DPDK patches and discussions
 help / color / mirror / Atom feed
From: "Jiang, Cheng1" <cheng1.jiang@intel.com>
To: "Xia, Chenbo" <chenbo.xia@intel.com>,
	"maxime.coquelin@redhat.com" <maxime.coquelin@redhat.com>,
	"Wang, Zhihong" <zhihong.wang@intel.com>
Cc: "dev@dpdk.org" <dev@dpdk.org>,
	"Fu, Patrick" <patrick.fu@intel.com>,
	"Liang, Cunming" <cunming.liang@intel.com>
Subject: Re: [dpdk-dev] [RFC] example/vhost: add support for vhost async data path
Date: Wed, 8 Jul 2020 03:19:13 +0000
Message-ID: <CY4PR11MB1639A34661F339AEB1E4B586DC670@CY4PR11MB1639.namprd11.prod.outlook.com> (raw)
In-Reply-To: <MN2PR11MB40639F0C895DBEBDB71DF8CB9C940@MN2PR11MB4063.namprd11.prod.outlook.com>

Hi Chenbo,

> -----Original Message-----
> From: Xia, Chenbo <chenbo.xia@intel.com>
> Sent: Tuesday, June 23, 2020 9:54 AM
> To: Jiang, Cheng1 <cheng1.jiang@intel.com>; maxime.coquelin@redhat.com;
> Wang, Zhihong <zhihong.wang@intel.com>
> Cc: dev@dpdk.org; Fu, Patrick <patrick.fu@intel.com>; Liang, Cunming
> <cunming.liang@intel.com>
> Subject: RE: [RFC] example/vhost: add support for vhost async data path
> 
> Hi Cheng,
> 
> > -----Original Message-----
> > From: Jiang, Cheng1 <cheng1.jiang@intel.com>
> > Sent: Monday, June 22, 2020 10:59 AM
> > To: maxime.coquelin@redhat.com; Xia, Chenbo <chenbo.xia@intel.com>;
> > Wang, Zhihong <zhihong.wang@intel.com>
> > Cc: dev@dpdk.org; Fu, Patrick <patrick.fu@intel.com>; Liang, Cunming
> > <cunming.liang@intel.com>; Jiang, Cheng1 <cheng1.jiang@intel.com>
> > Subject: [RFC] example/vhost: add support for vhost async data path
> >
> > This patch makes vhost-vswitch be able to use vhost asynchronous api
> > for enqueue operations. Demonstrated how the application leverage IOAT
> > DMA channel with vhost async api. Since this is an early preview
> > patch, the performance has not been fully optimized and it's not
> > suggested to use this patch as a tool for benchmark.
> >
> > We introduce two parameters to enable DMA acceleration for Tx
> > operations of
> > queues:
> > –async_vhost_driver Async vhost-user net driver which demonstrates how
> > to use the async vhost APIs will be used when this option is given. It
> > is disabled by default.
> >
> > -dmas This parameter is used to specify the assigned DMA device of a
> queue.
> >
> > This patch depends on following patch set:
> > http://patches.dpdk.org/cover/71265/
> >
> > Signed-off-by: Cheng Jiang <Cheng1.jiang@intel.com>
> > ---
> >  examples/vhost/main.c | 246
> > +++++++++++++++++++++++++++++++++++++++++-
> >  examples/vhost/main.h |   1 +
> >  2 files changed, 243 insertions(+), 4 deletions(-)
> >
> > diff --git a/examples/vhost/main.c b/examples/vhost/main.c index
> > ab649bf14..46dd282e0 100644
> > --- a/examples/vhost/main.c
> > +++ b/examples/vhost/main.c
> > @@ -24,11 +24,15 @@
> >  #include <rte_ip.h>
> >  #include <rte_tcp.h>
> >  #include <rte_pause.h>
> > +#include <rte_vhost_async.h>
> > +#include <rte_rawdev.h>
> > +#include <rte_ioat_rawdev.h>
> > +#include <rte_pci.h>
> >
> >  #include "main.h"
> >
> >  #ifndef MAX_QUEUES
> > -#define MAX_QUEUES 128
> > +#define MAX_QUEUES 512
> >  #endif
> >
> >  /* the maximum number of external ports supported */ @@ -58,6 +62,10
> > @@
> >  /* Maximum long option length for option parsing. */  #define
> > MAX_LONG_OPT_SZ 64
> >
> > +#define IOAT_RING_SIZE 4096
> > +
> > +#define MAX_ENQUEUED_SIZE 2048
> > +
> >  /* mask of enabled ports */
> >  static uint32_t enabled_port_mask = 0;
> >
> > @@ -96,6 +104,20 @@ static int dequeue_zero_copy;
> >
> >  static int builtin_net_driver;
> >
> > +static int async_vhost_driver;
> > +
> > +struct dma_info {
> > +	struct rte_pci_addr addr;
> > +	uint16_t dev_id;
> > +	bool is_valid;
> > +};
> > +
> > +struct dma_info_input {
> > +	struct dma_info dmas[RTE_MAX_QUEUES_PER_PORT * 2];
> > +	uint16_t nr;
> > +};
> > +
> > +static struct dma_info_input dma_bind[20];
> 
> Should '20' be MAX_VHOST_DEVICE as this is indexed by vid?
> 
Sure, you are right, MAX_VHOST_DEVICE will be used in the next version.

