d/numam-dpdk
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

compress/mlx5: add memory region management

Browse Source 
Mellanox user space drivers don't deal with physical addresses, that's
why any mbuf virtual address moved directly to the HW descriptor(WQE).

The mapping between the virtual address to the physical address is saved
in MR configured by the kernel to the HW.

Each MR has a key that should also be moved to the WQE by the SW.

When the SW see address which is not mapped, it extends the address
range and creates a MR using a system call.

Add memory region cache management:
	- 2 level cache per queue-pair - no locks.
	- 1 shared cache between all the queues using a lock.

Using this way, the MR key search per data-path address is optimized.

Signed-off-by: Matan Azrad <matan@nvidia.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@nvidia.com>
This commit is contained in:
Matan Azrad 2021-01-20 11:29:31 +00:00 committed by Akhil Goyal
parent 39a2c8715f
commit 0165bccdb4
1 changed files with 22 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
drivers/compress/mlx5/mlx5_compress.c
View File

@ -48,6 +48,7 @@ struct mlx5_compress_priv {
struct rte_compressdev_config dev_config;
LIST_HEAD(xform_list, mlx5_compress_xform) xform_list;
rte_spinlock_t xform_sl;
struct mlx5_mr_share_cache mr_scache; /* Global shared MR cache. */
};
struct mlx5_compress_qp {
@ -56,6 +57,7 @@ struct mlx5_compress_qp {
uint16_t pi;
uint16_t ci;
volatile uint64_t *uar_addr;
struct mlx5_mr_ctrl mr_ctrl;
int socket_id;
struct mlx5_devx_cq cq;
struct mlx5_devx_sq sq;
@ -121,6 +123,7 @@ mlx5_compress_qp_release(struct rte_compressdev *dev, uint16_t qp_id)
if (opaq != NULL)
rte_free(opaq);
}
mlx5_mr_btree_free(&qp->mr_ctrl.cache_bh);
rte_free(qp);
dev->data->queue_pairs[qp_id] = NULL;
return 0;
@ -190,6 +193,13 @@ mlx5_compress_qp_setup(struct rte_compressdev *dev, uint16_t qp_id,
rte_errno = ENOMEM;
goto err;
}
if (mlx5_mr_btree_init(&qp->mr_ctrl.cache_bh, MLX5_MR_BTREE_CACHE_N,
priv->dev_config.socket_id)) {
DRV_LOG(ERR, "Cannot allocate MR Btree for qp %u.",
(uint32_t)qp_id);
rte_errno = ENOMEM;
goto err;
}
qp->entries_n = 1 << log_ops_n;
qp->socket_id = socket_id;
qp->qp_id = qp_id;
@ -523,6 +533,17 @@ mlx5_compress_pci_probe(struct rte_pci_driver *pci_drv,
claim_zero(mlx5_glue->close_device(priv->ctx));
return -1;
}
if (mlx5_mr_btree_init(&priv->mr_scache.cache,
MLX5_MR_BTREE_CACHE_N * 2, rte_socket_id()) != 0) {
DRV_LOG(ERR, "Failed to allocate shared cache MR memory.");
mlx5_compress_hw_global_release(priv);
rte_compressdev_pmd_destroy(priv->cdev);
claim_zero(mlx5_glue->close_device(priv->ctx));
rte_errno = ENOMEM;
return -rte_errno;
}
priv->mr_scache.reg_mr_cb = mlx5_common_verbs_reg_mr;
priv->mr_scache.dereg_mr_cb = mlx5_common_verbs_dereg_mr;
pthread_mutex_lock(&priv_list_lock);
TAILQ_INSERT_TAIL(&mlx5_compress_priv_list, priv, next);
pthread_mutex_unlock(&priv_list_lock);
@ -553,6 +574,7 @@ mlx5_compress_pci_remove(struct rte_pci_device *pdev)
TAILQ_REMOVE(&mlx5_compress_priv_list, priv, next);
pthread_mutex_unlock(&priv_list_lock);
if (priv) {
mlx5_mr_release_cache(&priv->mr_scache);
mlx5_compress_hw_global_release(priv);
rte_compressdev_pmd_destroy(priv->cdev);
claim_zero(mlx5_glue->close_device(priv->ctx));