[dpdk-dev] [PATCH 2/5] net/mlx5: remove Memory Region support
Shahaf Shuler
shahafs at mellanox.com
Sun May 6 08:41:37 CEST 2018
Thursday, May 3, 2018 2:17 AM, Yongseok Koh:
> Subject: [dpdk-dev] [PATCH 2/5] net/mlx5: remove Memory Region support
>
> This patch removes current support of Memory Region (MR) in order to
> accommodate the dynamic memory hotplug patch. This patch can be
> compiled but traffic can't flow and HW will raise faults. Subsequent patches
> will add new MR support.
>
> Signed-off-by: Yongseok Koh <yskoh at mellanox.com>
Acked-by: Shahaf Shuler <shahafs at mellanox.com>
> ---
> config/common_base | 1 -
> doc/guides/nics/mlx5.rst | 8 -
> drivers/net/mlx5/Makefile | 4 -
> drivers/net/mlx5/mlx5.c | 4 -
> drivers/net/mlx5/mlx5.h | 10 --
> drivers/net/mlx5/mlx5_defs.h | 11 --
> drivers/net/mlx5/mlx5_mr.c | 343 ----------------------------------------
> drivers/net/mlx5/mlx5_rxq.c | 24 +--
> drivers/net/mlx5/mlx5_rxtx.h | 90 +----------
> drivers/net/mlx5/mlx5_trigger.c | 14 --
> drivers/net/mlx5/mlx5_txq.c | 17 --
> 11 files changed, 5 insertions(+), 521 deletions(-)
>
> diff --git a/config/common_base b/config/common_base index
> 03a8688b5..bf7d5e785 100644
> --- a/config/common_base
> +++ b/config/common_base
> @@ -296,7 +296,6 @@ CONFIG_RTE_LIBRTE_MLX4_TX_MP_CACHE=8
> CONFIG_RTE_LIBRTE_MLX5_PMD=n
> CONFIG_RTE_LIBRTE_MLX5_DEBUG=n
> CONFIG_RTE_LIBRTE_MLX5_DLOPEN_DEPS=n
> -CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE=8
>
> #
> # Compile burst-oriented Netronome NFP PMD driver diff --git
> a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst index
> 853c48f81..0fe6e1835 100644
> --- a/doc/guides/nics/mlx5.rst
> +++ b/doc/guides/nics/mlx5.rst
> @@ -167,14 +167,6 @@ These options can be modified in the ``.config`` file.
> adds additional run-time checks and debugging messages at the cost of
> lower performance.
>
> -- ``CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE`` (default **8**)
> -
> - Maximum number of cached memory pools (MPs) per TX queue. Each MP
> from
> - which buffers are to be transmitted must be associated to memory regions
> - (MRs). This is a slow operation that must be cached.
> -
> - This value is always 1 for RX queues since they use a single MP.
> -
> Environment variables
> ~~~~~~~~~~~~~~~~~~~~~
>
> diff --git a/drivers/net/mlx5/Makefile b/drivers/net/mlx5/Makefile index
> 3c5b4943a..13f079334 100644
> --- a/drivers/net/mlx5/Makefile
> +++ b/drivers/net/mlx5/Makefile
> @@ -82,10 +82,6 @@ else
> CFLAGS += -DNDEBUG -UPEDANTIC
> endif
>
> -ifdef CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACHE
> -CFLAGS += -
> DMLX5_PMD_TX_MP_CACHE=$(CONFIG_RTE_LIBRTE_MLX5_TX_MP_CACH
> E)
> -endif
> -
> include $(RTE_SDK)/mk/rte.lib.mk
>
> # Generate and clean-up mlx5_autoconf.h.