> >  /* Specify timeout (in useconds) between retries on RX. */  static
> > uint32_t burst_rx_delay_time = BURST_RX_WAIT_US;
> >  /* Specify the number of retries on RX. */ @@ -141,6 +163,61 @@
> > static struct rte_eth_conf vmdq_conf_default = {
> >  	},
> >  };
> >
> > +static int
> > +ioat_transfer_data_cb(int vid, uint16_t queue_id, struct
> > +dma_trans_desc
> > *descs,
> > +		struct dma_trans_status *opaque_data, uint16_t count) {
> > +	int ret;
> > +	uint16_t i_desc;
> > +
> > +	struct iov_it *src = NULL;
> > +	struct iov_it *dst = NULL;
> > +	unsigned long i_seg;
> > +
> > +	int dev_id = dma_bind[vid].dmas[queue_id * 2 +
> VIRTIO_RXQ].dev_id;
> > +	if (likely(!opaque_data)) {
> > +		for (i_desc = 0; i_desc < count; i_desc++) {
> > +			src = descs[i_desc].src;
> > +			dst = descs[i_desc].dst;
> > +			i_seg = 0;
> > +			while (i_seg < src->nr_segs) {
> > +				ret = rte_ioat_enqueue_copy(dev_id,
> > +					(uintptr_t)(src->iov[i_seg].iov_base)
> > +						+ src->offset,
> > +					(uintptr_t)(dst->iov[i_seg].iov_base)
> > +						+ dst->offset,
> > +					src->iov[i_seg].iov_len,
> > +					0,
> > +					0,
> > +					0);
> > +				if (ret != 1)
> > +					break;
> > +				i_seg++;
> > +			}
> > +		}
> > +	} else {
> > +		/* Opaque data is not supported */
> > +		return -1;
> > +	}
> > +	/* ring the doolbell */
> 
> s/doolbell/doorbell
> 
I'll fix in the next version.

