibility, and can also be used - * to achieve zero-copy messaging. - * - * Note that these numbers are purely a decision of this driver - we - * can change this without changing anything in the firmware of the remote - * processor. - */ -#define MAX_RPMSG_NUM_BUFS (512) -#define MAX_RPMSG_BUF_SIZE (512) - -/* * Local addresses are dynamically allocated on-demand. * We do not dynamically assign addresses from the low 1024 range, * in order to reserve that address range for predefined services. */ #define RPMSG_RESERVED_ADDRESSES (1024) -/* Address 53 is reserved for advertising remote services */ -#define RPMSG_NS_ADDR (53) - static void virtio_rpmsg_destroy_ept(struct rpmsg_endpoint *ept); static int virtio_rpmsg_send(struct rpmsg_endpoint *ept, void *data, int len); static int virtio_rpmsg_sendto(struct rpmsg_endpoint *ept, void *data, int len, diff --git a/include/linux/virtio_rpmsg.h b/include/linux/virtio_rpmsg.h new file mode 100644 index 00000000..679be8b --- /dev/null +++ b/include/linux/virtio_rpmsg.h @@ -0,0 +1,81 @@ +/* SPDX-License-Identifier: GPL-2.0 */ + +#ifndef _LINUX_VIRTIO_RPMSG_H +#define _LINUX_VIRTIO_RPMSG_H + +#include <linux/mod_devicetable.h> + +/** + * struct rpmsg_hdr - common header for all rpmsg messages + * @src: source address + * @dst: destination address + * @reserved: reserved for future use + * @len: length of payload (in bytes) + * @flags: message flags + * @data: @len bytes of message payload data + * + * Every message sent(/received) on the rpmsg bus begins with this header. + */ +struct rpmsg_hdr { + u32 src; + u32 dst; + u32 reserved; + u16 len; + u16 flags; + u8 data[]; +} __packed; + +/** + * struct rpmsg_ns_msg - dynamic name service announcement message + * @name: name of remote service that is published + * @addr: address of remote service that is published + * @flags: indicates whether service is created or destroyed + * + * This message is sent across to publish a new service, or announce + * about its removal. When we receive these messages, an appropriate + * rpmsg channel (i.e device) is created/destroyed. In turn, the ->probe() + * or ->remove() handler of the appropriate rpmsg driver will be invoked + * (if/as-soon-as one is registered). + */ +struct rpmsg_ns_msg { + char name[RPMSG_NAME_SIZE]; + u32 addr; + u32 flags; +} __packed; + +/** + * enum rpmsg_ns_flags - dynamic name service announcement flags + * + * @RPMSG_NS_CREATE: a new remote service was just created + * @RPMSG_NS_DESTROY: a known remote service was just destroyed + */ +enum rpmsg_ns_flags { + RPMSG_NS_CREATE = 0, + RPMSG_NS_DESTROY = 1, +}; + +/* + * We're allocating buffers of 512 bytes each for communications. The + * number of buffers will be computed from the number of buffers supported + * by the vring, upto a maximum of 512 buffers (256 in each direction). + * + * Each buffer will have 16 bytes for the msg header and 496 bytes for + * the payload. + * + * This will utilize a maximum total space of 256KB for the buffers. + * + * We might also want to add support for user-provided buffers in time. + * This will allow bigger buffer size flexibility, and can also be used + * to achieve zero-copy messaging. + * + * Note that these numbers are purely a decision of this driver - we + * can change this without changing anything in the firmware of the remote + * processor. + */ +#define MAX_RPMSG_NUM_BUFS 512 +#define MAX_RPMSG_BUF_SIZE 512 + +/* Address 53 is reserved for advertising remote services */ +#define RPMSG_NS_ADDR 53 + +#endif diff --git a/include/uapi/linux/rpmsg.h b/include/uapi/linux/rpmsg.h index e14c6da..d669c04 100644 --- a/include/uapi/linux/rpmsg.h +++ b/include/uapi/linux/rpmsg.h @@ -24,4 +24,7 @@ struct rpmsg_endpoint_info { #define RPMSG_CREATE_EPT_IOCTL _IOW(0xb5, 0x1, struct rpmsg_endpoint_info) #define RPMSG_DESTROY_EPT_IOCTL _IO(0xb5, 0x2) +/* The feature bitmap for virtio rpmsg */ +#define VIRTIO_RPMSG_F_NS 0 /* RP supports name service notifications */ + #endif -- 1.9.3
rpmsg_create_ept() takes struct rpmsg_channel_info chinfo as its last argument, not a u32 value. The first two arguments are also updated. Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- Documentation/rpmsg.txt | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/Documentation/rpmsg.txt b/Documentation/rpmsg.txt index 24b7a9e..1ce353c 100644 --- a/Documentation/rpmsg.txt +++ b/Documentation/rpmsg.txt @@ -192,9 +192,9 @@ Returns 0 on success and an appropriate error value on failure. :: - struct rpmsg_endpoint *rpmsg_create_ept(struct rpmsg_channel *rpdev, - void (*cb)(struct rpmsg_channel *, void *, int, void *, u32), - void *priv, u32 addr); + struct rpmsg_endpoint *rpmsg_create_ept(struct rpmsg_device *rpdev, + rpmsg_rx_cb_t cb, void *priv, + struct rpmsg_channel_info chinfo); every rpmsg address in the system is bound to an rx callback (so when inbound messages arrive, they are dispatched by the rpmsg bus using the -- 1.9.3
On Wed, 27 May 2020 at 12:05, Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> wrote:
rpmsg_create_ept() takes struct rpmsg_channel_info chinfo as its last argument, not a u32 value. The first two arguments are also updated.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- Documentation/rpmsg.txt | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/Documentation/rpmsg.txt b/Documentation/rpmsg.txt index 24b7a9e..1ce353c 100644 --- a/Documentation/rpmsg.txt +++ b/Documentation/rpmsg.txt @@ -192,9 +192,9 @@ Returns 0 on success and an appropriate error value on failure.