> diff --git a/drivers/net/mlx5/mlx5.c b/drivers/net/mlx5/mlx5.c index
> 6c4a571ab..01d554758 100644
> --- a/drivers/net/mlx5/mlx5.c
> +++ b/drivers/net/mlx5/mlx5.c
> @@ -245,10 +245,6 @@ mlx5_dev_close(struct rte_eth_dev *dev)
> if (ret)
> DRV_LOG(WARNING, "port %u some flows still remain",
> dev->data->port_id);
> - ret = mlx5_mr_verify(dev);
> - if (ret)
> - DRV_LOG(WARNING, "port %u some memory region still
> remain",
> - dev->data->port_id);
> memset(priv, 0, sizeof(*priv));
> }
>
> diff --git a/drivers/net/mlx5/mlx5.h b/drivers/net/mlx5/mlx5.h index
> 3ab16bfa2..47d266c90 100644
> --- a/drivers/net/mlx5/mlx5.h
> +++ b/drivers/net/mlx5/mlx5.h
> @@ -26,7 +26,6 @@
> #include <rte_pci.h>
> #include <rte_ether.h>
> #include <rte_ethdev_driver.h>
> -#include <rte_spinlock.h>
> #include <rte_interrupts.h>
> #include <rte_errno.h>
> #include <rte_flow.h>
> @@ -147,7 +146,6 @@ struct priv {
> struct mlx5_hrxq_drop *flow_drop_queue; /* Flow drop queue. */
> struct mlx5_flows flows; /* RTE Flow rules. */
> struct mlx5_flows ctrl_flows; /* Control flow rules. */
> - LIST_HEAD(mr, mlx5_mr) mr; /* Memory region. */
> LIST_HEAD(rxq, mlx5_rxq_ctrl) rxqsctrl; /* DPDK Rx queues. */
> LIST_HEAD(rxqibv, mlx5_rxq_ibv) rxqsibv; /* Verbs Rx queues. */
> LIST_HEAD(hrxq, mlx5_hrxq) hrxqs; /* Verbs Hash Rx queues. */
> @@ -157,7 +155,6 @@ struct priv {
> LIST_HEAD(ind_tables, mlx5_ind_table_ibv) ind_tbls;
> uint32_t link_speed_capa; /* Link speed capabilities. */
> struct mlx5_xstats_ctrl xstats_ctrl; /* Extended stats control. */
> - rte_spinlock_t mr_lock; /* MR Lock. */
> int primary_socket; /* Unix socket for primary process. */
> void *uar_base; /* Reserved address space for UAR mapping */
> struct rte_intr_handle intr_handle_socket; /* Interrupt handler. */
> @@ -309,13 +306,6 @@ void mlx5_socket_uninit(struct rte_eth_dev *priv);
> void mlx5_socket_handle(struct rte_eth_dev *priv); int
> mlx5_socket_connect(struct rte_eth_dev *priv);
>
> -/* mlx5_mr.c */
> -
> -struct mlx5_mr *mlx5_mr_new(struct rte_eth_dev *dev, struct
> rte_mempool *mp); -struct mlx5_mr *mlx5_mr_get(struct rte_eth_dev
> *dev, struct rte_mempool *mp); -int mlx5_mr_release(struct mlx5_mr *mr);
> -int mlx5_mr_verify(struct rte_eth_dev *dev);
> -
> /* mlx5_nl.c */
>
> int mlx5_nl_init(uint32_t nlgroups);
> diff --git a/drivers/net/mlx5/mlx5_defs.h b/drivers/net/mlx5/mlx5_defs.h
> index 55a86957d..f9093777d 100644
> --- a/drivers/net/mlx5/mlx5_defs.h
> +++ b/drivers/net/mlx5/mlx5_defs.h
> @@ -38,17 +38,6 @@
> #define MLX5_TX_COMP_THRESH_INLINE_DIV (1 << 3)
>
> /*
> - * Maximum number of cached Memory Pools (MPs) per TX queue. Each
> RTE MP
> - * from which buffers are to be transmitted will have to be mapped by this
> - * driver to their own Memory Region (MR). This is a slow operation.
> - *
> - * This value is always 1 for RX queues.
> - */
> -#ifndef MLX5_PMD_TX_MP_CACHE
> -#define MLX5_PMD_TX_MP_CACHE 8
> -#endif
> -
> -/*
> * If defined, only use software counters. The PMD will never ask the
> hardware
> * for these, and many of them won't be available.
> */
> diff --git a/drivers/net/mlx5/mlx5_mr.c b/drivers/net/mlx5/mlx5_mr.c index
> 6613bd6b9..736c40ae4 100644
> --- a/drivers/net/mlx5/mlx5_mr.c
> +++ b/drivers/net/mlx5/mlx5_mr.c
> @@ -18,346 +18,3 @@
> #include "mlx5_rxtx.h"
> #include "mlx5_glue.h"
>
> -struct mlx5_check_mempool_data {
> - int ret;
> - char *start;
> - char *end;
> -};
> -
> -/* Called by mlx5_check_mempool() when iterating the memory chunks. */
> -static void -mlx5_check_mempool_cb(struct rte_mempool *mp
> __rte_unused,
> - void *opaque, struct rte_mempool_memhdr *memhdr,
> - unsigned int mem_idx __rte_unused)
> -{
> - struct mlx5_check_mempool_data *data = opaque;
> -
> - /* It already failed, skip the next chunks. */
> - if (data->ret != 0)
> - return;
> - /* It is the first chunk. */
> - if (data->start == NULL && data->end == NULL) {
> - data->start = memhdr->addr;
> - data->end = data->start + memhdr->len;
> - return;
> - }
> - if (data->end == memhdr->addr) {
> - data->end += memhdr->len;
> - return;
> - }
> - if (data->start == (char *)memhdr->addr + memhdr->len) {
> - data->start -= memhdr->len;
> - return;
> - }
> - /* Error, mempool is not virtually contiguous. */
> - data->ret = -1;
> -}
> -
> -/**
> - * Check if a mempool can be used: it must be virtually contiguous.
