DPDK patches and discussions
 help / color / mirror / Atom feed
From: Slava Ovsiienko <viacheslavo@mellanox.com>
To: Vu Pham <vuhuong@mellanox.com>, "dev@dpdk.org" <dev@dpdk.org>
Cc: Ori Kam <orika@mellanox.com>, Matan Azrad <matan@mellanox.com>,
	Raslan Darawsheh <rasland@mellanox.com>,
	Vu Pham <vuhuong@mellanox.com>
Subject: Re: [dpdk-dev] [PATCH v3 3/4] common/mlx5: refactor memory management codes
Date: Wed, 8 Apr 2020 09:04:41 +0000	[thread overview]
Message-ID: <AM4PR05MB32652EE4B9324560E4170210D2C00@AM4PR05MB3265.eurprd05.prod.outlook.com> (raw)
In-Reply-To: <20200407170058.9274-4-vuhuong@mellanox.com>

> -----Original Message-----
> From: Vu Pham <vuhuong@mellanox.com>
> Sent: Tuesday, April 7, 2020 20:01
> To: dev@dpdk.org
> Cc: Slava Ovsiienko <viacheslavo@mellanox.com>; Ori Kam
> <orika@mellanox.com>; Matan Azrad <matan@mellanox.com>; Raslan
> Darawsheh <rasland@mellanox.com>; Vu Pham <vuhuong@mellanox.com>
> Subject: [PATCH v3 3/4] common/mlx5: refactor memory management codes
> 
> Refactor common memory btree and cache management to common driver.
> Replace some input parameters of MR APIs to more common datastructure
> like PD, port_id, share_cache,... so that muliptle PMD drivers can use those
> MR APIs.
> 
> Signed-off-by: Vu Pham <vuhuong@mellanox.com>
Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>

> ---
>  drivers/common/mlx5/mlx5_common_mr.c            | 1108
> +++++++++++++++++++++++
>  drivers/common/mlx5/mlx5_common_mr.h            |  160 ++++
>  drivers/common/mlx5/rte_common_mlx5_version.map |   14 +
>  3 files changed, 1282 insertions(+)
>  create mode 100644 drivers/common/mlx5/mlx5_common_mr.c
>  create mode 100644 drivers/common/mlx5/mlx5_common_mr.h
> 
> diff --git a/drivers/common/mlx5/mlx5_common_mr.c
> b/drivers/common/mlx5/mlx5_common_mr.c
> new file mode 100644
> index 0000000000..9d4a06dd5b
> --- /dev/null
> +++ b/drivers/common/mlx5/mlx5_common_mr.c
> @@ -0,0 +1,1108 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright 2016 6WIND S.A.
> + * Copyright 2020 Mellanox Technologies, Ltd  */ #include
> +<rte_eal_memconfig.h> #include <rte_errno.h> #include <rte_mempool.h>
> +#include <rte_malloc.h> #include <rte_rwlock.h>
> +
> +#include "mlx5_glue.h"
> +#include "mlx5_common_mp.h"
> +#include "mlx5_common_mr.h"
> +#include "mlx5_common_utils.h"
> +
> +struct mr_find_contig_memsegs_data {
> +	uintptr_t addr;
> +	uintptr_t start;
> +	uintptr_t end;
> +	const struct rte_memseg_list *msl;
> +};
> +
> +/**
> + * Expand B-tree table to a given size. Can't be called with holding
> + * memory_hotplug_lock or share_cache.rwlock due to rte_realloc().
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + * @param n
> + *   Number of entries for expansion.
> + *
> + * @return
> + *   0 on success, -1 on failure.
> + */
> +static int
> +mr_btree_expand(struct mlx5_mr_btree *bt, int n) {
> +	void *mem;
> +	int ret = 0;
> +
> +	if (n <= bt->size)
> +		return ret;
> +	/*
> +	 * Downside of directly using rte_realloc() is that SOCKET_ID_ANY is
> +	 * used inside if there's no room to expand. Because this is a quite
> +	 * rare case and a part of very slow path, it is very acceptable.
> +	 * Initially cache_bh[] will be given practically enough space and once
> +	 * it is expanded, expansion wouldn't be needed again ever.
> +	 */
> +	mem = rte_realloc(bt->table, n * sizeof(struct mr_cache_entry), 0);
> +	if (mem == NULL) {
> +		/* Not an error, B-tree search will be skipped. */
> +		DRV_LOG(WARNING, "failed to expand MR B-tree (%p) table",
> +			(void *)bt);
> +		ret = -1;
> +	} else {
> +		DRV_LOG(DEBUG, "expanded MR B-tree table (size=%u)", n);
> +		bt->table = mem;
> +		bt->size = n;
> +	}
> +	return ret;
> +}
> +
> +/**
> + * Look up LKey from given B-tree lookup table, store the last index
> +and return
> + * searched LKey.
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + * @param[out] idx
> + *   Pointer to index. Even on search failure, returns index where it stops
> + *   searching so that index can be used when inserting a new entry.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on no match.
> + */
> +static uint32_t
> +mr_btree_lookup(struct mlx5_mr_btree *bt, uint16_t *idx, uintptr_t
> +addr) {
> +	struct mr_cache_entry *lkp_tbl;
> +	uint16_t n;
> +	uint16_t base = 0;
> +
> +	MLX5_ASSERT(bt != NULL);
> +	lkp_tbl = *bt->table;
> +	n = bt->len;
> +	/* First entry must be NULL for comparison. */
> +	MLX5_ASSERT(bt->len > 0 || (lkp_tbl[0].start == 0 &&
> +				    lkp_tbl[0].lkey == UINT32_MAX));
> +	/* Binary search. */
> +	do {
> +		register uint16_t delta = n >> 1;
> +
> +		if (addr < lkp_tbl[base + delta].start) {
> +			n = delta;
> +		} else {
> +			base += delta;
> +			n -= delta;
> +		}
> +	} while (n > 1);
> +	MLX5_ASSERT(addr >= lkp_tbl[base].start);
> +	*idx = base;
> +	if (addr < lkp_tbl[base].end)
> +		return lkp_tbl[base].lkey;
> +	/* Not found. */
> +	return UINT32_MAX;
> +}
> +
> +/**
> + * Insert an entry to B-tree lookup table.
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + * @param entry
> + *   Pointer to new entry to insert.
> + *
> + * @return
> + *   0 on success, -1 on failure.