::
- struct rpmsg_endpoint *rpmsg_create_ept(struct rpmsg_channel *rpdev, - void (*cb)(struct rpmsg_channel *, void *, int, void *, u32), - void *priv, u32 addr); + struct rpmsg_endpoint *rpmsg_create_ept(struct rpmsg_device *rpdev, + rpmsg_rx_cb_t cb, void *priv, + struct rpmsg_channel_info chinfo);
Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
every rpmsg address in the system is bound to an rx callback (so when inbound messages arrive, they are dispatched by the rpmsg bus using the -- 1.9.3
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 | 382 ++++++++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost_rpmsg.h | 74 +++++++++ 4 files changed, 466 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 2c75d16..8b91f3e 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 + depends on 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 f3e1897..9cf459d 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 00000000..ea77e1f --- /dev/null +++ b/drivers/vhost/rpmsg.c @@ -0,0 +1,382 @@ +// 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 + */ +static int vhost_rpmsg_get_single(struct vhost_virtqueue *vq) +{ + 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] && (out || in != 1)) { + vq_err(vq, + "%s(): invalid %d input and %d output in response queue\n", + __func__, in, out); + goto return_buf; + } + + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST] && (in || out != 1)) { + vq_err(vq, + "%s(): invalid %d input and %d output in request queue\n", + __func__, in, out); + goto return_buf; + } + + return head; + +return_buf: + /* + * FIXME: might need to return the buffer using vhost_add_used() + * or vhost_discard_vq_desc(). vhost_discard_vq_desc() is + * described as "being useful for error handling," but it makes + * the thus discarded buffers "unseen," so next time we look we + * retrieve them again? + */ + 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; + 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_single(vq); + if (iter->head == vq->num) + iter->head = -EAGAIN; + + if (iter->head < 0) + goto unlock; + + tmp = vq->iov[0].iov_len; + if (tmp < sizeof(iter->rhdr)) { + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); + iter->head = -ENOBUFS; + goto return_buf; + } + + switch (qid) { + case VIRTIO_RPMSG_REQUEST: + if (len < 0) { + len = tmp - sizeof(iter->rhdr); + } else if (tmp < sizeof(iter->rhdr) + len) { + iter->head = -ENOBUFS; + goto return_buf; + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, WRITE, + vq->iov, 1, sizeof(iter->rhdr) + len); + + /* 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)); + iter->head = -EIO; + goto return_buf; + } + + iter->ept = vhost_rpmsg_ept_find(vr, iter->rhdr.dst); + if (!iter->ept) { + vq_err(vq, "%s(): no endpoint with address %d\n", + __func__, iter->rhdr.dst); + iter->head = -ENOENT; + goto return_buf; + } + + /* Let the endpoint read the payload */ + if (iter->ept->read) { + ssize_t ret = iter->ept->read(vr, iter); + + if (ret < 0) { + iter->head = ret; + goto return_buf; + } + + iter->rhdr.len = ret; + } else { + iter->rhdr.len = 0; + } + + /* Prepare for the response phase */ + iter->rhdr.dst = iter->rhdr.src; + iter->rhdr.src = iter->ept->addr; + + break; + case VIRTIO_RPMSG_RESPONSE: + if (!iter->ept && iter->rhdr.dst != 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, iter->rhdr.src); + if (!iter->ept) { + iter->head = -ENOENT; + goto return_buf; + } + } + + if (len < 0) { + len = tmp - sizeof(iter->rhdr); + } else if (tmp < sizeof(iter->rhdr) + len) { + iter->head = -ENOBUFS; + goto return_buf; + } else { + iter->rhdr.len = len; + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, READ, + vq->iov, 1, sizeof(iter->rhdr) + len); + + /* Write the RPMSG header with endpoint addresses */ + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), + &iter->iov_iter); + if (tmp != sizeof(iter->rhdr)) { + iter->head = -EIO; + goto return_buf; + } + + /* Let the endpoint write the payload */ + if (iter->ept && iter->ept->write) { + ssize_t ret = iter->ept->write(vr, iter); + + if (ret < 0) { + iter->head = ret; + goto return_buf; + } + } + + break; + } + + return 0; + +return_buf: + /* + * FIXME: vhost_discard_vq_desc() or vhost_add_used(), see comment in + * vhost_rpmsg_get_single() + */ +unlock: + vhost_enable_notify(&vr->dev, vq); + mutex_unlock(&vq->mutex); + + return iter->head; +} +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, + 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, 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_rpmsg_iter iter = { + .rhdr = { + .src = 0, + .dst = RPMSG_NS_ADDR, + .flags = RPMSG_NS_CREATE, /* rpmsg_recv_single() */ + }, + }; + struct rpmsg_ns_msg ns = { + .addr = src, + .flags = 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 00000000..a3d0dda --- /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(_src, _dst) { \ + .rhdr = { \ + .src = _src, \ + .dst = _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 -- 1.9.3
On Wed, 27 May 2020 at 12:05, Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> 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.