> - *
> - * @param[in] mp
> - * Pointer to memory pool.
> - * @param[out] start
> - * Pointer to the start address of the mempool virtual memory area
> - * @param[out] end
> - * Pointer to the end address of the mempool virtual memory area
> - *
> - * @return
> - * 0 on success (mempool is virtually contiguous), -1 on error.
> - */
> -static int
> -mlx5_check_mempool(struct rte_mempool *mp, uintptr_t *start,
> - uintptr_t *end)
> -{
> - struct mlx5_check_mempool_data data;
> -
> - memset(&data, 0, sizeof(data));
> - rte_mempool_mem_iter(mp, mlx5_check_mempool_cb, &data);
> - *start = (uintptr_t)data.start;
> - *end = (uintptr_t)data.end;
> - return data.ret;
> -}
> -
> -/**
> - * Register a Memory Region (MR) <-> Memory Pool (MP) association in
> - * txq->mp2mr[]. If mp2mr[] is full, remove an entry first.
> - *
> - * @param txq
> - * Pointer to TX queue structure.
> - * @param[in] mp
> - * Memory Pool for which a Memory Region lkey must be returned.
> - * @param idx
> - * Index of the next available entry.
> - *
> - * @return
> - * mr on success, NULL on failure and rte_errno is set.
> - */
> -struct mlx5_mr *
> -mlx5_txq_mp2mr_reg(struct mlx5_txq_data *txq, struct rte_mempool
> *mp,
> - unsigned int idx)
> -{
> - struct mlx5_txq_ctrl *txq_ctrl =
> - container_of(txq, struct mlx5_txq_ctrl, txq);
> - struct rte_eth_dev *dev;
> - struct mlx5_mr *mr;
> -
> - rte_spinlock_lock(&txq_ctrl->priv->mr_lock);
> - /* Add a new entry, register MR first. */
> - DRV_LOG(DEBUG, "port %u discovered new memory pool \"%s\"
> (%p)",
> - port_id(txq_ctrl->priv), mp->name, (void *)mp);
> - dev = eth_dev(txq_ctrl->priv);
> - mr = mlx5_mr_get(dev, mp);
> - if (mr == NULL) {
> - if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
> - DRV_LOG(DEBUG,
> - "port %u using unregistered mempool
> 0x%p(%s)"
> - " in secondary process, please create
> mempool"
> - " before rte_eth_dev_start()",
> - port_id(txq_ctrl->priv), (void *)mp, mp-
> >name);
> - rte_spinlock_unlock(&txq_ctrl->priv->mr_lock);
> - rte_errno = ENOTSUP;
> - return NULL;
> - }
> - mr = mlx5_mr_new(dev, mp);
> - }
> - if (unlikely(mr == NULL)) {
> - DRV_LOG(DEBUG,
> - "port %u unable to configure memory region,"
> - " ibv_reg_mr() failed.",
> - port_id(txq_ctrl->priv));
> - rte_spinlock_unlock(&txq_ctrl->priv->mr_lock);
> - return NULL;
> - }
> - if (unlikely(idx == RTE_DIM(txq->mp2mr))) {
> - /* Table is full, remove oldest entry. */
> - DRV_LOG(DEBUG,
> - "port %u memory region <-> memory pool table full,
> "
> - " dropping oldest entry",
> - port_id(txq_ctrl->priv));
> - --idx;
> - mlx5_mr_release(txq->mp2mr[0]);
> - memmove(&txq->mp2mr[0], &txq->mp2mr[1],
> - (sizeof(txq->mp2mr) - sizeof(txq->mp2mr[0])));
> - }
> - /* Store the new entry. */
> - txq_ctrl->txq.mp2mr[idx] = mr;
> - DRV_LOG(DEBUG,
> - "port %u new memory region lkey for MP \"%s\" (%p):
> 0x%08"
> - PRIu32,
> - port_id(txq_ctrl->priv), mp->name, (void *)mp,
> - txq_ctrl->txq.mp2mr[idx]->lkey);
> - rte_spinlock_unlock(&txq_ctrl->priv->mr_lock);
> - return mr;
> -}
> -
> -struct mlx5_mp2mr_mbuf_check_data {
> - int ret;
> -};
> -
> -/**
> - * Callback function for rte_mempool_obj_iter() to check whether a given
> - * mempool object looks like a mbuf.