> + */
> +static int
> +mr_btree_insert(struct mlx5_mr_btree *bt, struct mr_cache_entry *entry)
> +{
> +	struct mr_cache_entry *lkp_tbl;
> +	uint16_t idx = 0;
> +	size_t shift;
> +
> +	MLX5_ASSERT(bt != NULL);
> +	MLX5_ASSERT(bt->len <= bt->size);
> +	MLX5_ASSERT(bt->len > 0);
> +	lkp_tbl = *bt->table;
> +	/* Find out the slot for insertion. */
> +	if (mr_btree_lookup(bt, &idx, entry->start) != UINT32_MAX) {
> +		DRV_LOG(DEBUG,
> +			"abort insertion to B-tree(%p): already exist at"
> +			" idx=%u [0x%" PRIxPTR ", 0x%" PRIxPTR ") lkey=0x%x",
> +			(void *)bt, idx, entry->start, entry->end, entry->lkey);
> +		/* Already exist, return. */
> +		return 0;
> +	}
> +	/* If table is full, return error. */
> +	if (unlikely(bt->len == bt->size)) {
> +		bt->overflow = 1;
> +		return -1;
> +	}
> +	/* Insert entry. */
> +	++idx;
> +	shift = (bt->len - idx) * sizeof(struct mr_cache_entry);
> +	if (shift)
> +		memmove(&lkp_tbl[idx + 1], &lkp_tbl[idx], shift);
> +	lkp_tbl[idx] = *entry;
> +	bt->len++;
> +	DRV_LOG(DEBUG,
> +		"inserted B-tree(%p)[%u],"
> +		" [0x%" PRIxPTR ", 0x%" PRIxPTR ") lkey=0x%x",
> +		(void *)bt, idx, entry->start, entry->end, entry->lkey);
> +	return 0;
> +}
> +
> +/**
> + * Initialize B-tree and allocate memory for lookup table.
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + * @param n
> + *   Number of entries to allocate.
> + * @param socket
> + *   NUMA socket on which memory must be allocated.
> + *
> + * @return
> + *   0 on success, a negative errno value otherwise and rte_errno is set.
> + */
> +int
> +mlx5_mr_btree_init(struct mlx5_mr_btree *bt, int n, int socket) {
> +	if (bt == NULL) {
> +		rte_errno = EINVAL;
> +		return -rte_errno;
> +	}
> +	MLX5_ASSERT(!bt->table && !bt->size);
> +	memset(bt, 0, sizeof(*bt));
> +	bt->table = rte_calloc_socket("B-tree table",
> +				      n, sizeof(struct mr_cache_entry),
> +				      0, socket);
> +	if (bt->table == NULL) {
> +		rte_errno = ENOMEM;
> +		DEBUG("failed to allocate memory for btree cache on socket
> %d",
> +		      socket);
> +		return -rte_errno;
> +	}
> +	bt->size = n;
> +	/* First entry must be NULL for binary search. */
> +	(*bt->table)[bt->len++] = (struct mr_cache_entry) {
> +		.lkey = UINT32_MAX,
> +	};
> +	DEBUG("initialized B-tree %p with table %p",
> +	      (void *)bt, (void *)bt->table);
> +	return 0;
> +}
> +
> +/**
> + * Free B-tree resources.
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + */
> +void
> +mlx5_mr_btree_free(struct mlx5_mr_btree *bt) {
> +	if (bt == NULL)
> +		return;
> +	DEBUG("freeing B-tree %p with table %p",
> +	      (void *)bt, (void *)bt->table);
> +	rte_free(bt->table);
> +	memset(bt, 0, sizeof(*bt));
> +}
> +
> +/**
> + * Dump all the entries in a B-tree
> + *
> + * @param bt
> + *   Pointer to B-tree structure.
> + */
> +void
> +mlx5_mr_btree_dump(struct mlx5_mr_btree *bt __rte_unused) { #ifdef
> +RTE_LIBRTE_MLX5_DEBUG
> +	int idx;
> +	struct mr_cache_entry *lkp_tbl;
> +
> +	if (bt == NULL)
> +		return;
> +	lkp_tbl = *bt->table;
> +	for (idx = 0; idx < bt->len; ++idx) {
> +		struct mr_cache_entry *entry = &lkp_tbl[idx];
> +
> +		DEBUG("B-tree(%p)[%u],"
> +		      " [0x%" PRIxPTR ", 0x%" PRIxPTR ") lkey=0x%x",
> +		      (void *)bt, idx, entry->start, entry->end, entry->lkey);
> +	}
> +#endif
> +}
> +
> +/**
> + * Find virtually contiguous memory chunk in a given MR.
> + *
> + * @param dev
> + *   Pointer to MR structure.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry. If not found, this will not be
> + *   updated.
> + * @param start_idx
> + *   Start index of the memseg bitmap.
> + *
> + * @return
> + *   Next index to go on lookup.
> + */
> +static int
> +mr_find_next_chunk(struct mlx5_mr *mr, struct mr_cache_entry *entry,
> +		   int base_idx)
> +{
> +	uintptr_t start = 0;
> +	uintptr_t end = 0;
> +	uint32_t idx = 0;
> +
> +	/* MR for external memory doesn't have memseg list. */
> +	if (mr->msl == NULL) {
> +		struct ibv_mr *ibv_mr = mr->ibv_mr;
> +
> +		MLX5_ASSERT(mr->ms_bmp_n == 1);
> +		MLX5_ASSERT(mr->ms_n == 1);
> +		MLX5_ASSERT(base_idx == 0);
> +		/*
> +		 * Can't search it from memseg list but get it directly from
> +		 * verbs MR as there's only one chunk.
> +		 */
> +		entry->start = (uintptr_t)ibv_mr->addr;
> +		entry->end = (uintptr_t)ibv_mr->addr + mr->ibv_mr->length;
> +		entry->lkey = rte_cpu_to_be_32(mr->ibv_mr->lkey);
> +		/* Returning 1 ends iteration. */
> +		return 1;
> +	}
> +	for (idx = base_idx; idx < mr->ms_bmp_n; ++idx) {
> +		if (rte_bitmap_get(mr->ms_bmp, idx)) {
> +			const struct rte_memseg_list *msl;
> +			const struct rte_memseg *ms;
> +
> +			msl = mr->msl;
> +			ms = rte_fbarray_get(&msl->memseg_arr,
> +					     mr->ms_base_idx + idx);
> +			MLX5_ASSERT(msl->page_sz == ms->hugepage_sz);
> +			if (!start)
> +				start = ms->addr_64;
> +			end = ms->addr_64 + ms->hugepage_sz;
> +		} else if (start) {
> +			/* Passed the end of a fragment. */
> +			break;
> +		}
> +	}
> +	if (start) {
> +		/* Found one chunk. */
> +		entry->start = start;
> +		entry->end = end;
> +		entry->lkey = rte_cpu_to_be_32(mr->ibv_mr->lkey);
> +	}
> +	return idx;
> +}
> +
> +/**
> + * Insert a MR to the global B-tree cache. It may fail due to low-on-memory.
> + * Then, this entry will have to be searched by mr_lookup_list() in
> + * mlx5_mr_create() on miss.
> + *
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param mr
> + *   Pointer to MR to insert.
> + *
> + * @return
> + *   0 on success, -1 on failure.