Signed-off-by: Guennadi Liakhovetski <guennadi.liakhovetski@linux.intel.com> --- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 382 ++++++++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost_rpmsg.h | 74 +++++++++ 4 files changed, 466 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 2c75d16..8b91f3e 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 + depends on 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 f3e1897..9cf459d 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 00000000..ea77e1f --- /dev/null +++ b/drivers/vhost/rpmsg.c @@ -0,0 +1,382 @@ +// 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 + */ +
Thank you for adding that. Acked-by: Mathieu Poirier <mathieu.poirier@linaro.org>
+#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 + */ +static int vhost_rpmsg_get_single(struct vhost_virtqueue *vq) +{ + 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] && (out || in != 1)) { + vq_err(vq, + "%s(): invalid %d input and %d output in response queue\n", + __func__, in, out); + goto return_buf; + } + + if (vq == &vr->vq[VIRTIO_RPMSG_REQUEST] && (in || out != 1)) { + vq_err(vq, + "%s(): invalid %d input and %d output in request queue\n", + __func__, in, out); + goto return_buf; + } + + return head; + +return_buf: + /* + * FIXME: might need to return the buffer using vhost_add_used() + * or vhost_discard_vq_desc(). vhost_discard_vq_desc() is + * described as "being useful for error handling," but it makes + * the thus discarded buffers "unseen," so next time we look we + * retrieve them again? + */ + 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; + 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_single(vq); + if (iter->head == vq->num) + iter->head = -EAGAIN; + + if (iter->head < 0) + goto unlock; + + tmp = vq->iov[0].iov_len; + if (tmp < sizeof(iter->rhdr)) { + vq_err(vq, "%s(): size %zu too small\n", __func__, tmp); + iter->head = -ENOBUFS; + goto return_buf; + } + + switch (qid) { + case VIRTIO_RPMSG_REQUEST: + if (len < 0) { + len = tmp - sizeof(iter->rhdr); + } else if (tmp < sizeof(iter->rhdr) + len) { + iter->head = -ENOBUFS; + goto return_buf; + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, WRITE, + vq->iov, 1, sizeof(iter->rhdr) + len); + + /* 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)); + iter->head = -EIO; + goto return_buf; + } + + iter->ept = vhost_rpmsg_ept_find(vr, iter->rhdr.dst); + if (!iter->ept) { + vq_err(vq, "%s(): no endpoint with address %d\n", + __func__, iter->rhdr.dst); + iter->head = -ENOENT; + goto return_buf; + } + + /* Let the endpoint read the payload */ + if (iter->ept->read) { + ssize_t ret = iter->ept->read(vr, iter); + + if (ret < 0) { + iter->head = ret; + goto return_buf; + } + + iter->rhdr.len = ret; + } else { + iter->rhdr.len = 0; + } + + /* Prepare for the response phase */ + iter->rhdr.dst = iter->rhdr.src; + iter->rhdr.src = iter->ept->addr; + + break; + case VIRTIO_RPMSG_RESPONSE: + if (!iter->ept && iter->rhdr.dst != 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, iter->rhdr.src); + if (!iter->ept) { + iter->head = -ENOENT; + goto return_buf; + } + } + + if (len < 0) { + len = tmp - sizeof(iter->rhdr); + } else if (tmp < sizeof(iter->rhdr) + len) { + iter->head = -ENOBUFS; + goto return_buf; + } else { + iter->rhdr.len = len; + } + + /* len is now the size of the payload */ + iov_iter_init(&iter->iov_iter, READ, + vq->iov, 1, sizeof(iter->rhdr) + len); + + /* Write the RPMSG header with endpoint addresses */ + tmp = copy_to_iter(&iter->rhdr, sizeof(iter->rhdr), + &iter->iov_iter); + if (tmp != sizeof(iter->rhdr)) { + iter->head = -EIO; + goto return_buf; + } + + /* Let the endpoint write the payload */ + if (iter->ept && iter->ept->write) { + ssize_t ret = iter->ept->write(vr, iter); + + if (ret < 0) { + iter->head = ret; + goto return_buf; + } + } + + break; + } + + return 0; + +return_buf: + /* + * FIXME: vhost_discard_vq_desc() or vhost_add_used(), see comment in + * vhost_rpmsg_get_single() + */ +unlock: + vhost_enable_notify(&vr->dev, vq); + mutex_unlock(&vq->mutex); + + return iter->head; +} +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, + 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, 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_rpmsg_iter iter = { + .rhdr = { + .src = 0, + .dst = RPMSG_NS_ADDR, + .flags = RPMSG_NS_CREATE, /* rpmsg_recv_single() */ + }, + }; + struct rpmsg_ns_msg ns = { + .addr = src, + .flags = 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 00000000..a3d0dda --- /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(_src, _dst) { \ + .rhdr = { \ + .src = _src, \ + .dst = _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 -- 1.9.3
On Wed, May 27, 2020 at 08:05:41PM +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 looks really useful, but why is it implemented as an API and not as a real vhost driver which implements an rpmsg bus? If you implement it as a vhost driver which implements rpmsg_device_ops and rpmsg_endpoint_ops, then wouldn't you be able to implement your vhost-sof driver using the normal rpmsg APIs? I tried quickly hooking up this code to such a vhost driver and I was able to communicate between host and guest systems with both rpmsg-client-sample and rpmsg-char which almost no modifications to those drivers.
Hi Vincent, On Wed, Jun 17, 2020 at 09:17:42PM +0200, Vincent Whitchurch wrote:
On Wed, May 27, 2020 at 08:05:41PM +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 looks really useful, but why is it implemented as an API and not as a real vhost driver which implements an rpmsg bus? If you implement it as a vhost driver which implements rpmsg_device_ops and rpmsg_endpoint_ops, then wouldn't you be able to implement your vhost-sof driver using the normal rpmsg APIs?
Sorry, not sure what you mean by the "normal rpmsg API?" Do you mean the VirtIO RPMsg API? But that's the opposite side of the link - that's the guest side in the VM case and the Linux side in the remoteproc case. What this API is adding is a vhost RPMsg API. The kernel vhost framework itself is essentially a library of functions. Kernel vhost drivers simply create a misc device and use the vhost functions for some common functionality. This RPMsg vhost API stays in the same concept and provides further functions for RPMsg specific vhost operation.
I tried quickly hooking up this code to such a vhost driver and I was able to communicate between host and guest systems with both rpmsg-client-sample and rpmsg-char which almost no modifications to those drivers.
You mean you used this patch to create RPMsg vhost drivers? Without creating a vhost RPMsg bus? Nice, glad to hear that! Thanks Guennadi
On Thu, Jun 18, 2020 at 11:03:42AM +0200, Guennadi Liakhovetski wrote:
On Wed, Jun 17, 2020 at 09:17:42PM +0200, Vincent Whitchurch wrote:
On Wed, May 27, 2020 at 08:05:41PM +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 looks really useful, but why is it implemented as an API and not as a real vhost driver which implements an rpmsg bus? If you implement it as a vhost driver which implements rpmsg_device_ops and rpmsg_endpoint_ops, then wouldn't you be able to implement your vhost-sof driver using the normal rpmsg APIs?