> - *
> - * @param[in] mp
> - * The mempool pointer
> - * @param[in] arg
> - * Context data (struct txq_mp2mr_mbuf_check_data). Contains the
> - * return value.
> - * @param[in] obj
> - * Object address.
> - * @param index
> - * Object index, unused.
> - */
> -static void
> -txq_mp2mr_mbuf_check(struct rte_mempool *mp, void *arg, void *obj,
> - uint32_t index __rte_unused)
> -{
> - struct mlx5_mp2mr_mbuf_check_data *data = arg;
> - struct rte_mbuf *buf = obj;
> -
> - /*
> - * Check whether mbuf structure fits element size and whether
> mempool
> - * pointer is valid.
> - */
> - if (sizeof(*buf) > mp->elt_size || buf->pool != mp)
> - data->ret = -1;
> -}
> -
> -/**
> - * Iterator function for rte_mempool_walk() to register existing mempools
> and
> - * fill the MP to MR cache of a TX queue.
> - *
> - * @param[in] mp
> - * Memory Pool to register.
> - * @param *arg
> - * Pointer to TX queue structure.
> - */
> -void
> -mlx5_mp2mr_iter(struct rte_mempool *mp, void *arg) -{
> - struct priv *priv = (struct priv *)arg;
> - struct mlx5_mp2mr_mbuf_check_data data = {
> - .ret = 0,
> - };
> - struct mlx5_mr *mr;
> -
> - /* Register mempool only if the first element looks like a mbuf. */
> - if (rte_mempool_obj_iter(mp, txq_mp2mr_mbuf_check, &data) ==
> 0 ||
> - data.ret == -1)
> - return;
> - mr = mlx5_mr_get(eth_dev(priv), mp);
> - if (mr) {
> - mlx5_mr_release(mr);
> - return;
> - }
> - mr = mlx5_mr_new(eth_dev(priv), mp);
> - if (!mr)
> - DRV_LOG(ERR, "port %u cannot create memory region: %s",
> - port_id(priv), strerror(rte_errno));
> -}
> -
> -/**
> - * Register a new memory region from the mempool and store it in the
> memory
> - * region list.
> - *
> - * @param dev
> - * Pointer to Ethernet device.
> - * @param mp
> - * Pointer to the memory pool to register.
> - *
> - * @return
> - * The memory region on success, NULL on failure and rte_errno is set.
> - */
> -struct mlx5_mr *
> -mlx5_mr_new(struct rte_eth_dev *dev, struct rte_mempool *mp) -{
> - struct priv *priv = dev->data->dev_private;
> - const struct rte_memseg *ms;
> - uintptr_t start;
> - uintptr_t end;
> - struct mlx5_mr *mr;
> -
> - mr = rte_zmalloc_socket(__func__, sizeof(*mr), 0, mp->socket_id);
> - if (!mr) {
> - DRV_LOG(DEBUG,
> - "port %u unable to configure memory region,"
> - " ibv_reg_mr() failed.",
> - dev->data->port_id);
> - rte_errno = ENOMEM;
> - return NULL;
> - }
> - if (mlx5_check_mempool(mp, &start, &end) != 0) {
> - DRV_LOG(ERR, "port %u mempool %p: not virtually
> contiguous",
> - dev->data->port_id, (void *)mp);
> - rte_errno = ENOMEM;
> - return NULL;
> - }
> - DRV_LOG(DEBUG, "port %u mempool %p area start=%p end=%p
> size=%zu",
> - dev->data->port_id, (void *)mp, (void *)start, (void *)end,
> - (size_t)(end - start));
> - /* Save original addresses for exact MR lookup. */
> - mr->start = start;
> - mr->end = end;
> -
> - /* Round start and end to page boundary if found in memory
> segments. */
> - ms = rte_mem_virt2memseg((void *)start, NULL);
> - if (ms != NULL)
> - start = RTE_ALIGN_FLOOR(start, ms->hugepage_sz);
> - end = RTE_ALIGN_CEIL(end, ms->hugepage_sz);
> - DRV_LOG(DEBUG,
> - "port %u mempool %p using start=%p end=%p size=%zu for
> memory"
> - " region",
> - dev->data->port_id, (void *)mp, (void *)start, (void *)end,
> - (size_t)(end - start));
> - mr->mr = mlx5_glue->reg_mr(priv->pd, (void *)start, end - start,
> - IBV_ACCESS_LOCAL_WRITE);
> - if (!mr->mr) {
> - rte_errno = ENOMEM;
> - return NULL;
> - }
> - mr->mp = mp;
> - mr->lkey = rte_cpu_to_be_32(mr->mr->lkey);
> - rte_atomic32_inc(&mr->refcnt);
> - DRV_LOG(DEBUG, "port %u new memory Region %p refcnt: %d",
> - dev->data->port_id, (void *)mr, rte_atomic32_read(&mr-
> >refcnt));
> - LIST_INSERT_HEAD(&priv->mr, mr, next);
> - return mr;
> -}
> -
> -/**
> - * Search the memory region object in the memory region list.
