Re: [dpdk-dev] [PATCH v2 11/13] vdpa/mlx5: support live migration

[Maxime Coquelin <maxime.coquelin@redhat.com>]

On 1/29/20 11:09 AM, Matan Azrad wrote:
> Add support for live migration feature by the HW:
> 	Create a single Mkey that maps the memory address space of the
> 		VHOST live migration log file.
> 	Modify VIRTIO_NET_Q object and provide vhost_log_page,
> 		dirty_bitmap_mkey, dirty_bitmap_size, dirty_bitmap_addr
> 		and dirty_bitmap_dump_enable.
> 	Modify VIRTIO_NET_Q object and move state to SUSPEND.
> 	Query VIRTIO_NET_Q and get hw_available_idx and hw_used_idx.
> 
> Signed-off-by: Matan Azrad <matan@mellanox.com>
> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
> ---
>  doc/guides/vdpadevs/features/mlx5.ini |   1 +
>  drivers/vdpa/mlx5/Makefile            |   1 +
>  drivers/vdpa/mlx5/meson.build         |   1 +
>  drivers/vdpa/mlx5/mlx5_vdpa.c         |  44 +++++++++++-
>  drivers/vdpa/mlx5/mlx5_vdpa.h         |  55 ++++++++++++++
>  drivers/vdpa/mlx5/mlx5_vdpa_lm.c      | 130 ++++++++++++++++++++++++++++++++++
>  drivers/vdpa/mlx5/mlx5_vdpa_virtq.c   |   7 +-
>  7 files changed, 236 insertions(+), 3 deletions(-)
>  create mode 100644 drivers/vdpa/mlx5/mlx5_vdpa_lm.c
> 
> diff --git a/doc/guides/vdpadevs/features/mlx5.ini b/doc/guides/vdpadevs/features/mlx5.ini
> index e4ee34b..1da9c1b 100644
> --- a/doc/guides/vdpadevs/features/mlx5.ini
> +++ b/doc/guides/vdpadevs/features/mlx5.ini
> @@ -9,6 +9,7 @@ guest csum           = Y
>  host tso4            = Y
>  host tso6            = Y
>  version 1            = Y
> +log all              = Y
>  any layout           = Y
>  guest announce       = Y
>  mq                   = Y
> diff --git a/drivers/vdpa/mlx5/Makefile b/drivers/vdpa/mlx5/Makefile
> index 2f70a98..4d1f528 100644
> --- a/drivers/vdpa/mlx5/Makefile
> +++ b/drivers/vdpa/mlx5/Makefile
> @@ -12,6 +12,7 @@ SRCS-$(CONFIG_RTE_LIBRTE_MLX5_VDPA_PMD) += mlx5_vdpa_mem.c
>  SRCS-$(CONFIG_RTE_LIBRTE_MLX5_VDPA_PMD) += mlx5_vdpa_event.c
>  SRCS-$(CONFIG_RTE_LIBRTE_MLX5_VDPA_PMD) += mlx5_vdpa_virtq.c
>  SRCS-$(CONFIG_RTE_LIBRTE_MLX5_VDPA_PMD) += mlx5_vdpa_steer.c
> +SRCS-$(CONFIG_RTE_LIBRTE_MLX5_VDPA_PMD) += mlx5_vdpa_lm.c
>  
>  
>  # Basic CFLAGS.