Sorry, not sure what you mean by the "normal rpmsg API?" Do you mean the VirtIO RPMsg API? But that's the opposite side of the link - that's the guest side in the VM case and the Linux side in the remoteproc case. What this API is adding is a vhost RPMsg API. The kernel vhost framework itself is essentially a library of functions. Kernel vhost drivers simply create a misc device and use the vhost functions for some common functionality. This RPMsg vhost API stays in the same concept and provides further functions for RPMsg specific vhost operation.
By the "normal rpmsg API" I mean register_rpmsg_driver(), rpmsg_send(), etc. That API is not tied to virtio in any way and there are other non-virtio backends for this API in the tree. So it seems quite natural to implement a vhost backend for this API so that both sides of the link can use the same API but different backends, instead of forcing them to use of different APIs.
I tried quickly hooking up this code to such a vhost driver and I was able to communicate between host and guest systems with both rpmsg-client-sample and rpmsg-char which almost no modifications to those drivers.
You mean you used this patch to create RPMsg vhost drivers? Without creating a vhost RPMsg bus? Nice, glad to hear that!
Not quite, I hacked togther a single generic vhost-rpmsg-bus driver which just wraps the API in this patch and implements a basic rpmsg_device_ops and rpmsg_endpoint_ops. Then with the following patches and no other vhost-specific API use, I was able to load and use the same rpmsg-char and rpmsg-client-sample drivers on both host and guest kernels. Userspace sets up the vhost using vhost-rpmsg-bus' misc device and triggers creation of an rpdev which leads to a probe of the (for example) rpmsg-client-sample driver on the host (server), which, in turn, via NS announcement, triggers a creation of an rpdev and a probe of the rpmsg-client-sample driver on the guest (client). diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c index a76b963a7e5..7a03978d002 100644 --- a/drivers/rpmsg/rpmsg_char.c +++ b/drivers/rpmsg/rpmsg_char.c @@ -104,6 +104,11 @@ static int rpmsg_ept_cb(struct rpmsg_device *rpdev, void *buf, int len, struct rpmsg_eptdev *eptdev = priv; struct sk_buff *skb; + if (rpdev->dst == RPMSG_ADDR_ANY) { + printk("%s: got client address %#x from first rx!\n", __func__, addr); + rpdev->dst = addr; + } + skb = alloc_skb(len, GFP_ATOMIC); if (!skb) return -ENOMEM; @@ -235,6 +240,12 @@ static ssize_t rpmsg_eptdev_write(struct file *filp, const char __user *buf, goto unlock_eptdev; } + if (eptdev->rpdev->dst == RPMSG_ADDR_ANY) { + ret = -EPIPE; + WARN(1, "Cannot write first on server, must wait for client!\n"); + goto unlock_eptdev; + } + if (filp->f_flags & O_NONBLOCK) ret = rpmsg_trysend(eptdev->ept, kbuf, len); else diff --git a/samples/rpmsg/rpmsg_client_sample.c b/samples/rpmsg/rpmsg_client_sample.c index f161dfd3e70..5d8ca84dce0 100644 --- a/samples/rpmsg/rpmsg_client_sample.c +++ b/samples/rpmsg/rpmsg_client_sample.c @@ -46,6 +46,9 @@ static int rpmsg_sample_cb(struct rpmsg_device *rpdev, void *data, int len, return 0; } + if (rpdev->dst == RPMSG_ADDR_ANY) + rpdev->dst = src; + /* send a new message now */ ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); if (ret) @@ -68,11 +71,13 @@ static int rpmsg_sample_probe(struct rpmsg_device *rpdev) dev_set_drvdata(&rpdev->dev, idata); - /* send a message to our remote processor */ - ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); - if (ret) { - dev_err(&rpdev->dev, "rpmsg_send failed: %d\n", ret); - return ret; + if (rpdev->dst != RPMSG_ADDR_ANY) { + /* send a message to our remote processor */ + ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); + if (ret) { + dev_err(&rpdev->dev, "rpmsg_send failed: %d\n", ret); + return ret; + } } return 0;
On Thu, Jun 18, 2020 at 11:33:24AM +0200, Vincent Whitchurch wrote:
On Thu, Jun 18, 2020 at 11:03:42AM +0200, Guennadi Liakhovetski wrote:
On Wed, Jun 17, 2020 at 09:17:42PM +0200, Vincent Whitchurch wrote:
On Wed, May 27, 2020 at 08:05:41PM +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 looks really useful, but why is it implemented as an API and not as a real vhost driver which implements an rpmsg bus? If you implement it as a vhost driver which implements rpmsg_device_ops and rpmsg_endpoint_ops, then wouldn't you be able to implement your vhost-sof driver using the normal rpmsg APIs?
Sorry, not sure what you mean by the "normal rpmsg API?" Do you mean the VirtIO RPMsg API? But that's the opposite side of the link - that's the guest side in the VM case and the Linux side in the remoteproc case. What this API is adding is a vhost RPMsg API. The kernel vhost framework itself is essentially a library of functions. Kernel vhost drivers simply create a misc device and use the vhost functions for some common functionality. This RPMsg vhost API stays in the same concept and provides further functions for RPMsg specific vhost operation.
By the "normal rpmsg API" I mean register_rpmsg_driver(), rpmsg_send(), etc. That API is not tied to virtio in any way and there are other non-virtio backends for this API in the tree. So it seems quite natural to implement a vhost backend for this API so that both sides of the link can use the same API but different backends, instead of forcing them to use of different APIs.
Ok, I see what you mean now. But I'm not sure this is useful or desired. I'm not an expert in KVM / VirtIO, I've only been working in the area for less than a year, so, I might well be wrong. You're proposing to use the rpmsg API in vhost drivers. As far as I understand so far that API was only designated for the Linux side (in case of AMPs) which corresponds to VM guests in virtualisation case. So, I'm not sure we want to use the same API for the hosts? This can be done as you have illustrated, but is it desirable? The vhost API is far enough from the VirtIO driver API, so I'm not sure why we want the same API for rpmsg? Thanks Guennadi
I tried quickly hooking up this code to such a vhost driver and I was able to communicate between host and guest systems with both rpmsg-client-sample and rpmsg-char which almost no modifications to those drivers.