> > +	rte_ioat_do_copies(dev_id);
> > +	return i_desc;
> > +}
> > +
> > +static int
> > +ioat_check_completed_copies_cb(int vid, uint16_t queue_id,
> > +		struct dma_trans_status *opaque_data,
> > +		uint16_t max_packets __rte_unused)
> > +{
> > +	if (!opaque_data) {
> > +		uintptr_t dump[255];
> > +		return
> > rte_ioat_completed_copies(dma_bind[vid].dmas[queue_id * 2
> > +			+ VIRTIO_RXQ].dev_id, 255, dump, dump);
> > +	} else {
> > +		/* Opaque data is not supported */
> > +		return -1;
> > +	}
> > +}
> >
> >  static unsigned lcore_ids[RTE_MAX_LCORE];  static uint16_t
> > ports[RTE_MAX_ETHPORTS]; @@ -186,6 +263,94 @@ struct mbuf_table
> > lcore_tx_queue[RTE_MAX_LCORE];
> >   * Builds up the correct configuration for VMDQ VLAN pool map
> >   * according to the pool & queue limits.
> >   */
> > +
> > +static inline int
> > +open_dma(const char *value, void *dma_bind_info) {
> > +	struct dma_info_input *dma_info = dma_bind_info;
> > +	char *input = strndup(value, strlen(value) + 1);
> > +	char *addrs = input;
> > +	char *ptrs[2];
> > +	char *start, *end, *substr;
> > +	int64_t qid, vring_id;
> > +	struct rte_ioat_rawdev_config config;
> > +	struct rte_rawdev_info info = { .dev_private = &config };
> > +	char name[32];
> > +	int dev_id;
> > +	int ret = 0;
> > +
> > +	while (isblank(*addrs))
> > +		addrs++;
> > +	if (*addrs == '\0') {
> > +		ret = -1;
> > +		goto out;
> > +	}
> > +
> > +	/* process DMA devices within bracket. */
> > +	addrs++;
> > +	substr = strtok(addrs, ";]");
> > +	if (!substr) {
> > +		ret = -1;
> > +		goto out;
> > +	}
> > +
> > +	do {
> > +		rte_strsplit(substr, strlen(substr), ptrs, 2, '@');
> > +
> > +		start = strstr(ptrs[0], "txq");
> > +		if (start == NULL) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +
> > +		start += 3;
> > +		qid = strtol(start, &end, 0);
> > +		if (end == start) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +
> > +		vring_id = qid * 2 + VIRTIO_RXQ;
> > +		if (rte_pci_addr_parse(ptrs[1],
> > +				       &dma_info->dmas[vring_id].addr) < 0) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +
> > +		rte_pci_device_name(&dma_info->dmas[vring_id].addr,
> > +				    name, sizeof(name));
> > +		dev_id = rte_rawdev_get_dev_id(name);
> > +		if (dev_id == (uint16_t)(-ENODEV) ||
> > +		    dev_id == (uint16_t)(-EINVAL)) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +
> > +		if (rte_rawdev_info_get(dev_id, &info) < 0 ||
> > +		    strstr(info.driver_name, "ioat") == NULL) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +
> > +		dma_info->dmas[vring_id].dev_id = dev_id;
> > +		dma_info->dmas[vring_id].is_valid = true;