> + */
> +int
> +mlx5_mr_insert_cache(struct mlx5_mr_share_cache *share_cache,
> +		     struct mlx5_mr *mr)
> +{
> +	unsigned int n;
> +
> +	DRV_LOG(DEBUG, "Inserting MR(%p) to global cache(%p)",
> +		(void *)mr, (void *)share_cache);
> +	for (n = 0; n < mr->ms_bmp_n; ) {
> +		struct mr_cache_entry entry;
> +
> +		memset(&entry, 0, sizeof(entry));
> +		/* Find a contiguous chunk and advance the index. */
> +		n = mr_find_next_chunk(mr, &entry, n);
> +		if (!entry.end)
> +			break;
> +		if (mr_btree_insert(&share_cache->cache, &entry) < 0) {
> +			/*
> +			 * Overflowed, but the global table cannot be
> expanded
> +			 * because of deadlock.
> +			 */
> +			return -1;
> +		}
> +	}
> +	return 0;
> +}
> +
> +/**
> + * Look up address in the original global MR list.
> + *
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry. If no match, this will not be
> updated.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Found MR on match, NULL otherwise.
> + */
> +struct mlx5_mr *
> +mlx5_mr_lookup_list(struct mlx5_mr_share_cache *share_cache,
> +		    struct mr_cache_entry *entry, uintptr_t addr) {
> +	struct mlx5_mr *mr;
> +
> +	/* Iterate all the existing MRs. */
> +	LIST_FOREACH(mr, &share_cache->mr_list, mr) {
> +		unsigned int n;
> +
> +		if (mr->ms_n == 0)
> +			continue;
> +		for (n = 0; n < mr->ms_bmp_n; ) {
> +			struct mr_cache_entry ret;
> +
> +			memset(&ret, 0, sizeof(ret));
> +			n = mr_find_next_chunk(mr, &ret, n);
> +			if (addr >= ret.start && addr < ret.end) {
> +				/* Found. */
> +				*entry = ret;
> +				return mr;
> +			}
> +		}
> +	}
> +	return NULL;
> +}
> +
> +/**
> + * Look up address on global MR cache.
> + *
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry. If no match, this will not be
> updated.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on failure and rte_errno is set.
> + */
> +uint32_t
> +mlx5_mr_lookup_cache(struct mlx5_mr_share_cache *share_cache,
> +		     struct mr_cache_entry *entry, uintptr_t addr) {
> +	uint16_t idx;
> +	uint32_t lkey = UINT32_MAX;
> +	struct mlx5_mr *mr;
> +
> +	/*
> +	 * If the global cache has overflowed since it failed to expand the
> +	 * B-tree table, it can't have all the existing MRs. Then, the address
> +	 * has to be searched by traversing the original MR list instead, which
> +	 * is very slow path. Otherwise, the global cache is all inclusive.
> +	 */
> +	if (!unlikely(share_cache->cache.overflow)) {
> +		lkey = mr_btree_lookup(&share_cache->cache, &idx, addr);
> +		if (lkey != UINT32_MAX)
> +			*entry = (*share_cache->cache.table)[idx];
> +	} else {
> +		/* Falling back to the slowest path. */
> +		mr = mlx5_mr_lookup_list(share_cache, entry, addr);
> +		if (mr != NULL)
> +			lkey = entry->lkey;
> +	}
> +	MLX5_ASSERT(lkey == UINT32_MAX || (addr >= entry->start &&
> +					   addr < entry->end));
> +	return lkey;
> +}
> +
> +/**
> + * Free MR resources. MR lock must not be held to avoid a deadlock.
> +rte_free()
> + * can raise memory free event and the callback function will spin on the
> lock.
> + *
> + * @param mr
> + *   Pointer to MR to free.
> + */
> +static void
> +mr_free(struct mlx5_mr *mr)
> +{
> +	if (mr == NULL)
> +		return;
> +	DRV_LOG(DEBUG, "freeing MR(%p):", (void *)mr);
> +	if (mr->ibv_mr != NULL)
> +		claim_zero(mlx5_glue->dereg_mr(mr->ibv_mr));
> +	if (mr->ms_bmp != NULL)
> +		rte_bitmap_free(mr->ms_bmp);
> +	rte_free(mr);
> +}
> +
> +void
> +mlx5_mr_rebuild_cache(struct mlx5_mr_share_cache *share_cache) {
> +	struct mlx5_mr *mr;
> +
> +	DRV_LOG(DEBUG, "Rebuild dev cache[] %p", (void *)share_cache);
> +	/* Flush cache to rebuild. */
> +	share_cache->cache.len = 1;
> +	share_cache->cache.overflow = 0;
> +	/* Iterate all the existing MRs. */
> +	LIST_FOREACH(mr, &share_cache->mr_list, mr)
> +		if (mlx5_mr_insert_cache(share_cache, mr) < 0)
> +			return;
> +}
> +
> +/**
> + * Release resources of detached MR having no online entry.
> + *
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + */
> +static void
> +mlx5_mr_garbage_collect(struct mlx5_mr_share_cache *share_cache) {
> +	struct mlx5_mr *mr_next;
> +	struct mlx5_mr_list free_list = LIST_HEAD_INITIALIZER(free_list);
> +
> +	/* Must be called from the primary process. */
> +	MLX5_ASSERT(rte_eal_process_type() == RTE_PROC_PRIMARY);
> +	/*
> +	 * MR can't be freed with holding the lock because rte_free() could
> call
> +	 * memory free callback function. This will be a deadlock situation.
> +	 */
> +	rte_rwlock_write_lock(&share_cache->rwlock);
> +	/* Detach the whole free list and release it after unlocking. */
> +	free_list = share_cache->mr_free_list;
> +	LIST_INIT(&share_cache->mr_free_list);
> +	rte_rwlock_write_unlock(&share_cache->rwlock);
> +	/* Release resources. */
> +	mr_next = LIST_FIRST(&free_list);
> +	while (mr_next != NULL) {
> +		struct mlx5_mr *mr = mr_next;
> +
> +		mr_next = LIST_NEXT(mr, mr);
> +		mr_free(mr);
> +	}
> +}
> +
> +/* Called during rte_memseg_contig_walk() by mlx5_mr_create(). */
> +static int mr_find_contig_memsegs_cb(const struct rte_memseg_list *msl,
> +			  const struct rte_memseg *ms, size_t len, void *arg) {
> +	struct mr_find_contig_memsegs_data *data = arg;
> +
> +	if (data->addr < ms->addr_64 || data->addr >= ms->addr_64 + len)
> +		return 0;
> +	/* Found, save it and stop walking. */
> +	data->start = ms->addr_64;
> +	data->end = ms->addr_64 + len;
> +	data->msl = msl;
> +	return 1;
> +}
> +
> +/**
> + * Create a new global Memory Region (MR) for a missing virtual address.
> + * This API should be called on a secondary process, then a request is
> +sent to
> + * the primary process in order to create a MR for the address. As the
> +global MR
> + * list is on the shared memory, following LKey lookup should succeed
> +unless the
> + * request fails.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry, found in the global cache or newly
> + *   created. If failed to create one, this will not be updated.
> + * @param addr
> + *   Target virtual address to register.
> + * @param mr_ext_memseg_en
> + *   Configurable flag about external memory segment enable or not.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on failure and rte_errno is set.