You mean you used this patch to create RPMsg vhost drivers? Without creating a vhost RPMsg bus? Nice, glad to hear that!
Not quite, I hacked togther a single generic vhost-rpmsg-bus driver which just wraps the API in this patch and implements a basic rpmsg_device_ops and rpmsg_endpoint_ops. Then with the following patches and no other vhost-specific API use, I was able to load and use the same rpmsg-char and rpmsg-client-sample drivers on both host and guest kernels.
Userspace sets up the vhost using vhost-rpmsg-bus' misc device and triggers creation of an rpdev which leads to a probe of the (for example) rpmsg-client-sample driver on the host (server), which, in turn, via NS announcement, triggers a creation of an rpdev and a probe of the rpmsg-client-sample driver on the guest (client).
diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c index a76b963a7e5..7a03978d002 100644 --- a/drivers/rpmsg/rpmsg_char.c +++ b/drivers/rpmsg/rpmsg_char.c @@ -104,6 +104,11 @@ static int rpmsg_ept_cb(struct rpmsg_device *rpdev, void *buf, int len, struct rpmsg_eptdev *eptdev = priv; struct sk_buff *skb;
+ if (rpdev->dst == RPMSG_ADDR_ANY) { + printk("%s: got client address %#x from first rx!\n", __func__, addr); + rpdev->dst = addr; + } + skb = alloc_skb(len, GFP_ATOMIC); if (!skb) return -ENOMEM; @@ -235,6 +240,12 @@ static ssize_t rpmsg_eptdev_write(struct file *filp, const char __user *buf, goto unlock_eptdev; }
+ if (eptdev->rpdev->dst == RPMSG_ADDR_ANY) { + ret = -EPIPE; + WARN(1, "Cannot write first on server, must wait for client!\n"); + goto unlock_eptdev; + } + if (filp->f_flags & O_NONBLOCK) ret = rpmsg_trysend(eptdev->ept, kbuf, len); else diff --git a/samples/rpmsg/rpmsg_client_sample.c b/samples/rpmsg/rpmsg_client_sample.c index f161dfd3e70..5d8ca84dce0 100644 --- a/samples/rpmsg/rpmsg_client_sample.c +++ b/samples/rpmsg/rpmsg_client_sample.c @@ -46,6 +46,9 @@ static int rpmsg_sample_cb(struct rpmsg_device *rpdev, void *data, int len, return 0; }
+ if (rpdev->dst == RPMSG_ADDR_ANY) + rpdev->dst = src; + /* send a new message now */ ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); if (ret) @@ -68,11 +71,13 @@ static int rpmsg_sample_probe(struct rpmsg_device *rpdev)
dev_set_drvdata(&rpdev->dev, idata);
- /* send a message to our remote processor */ - ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); - if (ret) { - dev_err(&rpdev->dev, "rpmsg_send failed: %d\n", ret); - return ret; + if (rpdev->dst != RPMSG_ADDR_ANY) { + /* send a message to our remote processor */ + ret = rpmsg_send(rpdev->ept, MSG, strlen(MSG)); + if (ret) { + dev_err(&rpdev->dev, "rpmsg_send failed: %d\n", ret); + return ret; + } }
return 0;
On Thu, Jun 18, 2020 at 12:39:40PM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 18, 2020 at 11:33:24AM +0200, Vincent Whitchurch wrote:
By the "normal rpmsg API" I mean register_rpmsg_driver(), rpmsg_send(), etc. That API is not tied to virtio in any way and there are other non-virtio backends for this API in the tree. So it seems quite natural to implement a vhost backend for this API so that both sides of the link can use the same API but different backends, instead of forcing them to use of different APIs.
Ok, I see what you mean now. But I'm not sure this is useful or desired. I'm not an expert in KVM / VirtIO, I've only been working in the area for less than a year, so, I might well be wrong.
You're proposing to use the rpmsg API in vhost drivers. As far as I understand so far that API was only designated for the Linux side (in case of AMPs) which corresponds to VM guests in virtualisation case. So, I'm not sure we want to use the same API for the hosts? This can be done as you have illustrated, but is it desirable? The vhost API is far enough from the VirtIO driver API, so I'm not sure why we want the same API for rpmsg?
Note that "the Linux side" is ambiguous for AMP since both sides can be Linux, as they happen to be in my case. I'm running virtio/rpmsg between two physical processors (of different architectures), both running Linux. virtio has distinct driver and device roles so the completely different APIs on each side are understandable. But I don't see that distinction in the rpmsg API which is why it seems like a good idea to me to make it work from both sides of the link and allow the reuse of drivers like rpmsg-char, instead of imposing virtio's distinction on rpmsg.
On Thu, Jun 18, 2020 at 03:52:42PM +0200, Vincent Whitchurch wrote:
On Thu, Jun 18, 2020 at 12:39:40PM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 18, 2020 at 11:33:24AM +0200, Vincent Whitchurch wrote:
By the "normal rpmsg API" I mean register_rpmsg_driver(), rpmsg_send(), etc. That API is not tied to virtio in any way and there are other non-virtio backends for this API in the tree. So it seems quite natural to implement a vhost backend for this API so that both sides of the link can use the same API but different backends, instead of forcing them to use of different APIs.
Ok, I see what you mean now. But I'm not sure this is useful or desired. I'm not an expert in KVM / VirtIO, I've only been working in the area for less than a year, so, I might well be wrong.
You're proposing to use the rpmsg API in vhost drivers. As far as I understand so far that API was only designated for the Linux side (in case of AMPs) which corresponds to VM guests in virtualisation case. So, I'm not sure we want to use the same API for the hosts? This can be done as you have illustrated, but is it desirable? The vhost API is far enough from the VirtIO driver API, so I'm not sure why we want the same API for rpmsg?
Note that "the Linux side" is ambiguous for AMP since both sides can be Linux, as they happen to be in my case. I'm running virtio/rpmsg between two physical processors (of different architectures), both running Linux.
