From: "Xia, Chenbo" <chenbo.xia@intel.com>
To: "Coquelin, Maxime" <maxime.coquelin@redhat.com>,
"dev@dpdk.org" <dev@dpdk.org>,
"david.marchand@redhat.com" <david.marchand@redhat.com>,
"eperezma@redhat.com" <eperezma@redhat.com>
Subject: RE: [PATCH v1 21/21] net/virtio-user: remove max queues limitation
Date: Tue, 31 Jan 2023 05:19:42 +0000 [thread overview]
Message-ID: <SN6PR11MB35046D1A8D2855D2A808D0FD9CD09@SN6PR11MB3504.namprd11.prod.outlook.com> (raw)
In-Reply-To: <20221130155639.150553-22-maxime.coquelin@redhat.com>
Hi Maxime,
> -----Original Message-----
> From: Maxime Coquelin <maxime.coquelin@redhat.com>
> Sent: Wednesday, November 30, 2022 11:57 PM
> To: dev@dpdk.org; Xia, Chenbo <chenbo.xia@intel.com>;
> david.marchand@redhat.com; eperezma@redhat.com
> Cc: Maxime Coquelin <maxime.coquelin@redhat.com>
> Subject: [PATCH v1 21/21] net/virtio-user: remove max queues limitation
>
> This patch removes the limitation of 8 queue pairs by
> dynamically allocating vring metadata once we know the
> maximum number of queue pairs supported by the backend.
>
> This is especially useful for Vhost-vDPA with physical
> devices, where the maximum queues supported may be much
> more than 8 pairs.
>
> Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
> drivers/net/virtio/virtio.h | 6 -
> .../net/virtio/virtio_user/virtio_user_dev.c | 118 ++++++++++++++----
> .../net/virtio/virtio_user/virtio_user_dev.h | 16 +--
> drivers/net/virtio/virtio_user_ethdev.c | 17 +--
> 4 files changed, 109 insertions(+), 48 deletions(-)
>
> diff --git a/drivers/net/virtio/virtio.h b/drivers/net/virtio/virtio.h
> index 5c8f71a44d..04a897bf51 100644
> --- a/drivers/net/virtio/virtio.h
> +++ b/drivers/net/virtio/virtio.h
> @@ -124,12 +124,6 @@
> VIRTIO_NET_HASH_TYPE_UDP_EX)
>
>
> -/*
> - * Maximum number of virtqueues per device.
> - */
> -#define VIRTIO_MAX_VIRTQUEUE_PAIRS 8
> -#define VIRTIO_MAX_VIRTQUEUES (VIRTIO_MAX_VIRTQUEUE_PAIRS * 2 + 1)
> -
> /* VirtIO device IDs. */
> #define VIRTIO_ID_NETWORK 0x01
> #define VIRTIO_ID_BLOCK 0x02
> diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> index 7c48c9bb29..aa24fdea70 100644
> --- a/drivers/net/virtio/virtio_user/virtio_user_dev.c
> +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.c
> @@ -17,6 +17,7 @@
> #include <rte_alarm.h>
> #include <rte_string_fns.h>
> #include <rte_eal_memconfig.h>
> +#include <rte_malloc.h>
>
> #include "vhost.h"
> #include "virtio_user_dev.h"
> @@ -58,8 +59,8 @@ virtio_user_kick_queue(struct virtio_user_dev *dev,
> uint32_t queue_sel)
> int ret;
> struct vhost_vring_file file;
> struct vhost_vring_state state;
> - struct vring *vring = &dev->vrings[queue_sel];
> - struct vring_packed *pq_vring = &dev->packed_vrings[queue_sel];
> + struct vring *vring = &dev->vrings.split[queue_sel];
> + struct vring_packed *pq_vring = &dev->vrings.packed[queue_sel];
> struct vhost_vring_addr addr = {
> .index = queue_sel,
> .log_guest_addr = 0,
> @@ -299,18 +300,6 @@ virtio_user_dev_init_max_queue_pairs(struct
> virtio_user_dev *dev, uint32_t user_
> return ret;
> }
>
> - if (dev->max_queue_pairs > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
> - /*
> - * If the device supports control queue, the control queue
> - * index is max_virtqueue_pairs * 2. Disable MQ if it happens.
