[PATCH v4 13/26] vhost: add API to set max queue pairs
Xia, Chenbo
chenbo.xia at intel.com
Mon Jun 5 09:56:23 CEST 2023
> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin at redhat.com>
> Sent: Friday, June 2, 2023 4:08 AM
> To: dev at dpdk.org; Xia, Chenbo <chenbo.xia at intel.com>;
> david.marchand at redhat.com; mkp at redhat.com; fbl at redhat.com;
> jasowang at redhat.com; Liang, Cunming <cunming.liang at intel.com>; Xie, Yongji
> <xieyongji at bytedance.com>; echaudro at redhat.com; eperezma at redhat.com;
> amorenoz at redhat.com; lulu at redhat.com
> Cc: Maxime Coquelin <maxime.coquelin at redhat.com>
> Subject: [PATCH v4 13/26] vhost: add API to set max queue pairs
>
> This patch introduces a new rte_vhost_driver_set_max_queues
> API as preliminary work for multiqueue support with VDUSE.
>
> Indeed, with VDUSE we need to pre-allocate the vrings at
> device creation time, so we need such API not to allocate
> the 128 queue pairs supported by the Vhost library.
>
> Calling the API is optional, 128 queue pairs remaining the
> default.
>
> Signed-off-by: Maxime Coquelin <maxime.coquelin at redhat.com>
> ---
> doc/guides/prog_guide/vhost_lib.rst | 4 +++
> doc/guides/rel_notes/release_23_07.rst | 5 ++++
> lib/vhost/rte_vhost.h | 17 ++++++++++++
> lib/vhost/socket.c | 36 ++++++++++++++++++++++++--
> lib/vhost/version.map | 1 +
> 5 files changed, 61 insertions(+), 2 deletions(-)
>
> diff --git a/doc/guides/prog_guide/vhost_lib.rst
> b/doc/guides/prog_guide/vhost_lib.rst
> index 0f964d7a4a..0c2b4d020a 100644
> --- a/doc/guides/prog_guide/vhost_lib.rst
> +++ b/doc/guides/prog_guide/vhost_lib.rst
> @@ -339,6 +339,10 @@ The following is an overview of some key Vhost API
> functions:
> Inject the offloaded interrupt received by the 'guest_notify' callback,
> into the vhost device's queue.
>
> +* ``rte_vhost_driver_set_max_queue_num(const char *path, uint32_t
> max_queue_pairs)``
> +
> + Set the maximum number of queue pairs supported by the device.
> +
> Vhost-user Implementations
> --------------------------
>
> diff --git a/doc/guides/rel_notes/release_23_07.rst
> b/doc/guides/rel_notes/release_23_07.rst
> index 3eed8ac561..7034fb664c 100644
> --- a/doc/guides/rel_notes/release_23_07.rst
> +++ b/doc/guides/rel_notes/release_23_07.rst
> @@ -62,6 +62,11 @@ New Features
> rte_vhost_notify_guest(), is added to raise the interrupt outside of
> the
> fast path.
>
> +* **Added Vhost API to set maximum queue pairs supported.**
> +
> + Introduced ``rte_vhost_driver_set_max_queue_num()`` to be able to limit
> the
> + maximum number of supported queue pairs, required for VDUSE support.
> +
>
> Removed Items
> -------------
> diff --git a/lib/vhost/rte_vhost.h b/lib/vhost/rte_vhost.h
> index 7a10bc36cf..7844c9f142 100644
> --- a/lib/vhost/rte_vhost.h
> +++ b/lib/vhost/rte_vhost.h
> @@ -609,6 +609,23 @@ rte_vhost_driver_get_protocol_features(const char
> *path,
> int
> rte_vhost_driver_get_queue_num(const char *path, uint32_t *queue_num);
>
> +/**
> + * @warning
> + * @b EXPERIMENTAL: this API may change, or be removed, without prior
> notice.
> + *
> + * Set the maximum number of queue pairs supported by the device.
