From: "Xia, Chenbo" <chenbo.xia@intel.com>
To: "Ma, WenwuX" <wenwux.ma@intel.com>, "dev@dpdk.org" <dev@dpdk.org>
Cc: "maxime.coquelin@redhat.com" <maxime.coquelin@redhat.com>,
"Jiang, Cheng1" <cheng1.jiang@intel.com>,
"Hu, Jiayu" <jiayu.hu@intel.com>,
"Pai G, Sunil" <sunil.pai.g@intel.com>,
"Yang, YvonneX" <yvonnex.yang@intel.com>
Subject: Re: [dpdk-dev] [PATCH 4/4] examples/vhost: support vhost async dequeue data path
Date: Wed, 15 Sep 2021 03:27:48 +0000 [thread overview]
Message-ID: <MN2PR11MB4063AD59425BC4C564B7F2399CDB9@MN2PR11MB4063.namprd11.prod.outlook.com> (raw)
In-Reply-To: <20210906204837.112466-5-wenwux.ma@intel.com>
Hi,
> -----Original Message-----
> From: Ma, WenwuX <wenwux.ma@intel.com>
> Sent: Tuesday, September 7, 2021 4:49 AM
> To: dev@dpdk.org
> Cc: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>; Jiang,
> Cheng1 <cheng1.jiang@intel.com>; Hu, Jiayu <jiayu.hu@intel.com>; Pai G, Sunil
> <sunil.pai.g@intel.com>; Yang, YvonneX <yvonnex.yang@intel.com>; Ma, WenwuX
> <wenwux.ma@intel.com>
> Subject: [PATCH 4/4] examples/vhost: support vhost async dequeue data path
>
> This patch is to add vhost async dequeue data-path in vhost sample.
> vswitch can leverage IOAT to accelerate vhost async dequeue data-path.
>
> Signed-off-by: Wenwu Ma <wenwux.ma@intel.com>
> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
> ---
> doc/guides/sample_app_ug/vhost.rst | 9 +-
> examples/vhost/ioat.c | 61 +++++++--
> examples/vhost/ioat.h | 25 ++++
> examples/vhost/main.c | 201 +++++++++++++++++++----------
> examples/vhost/main.h | 3 +-
> 5 files changed, 216 insertions(+), 83 deletions(-)
>
> diff --git a/doc/guides/sample_app_ug/vhost.rst
> b/doc/guides/sample_app_ug/vhost.rst
> index 9afde9c7f5..63dcf181e1 100644
> --- a/doc/guides/sample_app_ug/vhost.rst
> +++ b/doc/guides/sample_app_ug/vhost.rst
> @@ -169,9 +169,12 @@ demonstrates how to use the async vhost APIs. It's used
> in combination with dmas
> **--dmas**
> This parameter is used to specify the assigned DMA device of a vhost device.
> Async vhost-user net driver will be used if --dmas is set. For example
> ---dmas [txd0@00:04.0,txd1@00:04.1] means use DMA channel 00:04.0 for vhost
> -device 0 enqueue operation and use DMA channel 00:04.1 for vhost device 1
> -enqueue operation.
> +--dmas [txd0@00:04.0,txd1@00:04.1,rxd0@00:04.2,rxd1@00:04.3] means use
> +DMA channel 00:04.0/00:04.2 for vhost device 0 enqueue/dequeue operation
> +and use DMA channel 00:04.1/00:04.3 for vhost device 1 enqueue/dequeue
> +operation. The index of the device corresponds to the socket file in order,
> +that means vhost device 0 is created through the first socket file, vhost
> +device 1 is created through the second socket file, and so on.