> - *
> - * @param dev
> - * Pointer to Ethernet device.
> - * @param mp
> - * Pointer to the memory pool to register.
> - *
> - * @return
> - * The memory region on success.
> - */
> -struct mlx5_mr *
> -mlx5_mr_get(struct rte_eth_dev *dev, struct rte_mempool *mp) -{
> - struct priv *priv = dev->data->dev_private;
> - struct mlx5_mr *mr;
> -
> - assert(mp);
> - if (LIST_EMPTY(&priv->mr))
> - return NULL;
> - LIST_FOREACH(mr, &priv->mr, next) {
> - if (mr->mp == mp) {
> - rte_atomic32_inc(&mr->refcnt);
> - return mr;
> - }
> - }
> - return NULL;
> -}
> -
> -/**
> - * Release the memory region object.
> - *
> - * @param mr
> - * Pointer to memory region to release.
> - *
> - * @return
> - * 1 while a reference on it exists, 0 when freed.
> - */
> -int
> -mlx5_mr_release(struct mlx5_mr *mr)
> -{
> - assert(mr);
> - if (rte_atomic32_dec_and_test(&mr->refcnt)) {
> - DRV_LOG(DEBUG, "memory region %p refcnt: %d", (void
> *)mr,
> - rte_atomic32_read(&mr->refcnt));
> - claim_zero(mlx5_glue->dereg_mr(mr->mr));
> - LIST_REMOVE(mr, next);
> - rte_free(mr);
> - return 0;
> - }
> - return 1;
> -}
> -
> -/**
> - * Verify the flow list is empty
> - *
> - * @param dev
> - * Pointer to Ethernet device.
> - *
> - * @return
> - * The number of object not released.
> - */
> -int
> -mlx5_mr_verify(struct rte_eth_dev *dev) -{
> - struct priv *priv = dev->data->dev_private;
> - int ret = 0;
> - struct mlx5_mr *mr;
> -
> - LIST_FOREACH(mr, &priv->mr, next) {
> - DRV_LOG(DEBUG, "port %u memory region %p still
> referenced",
> - dev->data->port_id, (void *)mr);
> - ++ret;
> - }
> - return ret;
> -}
> diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c
> index d993e3846..d4fe1fed7 100644
> --- a/drivers/net/mlx5/mlx5_rxq.c
> +++ b/drivers/net/mlx5/mlx5_rxq.c
> @@ -649,16 +649,6 @@ mlx5_rxq_ibv_new(struct rte_eth_dev *dev,
> uint16_t idx)
> goto error;
> }
> tmpl->rxq_ctrl = rxq_ctrl;
> - /* Use the entire RX mempool as the memory region. */
> - tmpl->mr = mlx5_mr_get(dev, rxq_data->mp);
> - if (!tmpl->mr) {
> - tmpl->mr = mlx5_mr_new(dev, rxq_data->mp);
> - if (!tmpl->mr) {
> - DRV_LOG(ERR, "port %u: memeroy region creation
> failure",
> - dev->data->port_id);
> - goto error;
> - }
> - }
> if (rxq_ctrl->irq) {
> tmpl->channel = mlx5_glue->create_comp_channel(priv-
> >ctx);
> if (!tmpl->channel) {
> @@ -799,7 +789,7 @@ mlx5_rxq_ibv_new(struct rte_eth_dev *dev,
> uint16_t idx)
> .addr = rte_cpu_to_be_64(rte_pktmbuf_mtod(buf,
> uintptr_t)),
> .byte_count = rte_cpu_to_be_32(DATA_LEN(buf)),
> - .lkey = tmpl->mr->lkey,
> + .lkey = UINT32_MAX,
> };
> }
> rxq_data->rq_db = rwq.dbrec;
> @@ -835,8 +825,6 @@ mlx5_rxq_ibv_new(struct rte_eth_dev *dev,
> uint16_t idx)
> claim_zero(mlx5_glue->destroy_cq(tmpl->cq));
> if (tmpl->channel)
> claim_zero(mlx5_glue->destroy_comp_channel(tmpl-
> >channel));
> - if (tmpl->mr)
> - mlx5_mr_release(tmpl->mr);
> priv->verbs_alloc_ctx.type = MLX5_VERBS_ALLOC_TYPE_NONE;
> rte_errno = ret; /* Restore rte_errno. */
> return NULL;
> @@ -865,10 +853,8 @@ mlx5_rxq_ibv_get(struct rte_eth_dev *dev,
> uint16_t idx)
> if (!rxq_data)
> return NULL;
> rxq_ctrl = container_of(rxq_data, struct mlx5_rxq_ctrl, rxq);
> - if (rxq_ctrl->ibv) {
> - mlx5_mr_get(dev, rxq_data->mp);