> diff --git a/drivers/vdpa/mlx5/meson.build b/drivers/vdpa/mlx5/meson.build
> index 2849178..2e521b8 100644
> --- a/drivers/vdpa/mlx5/meson.build
> +++ b/drivers/vdpa/mlx5/meson.build
> @@ -16,6 +16,7 @@ sources = files(
>  	'mlx5_vdpa_event.c',
>  	'mlx5_vdpa_virtq.c',
>  	'mlx5_vdpa_steer.c',
> +	'mlx5_vdpa_lm.c',
>  )
>  cflags_options = [
>  	'-std=c11',
> diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.c b/drivers/vdpa/mlx5/mlx5_vdpa.c
> index 71189c4..4ce0ba0 100644
> --- a/drivers/vdpa/mlx5/mlx5_vdpa.c
> +++ b/drivers/vdpa/mlx5/mlx5_vdpa.c
> @@ -19,7 +19,8 @@
>  			    (1ULL << VIRTIO_F_ANY_LAYOUT) | \
>  			    (1ULL << VIRTIO_NET_F_MQ) | \
>  			    (1ULL << VIRTIO_NET_F_GUEST_ANNOUNCE) | \
> -			    (1ULL << VIRTIO_F_ORDER_PLATFORM))
> +			    (1ULL << VIRTIO_F_ORDER_PLATFORM) | \
> +			    (1ULL << VHOST_F_LOG_ALL))
>  
>  #define MLX5_VDPA_PROTOCOL_FEATURES \
>  			    ((1ULL << VHOST_USER_PROTOCOL_F_SLAVE_REQ) | \
> @@ -127,6 +128,45 @@
>  	return mlx5_vdpa_virtq_enable(virtq, state);
>  }
>  
> +static int
> +mlx5_vdpa_features_set(int vid)
> +{
> +	int did = rte_vhost_get_vdpa_device_id(vid);
> +	struct mlx5_vdpa_priv *priv = mlx5_vdpa_find_priv_resource_by_did(did);
> +	uint64_t log_base, log_size;
> +	uint64_t features;
> +	int ret;
> +
> +	if (priv == NULL) {
> +		DRV_LOG(ERR, "Invalid device id: %d.", did);
> +		return -EINVAL;
> +	}
> +	ret = rte_vhost_get_negotiated_features(vid, &features);
> +	if (ret) {
> +		DRV_LOG(ERR, "Failed to get negotiated features.");
> +		return ret;
> +	}
> +	if (RTE_VHOST_NEED_LOG(features)) {
> +		ret = rte_vhost_get_log_base(vid, &log_base, &log_size);
> +		if (ret) {
> +			DRV_LOG(ERR, "Failed to get log base.");
> +			return ret;
> +		}
> +		ret = mlx5_vdpa_dirty_bitmap_set(priv, log_base, log_size);
> +		if (ret) {
> +			DRV_LOG(ERR, "Failed to set dirty bitmap.");
> +			return ret;
> +		}
> +		DRV_LOG(INFO, "mlx5 vdpa: enabling dirty logging...");
> +		ret = mlx5_vdpa_logging_enable(priv, 1);
> +		if (ret) {
> +			DRV_LOG(ERR, "Failed t enable dirty logging.");
> +			return ret;
> +		}
> +	}
> +	return 0;
> +}
> +
>  static struct rte_vdpa_dev_ops mlx5_vdpa_ops = {
>  	.get_queue_num = mlx5_vdpa_get_queue_num,
>  	.get_features = mlx5_vdpa_get_vdpa_features,
> @@ -134,7 +174,7 @@
>  	.dev_conf = NULL,
>  	.dev_close = NULL,
>  	.set_vring_state = mlx5_vdpa_set_vring_state,
> -	.set_features = NULL,
> +	.set_features = mlx5_vdpa_features_set,
>  	.migration_done = NULL,
>  	.get_vfio_group_fd = NULL,
>  	.get_vfio_device_fd = NULL,
> diff --git a/drivers/vdpa/mlx5/mlx5_vdpa.h b/drivers/vdpa/mlx5/mlx5_vdpa.h
> index af78ea1..70264e4 100644
> --- a/drivers/vdpa/mlx5/mlx5_vdpa.h
> +++ b/drivers/vdpa/mlx5/mlx5_vdpa.h
> @@ -244,4 +244,59 @@ int mlx5_vdpa_event_qp_create(struct mlx5_vdpa_priv *priv, uint16_t desc_n,
>   */
>  int mlx5_vdpa_steer_setup(struct mlx5_vdpa_priv *priv);
>  
> +/**
> + * Enable\Disable live migration logging.
> + *
> + * @param[in] priv
> + *   The vdpa driver private structure.