>
> Common Issues
> -------------
> diff --git a/examples/vhost/ioat.c b/examples/vhost/ioat.c
> index 6adc30b622..540b61fff6 100644
> --- a/examples/vhost/ioat.c
> +++ b/examples/vhost/ioat.c
> @@ -21,6 +21,8 @@ struct packet_tracker {
>
> struct packet_tracker cb_tracker[MAX_VHOST_DEVICE];
>
> +int vid2socketid[MAX_VHOST_DEVICE];
> +
> int
> open_ioat(const char *value)
> {
> @@ -29,7 +31,7 @@ open_ioat(const char *value)
> char *addrs = input;
> char *ptrs[2];
> char *start, *end, *substr;
> - int64_t vid, vring_id;
> + int64_t socketid, vring_id;
> struct rte_ioat_rawdev_config config;
> struct rte_rawdev_info info = { .dev_private = &config };
> char name[32];
> @@ -60,6 +62,7 @@ open_ioat(const char *value)
> goto out;
> }
> while (i < args_nr) {
> + bool is_txd;
> char *arg_temp = dma_arg[i];
> uint8_t sub_nr;
> sub_nr = rte_strsplit(arg_temp, strlen(arg_temp), ptrs, 2, '@');
> @@ -68,27 +71,39 @@ open_ioat(const char *value)
> goto out;
> }
>
> - start = strstr(ptrs[0], "txd");
> - if (start == NULL) {
> + int async_flag;
> + char *txd, *rxd;
> + txd = strstr(ptrs[0], "txd");
> + rxd = strstr(ptrs[0], "rxd");
> + if (txd) {
> + is_txd = true;
> + start = txd;
> + async_flag = ASYNC_ENQUEUE_VHOST;
> + } else if (rxd) {
> + is_txd = false;
> + start = rxd;
> + async_flag = ASYNC_DEQUEUE_VHOST;
> + } else {
> ret = -1;
> goto out;
> }
>
> start += 3;
> - vid = strtol(start, &end, 0);
> + socketid = strtol(start, &end, 0);
> if (end == start) {
> ret = -1;
> goto out;
> }
>
> - vring_id = 0 + VIRTIO_RXQ;
> + vring_id = is_txd ? VIRTIO_RXQ : VIRTIO_TXQ;
> +
> if (rte_pci_addr_parse(ptrs[1],
> - &(dma_info + vid)->dmas[vring_id].addr) < 0) {
> + &(dma_info + socketid)->dmas[vring_id].addr) < 0) {
> ret = -1;
> goto out;
> }
>
> - rte_pci_device_name(&(dma_info + vid)->dmas[vring_id].addr,
> + rte_pci_device_name(&(dma_info + socketid)->dmas[vring_id].addr,
> name, sizeof(name));
> dev_id = rte_rawdev_get_dev_id(name);
> if (dev_id == (uint16_t)(-ENODEV) ||
> @@ -103,8 +118,9 @@ open_ioat(const char *value)
> goto out;
> }
>
> - (dma_info + vid)->dmas[vring_id].dev_id = dev_id;
> - (dma_info + vid)->dmas[vring_id].is_valid = true;
> + (dma_info + socketid)->dmas[vring_id].dev_id = dev_id;
> + (dma_info + socketid)->dmas[vring_id].is_valid = true;
> + (dma_info + socketid)->async_flag |= async_flag;
> config.ring_size = IOAT_RING_SIZE;
> config.hdls_disable = true;
> if (rte_rawdev_configure(dev_id, &info, sizeof(config)) < 0) {
> @@ -126,13 +142,16 @@ ioat_transfer_data_cb(int vid, uint16_t queue_id,
> struct rte_vhost_async_status *opaque_data, uint16_t count)
> {
> uint32_t i_desc;
> - uint16_t dev_id = dma_bind[vid].dmas[queue_id * 2 + VIRTIO_RXQ].dev_id;
> struct rte_vhost_iov_iter *src = NULL;
> struct rte_vhost_iov_iter *dst = NULL;
> unsigned long i_seg;
> unsigned short mask = MAX_ENQUEUED_SIZE - 1;