> + if (rxq_ctrl->ibv)
> rte_atomic32_inc(&rxq_ctrl->ibv->refcnt);
> - }
> return rxq_ctrl->ibv;
> }
>
> @@ -884,15 +870,9 @@ mlx5_rxq_ibv_get(struct rte_eth_dev *dev,
> uint16_t idx) int mlx5_rxq_ibv_release(struct mlx5_rxq_ibv *rxq_ibv) {
> - int ret;
> -
> assert(rxq_ibv);
> assert(rxq_ibv->wq);
> assert(rxq_ibv->cq);
> - assert(rxq_ibv->mr);
> - ret = mlx5_mr_release(rxq_ibv->mr);
> - if (!ret)
> - rxq_ibv->mr = NULL;
> if (rte_atomic32_dec_and_test(&rxq_ibv->refcnt)) {
> DRV_LOG(DEBUG, "port %u Verbs Rx queue %u: refcnt %d",
> port_id(rxq_ibv->rxq_ctrl->priv),
> diff --git a/drivers/net/mlx5/mlx5_rxtx.h b/drivers/net/mlx5/mlx5_rxtx.h
> index 2fc12a186..e8cad51aa 100644
> --- a/drivers/net/mlx5/mlx5_rxtx.h
> +++ b/drivers/net/mlx5/mlx5_rxtx.h
> @@ -54,17 +54,6 @@ struct mlx5_txq_stats {
>
> struct priv;
>
> -/* Memory region queue object. */
> -struct mlx5_mr {
> - LIST_ENTRY(mlx5_mr) next; /**< Pointer to the next element. */
> - rte_atomic32_t refcnt; /*<< Reference counter. */
> - uint32_t lkey; /*<< rte_cpu_to_be_32(mr->lkey) */
> - uintptr_t start; /* Start address of MR */
> - uintptr_t end; /* End address of MR */
> - struct ibv_mr *mr; /*<< Memory Region. */
> - struct rte_mempool *mp; /*<< Memory Pool. */
> -};
> -
> /* Compressed CQE context. */
> struct rxq_zip {
> uint16_t ai; /* Array index. */
> @@ -114,7 +103,6 @@ struct mlx5_rxq_ibv {
> struct ibv_cq *cq; /* Completion Queue. */
> struct ibv_wq *wq; /* Work Queue. */
> struct ibv_comp_channel *channel;
> - struct mlx5_mr *mr; /* Memory Region (for mp). */
> };
>
> /* RX queue control descriptor. */
> @@ -175,7 +163,6 @@ struct mlx5_txq_data {
> uint16_t mpw_hdr_dseg:1; /* Enable DSEGs in the title WQEBB. */
> uint16_t max_inline; /* Multiple of RTE_CACHE_LINE_SIZE to inline.
> */
> uint16_t inline_max_packet_sz; /* Max packet size for inlining. */
> - uint16_t mr_cache_idx; /* Index of last hit entry. */
> uint32_t qp_num_8s; /* QP number shifted by 8. */
> uint64_t offloads; /* Offloads for Tx Queue. */
> volatile struct mlx5_cqe (*cqes)[]; /* Completion queue. */ @@ -
> 183,7 +170,6 @@ struct mlx5_txq_data {
> volatile uint32_t *qp_db; /* Work queue doorbell. */
> volatile uint32_t *cq_db; /* Completion queue doorbell. */
> volatile void *bf_reg; /* Blueflame register remapped. */
> - struct mlx5_mr *mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MR
> translation table. */
> struct rte_mbuf *(*elts)[]; /* TX elements. */
> struct mlx5_txq_stats stats; /* TX queue counters. */ }
> __rte_cache_aligned; @@ -322,12 +308,6 @@ uint16_t
> mlx5_tx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t
> mlx5_rx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts,
> uint16_t pkts_n);
>
> -/* mlx5_mr.c */
> -
> -void mlx5_mp2mr_iter(struct rte_mempool *mp, void *arg); -struct
> mlx5_mr *mlx5_txq_mp2mr_reg(struct mlx5_txq_data *txq,
> - struct rte_mempool *mp, unsigned int idx);
> -
> #ifndef NDEBUG
> /**
> * Verify or set magic value in CQE.
> @@ -513,76 +493,12 @@ mlx5_tx_complete(struct mlx5_txq_data *txq)
> *txq->cq_db = rte_cpu_to_be_32(cq_ci); }
>
> -/**
> - * Get Memory Pool (MP) from mbuf. If mbuf is indirect, the pool from
> which
> - * the cloned mbuf is allocated is returned instead.