> + *
> + * @param path
> + * The vhost-user socket file path
> + * @param max_queue_pairs
> + * The maximum number of queue pairs
> + * @return
> + * 0 on success, -1 on failure
> + */
> +__rte_experimental
> +int
> +rte_vhost_driver_set_max_queue_num(const char *path, uint32_t
> max_queue_pairs);
> +
> /**
> * Get the feature bits after negotiation
> *
> diff --git a/lib/vhost/socket.c b/lib/vhost/socket.c
> index 407d0011c3..29f7a8cece 100644
> --- a/lib/vhost/socket.c
> +++ b/lib/vhost/socket.c
> @@ -57,6 +57,8 @@ struct vhost_user_socket {
>
> uint64_t protocol_features;
>
> + uint32_t max_queue_pairs;
> +
> struct rte_vdpa_device *vdpa_dev;
>
> struct rte_vhost_device_ops const *notify_ops;
> @@ -823,7 +825,7 @@ rte_vhost_driver_get_queue_num(const char *path,
> uint32_t *queue_num)
>
> vdpa_dev = vsocket->vdpa_dev;
> if (!vdpa_dev) {
> - *queue_num = VHOST_MAX_QUEUE_PAIRS;
> + *queue_num = vsocket->max_queue_pairs;
> goto unlock_exit;
> }
>
> @@ -833,7 +835,36 @@ rte_vhost_driver_get_queue_num(const char *path,
> uint32_t *queue_num)
> goto unlock_exit;
> }
>
> - *queue_num = RTE_MIN((uint32_t)VHOST_MAX_QUEUE_PAIRS,
> vdpa_queue_num);
> + *queue_num = RTE_MIN(vsocket->max_queue_pairs, vdpa_queue_num);
> +
> +unlock_exit:
> + pthread_mutex_unlock(&vhost_user.mutex);
> + return ret;
> +}
> +
> +int
> +rte_vhost_driver_set_max_queue_num(const char *path, uint32_t
> max_queue_pairs)
> +{
> + struct vhost_user_socket *vsocket;
> + int ret = 0;
> +
> + VHOST_LOG_CONFIG(path, INFO, "Setting max queue pairs to %u\n",
> max_queue_pairs);
> +
> + if (max_queue_pairs > VHOST_MAX_QUEUE_PAIRS) {
> + VHOST_LOG_CONFIG(path, ERR, "Library only supports up to %u
> queue pairs\n",
> + VHOST_MAX_QUEUE_PAIRS);
> + return -1;
> + }
> +
> + pthread_mutex_lock(&vhost_user.mutex);
> + vsocket = find_vhost_user_socket(path);
> + if (!vsocket) {
> + VHOST_LOG_CONFIG(path, ERR, "socket file is not registered
> yet.\n");
> + ret = -1;
> + goto unlock_exit;
> + }
> +
> + vsocket->max_queue_pairs = max_queue_pairs;
>
> unlock_exit:
> pthread_mutex_unlock(&vhost_user.mutex);
> @@ -889,6 +920,7 @@ rte_vhost_driver_register(const char *path, uint64_t
> flags)
> goto out_free;
> }
> vsocket->vdpa_dev = NULL;
> + vsocket->max_queue_pairs = VHOST_MAX_QUEUE_PAIRS;
> vsocket->extbuf = flags & RTE_VHOST_USER_EXTBUF_SUPPORT;
> vsocket->linearbuf = flags & RTE_VHOST_USER_LINEARBUF_SUPPORT;
> vsocket->async_copy = flags & RTE_VHOST_USER_ASYNC_COPY;
> diff --git a/lib/vhost/version.map b/lib/vhost/version.map
> index 7bcbfd12cf..5051c29022 100644
> --- a/lib/vhost/version.map
> +++ b/lib/vhost/version.map
> @@ -107,6 +107,7 @@ EXPERIMENTAL {
>
> # added in 23.07
> rte_vhost_notify_guest;
> + rte_vhost_driver_set_max_queue_num;
> };
>
> INTERNAL {
> --
> 2.40.1
Reviewed-by: Chenbo Xia <chenbo.xia at intel.com>
More information about the dev
mailing list