> + */
> +static uint32_t
> +mlx5_mr_create_secondary(struct ibv_pd *pd __rte_unused,
> +			 struct mlx5_mp_id *mp_id,
> +			 struct mlx5_mr_share_cache *share_cache,
> +			 struct mr_cache_entry *entry, uintptr_t addr,
> +			 unsigned int mr_ext_memseg_en __rte_unused) {
> +	int ret;
> +
> +	DEBUG("port %u requesting MR creation for address (%p)",
> +	      mp_id->port_id, (void *)addr);
> +	ret = mlx5_mp_req_mr_create(mp_id, addr);
> +	if (ret) {
> +		DEBUG("Fail to request MR creation for address (%p)",
> +		      (void *)addr);
> +		return UINT32_MAX;
> +	}
> +	rte_rwlock_read_lock(&share_cache->rwlock);
> +	/* Fill in output data. */
> +	mlx5_mr_lookup_cache(share_cache, entry, addr);
> +	/* Lookup can't fail. */
> +	MLX5_ASSERT(entry->lkey != UINT32_MAX);
> +	rte_rwlock_read_unlock(&share_cache->rwlock);
> +	DEBUG("MR CREATED by primary process for %p:\n"
> +	      "  [0x%" PRIxPTR ", 0x%" PRIxPTR "), lkey=0x%x",
> +	      (void *)addr, entry->start, entry->end, entry->lkey);
> +	return entry->lkey;
> +}
> +
> +/**
> + * Create a new global Memory Region (MR) for a missing virtual address.
> + * Register entire virtually contiguous memory chunk around the address.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry, found in the global cache or newly
> + *   created. If failed to create one, this will not be updated.
> + * @param addr
> + *   Target virtual address to register.
> + * @param mr_ext_memseg_en
> + *   Configurable flag about external memory segment enable or not.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on failure and rte_errno is set.
> + */
> +uint32_t
> +mlx5_mr_create_primary(struct ibv_pd *pd,
> +		       struct mlx5_mr_share_cache *share_cache,
> +		       struct mr_cache_entry *entry, uintptr_t addr,
> +		       unsigned int mr_ext_memseg_en) {
> +	struct mr_find_contig_memsegs_data data = {.addr = addr, };
> +	struct mr_find_contig_memsegs_data data_re;
> +	const struct rte_memseg_list *msl;
> +	const struct rte_memseg *ms;
> +	struct mlx5_mr *mr = NULL;
> +	int ms_idx_shift = -1;
> +	uint32_t bmp_size;
> +	void *bmp_mem;
> +	uint32_t ms_n;
> +	uint32_t n;
> +	size_t len;
> +
> +	DRV_LOG(DEBUG, "Creating a MR using address (%p)", (void *)addr);
> +	/*
> +	 * Release detached MRs if any. This can't be called with holding
> either
> +	 * memory_hotplug_lock or share_cache->rwlock. MRs on the free list
> have
> +	 * been detached by the memory free event but it couldn't be
> released
> +	 * inside the callback due to deadlock. As a result, releasing resources
> +	 * is quite opportunistic.
> +	 */
> +	mlx5_mr_garbage_collect(share_cache);
> +	/*
> +	 * If enabled, find out a contiguous virtual address chunk in use, to
> +	 * which the given address belongs, in order to register maximum
> range.
> +	 * In the best case where mempools are not dynamically recreated
> and
> +	 * '--socket-mem' is specified as an EAL option, it is very likely to
> +	 * have only one MR(LKey) per a socket and per a hugepage-size even
> +	 * though the system memory is highly fragmented. As the whole
> memory
> +	 * chunk will be pinned by kernel, it can't be reused unless entire
> +	 * chunk is freed from EAL.
> +	 *
> +	 * If disabled, just register one memseg (page). Then, memory
> +	 * consumption will be minimized but it may drop performance if
> there
> +	 * are many MRs to lookup on the datapath.
> +	 */
> +	if (!mr_ext_memseg_en) {
> +		data.msl = rte_mem_virt2memseg_list((void *)addr);
> +		data.start = RTE_ALIGN_FLOOR(addr, data.msl->page_sz);
> +		data.end = data.start + data.msl->page_sz;
> +	} else if (!rte_memseg_contig_walk(mr_find_contig_memsegs_cb,
> &data)) {
> +		DRV_LOG(WARNING,
> +			"Unable to find virtually contiguous"
> +			" chunk for address (%p)."
> +			" rte_memseg_contig_walk() failed.", (void *)addr);
> +		rte_errno = ENXIO;
> +		goto err_nolock;
> +	}
> +alloc_resources:
> +	/* Addresses must be page-aligned. */
> +	MLX5_ASSERT(data.msl);
> +	MLX5_ASSERT(rte_is_aligned((void *)data.start, data.msl->page_sz));
> +	MLX5_ASSERT(rte_is_aligned((void *)data.end, data.msl->page_sz));
> +	msl = data.msl;
> +	ms = rte_mem_virt2memseg((void *)data.start, msl);
> +	len = data.end - data.start;
> +	MLX5_ASSERT(ms);
> +	MLX5_ASSERT(msl->page_sz == ms->hugepage_sz);
> +	/* Number of memsegs in the range. */
> +	ms_n = len / msl->page_sz;
> +	DEBUG("Extending %p to [0x%" PRIxPTR ", 0x%" PRIxPTR "),"
> +	      " page_sz=0x%" PRIx64 ", ms_n=%u",
> +	      (void *)addr, data.start, data.end, msl->page_sz, ms_n);
> +	/* Size of memory for bitmap. */
> +	bmp_size = rte_bitmap_get_memory_footprint(ms_n);
> +	mr = rte_zmalloc_socket(NULL,
> +				RTE_ALIGN_CEIL(sizeof(*mr),
> +					       RTE_CACHE_LINE_SIZE) +
> +				bmp_size,
> +				RTE_CACHE_LINE_SIZE, msl->socket_id);
> +	if (mr == NULL) {
> +		DEBUG("Unable to allocate memory for a new MR of"
> +		      " address (%p).", (void *)addr);
> +		rte_errno = ENOMEM;
> +		goto err_nolock;
> +	}
> +	mr->msl = msl;
> +	/*
> +	 * Save the index of the first memseg and initialize memseg bitmap.
> To
> +	 * see if a memseg of ms_idx in the memseg-list is still valid, check:
> +	 *	rte_bitmap_get(mr->bmp, ms_idx - mr->ms_base_idx)
> +	 */
> +	mr->ms_base_idx = rte_fbarray_find_idx(&msl->memseg_arr, ms);
> +	bmp_mem = RTE_PTR_ALIGN_CEIL(mr + 1, RTE_CACHE_LINE_SIZE);
> +	mr->ms_bmp = rte_bitmap_init(ms_n, bmp_mem, bmp_size);
> +	if (mr->ms_bmp == NULL) {
> +		DEBUG("Unable to initialize bitmap for a new MR of"
> +		      " address (%p).", (void *)addr);
> +		rte_errno = EINVAL;
> +		goto err_nolock;
> +	}
> +	/*
> +	 * Should recheck whether the extended contiguous chunk is still
> valid.