> > +		config.ring_size = IOAT_RING_SIZE;
> > +		if (rte_rawdev_configure(dev_id, &info) < 0) {
> > +			ret = -1;
> > +			goto out;
> > +		}
> > +		rte_rawdev_start(dev_id);
> > +
> > +		dma_info->nr++;
> > +
> > +		substr = strtok(NULL, ";]");
> > +	} while (substr);
> > +
> > +out:
> > +	free(input);
> > +	return ret;
> > +}
> > +
> >  static inline int
> >  get_eth_conf(struct rte_eth_conf *eth_conf, uint32_t num_devices)  {
> > @@ -
> > 488,6 +653,8 @@ us_vhost_parse_args(int argc, char **argv)
> >  		{"client", no_argument, &client_mode, 1},
> >  		{"dequeue-zero-copy", no_argument,
> &dequeue_zero_copy, 1},
> >  		{"builtin-net-driver", no_argument, &builtin_net_driver, 1},
> > +		{"async_vhost_driver", no_argument, &async_vhost_driver,
> 1},
> > +		{"dmas", required_argument, NULL, 0},
> >  		{NULL, 0, 0, 0},
> >  	};
> >
> > @@ -623,13 +790,25 @@ us_vhost_parse_args(int argc, char **argv)
> >  						"socket-file",
> > MAX_LONG_OPT_SZ)) {
> >  				if (us_vhost_parse_socket_path(optarg) == -
> 1) {
> >  					RTE_LOG(INFO, VHOST_CONFIG,
> > -					"Invalid argument for socket name
> > (Max %d characters)\n",
> > -					PATH_MAX);
> > +						"Invalid argument for socket
> > name (Max %d characters)\n",
> > +						PATH_MAX);
> >  					us_vhost_usage(prgname);
> >  					return -1;
> >  				}
> >  			}
> >
> > +			if (!strncmp(long_option[option_index].name,
> > +						"dmas",
> MAX_LONG_OPT_SZ))
> > {
> > +				if (open_dma(optarg, &(dma_bind[0])) == -1)
> {
> > +					if (*optarg == -1) {
> > +						RTE_LOG(INFO,
> > VHOST_CONFIG,
> > +							"Wrong DMA args\n");
> > +						us_vhost_usage(prgname);
> > +						return -1;
> > +					}
> > +				}
> > +			}
> > +
> >  			break;
> >
> >  			/* Invalid option - print options. */ @@ -785,9
> +964,26 @@
> > virtio_xmit(struct vhost_dev *dst_vdev, struct vhost_dev *src_vdev,
> >  	    struct rte_mbuf *m)
> >  {
> >  	uint16_t ret;
> > +	struct rte_mbuf *m_cpl[1];
> >
> >  	if (builtin_net_driver) {
> >  		ret = vs_enqueue_pkts(dst_vdev, VIRTIO_RXQ, &m, 1);
> > +	} else if (async_vhost_driver) {
> > +		ret = rte_vhost_submit_enqueue_burst(dst_vdev->vid,
> > VIRTIO_RXQ,
> > +						&m, 1);
> > +
> > +		if (likely(ret)) {
> > +			dst_vdev->nr_async_pkts++;
> > +			rte_mbuf_refcnt_update(m, 1);
> > +		}
> > +
> > +		while (likely(dst_vdev->nr_async_pkts)) {
> > +			dst_vdev->nr_async_pkts =
> > +
> 	rte_vhost_poll_enqueue_completed(dst_vdev-
> > >vid,
> > +					VIRTIO_RXQ, m_cpl, 1);
> 
> I think nr_async_pkts should be changed only when
> rte_vhost_poll_enqueue_completed succeed?
> 
> Thanks!
> Chenbo
> 
Yes, make sense, I'll update in the next version.