> + * @param[in] enable
> + *   Set for enable, unset for disable.
> + *
> + * @return
> + *   0 on success, a negative value otherwise.
> + */
> +int mlx5_vdpa_logging_enable(struct mlx5_vdpa_priv *priv, int enable);
> +
> +/**
> + * Set dirty bitmap logging to allow live migration.
> + *
> + * @param[in] priv
> + *   The vdpa driver private structure.
> + * @param[in] log_base
> + *   Vhost log base.
> + * @param[in] log_size
> + *   Vhost log size.
> + *
> + * @return
> + *   0 on success, a negative value otherwise.
> + */
> +int mlx5_vdpa_dirty_bitmap_set(struct mlx5_vdpa_priv *priv, uint64_t log_base,
> +			       uint64_t log_size);
> +
> +/**
> + * Log all virtqs information for live migration.
> + *
> + * @param[in] priv
> + *   The vdpa driver private structure.
> + * @param[in] enable
> + *   Set for enable, unset for disable.
> + *
> + * @return
> + *   0 on success, a negative value otherwise.
> + */
> +int mlx5_vdpa_lm_log(struct mlx5_vdpa_priv *priv);
> +
> +/**
> + * Modify virtq state to be ready or suspend.
> + *
> + * @param[in] virtq
> + *   The vdpa driver private virtq structure.
> + * @param[in] state
> + *   Set for ready, otherwise suspend.
> + *
> + * @return
> + *   0 on success, a negative value otherwise.
> + */
> +int mlx5_vdpa_virtq_modify(struct mlx5_vdpa_virtq *virtq, int state);
> +
>  #endif /* RTE_PMD_MLX5_VDPA_H_ */
> diff --git a/drivers/vdpa/mlx5/mlx5_vdpa_lm.c b/drivers/vdpa/mlx5/mlx5_vdpa_lm.c
> new file mode 100644
> index 0000000..cfeec5f
> --- /dev/null
> +++ b/drivers/vdpa/mlx5/mlx5_vdpa_lm.c
> @@ -0,0 +1,130 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright 2019 Mellanox Technologies, Ltd
> + */
> +#include <rte_malloc.h>
> +#include <rte_errno.h>
> +
> +#include "mlx5_vdpa_utils.h"
> +#include "mlx5_vdpa.h"
> +
> +
> +int
> +mlx5_vdpa_logging_enable(struct mlx5_vdpa_priv *priv, int enable)
> +{
> +	struct mlx5_devx_virtq_attr attr = {
> +		.type = MLX5_VIRTQ_MODIFY_TYPE_DIRTY_BITMAP_DUMP_ENABLE,
> +		.dirty_bitmap_dump_enable = enable,
> +	};
> +	struct mlx5_vdpa_virtq *virtq;
> +
> +	SLIST_FOREACH(virtq, &priv->virtq_list, next) {
> +		attr.queue_index = virtq->index;
> +		if (mlx5_devx_cmd_modify_virtq(virtq->virtq, &attr)) {
> +			DRV_LOG(ERR, "Failed to modify virtq %d logging.",
> +				virtq->index);
> +			return -1;
> +		}
> +	}
> +	return 0;
> +}
> +
> +int
> +mlx5_vdpa_dirty_bitmap_set(struct mlx5_vdpa_priv *priv, uint64_t log_base,
> +			   uint64_t log_size)
> +{
> +	struct mlx5_devx_mkey_attr mkey_attr = {
> +			.addr = (uintptr_t)log_base,
> +			.size = log_size,
> +			.pd = priv->pdn,
> +			.pg_access = 1,
> +			.klm_array = NULL,
> +			.klm_num = 0,
> +	};
> +	struct mlx5_devx_virtq_attr attr = {
> +		.type = MLX5_VIRTQ_MODIFY_TYPE_DIRTY_BITMAP_PARAMS,
> +		.dirty_bitmap_addr = log_base,