> +	 * Because memory_hotplug_lock can't be held if there's any memory
> +	 * related calls in a critical path, resource allocation above can't be
> +	 * locked. If the memory has been changed at this point, try again
> with
> +	 * just single page. If not, go on with the big chunk atomically from
> +	 * here.
> +	 */
> +	rte_mcfg_mem_read_lock();
> +	data_re = data;
> +	if (len > msl->page_sz &&
> +	    !rte_memseg_contig_walk(mr_find_contig_memsegs_cb, &data_re))
> {
> +		DEBUG("Unable to find virtually contiguous"
> +		      " chunk for address (%p)."
> +		      " rte_memseg_contig_walk() failed.", (void *)addr);
> +		rte_errno = ENXIO;
> +		goto err_memlock;
> +	}
> +	if (data.start != data_re.start || data.end != data_re.end) {
> +		/*
> +		 * The extended contiguous chunk has been changed. Try
> again
> +		 * with single memseg instead.
> +		 */
> +		data.start = RTE_ALIGN_FLOOR(addr, msl->page_sz);
> +		data.end = data.start + msl->page_sz;
> +		rte_mcfg_mem_read_unlock();
> +		mr_free(mr);
> +		goto alloc_resources;
> +	}
> +	MLX5_ASSERT(data.msl == data_re.msl);
> +	rte_rwlock_write_lock(&share_cache->rwlock);
> +	/*
> +	 * Check the address is really missing. If other thread already created
> +	 * one or it is not found due to overflow, abort and return.
> +	 */
> +	if (mlx5_mr_lookup_cache(share_cache, entry, addr) != UINT32_MAX)
> {
> +		/*
> +		 * Insert to the global cache table. It may fail due to
> +		 * low-on-memory. Then, this entry will have to be searched
> +		 * here again.
> +		 */
> +		mr_btree_insert(&share_cache->cache, entry);
> +		DEBUG("Found MR for %p on final lookup, abort", (void
> *)addr);
> +		rte_rwlock_write_unlock(&share_cache->rwlock);
> +		rte_mcfg_mem_read_unlock();
> +		/*
> +		 * Must be unlocked before calling rte_free() because
> +		 * mlx5_mr_mem_event_free_cb() can be called inside.
> +		 */
> +		mr_free(mr);
> +		return entry->lkey;
> +	}
> +	/*
> +	 * Trim start and end addresses for verbs MR. Set bits for registering
> +	 * memsegs but exclude already registered ones. Bitmap can be
> +	 * fragmented.
> +	 */
> +	for (n = 0; n < ms_n; ++n) {
> +		uintptr_t start;
> +		struct mr_cache_entry ret;
> +
> +		memset(&ret, 0, sizeof(ret));
> +		start = data_re.start + n * msl->page_sz;
> +		/* Exclude memsegs already registered by other MRs. */
> +		if (mlx5_mr_lookup_cache(share_cache, &ret, start) ==
> +		    UINT32_MAX) {
> +			/*
> +			 * Start from the first unregistered memseg in the
> +			 * extended range.
> +			 */
> +			if (ms_idx_shift == -1) {
> +				mr->ms_base_idx += n;
> +				data.start = start;
> +				ms_idx_shift = n;
> +			}
> +			data.end = start + msl->page_sz;
> +			rte_bitmap_set(mr->ms_bmp, n - ms_idx_shift);
> +			++mr->ms_n;
> +		}
> +	}
> +	len = data.end - data.start;
> +	mr->ms_bmp_n = len / msl->page_sz;
> +	MLX5_ASSERT(ms_idx_shift + mr->ms_bmp_n <= ms_n);
> +	/*
> +	 * Finally create a verbs MR for the memory chunk. ibv_reg_mr() can
> be
> +	 * called with holding the memory lock because it doesn't use
> +	 * mlx5_alloc_buf_extern() which eventually calls rte_malloc_socket()
> +	 * through mlx5_alloc_verbs_buf().
> +	 */
> +	mr->ibv_mr = mlx5_glue->reg_mr(pd, (void *)data.start, len,
> +				       IBV_ACCESS_LOCAL_WRITE |
> +					   IBV_ACCESS_RELAXED_ORDERING);
> +	if (mr->ibv_mr == NULL) {
> +		DEBUG("Fail to create a verbs MR for address (%p)",
> +		      (void *)addr);
> +		rte_errno = EINVAL;
> +		goto err_mrlock;
> +	}
> +	MLX5_ASSERT((uintptr_t)mr->ibv_mr->addr == data.start);
> +	MLX5_ASSERT(mr->ibv_mr->length == len);
> +	LIST_INSERT_HEAD(&share_cache->mr_list, mr, mr);
> +	DEBUG("MR CREATED (%p) for %p:\n"
> +	      "  [0x%" PRIxPTR ", 0x%" PRIxPTR "),"
> +	      " lkey=0x%x base_idx=%u ms_n=%u, ms_bmp_n=%u",
> +	      (void *)mr, (void *)addr, data.start, data.end,
> +	      rte_cpu_to_be_32(mr->ibv_mr->lkey),
> +	      mr->ms_base_idx, mr->ms_n, mr->ms_bmp_n);
> +	/* Insert to the global cache table. */
> +	mlx5_mr_insert_cache(share_cache, mr);
> +	/* Fill in output data. */
> +	mlx5_mr_lookup_cache(share_cache, entry, addr);
> +	/* Lookup can't fail. */
> +	MLX5_ASSERT(entry->lkey != UINT32_MAX);
> +	rte_rwlock_write_unlock(&share_cache->rwlock);
> +	rte_mcfg_mem_read_unlock();
> +	return entry->lkey;
> +err_mrlock:
> +	rte_rwlock_write_unlock(&share_cache->rwlock);
> +err_memlock:
> +	rte_mcfg_mem_read_unlock();
> +err_nolock:
> +	/*
> +	 * In case of error, as this can be called in a datapath, a warning
> +	 * message per an error is preferable instead. Must be unlocked
> before
> +	 * calling rte_free() because mlx5_mr_mem_event_free_cb() can be
> called
> +	 * inside.
> +	 */
> +	mr_free(mr);
> +	return UINT32_MAX;
> +}
> +
> +/**
> + * Create a new global Memory Region (MR) for a missing virtual address.
> + * This can be called from primary and secondary process.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry, found in the global cache or newly
> + *   created. If failed to create one, this will not be updated.
> + * @param addr
> + *   Target virtual address to register.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on failure and rte_errno is set.