Ok, interesting, I didn't know such configurations were used too. I understood the Linux rpmsg implementation in the way, that it's assumed, that the "host" has to boot the "device" by sending an ELF formatted executable image to it, is that optional? You aren't sending a complete Linux image to the device side, are you?
virtio has distinct driver and device roles so the completely different APIs on each side are understandable. But I don't see that distinction in the rpmsg API which is why it seems like a good idea to me to make it work from both sides of the link and allow the reuse of drivers like rpmsg-char, instead of imposing virtio's distinction on rpmsg.
Understand. In principle I'm open to this idea, but before I implement it it would be good to know what maintainers think? Thanks Guennadi
On Thu, Jun 18, 2020 at 04:14:12PM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 18, 2020 at 03:52:42PM +0200, Vincent Whitchurch wrote:
Note that "the Linux side" is ambiguous for AMP since both sides can be Linux, as they happen to be in my case. I'm running virtio/rpmsg between two physical processors (of different architectures), both running Linux.
Ok, interesting, I didn't know such configurations were used too. I understood the Linux rpmsg implementation in the way, that it's assumed, that the "host" has to boot the "device" by sending an ELF formatted executable image to it, is that optional? You aren't sending a complete Linux image to the device side, are you?
I do pack the zImage, the dtb, and the initramfs into an ELF (along with a tiny "bootloader" with just a handful of instructions), but the remoteproc framework is not tied to the ELF format since ->parse_fw() and friends are overridable by the remoteproc driver.
virtio has distinct driver and device roles so the completely different APIs on each side are understandable. But I don't see that distinction in the rpmsg API which is why it seems like a good idea to me to make it work from both sides of the link and allow the reuse of drivers like rpmsg-char, instead of imposing virtio's distinction on rpmsg.
Understand. In principle I'm open to this idea, but before I implement it it would be good to know what maintainers think?
Certainly.
On 2020/5/28 上午2:05, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
Hi: It would be hard to evaluate this series without a real user. So if possible, I suggest to post the actual user for vhost rpmsg API. Thanks
Thanks Guennadi
Hi Jason, On Fri, May 29, 2020 at 02:01:53PM +0800, Jason Wang wrote:
On 2020/5/28 上午2:05, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
Hi:
It would be hard to evaluate this series without a real user. So if possible, I suggest to post the actual user for vhost rpmsg API.
Sure, the whole series is available at https://github.com/thesofproject/linux/pull/1501/commits or would you prefer the missing patches posted to the lists too? Thanks Guennadi
On 2020/5/29 下午2:50, Guennadi Liakhovetski wrote:
Hi Jason,
On 2020/5/28 上午2:05, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
Hi:
It would be hard to evaluate this series without a real user. So if possible, I suggest to post the actual user for vhost rpmsg API. Sure, the whole series is available at https://github.com/thesofproject/linux/pull/1501/commits or would you
On Fri, May 29, 2020 at 02:01:53PM +0800, Jason Wang wrote: prefer the missing patches posted to the lists too?
Yes, since I see new virtio and vhost drivers there. Thanks
Thanks Guennadi
On Wed, May 27, 2020 at 08:05:36PM +0200, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
RPMsg over virtio has several problems. One is that it's not specced at all. Before we add more stuff, I'd like so see at least an attempt at describing what it's supposed to do. Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert. It's great to see it's seeing active development finally. Do you think you will have time to address these?
Thanks Guennadi
Guennadi Liakhovetski (5): vhost: convert VHOST_VSOCK_SET_RUNNING to a generic ioctl vhost: (cosmetic) remove a superfluous variable initialisation rpmsg: move common structures and defines to headers rpmsg: update documentation vhost: add an RPMsg API
Documentation/rpmsg.txt | 6 +- drivers/rpmsg/virtio_rpmsg_bus.c | 78 +------- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 382 +++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.c | 2 +- drivers/vhost/vhost_rpmsg.h | 74 ++++++++ include/linux/virtio_rpmsg.h | 81 +++++++++ include/uapi/linux/rpmsg.h | 3 + include/uapi/linux/vhost.h | 4 +- 10 files changed, 559 insertions(+), 81 deletions(-) create mode 100644 drivers/vhost/rpmsg.c create mode 100644 drivers/vhost/vhost_rpmsg.h create mode 100644 include/linux/virtio_rpmsg.h
-- 1.9.3
Hi Michael, Thanks for your review. On Thu, Jun 04, 2020 at 03:23:37PM -0400, Michael S. Tsirkin wrote:
On Wed, May 27, 2020 at 08:05:36PM +0200, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
RPMsg over virtio has several problems. One is that it's not specced at all. Before we add more stuff, I'd like so see at least an attempt at describing what it's supposed to do.
Sure, I can work on this with the original authors of the virtio-rpmsg implementation.
Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert.
Yes, I have to fix this, thanks.
It's great to see it's seeing active development finally. Do you think you will have time to address these?
Sure, I'll try to take care of them. Thanks Guennadi
Guennadi Liakhovetski (5): vhost: convert VHOST_VSOCK_SET_RUNNING to a generic ioctl vhost: (cosmetic) remove a superfluous variable initialisation rpmsg: move common structures and defines to headers rpmsg: update documentation vhost: add an RPMsg API
Documentation/rpmsg.txt | 6 +- drivers/rpmsg/virtio_rpmsg_bus.c | 78 +------- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 382 +++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.c | 2 +- drivers/vhost/vhost_rpmsg.h | 74 ++++++++ include/linux/virtio_rpmsg.h | 81 +++++++++ include/uapi/linux/rpmsg.h | 3 + include/uapi/linux/vhost.h | 4 +- 10 files changed, 559 insertions(+), 81 deletions(-) create mode 100644 drivers/vhost/rpmsg.c create mode 100644 drivers/vhost/vhost_rpmsg.h create mode 100644 include/linux/virtio_rpmsg.h
-- 1.9.3
Hi Michael, On Fri, Jun 05, 2020 at 08:34:35AM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 04, 2020 at 03:23:37PM -0400, Michael S. Tsirkin wrote:
[snip]
Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert.
Yes, I have to fix this, thanks.