> - */
> - PMD_DRV_LOG(ERR, "(%s) Device advertises too many queues (%u,
> max supported %u)",
> - dev->path, dev->max_queue_pairs,
> VIRTIO_MAX_VIRTQUEUE_PAIRS);
> - dev->max_queue_pairs = 1;
> -
> - return -1;
> - }
> -
> return 0;
> }
>
> @@ -579,6 +568,86 @@ virtio_user_dev_setup(struct virtio_user_dev *dev)
> return 0;
> }
>
> +static int
> +virtio_user_alloc_vrings(struct virtio_user_dev *dev)
> +{
> + int i, size, nr_vrings;
> +
> + nr_vrings = dev->max_queue_pairs * 2;
> + if (dev->hw_cvq)
> + nr_vrings++;
> +
> + dev->callfds = rte_zmalloc("virtio_user_dev", nr_vrings *
> sizeof(*dev->callfds), 0);
> + if (!dev->callfds) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to alloc callfds", dev->path);
> + return -1;
> + }
> +
> + dev->kickfds = rte_zmalloc("virtio_user_dev", nr_vrings *
> sizeof(*dev->kickfds), 0);
> + if (!dev->kickfds) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to alloc kickfds", dev->path);
> + goto free_callfds;
> + }
> +
> + for (i = 0; i < nr_vrings; i++) {
> + dev->callfds[i] = -1;
> + dev->kickfds[i] = -1;
> + }
> +
> + size = RTE_MAX(sizeof(*dev->vrings.split), sizeof(*dev-
> >vrings.packed));
> + dev->vrings.ptr = rte_zmalloc("virtio_user_dev", nr_vrings * size,
> 0);
> + if (!dev->vrings.ptr) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to alloc vrings metadata", dev-
> >path);
> + goto free_kickfds;
> + }
> +
> + dev->packed_queues = rte_zmalloc("virtio_user_dev",
> + nr_vrings * sizeof(*dev->packed_queues), 0);
Should we pass the info of packed vq or not to save the alloc of
dev->packed_queues, also to know correct size of dev->vrings.ptr.
Thanks,
Chenbo
> + if (!dev->packed_queues) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to alloc packed queues
> metadata", dev->path);
> + goto free_vrings;
> + }
> +
> + dev->qp_enabled = rte_zmalloc("virtio_user_dev",
> + dev->max_queue_pairs * sizeof(*dev->qp_enabled), 0);
> + if (!dev->qp_enabled) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to alloc QP enable states",
> dev->path);
> + goto free_packed_queues;
> + }
> +
> + return 0;
> +
> +free_packed_queues:
> + rte_free(dev->packed_queues);
> + dev->packed_queues = NULL;
> +free_vrings:
> + rte_free(dev->vrings.ptr);
> + dev->vrings.ptr = NULL;
> +free_kickfds:
> + rte_free(dev->kickfds);
> + dev->kickfds = NULL;
> +free_callfds:
> + rte_free(dev->callfds);
> + dev->callfds = NULL;
> +
> + return -1;
> +}
> +
> +static void
> +virtio_user_free_vrings(struct virtio_user_dev *dev)
> +{
> + rte_free(dev->qp_enabled);
> + dev->qp_enabled = NULL;
> + rte_free(dev->packed_queues);
> + dev->packed_queues = NULL;
> + rte_free(dev->vrings.ptr);
> + dev->vrings.ptr = NULL;
> + rte_free(dev->kickfds);
> + dev->kickfds = NULL;
> + rte_free(dev->callfds);
> + dev->callfds = NULL;
> +}
> +
> /* Use below macro to filter features from vhost backend */
> #define VIRTIO_USER_SUPPORTED_FEATURES \
> (1ULL << VIRTIO_NET_F_MAC | \
> @@ -607,16 +676,10 @@ virtio_user_dev_init(struct virtio_user_dev *dev,
> char *path, uint16_t queues,
> enum virtio_user_backend_type backend_type)
> {
> uint64_t backend_features;
> - int i;
>
> pthread_mutex_init(&dev->mutex, NULL);
> strlcpy(dev->path, path, PATH_MAX);
>
> - for (i = 0; i < VIRTIO_MAX_VIRTQUEUES; i++) {
> - dev->kickfds[i] = -1;
> - dev->callfds[i] = -1;
> - }
> -
> dev->started = 0;
> dev->queue_pairs = 1; /* mq disabled by default */
> dev->queue_size = queue_size;
> @@ -661,9 +724,14 @@ virtio_user_dev_init(struct virtio_user_dev *dev,
> char *path, uint16_t queues,
> if (dev->max_queue_pairs > 1)
> cq = 1;
>
> + if (virtio_user_alloc_vrings(dev) < 0) {
> + PMD_INIT_LOG(ERR, "(%s) Failed to allocate vring metadata",
> dev->path);
> + goto destroy;
> + }
> +
> if (virtio_user_dev_init_notify(dev) < 0) {
> PMD_INIT_LOG(ERR, "(%s) Failed to init notifiers", dev->path);
> - goto destroy;
> + goto free_vrings;
> }
>
> if (virtio_user_fill_intr_handle(dev) < 0) {