> + */
> +static uint32_t
> +mlx5_mr_create(struct ibv_pd *pd, struct mlx5_mp_id *mp_id,
> +	       struct mlx5_mr_share_cache *share_cache,
> +	       struct mr_cache_entry *entry, uintptr_t addr,
> +	       unsigned int mr_ext_memseg_en)
> +{
> +	uint32_t ret = 0;
> +
> +	switch (rte_eal_process_type()) {
> +	case RTE_PROC_PRIMARY:
> +		ret = mlx5_mr_create_primary(pd, share_cache, entry,
> +					     addr, mr_ext_memseg_en);
> +		break;
> +	case RTE_PROC_SECONDARY:
> +		ret = mlx5_mr_create_secondary(pd, mp_id, share_cache,
> entry,
> +					       addr, mr_ext_memseg_en);
> +		break;
> +	default:
> +		break;
> +	}
> +	return ret;
> +}
> +
> +/**
> + * Look up address in the global MR cache table. If not found, create a new
> MR.
> + * Insert the found/created entry to local bottom-half cache table.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param mr_ctrl
> + *   Pointer to per-queue MR control structure.
> + * @param[out] entry
> + *   Pointer to returning MR cache entry, found in the global cache or newly
> + *   created. If failed to create one, this is not written.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on no match.
> + */
> +static uint32_t
> +mr_lookup_caches(struct ibv_pd *pd, struct mlx5_mp_id *mp_id,
> +		 struct mlx5_mr_share_cache *share_cache,
> +		 struct mlx5_mr_ctrl *mr_ctrl,
> +		 struct mr_cache_entry *entry, uintptr_t addr,
> +		 unsigned int mr_ext_memseg_en)
> +{
> +	struct mlx5_mr_btree *bt = &mr_ctrl->cache_bh;
> +	uint32_t lkey;
> +	uint16_t idx;
> +
> +	/* If local cache table is full, try to double it. */
> +	if (unlikely(bt->len == bt->size))
> +		mr_btree_expand(bt, bt->size << 1);
> +	/* Look up in the global cache. */
> +	rte_rwlock_read_lock(&share_cache->rwlock);
> +	lkey = mr_btree_lookup(&share_cache->cache, &idx, addr);
> +	if (lkey != UINT32_MAX) {
> +		/* Found. */
> +		*entry = (*share_cache->cache.table)[idx];
> +		rte_rwlock_read_unlock(&share_cache->rwlock);
> +		/*
> +		 * Update local cache. Even if it fails, return the found entry
> +		 * to update top-half cache. Next time, this entry will be
> found
> +		 * in the global cache.
> +		 */
> +		mr_btree_insert(bt, entry);
> +		return lkey;
> +	}
> +	rte_rwlock_read_unlock(&share_cache->rwlock);
> +	/* First time to see the address? Create a new MR. */
> +	lkey = mlx5_mr_create(pd, mp_id, share_cache, entry, addr,
> +			      mr_ext_memseg_en);
> +	/*
> +	 * Update the local cache if successfully created a new global MR.
> Even
> +	 * if failed to create one, there's no action to take in this datapath
> +	 * code. As returning LKey is invalid, this will eventually make HW
> +	 * fail.
> +	 */
> +	if (lkey != UINT32_MAX)
> +		mr_btree_insert(bt, entry);
> +	return lkey;
> +}
> +
> +/**
> + * Bottom-half of LKey search on datapath. First search in cache_bh[]
> +and if
> + * misses, search in the global MR cache table and update the new entry
> +to
> + * per-queue local caches.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + * @param mr_ctrl
> + *   Pointer to per-queue MR control structure.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on no match.
> + */
> +uint32_t mlx5_mr_addr2mr_bh(struct ibv_pd *pd, struct mlx5_mp_id
> *mp_id,
> +			    struct mlx5_mr_share_cache *share_cache,
> +			    struct mlx5_mr_ctrl *mr_ctrl,
> +			    uintptr_t addr, unsigned int mr_ext_memseg_en) {
> +	uint32_t lkey;
> +	uint16_t bh_idx = 0;
> +	/* Victim in top-half cache to replace with new entry. */
> +	struct mr_cache_entry *repl = &mr_ctrl->cache[mr_ctrl->head];
> +
> +	/* Binary-search MR translation table. */
> +	lkey = mr_btree_lookup(&mr_ctrl->cache_bh, &bh_idx, addr);
> +	/* Update top-half cache. */
> +	if (likely(lkey != UINT32_MAX)) {
> +		*repl = (*mr_ctrl->cache_bh.table)[bh_idx];
> +	} else {
> +		/*
> +		 * If missed in local lookup table, search in the global cache
> +		 * and local cache_bh[] will be updated inside if possible.
> +		 * Top-half cache entry will also be updated.
> +		 */
> +		lkey = mr_lookup_caches(pd, mp_id, share_cache, mr_ctrl,
> +					repl, addr, mr_ext_memseg_en);
> +		if (unlikely(lkey == UINT32_MAX))
> +			return UINT32_MAX;
> +	}
> +	/* Update the most recently used entry. */
> +	mr_ctrl->mru = mr_ctrl->head;
> +	/* Point to the next victim, the oldest. */
> +	mr_ctrl->head = (mr_ctrl->head + 1) % MLX5_MR_CACHE_N;
> +	return lkey;
> +}
> +
> +/**
> + * Release all the created MRs and resources on global MR cache of a device.
> + * list.
> + *
> + * @param share_cache
> + *   Pointer to a global shared MR cache.
> + */
> +void
> +mlx5_mr_release_cache(struct mlx5_mr_share_cache *share_cache) {
> +	struct mlx5_mr *mr_next;
> +
> +	rte_rwlock_write_lock(&share_cache->rwlock);
> +	/* Detach from MR list and move to free list. */
> +	mr_next = LIST_FIRST(&share_cache->mr_list);
> +	while (mr_next != NULL) {
> +		struct mlx5_mr *mr = mr_next;
> +
> +		mr_next = LIST_NEXT(mr, mr);
> +		LIST_REMOVE(mr, mr);
> +		LIST_INSERT_HEAD(&share_cache->mr_free_list, mr, mr);
> +	}
> +	LIST_INIT(&share_cache->mr_list);
> +	/* Free global cache. */
> +	mlx5_mr_btree_free(&share_cache->cache);
> +	rte_rwlock_write_unlock(&share_cache->rwlock);
> +	/* Free all remaining MRs. */
> +	mlx5_mr_garbage_collect(share_cache);
> +}
> +
> +/**
> + * Flush all of the local cache entries.
> + *
> + * @param mr_ctrl
> + *   Pointer to per-queue MR local cache.
> + */
> +void
> +mlx5_mr_flush_local_cache(struct mlx5_mr_ctrl *mr_ctrl) {
> +	/* Reset the most-recently-used index. */
> +	mr_ctrl->mru = 0;
> +	/* Reset the linear search array. */
> +	mr_ctrl->head = 0;
> +	memset(mr_ctrl->cache, 0, sizeof(mr_ctrl->cache));
> +	/* Reset the B-tree table. */
> +	mr_ctrl->cache_bh.len = 1;
> +	mr_ctrl->cache_bh.overflow = 0;
> +	/* Update the generation number. */
> +	mr_ctrl->cur_gen = *mr_ctrl->dev_gen_ptr;
> +	DRV_LOG(DEBUG, "mr_ctrl(%p): flushed, cur_gen=%d",
> +		(void *)mr_ctrl, mr_ctrl->cur_gen);
> +}
> +
> +/**
> + * Creates a memory region for external memory, that is memory which is
> +not
> + * part of the DPDK memory segments.