> - unsigned short write = cb_tracker[dev_id].next_write;
>
> + if (queue_id >= MAX_RING_COUNT)
> + return -1;
> +
> + uint16_t dev_id = dma_bind[vid2socketid[vid]].dmas[queue_id].dev_id;
> + unsigned short write = cb_tracker[dev_id].next_write;
> if (!opaque_data) {
> for (i_desc = 0; i_desc < count; i_desc++) {
> src = descs[i_desc].src;
> @@ -170,16 +189,16 @@ ioat_check_completed_copies_cb(int vid, uint16_t
> queue_id,
> struct rte_vhost_async_status *opaque_data,
> uint16_t max_packets)
> {
> - if (!opaque_data) {
> + if (!opaque_data && (queue_id < MAX_RING_COUNT)) {
Should be: if (!opaque_data && queue_id < MAX_RING_COUNT) {
> uintptr_t dump[255];
> int n_seg;
> unsigned short read, write;
> unsigned short nb_packet = 0;
> unsigned short mask = MAX_ENQUEUED_SIZE - 1;
> unsigned short i;
> + uint16_t dev_id;
>
> - uint16_t dev_id = dma_bind[vid].dmas[queue_id * 2
> - + VIRTIO_RXQ].dev_id;
> + dev_id = dma_bind[vid2socketid[vid]].dmas[queue_id].dev_id;
> n_seg = rte_ioat_completed_ops(dev_id, 255, NULL, NULL, dump,
> dump);
> if (n_seg < 0) {
> RTE_LOG(ERR,
> @@ -215,4 +234,18 @@ ioat_check_completed_copies_cb(int vid, uint16_t queue_id,
> return -1;
> }
>
> +uint32_t get_async_flag_by_vid(int vid)
> +{
> + return dma_bind[vid2socketid[vid]].async_flag;
async_flag is sometimes int and sometimes uint32_t. Please check code that uses
the flag and make it all uint32_t.
> +}
> +
> +uint32_t get_async_flag_by_socketid(int socketid)
> +{
> + return dma_bind[socketid].async_flag;
> +}
> +
> +void init_vid2socketid_array(int vid, int socketid)
> +{
> + vid2socketid[vid] = socketid;
> +}
> #endif /* RTE_RAW_IOAT */
> diff --git a/examples/vhost/ioat.h b/examples/vhost/ioat.h
> index 62e163c585..fa5086e662 100644
> --- a/examples/vhost/ioat.h
> +++ b/examples/vhost/ioat.h
> @@ -12,6 +12,9 @@
> #define MAX_VHOST_DEVICE 1024
> #define IOAT_RING_SIZE 4096
> #define MAX_ENQUEUED_SIZE 4096
> +#define MAX_RING_COUNT 2
> +#define ASYNC_ENQUEUE_VHOST 1
> +#define ASYNC_DEQUEUE_VHOST 2
>
> struct dma_info {
> struct rte_pci_addr addr;
> @@ -20,6 +23,7 @@ struct dma_info {
> };
>
> struct dma_for_vhost {
> + int async_flag;
> struct dma_info dmas[RTE_MAX_QUEUES_PER_PORT * 2];
> uint16_t nr;
> };
> @@ -36,6 +40,10 @@ int32_t
> ioat_check_completed_copies_cb(int vid, uint16_t queue_id,
> struct rte_vhost_async_status *opaque_data,
> uint16_t max_packets);
> +
> +uint32_t get_async_flag_by_vid(int vid);
> +uint32_t get_async_flag_by_socketid(int socketid);
> +void init_vid2socketid_array(int vid, int socketid);
> #else
> static int open_ioat(const char *value __rte_unused)
> {
> @@ -59,5 +67,22 @@ ioat_check_completed_copies_cb(int vid __rte_unused,
> {
> return -1;
> }
> +
> +static uint32_t
> +get_async_flag_by_vid(int vid __rte_unused)
> +{
> + return 0;
> +}
> +
> +static uint32_t
> +get_async_flag_by_socketid(int socketid __rte_unused)