> @@ -722,6 +790,8 @@ virtio_user_dev_init(struct virtio_user_dev *dev, char
> *path, uint16_t queues,
>
> notify_uninit:
> virtio_user_dev_uninit_notify(dev);
> +free_vrings:
> + virtio_user_free_vrings(dev);
> destroy:
> dev->ops->destroy(dev);
>
> @@ -742,6 +812,8 @@ virtio_user_dev_uninit(struct virtio_user_dev *dev)
>
> virtio_user_dev_uninit_notify(dev);
>
> + virtio_user_free_vrings(dev);
> +
> free(dev->ifname);
>
> if (dev->is_server)
> @@ -897,7 +969,7 @@ static void
> virtio_user_handle_cq_packed(struct virtio_user_dev *dev, uint16_t
> queue_idx)
> {
> struct virtio_user_queue *vq = &dev->packed_queues[queue_idx];
> - struct vring_packed *vring = &dev->packed_vrings[queue_idx];
> + struct vring_packed *vring = &dev->vrings.packed[queue_idx];
> uint16_t n_descs, flags;
>
> /* Perform a load-acquire barrier in desc_is_avail to
> @@ -931,7 +1003,7 @@ virtio_user_handle_cq_split(struct virtio_user_dev
> *dev, uint16_t queue_idx)
> uint16_t avail_idx, desc_idx;
> struct vring_used_elem *uep;
> uint32_t n_descs;
> - struct vring *vring = &dev->vrings[queue_idx];
> + struct vring *vring = &dev->vrings.split[queue_idx];
>
> /* Consume avail ring, using used ring idx as first one */
> while (__atomic_load_n(&vring->used->idx, __ATOMIC_RELAXED)
> diff --git a/drivers/net/virtio/virtio_user/virtio_user_dev.h
> b/drivers/net/virtio/virtio_user/virtio_user_dev.h
> index e8753f6019..7323d88302 100644
> --- a/drivers/net/virtio/virtio_user/virtio_user_dev.h
> +++ b/drivers/net/virtio/virtio_user/virtio_user_dev.h
> @@ -29,8 +29,8 @@ struct virtio_user_dev {
> enum virtio_user_backend_type backend_type;
> bool is_server; /* server or client mode */
>
> - int callfds[VIRTIO_MAX_VIRTQUEUES];
> - int kickfds[VIRTIO_MAX_VIRTQUEUES];
> + int *callfds;
> + int *kickfds;
> int mac_specified;
> uint16_t max_queue_pairs;
> uint16_t queue_pairs;
> @@ -48,11 +48,13 @@ struct virtio_user_dev {
> char *ifname;
>
> union {
> - struct vring vrings[VIRTIO_MAX_VIRTQUEUES];
> - struct vring_packed packed_vrings[VIRTIO_MAX_VIRTQUEUES];
> - };
> - struct virtio_user_queue packed_queues[VIRTIO_MAX_VIRTQUEUES];
> - bool qp_enabled[VIRTIO_MAX_VIRTQUEUE_PAIRS];
> + void *ptr;
> + struct vring *split;
> + struct vring_packed *packed;
> + } vrings;
> +
> + struct virtio_user_queue *packed_queues;
> + bool *qp_enabled;
>
> struct virtio_user_backend_ops *ops;
> pthread_mutex_t mutex;
> diff --git a/drivers/net/virtio/virtio_user_ethdev.c
> b/drivers/net/virtio/virtio_user_ethdev.c
> index d23959e836..b1fc4d5d30 100644
> --- a/drivers/net/virtio/virtio_user_ethdev.c
> +++ b/drivers/net/virtio/virtio_user_ethdev.c
> @@ -186,7 +186,7 @@ virtio_user_setup_queue_packed(struct virtqueue *vq,
> uint64_t used_addr;
> uint16_t i;
>
> - vring = &dev->packed_vrings[queue_idx];
> + vring = &dev->vrings.packed[queue_idx];
> desc_addr = (uintptr_t)vq->vq_ring_virt_mem;
> avail_addr = desc_addr + vq->vq_nentries *
> sizeof(struct vring_packed_desc);
> @@ -216,10 +216,10 @@ virtio_user_setup_queue_split(struct virtqueue *vq,
> struct virtio_user_dev *dev)
> ring[vq->vq_nentries]),
> VIRTIO_VRING_ALIGN);
>
> - dev->vrings[queue_idx].num = vq->vq_nentries;
> - dev->vrings[queue_idx].desc = (void *)(uintptr_t)desc_addr;
> - dev->vrings[queue_idx].avail = (void *)(uintptr_t)avail_addr;
> - dev->vrings[queue_idx].used = (void *)(uintptr_t)used_addr;
> + dev->vrings.split[queue_idx].num = vq->vq_nentries;
> + dev->vrings.split[queue_idx].desc = (void *)(uintptr_t)desc_addr;
> + dev->vrings.split[queue_idx].avail = (void *)(uintptr_t)avail_addr;
> + dev->vrings.split[queue_idx].used = (void *)(uintptr_t)used_addr;
> }
>
> static int
> @@ -619,13 +619,6 @@ virtio_user_pmd_probe(struct rte_vdev_device *vdev)
> }
> }
>
> - if (queues > VIRTIO_MAX_VIRTQUEUE_PAIRS) {
> - PMD_INIT_LOG(ERR, "arg %s %" PRIu64 " exceeds the limit %u",
> - VIRTIO_USER_ARG_QUEUES_NUM, queues,
> - VIRTIO_MAX_VIRTQUEUE_PAIRS);
> - goto end;
> - }
> -
> if (rte_kvargs_count(kvlist, VIRTIO_USER_ARG_MRG_RXBUF) == 1) {