> - *
> - * @param buf
> - * Pointer to mbuf.
> - *
> - * @return
> - * Memory pool where data is located for given mbuf.
> - */
> -static struct rte_mempool *
> -mlx5_tx_mb2mp(struct rte_mbuf *buf)
> -{
> - if (unlikely(RTE_MBUF_INDIRECT(buf)))
> - return rte_mbuf_from_indirect(buf)->pool;
> - return buf->pool;
> -}
> -
> -/**
> - * Get Memory Region (MR) <-> rte_mbuf association from txq->mp2mr[].
> - * Add MP to txq->mp2mr[] if it's not registered yet. If mp2mr[] is full,
> - * remove an entry first.
> - *
> - * @param txq
> - * Pointer to TX queue structure.
> - * @param[in] mp
> - * Memory Pool for which a Memory Region lkey must be returned.
> - *
> - * @return
> - * mr->lkey on success, (uint32_t)-1 on failure.
> - */
> static __rte_always_inline uint32_t
> mlx5_tx_mb2mr(struct mlx5_txq_data *txq, struct rte_mbuf *mb) {
> - uint16_t i = txq->mr_cache_idx;
> - uintptr_t addr = rte_pktmbuf_mtod(mb, uintptr_t);
> - struct mlx5_mr *mr;
> -
> - assert(i < RTE_DIM(txq->mp2mr));
> - if (likely(txq->mp2mr[i]->start <= addr && txq->mp2mr[i]->end >
> addr))
> - return txq->mp2mr[i]->lkey;
> - for (i = 0; (i != RTE_DIM(txq->mp2mr)); ++i) {
> - if (unlikely(txq->mp2mr[i] == NULL ||
> - txq->mp2mr[i]->mr == NULL)) {
> - /* Unknown MP, add a new MR for it. */
> - break;
> - }
> - if (txq->mp2mr[i]->start <= addr &&
> - txq->mp2mr[i]->end > addr) {
> - assert(txq->mp2mr[i]->lkey != (uint32_t)-1);
> - txq->mr_cache_idx = i;
> - return txq->mp2mr[i]->lkey;
> - }
> - }
> - mr = mlx5_txq_mp2mr_reg(txq, mlx5_tx_mb2mp(mb), i);
> - /*
> - * Request the reference to use in this queue, the original one is
> - * kept by the control plane.
> - */
> - if (mr) {
> - rte_atomic32_inc(&mr->refcnt);
> - txq->mr_cache_idx = i >= RTE_DIM(txq->mp2mr) ? i - 1 : i;
> - return mr->lkey;
> - } else {
> - struct rte_mempool *mp = mlx5_tx_mb2mp(mb);
> -
> - DRV_LOG(WARNING, "failed to register mempool
> 0x%p(%s)",
> - (void *)mp, mp->name);
> - }
> - return (uint32_t)-1;
> + (void)txq;
> + (void)mb;
> + return UINT32_MAX;
> }
>
> /**
> diff --git a/drivers/net/mlx5/mlx5_trigger.c
> b/drivers/net/mlx5/mlx5_trigger.c index fc56d1ee8..3db6c3f35 100644
> --- a/drivers/net/mlx5/mlx5_trigger.c
> +++ b/drivers/net/mlx5/mlx5_trigger.c
> @@ -48,17 +48,10 @@ mlx5_txq_start(struct rte_eth_dev *dev)
>
> /* Add memory regions to Tx queues. */
> for (i = 0; i != priv->txqs_n; ++i) {
> - unsigned int idx = 0;
> - struct mlx5_mr *mr;
> struct mlx5_txq_ctrl *txq_ctrl = mlx5_txq_get(dev, i);
>
> if (!txq_ctrl)
> continue;
> - LIST_FOREACH(mr, &priv->mr, next) {
> - mlx5_txq_mp2mr_reg(&txq_ctrl->txq, mr->mp,
> idx++);
> - if (idx == MLX5_PMD_TX_MP_CACHE)
> - break;
> - }
> txq_alloc_elts(txq_ctrl);
> txq_ctrl->ibv = mlx5_txq_ibv_new(dev, i);
> if (!txq_ctrl->ibv) {
> @@ -144,13 +137,11 @@ int
> mlx5_dev_start(struct rte_eth_dev *dev) {
> struct priv *priv = dev->data->dev_private;
> - struct mlx5_mr *mr = NULL;
> int ret;
>
> dev->data->dev_started = 1;
> DRV_LOG(DEBUG, "port %u allocating and configuring hash Rx
> queues",
> dev->data->port_id);