> +{
> + return 0;
> +}
> +
> +static void
> +init_vid2socketid_array(int vid __rte_unused, int socketid __rte_unused)
> +{
> +}
> #endif
> #endif /* _IOAT_H_ */
> diff --git a/examples/vhost/main.c b/examples/vhost/main.c
> index e246b640ea..b34534111d 100644
> --- a/examples/vhost/main.c
> +++ b/examples/vhost/main.c
> @@ -93,8 +93,6 @@ static int client_mode;
>
> static int builtin_net_driver;
>
> -static int async_vhost_driver;
> -
> static char *dma_type;
>
> /* Specify timeout (in useconds) between retries on RX. */
> @@ -679,7 +677,6 @@ us_vhost_parse_args(int argc, char **argv)
> us_vhost_usage(prgname);
> return -1;
> }
> - async_vhost_driver = 1;
> break;
>
> case OPT_CLIENT_NUM:
> @@ -855,7 +852,8 @@ complete_async_pkts(struct vhost_dev *vdev)
> VIRTIO_RXQ, p_cpl, MAX_PKT_BURST);
> if (complete_count) {
> free_pkts(p_cpl, complete_count);
> - __atomic_sub_fetch(&vdev->pkts_inflight, complete_count,
> __ATOMIC_SEQ_CST);
> + __atomic_sub_fetch(&vdev->pkts_enq_inflight,
> + complete_count, __ATOMIC_SEQ_CST);
> }
>
> }
> @@ -900,7 +898,7 @@ drain_vhost(struct vhost_dev *vdev)
> __ATOMIC_SEQ_CST);
> }
>
> - if (!async_vhost_driver)
> + if ((get_async_flag_by_vid(vdev->vid) & ASYNC_ENQUEUE_VHOST) == 0)
> free_pkts(m, nr_xmit);
> }
>
> @@ -1180,8 +1178,8 @@ async_enqueue_pkts(struct vhost_dev *vdev, uint16_t
> queue_id,
> complete_async_pkts(vdev);
> enqueue_count = rte_vhost_submit_enqueue_burst(vdev->vid,
> queue_id, pkts, rx_count);
> - __atomic_add_fetch(&vdev->pkts_inflight, enqueue_count,
> - __ATOMIC_SEQ_CST);
> + __atomic_add_fetch(&vdev->pkts_enq_inflight,
> + enqueue_count, __ATOMIC_SEQ_CST);
>
> enqueue_fail = rx_count - enqueue_count;
> if (enqueue_fail)
> @@ -1237,10 +1235,23 @@ drain_eth_rx(struct vhost_dev *vdev)
> __ATOMIC_SEQ_CST);
> }
>
> - if (!async_vhost_driver)
> + if ((get_async_flag_by_vid(vdev->vid) & ASYNC_ENQUEUE_VHOST) == 0)
> free_pkts(pkts, rx_count);
> }
>
> +uint16_t async_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
> + struct rte_mempool *mbuf_pool,
> + struct rte_mbuf **pkts, uint16_t count)
> +{
> + int nr_inflight;
> + uint16_t dequeue_count;
> + dequeue_count = rte_vhost_async_try_dequeue_burst(dev->vid, queue_id,
> + mbuf_pool, pkts, count, &nr_inflight);
> + if (likely(nr_inflight != -1))
> + dev->pkts_deq_inflight = nr_inflight;
> + return dequeue_count;
> +}
> +
> uint16_t sync_dequeue_pkts(struct vhost_dev *dev, uint16_t queue_id,
> struct rte_mempool *mbuf_pool,
> struct rte_mbuf **pkts, uint16_t count)
> @@ -1336,6 +1347,32 @@ switch_worker(void *arg __rte_unused)
> return 0;
> }
>
> +static void
> +vhost_clear_queue_thread_unsafe(struct vhost_dev *vdev, uint16_t queue_id)
> +{
> + uint16_t n_pkt = 0;
> + struct rte_mbuf *m_enq_cpl[vdev->pkts_enq_inflight];
> + struct rte_mbuf *m_deq_cpl[vdev->pkts_deq_inflight];
> +
> + if ((queue_id % VIRTIO_QNUM) == 0) {