> +		.dirty_bitmap_size = log_size,
> +	};
> +	struct mlx5_vdpa_query_mr *mr = rte_malloc(__func__, sizeof(*mr), 0);
> +	struct mlx5_vdpa_virtq *virtq;
> +
> +	if (!mr) {
> +		DRV_LOG(ERR, "Failed to allocate mem for lm mr.");
> +		return -1;
> +	}
> +	mr->umem = mlx5_glue->devx_umem_reg(priv->ctx,
> +					    (void *)(uintptr_t)log_base,
> +					    log_size, IBV_ACCESS_LOCAL_WRITE);
> +	if (!mr->umem) {
> +		DRV_LOG(ERR, "Failed to register umem for lm mr.");
> +		goto err;
> +	}
> +	mkey_attr.umem_id = mr->umem->umem_id;
> +	mr->mkey = mlx5_devx_cmd_mkey_create(priv->ctx, &mkey_attr);
> +	if (!mr->mkey) {
> +		DRV_LOG(ERR, "Failed to create Mkey for lm.");
> +		goto err;
> +	}
> +	attr.dirty_bitmap_mkey = mr->mkey->id;
> +	SLIST_FOREACH(virtq, &priv->virtq_list, next) {
> +		attr.queue_index = virtq->index;
> +		if (mlx5_devx_cmd_modify_virtq(virtq->virtq, &attr)) {
> +			DRV_LOG(ERR, "Failed to modify virtq %d for lm.",
> +				virtq->index);
> +			goto err;
> +		}
> +	}
> +	mr->is_indirect = 0;
> +	SLIST_INSERT_HEAD(&priv->mr_list, mr, next);
> +	return 0;
> +err:
> +	if (mr->mkey)
> +		mlx5_devx_cmd_destroy(mr->mkey);
> +	if (mr->umem)
> +		mlx5_glue->devx_umem_dereg(mr->umem);
> +	rte_free(mr);
> +	return -1;
> +}
> +
> +#define MLX5_VDPA_USED_RING_LEN(size) \
> +	((size) * sizeof(struct vring_used_elem) + sizeof(uint16_t) * 3)
> +
> +int
> +mlx5_vdpa_lm_log(struct mlx5_vdpa_priv *priv)
> +{
> +	struct mlx5_devx_virtq_attr attr = {0};
> +	struct mlx5_vdpa_virtq *virtq;
> +	uint64_t features;
> +	int ret = rte_vhost_get_negotiated_features(priv->vid, &features);
> +
> +	if (ret) {
> +		DRV_LOG(ERR, "Failed to get negotiated features.");
> +		return -1;
> +	}
> +	if (RTE_VHOST_NEED_LOG(features)) {
> +		SLIST_FOREACH(virtq, &priv->virtq_list, next) {
> +			ret = mlx5_vdpa_virtq_modify(virtq, 0);
> +			if (ret)
> +				return -1;
> +			if (mlx5_devx_cmd_query_virtq(virtq->virtq, &attr)) {
> +				DRV_LOG(ERR, "Failed to query virtq %d.",
> +					virtq->index);
> +				return -1;
> +			}
> +			DRV_LOG(INFO, "Query vid %d vring %d: hw_available_idx="
> +				"%d, hw_used_index=%d", priv->vid, virtq->index,
> +				attr.hw_available_index, attr.hw_used_index);
> +			ret = rte_vhost_set_vring_base(priv->vid, virtq->index,
> +						       attr.hw_available_index,
> +						       attr.hw_used_index);
> +			if (ret) {
> +				DRV_LOG(ERR, "Failed to set virtq %d base.",
> +					virtq->index);
> +				return -1;
> +			}
> +			rte_vhost_log_used_vring(priv->vid, virtq->index, 0,
> +				       MLX5_VDPA_USED_RING_LEN(virtq->vq_size));
> +		}
> +	}
> +	return 0;
> +}

To avoid one more level of indentation, I would do:
if (!RTE_VHOST_NEED_LOG(features))
	return 0;

Other than that:
Acked-by: Maxime Coquelin <maxime.coquelin@redhat.com>

Thanks,
Maxime