Just to make sure my understanding is correct: this would involve also modifying the current virtio_rpmsg_bus.c implementation to add endianness conversions. That's what you meant, right? Thanks Guennadi
On Mon, Jun 08, 2020 at 09:37:15AM +0200, Guennadi Liakhovetski wrote:
Hi Michael,
On Fri, Jun 05, 2020 at 08:34:35AM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 04, 2020 at 03:23:37PM -0400, Michael S. Tsirkin wrote:
[snip]
Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert.
Yes, I have to fix this, thanks.
Just to make sure my understanding is correct: this would involve also modifying the current virtio_rpmsg_bus.c implementation to add endianness conversions. That's what you meant, right?
Thanks Guennadi
right and if there are legacy compat considerations, using _virtio16 and friends types, as well as virtio16_to_cpu and friends functions. -- MST
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet. Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data. So I think the same should be done for RPMsg instead of enforcing LE? Thanks Guennadi On Mon, Jun 08, 2020 at 09:37:15AM +0200, Guennadi Liakhovetski wrote:
Hi Michael,
On Fri, Jun 05, 2020 at 08:34:35AM +0200, Guennadi Liakhovetski wrote:
On Thu, Jun 04, 2020 at 03:23:37PM -0400, Michael S. Tsirkin wrote:
[snip]
Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert.
Yes, I have to fix this, thanks.
Just to make sure my understanding is correct: this would involve also modifying the current virtio_rpmsg_bus.c implementation to add endianness conversions. That's what you meant, right?
On Mon, Jun 08, 2020 at 11:11:00AM +0200, Guennadi Liakhovetski wrote:
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet.
Yes. And it would be very good to have some standartization before we keep adding things. For example without any spec if host code breaks with some guests, how do we know which side should be fixed?
Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data.
They really don't or shouldn't. That's limited to legacy chapters. Some definitions could have slipped through but it's not the norm. I just quickly looked through the 1.1 spec and could not find any instances that specify "guest native endianness" but feel free to point them out to me.
So I think the same should be done for RPMsg instead of enforcing LE?
Thanks Guennadi
That makes hardware implementations as well as any cross-endian hypervisors tricky. -- MST
On Mon, Jun 08, 2020 at 05:19:06AM -0400, Michael S. Tsirkin wrote:
On Mon, Jun 08, 2020 at 11:11:00AM +0200, Guennadi Liakhovetski wrote:
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet.
Yes. And it would be very good to have some standartization before we keep adding things. For example without any spec if host code breaks with some guests, how do we know which side should be fixed?
Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data.
They really don't or shouldn't. That's limited to legacy chapters. Some definitions could have slipped through but it's not the norm. I just quickly looked through the 1.1 spec and could not find any instances that specify "guest native endianness" but feel free to point them out to me.
Oh, there you go. No, sorry, my fault, it's the other way round: "guest native" is for legacy and LE is for current / v1.0 and up.
So I think the same should be done for RPMsg instead of enforcing LE?
That makes hardware implementations as well as any cross-endian hypervisors tricky.
Yes, LE it is then. And we need to add some text to the spec. In theory there could be a backward compatibility issue - in case someone was already using virtio_rpmsg_bus.c in BE mode, but I very much doubt that... Thanks Guennadi
On Mon, Jun 08, 2020 at 12:15:26PM +0200, Guennadi Liakhovetski wrote:
On Mon, Jun 08, 2020 at 05:19:06AM -0400, Michael S. Tsirkin wrote:
On Mon, Jun 08, 2020 at 11:11:00AM +0200, Guennadi Liakhovetski wrote:
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet.
Yes. And it would be very good to have some standartization before we keep adding things. For example without any spec if host code breaks with some guests, how do we know which side should be fixed?
Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data.
They really don't or shouldn't. That's limited to legacy chapters. Some definitions could have slipped through but it's not the norm. I just quickly looked through the 1.1 spec and could not find any instances that specify "guest native endianness" but feel free to point them out to me.
Oh, there you go. No, sorry, my fault, it's the other way round: "guest native" is for legacy and LE is for current / v1.0 and up.
So I think the same should be done for RPMsg instead of enforcing LE?
That makes hardware implementations as well as any cross-endian hypervisors tricky.
Yes, LE it is then. And we need to add some text to the spec.
I found the protocol and the message format definition: https://github.com/OpenAMP/open-amp/wiki/RPMsg-Messaging-Protocol#transport-... Don't know what the best way for referencing it in the VirtIO standard would be: just a link to the source or a quote. Thanks Guennadi
On Mon, Jun 08, 2020 at 01:16:38PM +0200, Guennadi Liakhovetski wrote:
On Mon, Jun 08, 2020 at 12:15:26PM +0200, Guennadi Liakhovetski wrote:
On Mon, Jun 08, 2020 at 05:19:06AM -0400, Michael S. Tsirkin wrote:
On Mon, Jun 08, 2020 at 11:11:00AM +0200, Guennadi Liakhovetski wrote:
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet.
Yes. And it would be very good to have some standartization before we keep adding things. For example without any spec if host code breaks with some guests, how do we know which side should be fixed?
Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data.
They really don't or shouldn't. That's limited to legacy chapters. Some definitions could have slipped through but it's not the norm. I just quickly looked through the 1.1 spec and could not find any instances that specify "guest native endianness" but feel free to point them out to me.
Oh, there you go. No, sorry, my fault, it's the other way round: "guest native" is for legacy and LE is for current / v1.0 and up.
So I think the same should be done for RPMsg instead of enforcing LE?
That makes hardware implementations as well as any cross-endian hypervisors tricky.
Yes, LE it is then. And we need to add some text to the spec.
I found the protocol and the message format definition: https://github.com/OpenAMP/open-amp/wiki/RPMsg-Messaging-Protocol#transport-... Don't know what the best way for referencing it in the VirtIO standard would be: just a link to the source or a quote.