You are assuming VIRTIO_QNUM equals 2 here. The correct logic should be
'queue_id % 2 == 0' or 'queue_id & 0x1 == 0'
Thanks,
Chenbo
> + while (vdev->pkts_enq_inflight) {
> + n_pkt = rte_vhost_clear_queue_thread_unsafe(vdev->vid,
> + queue_id, m_enq_cpl, vdev->pkts_enq_inflight);
> + free_pkts(m_enq_cpl, n_pkt);
> + __atomic_sub_fetch(&vdev->pkts_enq_inflight,
> + n_pkt, __ATOMIC_SEQ_CST);
> + }
next prev parent reply other threads:[~2021-09-15 3:28 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-06 20:48 [dpdk-dev] [PATCH 0/4] support async dequeue for split ring Wenwu Ma
2021-09-06 20:48 ` [dpdk-dev] [PATCH 1/4] vhost: " Wenwu Ma
2021-09-10 7:36 ` Yang, YvonneX
2021-09-15 2:51 ` Xia, Chenbo
[not found] ` <CO1PR11MB4897F3D5ABDE7133DB99791385DB9@CO1PR11MB4897.namprd11.prod.outlook.com>
2021-09-15 11:35 ` Xia, Chenbo
2021-09-06 20:48 ` [dpdk-dev] [PATCH 2/4] examples/vhost: refactor vhost enqueue and dequeue datapaths Wenwu Ma
2021-09-10 7:38 ` Yang, YvonneX
2021-09-15 3:02 ` Xia, Chenbo
2021-09-06 20:48 ` [dpdk-dev] [PATCH 3/4] examples/vhost: use a new API to query remaining ring space Wenwu Ma
2021-09-10 7:38 ` Yang, YvonneX
2021-09-15 3:04 ` Xia, Chenbo
2021-09-06 20:48 ` [dpdk-dev] [PATCH 4/4] examples/vhost: support vhost async dequeue data path Wenwu Ma
2021-09-10 7:39 ` Yang, YvonneX
2021-09-15 3:27 ` Xia, Chenbo [this message]
2021-09-10 7:33 ` [dpdk-dev] [PATCH 0/4] support async dequeue for split ring Yang, YvonneX
2021-09-17 19:26 ` [dpdk-dev] [PATCH v2 " Wenwu Ma
2021-09-17 19:27 ` [dpdk-dev] [PATCH v2 1/4] vhost: " Wenwu Ma
2021-09-27 6:33 ` Jiang, Cheng1
2021-09-17 19:27 ` [dpdk-dev] [PATCH v2 2/4] examples/vhost: refactor vhost enqueue and dequeue datapaths Wenwu Ma
2021-09-27 6:56 ` Jiang, Cheng1
2021-09-17 19:27 ` [dpdk-dev] [PATCH v2 3/4] examples/vhost: use a new API to query remaining ring space Wenwu Ma
2021-09-17 19:27 ` [dpdk-dev] [PATCH v2 4/4] examples/vhost: support vhost async dequeue data path Wenwu Ma
2021-09-28 18:56 ` [dpdk-dev] [PATCH v3 0/4] support async dequeue for split ring Wenwu Ma
2021-09-28 18:56 ` [dpdk-dev] [PATCH v3 1/4] vhost: " Wenwu Ma
2021-09-28 18:56 ` [dpdk-dev] [PATCH v3 2/4] examples/vhost: refactor vhost enqueue and dequeue datapaths Wenwu Ma
2021-09-28 18:56 ` [dpdk-dev] [PATCH v3 3/4] examples/vhost: use a new API to query remaining ring space Wenwu Ma
2021-09-28 18:56 ` [dpdk-dev] [PATCH v3 4/4] examples/vhost: support vhost async dequeue data path Wenwu Ma
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=MN2PR11MB4063AD59425BC4C564B7F2399CDB9@MN2PR11MB4063.namprd11.prod.outlook.com \
--to=chenbo.xia@intel.com \
--cc=cheng1.jiang@intel.com \
--cc=dev@dpdk.org \
--cc=jiayu.hu@intel.com \
--cc=maxime.coquelin@redhat.com \
--cc=sunil.pai.g@intel.com \
--cc=wenwux.ma@intel.com \
--cc=yvonnex.yang@intel.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).