> - rte_mempool_walk(mlx5_mp2mr_iter, priv);
> ret = mlx5_txq_start(dev);
> if (ret) {
> DRV_LOG(ERR, "port %u Tx queue allocation failed: %s", @@
> -190,8 +181,6 @@ mlx5_dev_start(struct rte_eth_dev *dev)
> ret = rte_errno; /* Save rte_errno before cleanup. */
> /* Rollback. */
> dev->data->dev_started = 0;
> - for (mr = LIST_FIRST(&priv->mr); mr; mr = LIST_FIRST(&priv->mr))
> - mlx5_mr_release(mr);
> mlx5_flow_stop(dev, &priv->flows);
> mlx5_traffic_disable(dev);
> mlx5_txq_stop(dev);
> @@ -212,7 +201,6 @@ void
> mlx5_dev_stop(struct rte_eth_dev *dev)
> {
> struct priv *priv = dev->data->dev_private;
> - struct mlx5_mr *mr;
>
> dev->data->dev_started = 0;
> /* Prevent crashes when queues are still in use. */ @@ -228,8 +216,6
> @@ mlx5_dev_stop(struct rte_eth_dev *dev)
> mlx5_dev_interrupt_handler_uninstall(dev);
> mlx5_txq_stop(dev);
> mlx5_rxq_stop(dev);
> - for (mr = LIST_FIRST(&priv->mr); mr; mr = LIST_FIRST(&priv->mr))
> - mlx5_mr_release(mr);
> }
>
> /**
> diff --git a/drivers/net/mlx5/mlx5_txq.c b/drivers/net/mlx5/mlx5_txq.c
> index 3f4b5fea5..a71f3d0f0 100644
> --- a/drivers/net/mlx5/mlx5_txq.c
> +++ b/drivers/net/mlx5/mlx5_txq.c
> @@ -409,7 +409,6 @@ mlx5_txq_ibv_new(struct rte_eth_dev *dev,
> uint16_t idx)
> return NULL;
> }
> memset(&tmpl, 0, sizeof(struct mlx5_txq_ibv));
> - /* MRs will be registered in mp2mr[] later. */
> attr.cq = (struct ibv_cq_init_attr_ex){
> .comp_mask = 0,
> };
> @@ -812,7 +811,6 @@ mlx5_txq_new(struct rte_eth_dev *dev, uint16_t
> idx, uint16_t desc,
> tmpl->txq.elts_n = log2above(desc);
> tmpl->idx = idx;
> txq_set_params(tmpl);
> - /* MRs will be registered in mp2mr[] later. */
> DRV_LOG(DEBUG, "port %u priv->device_attr.max_qp_wr is %d",
> dev->data->port_id, priv-
> >device_attr.orig_attr.max_qp_wr);
> DRV_LOG(DEBUG, "port %u priv->device_attr.max_sge is %d", @@ -
> 847,15 +845,7 @@ mlx5_txq_get(struct rte_eth_dev *dev, uint16_t idx)
> if ((*priv->txqs)[idx]) {
> ctrl = container_of((*priv->txqs)[idx], struct mlx5_txq_ctrl,
> txq);
> - unsigned int i;
> -
> mlx5_txq_ibv_get(dev, idx);
> - for (i = 0; i != MLX5_PMD_TX_MP_CACHE; ++i) {
> - if (ctrl->txq.mp2mr[i])
> - claim_nonzero
> - (mlx5_mr_get(dev,
> - ctrl->txq.mp2mr[i]->mp));
> - }
> rte_atomic32_inc(&ctrl->refcnt);
> }
> return ctrl;
> @@ -876,7 +866,6 @@ int
> mlx5_txq_release(struct rte_eth_dev *dev, uint16_t idx) {
> struct priv *priv = dev->data->dev_private;
> - unsigned int i;
> struct mlx5_txq_ctrl *txq;
> size_t page_size = sysconf(_SC_PAGESIZE);
>
> @@ -885,12 +874,6 @@ mlx5_txq_release(struct rte_eth_dev *dev,
> uint16_t idx)
> txq = container_of((*priv->txqs)[idx], struct mlx5_txq_ctrl, txq);
> if (txq->ibv && !mlx5_txq_ibv_release(txq->ibv))
> txq->ibv = NULL;
> - for (i = 0; i != MLX5_PMD_TX_MP_CACHE; ++i) {
> - if (txq->txq.mp2mr[i]) {
> - mlx5_mr_release(txq->txq.mp2mr[i]);
> - txq->txq.mp2mr[i] = NULL;
> - }
> - }
> if (priv->uar_base)
> munmap((void *)RTE_ALIGN_FLOOR((uintptr_t)txq-
> >txq.bf_reg,
> page_size), page_size);
> --
> 2.11.0
More information about the dev
mailing list