> + *
> + * @param pd
> + *   Pointer to ibv_pd of a device (net, regex, vdpa,...).
> + * @param addr
> + *   Starting virtual address of memory.
> + * @param len
> + *   Length of memory segment being mapped.
> + * @param socked_id
> + *   Socket to allocate heap memory for the control structures.
> + *
> + * @return
> + *   Pointer to MR structure on success, NULL otherwise.
> + */
> +struct mlx5_mr *
> +mlx5_create_mr_ext(struct ibv_pd *pd, uintptr_t addr, size_t len, int
> +socket_id) {
> +	struct mlx5_mr *mr = NULL;
> +
> +	mr = rte_zmalloc_socket(NULL,
> +				RTE_ALIGN_CEIL(sizeof(*mr),
> +					       RTE_CACHE_LINE_SIZE),
> +				RTE_CACHE_LINE_SIZE, socket_id);
> +	if (mr == NULL)
> +		return NULL;
> +	mr->ibv_mr = mlx5_glue->reg_mr(pd, (void *)addr, len,
> +				       IBV_ACCESS_LOCAL_WRITE |
> +					   IBV_ACCESS_RELAXED_ORDERING);
> +	if (mr->ibv_mr == NULL) {
> +		DRV_LOG(WARNING,
> +			"Fail to create a verbs MR for address (%p)",
> +			(void *)addr);
> +		rte_free(mr);
> +		return NULL;
> +	}
> +	mr->msl = NULL; /* Mark it is external memory. */
> +	mr->ms_bmp = NULL;
> +	mr->ms_n = 1;
> +	mr->ms_bmp_n = 1;
> +	DRV_LOG(DEBUG,
> +		"MR CREATED (%p) for external memory %p:\n"
> +		"  [0x%" PRIxPTR ", 0x%" PRIxPTR "),"
> +		" lkey=0x%x base_idx=%u ms_n=%u, ms_bmp_n=%u",
> +		(void *)mr, (void *)addr,
> +		addr, addr + len, rte_cpu_to_be_32(mr->ibv_mr->lkey),
> +		mr->ms_base_idx, mr->ms_n, mr->ms_bmp_n);
> +	return mr;
> +}
> +
> +/**
> + * Dump all the created MRs and the global cache entries.
> + *
> + * @param sh
> + *   Pointer to Ethernet device shared context.
> + */
> +void
> +mlx5_mr_dump_cache(struct mlx5_mr_share_cache *share_cache
> +__rte_unused) { #ifdef RTE_LIBRTE_MLX5_DEBUG
> +	struct mlx5_mr *mr;
> +	int mr_n = 0;
> +	int chunk_n = 0;
> +
> +	rte_rwlock_read_lock(&share_cache->rwlock);
> +	/* Iterate all the existing MRs. */
> +	LIST_FOREACH(mr, &share_cache->mr_list, mr) {
> +		unsigned int n;
> +
> +		DEBUG("MR[%u], LKey = 0x%x, ms_n = %u, ms_bmp_n = %u",
> +		      mr_n++, rte_cpu_to_be_32(mr->ibv_mr->lkey),
> +		      mr->ms_n, mr->ms_bmp_n);
> +		if (mr->ms_n == 0)
> +			continue;
> +		for (n = 0; n < mr->ms_bmp_n; ) {
> +			struct mr_cache_entry ret = { 0, };
> +
> +			n = mr_find_next_chunk(mr, &ret, n);
> +			if (!ret.end)
> +				break;
> +			DEBUG("  chunk[%u], [0x%" PRIxPTR ", 0x%" PRIxPTR
> ")",
> +			      chunk_n++, ret.start, ret.end);
> +		}
> +	}
> +	DEBUG("Dumping global cache %p", (void *)share_cache);
> +	mlx5_mr_btree_dump(&share_cache->cache);
> +	rte_rwlock_read_unlock(&share_cache->rwlock);
> +#endif
> +}
> diff --git a/drivers/common/mlx5/mlx5_common_mr.h
> b/drivers/common/mlx5/mlx5_common_mr.h
> new file mode 100644
> index 0000000000..e805f96375
> --- /dev/null
> +++ b/drivers/common/mlx5/mlx5_common_mr.h
> @@ -0,0 +1,160 @@
> +/* SPDX-License-Identifier: BSD-3-Clause
> + * Copyright 2018 6WIND S.A.
> + * Copyright 2018 Mellanox Technologies, Ltd  */
> +
> +#ifndef RTE_PMD_MLX5_COMMON_MR_H_
> +#define RTE_PMD_MLX5_COMMON_MR_H_
> +
> +#include <stddef.h>
> +#include <stdint.h>
> +#include <sys/queue.h>
> +
> +/* Verbs header. */
> +/* ISO C doesn't support unnamed structs/unions, disabling -pedantic.
> +*/ #ifdef PEDANTIC #pragma GCC diagnostic ignored "-Wpedantic"
> +#endif
> +#include <infiniband/verbs.h>
> +#include <infiniband/mlx5dv.h>
> +#ifdef PEDANTIC
> +#pragma GCC diagnostic error "-Wpedantic"
> +#endif
> +
> +#include <rte_rwlock.h>
> +#include <rte_bitmap.h>
> +#include <rte_memory.h>
> +
> +#include "mlx5_common_mp.h"
> +
> +/* Size of per-queue MR cache array for linear search. */ #define
> +MLX5_MR_CACHE_N 8 #define MLX5_MR_BTREE_CACHE_N 256
> +
> +/* Memory Region object. */
> +struct mlx5_mr {
> +	LIST_ENTRY(mlx5_mr) mr; /**< Pointer to the prev/next entry. */
> +	struct ibv_mr *ibv_mr; /* Verbs Memory Region. */
> +	const struct rte_memseg_list *msl;
> +	int ms_base_idx; /* Start index of msl->memseg_arr[]. */
> +	int ms_n; /* Number of memsegs in use. */
> +	uint32_t ms_bmp_n; /* Number of bits in memsegs bit-mask. */
> +	struct rte_bitmap *ms_bmp; /* Bit-mask of memsegs belonged to MR.