Thanks Guennadi
I wasn't aware of that one, thanks! OK so that's good. Ideally we'd have RPMsg Header Definition, RPMsg Channel and RPMsg Endppint in the spec proper. This link is informal so can't be copied into spec as is but can be used as a basis. We'd also need approval from authors for inclusion in the spec, sent to the TC mailing list. -- MST
On Mon, Jun 08, 2020 at 12:15:27PM +0200, Guennadi Liakhovetski wrote:
On Mon, Jun 08, 2020 at 05:19:06AM -0400, Michael S. Tsirkin wrote:
On Mon, Jun 08, 2020 at 11:11:00AM +0200, Guennadi Liakhovetski wrote:
Update: I looked through VirtIO 1.0 and 1.1 specs, data format their, including byte order, is defined on a per-device type basis. RPMsg is indeed included in the spec as device type 7, but that's the only mention of it in both versions. It seems RPMsg over VirtIO isn't standardised yet.
Yes. And it would be very good to have some standartization before we keep adding things. For example without any spec if host code breaks with some guests, how do we know which side should be fixed?
Also it looks like newer interface definitions specify using "guest native endianness" for Virtual Queue data.
They really don't or shouldn't. That's limited to legacy chapters. Some definitions could have slipped through but it's not the norm. I just quickly looked through the 1.1 spec and could not find any instances that specify "guest native endianness" but feel free to point them out to me.
Oh, there you go. No, sorry, my fault, it's the other way round: "guest native" is for legacy and LE is for current / v1.0 and up.
So I think the same should be done for RPMsg instead of enforcing LE?
That makes hardware implementations as well as any cross-endian hypervisors tricky.
Yes, LE it is then. And we need to add some text to the spec.
In theory there could be a backward compatibility issue - in case someone was already using virtio_rpmsg_bus.c in BE mode, but I very much doubt that...
Thanks Guennadi
It's probably easiest to use virtio wrappers and then we don't need to worry about it. -- MST
On Thu, 2020-06-04 at 15:23 -0400, Michael S. Tsirkin wrote:
On Wed, May 27, 2020 at 08:05:36PM +0200, Guennadi Liakhovetski wrote:
v3: - address several checkpatch warnings - address comments from Mathieu Poirier
v2: - update patch #5 with a correct vhost_dev_init() prototype - drop patch #6 - it depends on a different patch, that is currently an RFC - address comments from Pierre-Louis Bossart: * remove "default n" from Kconfig
Linux supports RPMsg over VirtIO for "remote processor" /AMP use cases. It can however also be used for virtualisation scenarios, e.g. when using KVM to run Linux on both the host and the guests. This patch set adds a wrapper API to facilitate writing vhost drivers for such RPMsg-based solutions. The first use case is an audio DSP virtualisation project, currently under development, ready for review and submission, available at https://github.com/thesofproject/linux/pull/1501/commits A further patch for the ADSP vhost RPMsg driver will be sent separately for review only since it cannot be merged without audio patches being upstreamed first.
RPMsg over virtio has several problems. One is that it's not specced at all. Before we add more stuff, I'd like so see at least an attempt at describing what it's supposed to do.
Sure, I'll add some more context here. The remote processor in this use case is any DSP (from any vendor) running SOF. The work from Guennadi virtualises the SOF mailbox and SOF doorbell mechanisms (which the platform driver abstracts) via rpmsg/virtio so the guest SOF drivers can send and receive SOF IPCs (just as the host SOF driver would do). It's 95% the same Linux driver on host or guest (for each feature). I would also add here (and it's maybe confusing in the SOF naming) that SOF is multi a feature FW, it's not just an audio FW, so we would also expect to see other guest drivers (e.g. sensing) that would use the same mechanism for IPC on guests. I would expect the feature driver count to increase as the FW features grow. The IPC ABI between the FW and host drivers continually evolves as features and new HW is added (not just from Intel, but from other SOF partners and external partners that supply proprietary audio processing). The only part of the interface that is specced is the rpmsg header, as the SOF message content will keep evolving (it's up to driver and FW to align on ABI version used - it does this already today). I guess it boils down to two goals here 1) virtualising the SOF features on any platform/guest/OS so that guests would be able to access any FW feature (provided guest was permitted) just as they would on host. 2) Supporting FW features and use cases from multiple parties without having to change driver core or driver virtualisation core. i.e. all the changes (for new features) would be in the edge drivers e.g. new audio features would impact audio driver only.
Another it's out of line with 1.0 spec passing guest endian data around. Won't work if host and guest endian-ness do not match. Should pass eveything in LE and convert.
I think Guennadi is working on this now.
It's great to see it's seeing active development finally. Do you think you will have time to address these?
Yes, of course. Let me know if you need any more background or context. Thanks Liam
Thanks Guennadi
Guennadi Liakhovetski (5): vhost: convert VHOST_VSOCK_SET_RUNNING to a generic ioctl vhost: (cosmetic) remove a superfluous variable initialisation rpmsg: move common structures and defines to headers rpmsg: update documentation vhost: add an RPMsg API
Documentation/rpmsg.txt | 6 +- drivers/rpmsg/virtio_rpmsg_bus.c | 78 +------- drivers/vhost/Kconfig | 7 + drivers/vhost/Makefile | 3 + drivers/vhost/rpmsg.c | 382 +++++++++++++++++++++++++++++++++++++++ drivers/vhost/vhost.c | 2 +- drivers/vhost/vhost_rpmsg.h | 74 ++++++++ include/linux/virtio_rpmsg.h | 81 +++++++++ include/uapi/linux/rpmsg.h | 3 + include/uapi/linux/vhost.h | 4 +- 10 files changed, 559 insertions(+), 81 deletions(-) create mode 100644 drivers/vhost/rpmsg.c create mode 100644 drivers/vhost/vhost_rpmsg.h create mode 100644 include/linux/virtio_rpmsg.h
-- 1.9.3
_______________________________________________ Sound-open-firmware mailing list Sound-open-firmware@alsa-project.org https://mailman.alsa-project.org/mailman/listinfo/sound-open-firmware
participants (6)
-
Guennadi Liakhovetski -
Jason Wang -
Liam Girdwood -
Mathieu Poirier -
Michael S. Tsirkin -
Vincent Whitchurch