Thanks,
Cheng

> > +			dst_vdev->nr_async_pkts--;
> > +			rte_pktmbuf_free(*m_cpl);
> > +		}
> >  	} else {
> >  		ret = rte_vhost_enqueue_burst(dst_vdev->vid, VIRTIO_RXQ,
> &m, 1);
> >  	}
> > @@ -1036,6 +1232,19 @@ drain_mbuf_table(struct mbuf_table *tx_q)
> >  	}
> >  }
> >
> > +static __rte_always_inline void
> > +complete_async_pkts(struct vhost_dev *vdev, uint16_t qid) {
> > +	struct rte_mbuf *p_cpl[MAX_PKT_BURST];
> > +	uint16_t complete_count;
> > +
> > +	complete_count = rte_vhost_poll_enqueue_completed(vdev->vid,
> > +						qid, p_cpl, MAX_PKT_BURST);
> > +	vdev->nr_async_pkts -= complete_count;
> > +	if (complete_count)
> > +		free_pkts(p_cpl, complete_count);
> > +}
> > +
> >  static __rte_always_inline void
> >  drain_eth_rx(struct vhost_dev *vdev)
> >  {
> > @@ -1044,6 +1253,10 @@ drain_eth_rx(struct vhost_dev *vdev)
> >
> >  	rx_count = rte_eth_rx_burst(ports[0], vdev->vmdq_rx_q,
> >  				    pkts, MAX_PKT_BURST);
> > +
> > +	while (likely(vdev->nr_async_pkts))
> > +		complete_async_pkts(vdev, VIRTIO_RXQ);
> > +
> >  	if (!rx_count)
> >  		return;
> >
> > @@ -1068,16 +1281,22 @@ drain_eth_rx(struct vhost_dev *vdev)
> >  	if (builtin_net_driver) {
> >  		enqueue_count = vs_enqueue_pkts(vdev, VIRTIO_RXQ,
> >  						pkts, rx_count);
> > +	} else if (async_vhost_driver) {
> > +		enqueue_count = rte_vhost_submit_enqueue_burst(vdev-
> >vid,
> > +					VIRTIO_RXQ, pkts, rx_count);
> > +		vdev->nr_async_pkts += enqueue_count;
> >  	} else {
> >  		enqueue_count = rte_vhost_enqueue_burst(vdev->vid,
> > VIRTIO_RXQ,
> >  						pkts, rx_count);
> >  	}
> > +
> >  	if (enable_stats) {
> >  		rte_atomic64_add(&vdev->stats.rx_total_atomic, rx_count);
> >  		rte_atomic64_add(&vdev->stats.rx_atomic,
> enqueue_count);
> >  	}
> >
> > -	free_pkts(pkts, rx_count);
> > +	if (!async_vhost_driver)
> > +		free_pkts(pkts, rx_count);
> >  }
> >
> >  static __rte_always_inline void
> > @@ -1224,6 +1443,9 @@ destroy_device(int vid)
> >  		"(%d) device has been removed from data core\n",
> >  		vdev->vid);
> >
> > +	if (async_vhost_driver)
> > +		rte_vhost_async_channel_unregister(vid, VIRTIO_RXQ);
> > +
> >  	rte_free(vdev);
> >  }
> >
> > @@ -1238,6 +1460,12 @@ new_device(int vid)
> >  	uint32_t device_num_min = num_devices;
> >  	struct vhost_dev *vdev;
> >
> > +	struct rte_vhost_async_channel_ops channel_ops = {
> > +		.transfer_data = ioat_transfer_data_cb,
> > +		.check_completed_copies =
> ioat_check_completed_copies_cb
> > +	};
> > +	struct dma_channel_features f;
> > +
> >  	vdev = rte_zmalloc("vhost device", sizeof(*vdev),
> > RTE_CACHE_LINE_SIZE);
> >  	if (vdev == NULL) {
> >  		RTE_LOG(INFO, VHOST_DATA,
> > @@ -1278,6 +1506,13 @@ new_device(int vid)
> >  		"(%d) device has been added to data core %d\n",
> >  		vid, vdev->coreid);
> >
> > +	if (async_vhost_driver) {
> > +		f.inorder = 1;
> > +		f.threshold = 256;
> > +		return rte_vhost_async_channel_register(vid, VIRTIO_RXQ,
> > +			f.intval, &channel_ops);
> > +	}
> > +
> >  	return 0;
> >  }
> >
> > @@ -1517,6 +1752,9 @@ main(int argc, char *argv[])
> >  	/* Register vhost user driver to handle vhost messages. */
> >  	for (i = 0; i < nb_sockets; i++) {
> >  		char *file = socket_files + i * PATH_MAX;
> > +		if (async_vhost_driver)
> > +			flags = flags | RTE_VHOST_USER_ASYNC_COPY;
> > +
> >  		ret = rte_vhost_driver_register(file, flags);
> >  		if (ret != 0) {
> >  			unregister_drivers(i);
> > diff --git a/examples/vhost/main.h b/examples/vhost/main.h index
> > 7cba0edbf..4317b6ae8 100644
> > --- a/examples/vhost/main.h
> > +++ b/examples/vhost/main.h
> > @@ -51,6 +51,7 @@ struct vhost_dev {
> >  	uint64_t features;
> >  	size_t hdr_len;
> >  	uint16_t nr_vrings;
> > +	uint16_t nr_async_pkts;
> >  	struct rte_vhost_memory *mem;
> >  	struct device_statistics stats;
> >  	TAILQ_ENTRY(vhost_dev) global_vdev_entry;
> > --
> > 2.26.2


  reply	other threads:[~2020-07-08  3:19 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-06-22  2:59 Cheng Jiang
2020-06-23  1:54 ` Xia, Chenbo
2020-07-08  3:19   ` Jiang, Cheng1 [this message]
2020-07-08  7:26 ` [dpdk-dev] [RFC v2] " Cheng Jiang

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=CY4PR11MB1639A34661F339AEB1E4B586DC670@CY4PR11MB1639.namprd11.prod.outlook.com \
    --to=cheng1.jiang@intel.com \
    --cc=chenbo.xia@intel.com \
    --cc=cunming.liang@intel.com \
    --cc=dev@dpdk.org \
    --cc=maxime.coquelin@redhat.com \
    --cc=patrick.fu@intel.com \
    --cc=zhihong.wang@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

DPDK patches and discussions

This inbox may be cloned and mirrored by anyone:

	git clone --mirror https://inbox.dpdk.org/dev/0 dev/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 dev dev/ https://inbox.dpdk.org/dev \
		dev@dpdk.org
	public-inbox-index dev

Example config snippet for mirrors.
Newsgroup available over NNTP:
	nntp://inbox.dpdk.org/inbox.dpdk.dev


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git