> if (rte_kvargs_process(kvlist, VIRTIO_USER_ARG_MRG_RXBUF,
> &get_integer_arg, &mrg_rxbuf) < 0) {
> --
> 2.38.1
next prev parent reply other threads:[~2023-01-31 5:19 UTC|newest]
Thread overview: 48+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-11-30 15:56 [PATCH v1 00/21] Add control queue & MQ support to Virtio-user vDPA Maxime Coquelin
2022-11-30 15:56 ` [PATCH v1 01/21] net/virtio: move CVQ code into a dedicated file Maxime Coquelin
2023-01-30 7:50 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 02/21] net/virtio: introduce notify callback for control queue Maxime Coquelin
2023-01-30 7:51 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 03/21] net/virtio: virtqueue headers alloc refactoring Maxime Coquelin
2023-01-30 7:51 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 04/21] net/virtio: remove port ID info from Rx queue Maxime Coquelin
2023-01-30 7:51 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 05/21] net/virtio: remove unused fields in Tx queue struct Maxime Coquelin
2023-01-30 7:51 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 06/21] net/virtio: remove unused queue ID field in Rx queue Maxime Coquelin
2023-01-30 7:52 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 07/21] net/virtio: remove unused Port ID in control queue Maxime Coquelin
2023-01-30 7:52 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 08/21] net/virtio: move vring memzone to virtqueue struct Maxime Coquelin
2023-01-30 7:52 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 09/21] net/virtio: refactor indirect desc headers init Maxime Coquelin
2023-01-30 7:52 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 10/21] net/virtio: alloc Rx SW ring only if vectorized path Maxime Coquelin
2023-01-30 7:49 ` Xia, Chenbo
2023-02-07 10:12 ` Maxime Coquelin
2022-11-30 15:56 ` [PATCH v1 11/21] net/virtio: extract virtqueue init from virtio queue init Maxime Coquelin
2023-01-30 7:53 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 12/21] net/virtio-user: fix device starting failure handling Maxime Coquelin
2023-01-31 5:20 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 13/21] net/virtio-user: simplify queues setup Maxime Coquelin
2023-01-31 5:21 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 14/21] net/virtio-user: use proper type for number of queue pairs Maxime Coquelin
2023-01-31 5:21 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 15/21] net/virtio-user: get max number of queue pairs from device Maxime Coquelin
2023-01-31 5:21 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 16/21] net/virtio-user: allocate shadow control queue Maxime Coquelin
2023-01-31 5:21 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 17/21] net/virtio-user: send shadow virtqueue info to the backend Maxime Coquelin
2023-01-31 5:22 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 18/21] net/virtio-user: add new callback to enable control queue Maxime Coquelin
2023-01-31 5:22 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 19/21] net/virtio-user: forward control messages to shadow queue Maxime Coquelin
2022-11-30 16:54 ` Stephen Hemminger
2022-12-06 12:58 ` Maxime Coquelin
2022-11-30 15:56 ` [PATCH v1 20/21] net/virtio-user: advertize control VQ support with vDPA Maxime Coquelin
2023-01-31 5:24 ` Xia, Chenbo
2022-11-30 15:56 ` [PATCH v1 21/21] net/virtio-user: remove max queues limitation Maxime Coquelin
2023-01-31 5:19 ` Xia, Chenbo [this message]
2023-02-07 14:14 ` Maxime Coquelin
2023-01-30 5:57 ` [PATCH v1 00/21] Add control queue & MQ support to Virtio-user vDPA Xia, Chenbo
2023-02-07 10:08 ` Maxime Coquelin
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=SN6PR11MB35046D1A8D2855D2A808D0FD9CD09@SN6PR11MB3504.namprd11.prod.outlook.com \
--to=chenbo.xia@intel.com \
--cc=david.marchand@redhat.com \
--cc=dev@dpdk.org \
--cc=eperezma@redhat.com \
--cc=maxime.coquelin@redhat.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).