> */
> +};
> +
> +/* Cache entry for Memory Region. */
> +struct mr_cache_entry {
> +	uintptr_t start; /* Start address of MR. */
> +	uintptr_t end; /* End address of MR. */
> +	uint32_t lkey; /* rte_cpu_to_be_32(ibv_mr->lkey). */ } __rte_packed;
> +
> +/* MR Cache table for Binary search. */ struct mlx5_mr_btree {
> +	uint16_t len; /* Number of entries. */
> +	uint16_t size; /* Total number of entries. */
> +	int overflow; /* Mark failure of table expansion. */
> +	struct mr_cache_entry (*table)[];
> +} __rte_packed;
> +
> +/* Per-queue MR control descriptor. */
> +struct mlx5_mr_ctrl {
> +	uint32_t *dev_gen_ptr; /* Generation number of device to poll. */
> +	uint32_t cur_gen; /* Generation number saved to flush caches. */
> +	uint16_t mru; /* Index of last hit entry in top-half cache. */
> +	uint16_t head; /* Index of the oldest entry in top-half cache. */
> +	struct mr_cache_entry cache[MLX5_MR_CACHE_N]; /* Cache for top-
> half. */
> +	struct mlx5_mr_btree cache_bh; /* Cache for bottom-half. */ }
> +__rte_packed;
> +
> +LIST_HEAD(mlx5_mr_list, mlx5_mr);
> +
> +/* Global per-device MR cache. */
> +struct mlx5_mr_share_cache {
> +	uint32_t dev_gen; /* Generation number to flush local caches. */
> +	rte_rwlock_t rwlock; /* MR cache Lock. */
> +	struct mlx5_mr_btree cache; /* Global MR cache table. */
> +	struct mlx5_mr_list mr_list; /* Registered MR list. */
> +	struct mlx5_mr_list mr_free_list; /* Freed MR list. */ } __rte_packed;
> +
> +/**
> + * Look up LKey from given lookup table by linear search. Firstly look
> +up the
> + * last-hit entry. If miss, the entire array is searched. If found,
> +update the
> + * last-hit index and return LKey.
> + *
> + * @param lkp_tbl
> + *   Pointer to lookup table.
> + * @param[in,out] cached_idx
> + *   Pointer to last-hit index.
> + * @param n
> + *   Size of lookup table.
> + * @param addr
> + *   Search key.
> + *
> + * @return
> + *   Searched LKey on success, UINT32_MAX on no match.
> + */
> +static __rte_always_inline uint32_t
> +mlx5_mr_lookup_lkey(struct mr_cache_entry *lkp_tbl, uint16_t *cached_idx,
> +		    uint16_t n, uintptr_t addr)
> +{
> +	uint16_t idx;
> +
> +	if (likely(addr >= lkp_tbl[*cached_idx].start &&
> +		   addr < lkp_tbl[*cached_idx].end))
> +		return lkp_tbl[*cached_idx].lkey;
> +	for (idx = 0; idx < n && lkp_tbl[idx].start != 0; ++idx) {
> +		if (addr >= lkp_tbl[idx].start &&
> +		    addr < lkp_tbl[idx].end) {
> +			/* Found. */
> +			*cached_idx = idx;
> +			return lkp_tbl[idx].lkey;
> +		}
> +	}
> +	return UINT32_MAX;
> +}
> +
> +__rte_experimental
> +int mlx5_mr_btree_init(struct mlx5_mr_btree *bt, int n, int socket);
> +__rte_experimental void mlx5_mr_btree_free(struct mlx5_mr_btree *bt);
> +__rte_experimental void mlx5_mr_btree_dump(struct mlx5_mr_btree *bt
> +__rte_unused); __rte_experimental uint32_t mlx5_mr_addr2mr_bh(struct
> +ibv_pd *pd, struct mlx5_mp_id *mp_id,
> +			    struct mlx5_mr_share_cache *share_cache,
> +			    struct mlx5_mr_ctrl *mr_ctrl,
> +			    uintptr_t addr, unsigned int mr_ext_memseg_en);
> +__rte_experimental void mlx5_mr_release_cache(struct
> +mlx5_mr_share_cache *mr_cache); __rte_experimental void
> +mlx5_mr_dump_cache(struct mlx5_mr_share_cache *share_cache
> +__rte_unused); __rte_experimental void mlx5_mr_rebuild_cache(struct
> +mlx5_mr_share_cache *share_cache); __rte_experimental void
> +mlx5_mr_flush_local_cache(struct mlx5_mr_ctrl *mr_ctrl);
> +__rte_experimental int mlx5_mr_insert_cache(struct mlx5_mr_share_cache
> +*share_cache,
> +		     struct mlx5_mr *mr);
> +__rte_experimental
> +uint32_t
> +mlx5_mr_lookup_cache(struct mlx5_mr_share_cache *share_cache,
> +		     struct mr_cache_entry *entry, uintptr_t addr);
> +__rte_experimental struct mlx5_mr * mlx5_mr_lookup_list(struct
> +mlx5_mr_share_cache *share_cache,
> +		    struct mr_cache_entry *entry, uintptr_t addr);
> __rte_experimental
> +struct mlx5_mr * mlx5_create_mr_ext(struct ibv_pd *pd, uintptr_t addr,
> +size_t len,
> +		   int socket_id);
> +__rte_experimental
> +uint32_t
> +mlx5_mr_create_primary(struct ibv_pd *pd,
> +		       struct mlx5_mr_share_cache *share_cache,
> +		       struct mr_cache_entry *entry, uintptr_t addr,
> +		       unsigned int mr_ext_memseg_en);
> +
> +#endif /* RTE_PMD_MLX5_COMMON_MR_H_ */
> diff --git a/drivers/common/mlx5/rte_common_mlx5_version.map
> b/drivers/common/mlx5/rte_common_mlx5_version.map
> index 265703d1c9..b58a378278 100644
> --- a/drivers/common/mlx5/rte_common_mlx5_version.map
> +++ b/drivers/common/mlx5/rte_common_mlx5_version.map
> @@ -61,4 +61,18 @@ EXPERIMENTAL {
>  	mlx5_mp_req_mr_create;
>  	mlx5_mp_req_queue_state_modify;
>  	mlx5_mp_req_verbs_cmd_fd;
> +
> +	mlx5_mr_btree_init;
> +	mlx5_mr_btree_free;
> +	mlx5_mr_btree_dump;
> +	mlx5_mr_addr2mr_bh;
> +	mlx5_mr_release_cache;
> +	mlx5_mr_dump_cache;
> +	mlx5_mr_rebuild_cache;
> +	mlx5_mr_insert_cache;
> +	mlx5_mr_lookup_cache;
> +	mlx5_mr_lookup_list;
> +	mlx5_create_mr_ext;
> +	mlx5_mr_create_primary;
> +	mlx5_mr_flush_local_cache;
>  };
> --
> 2.16.6


  reply	other threads:[~2020-04-08  9:04 UTC|newest]

Thread overview: 26+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-04-02 19:21 [dpdk-dev] [PATCH 0/4] refactor multi-process IPC and memory management codes to common driver Vu Pham
2020-04-02 19:21 ` [dpdk-dev] [PATCH 1/4] common/mlx5: refactor multi-process IPC handling " Vu Pham
2020-04-02 19:21 ` [dpdk-dev] [PATCH 2/4] net/mlx5: modify net PMD to use common multi-process APIs Vu Pham
2020-04-02 19:21 ` [dpdk-dev] [PATCH 3/4] common/mlx5: refactor memory management codes Vu Pham
2020-04-02 19:21 ` [dpdk-dev] [PATCH 4/4] net/mlx5: modify net PMD to use common memory management driver Vu Pham
2020-04-07 16:48 ` [dpdk-dev] [PATCH v2 0/4] refactor multi-process IPC and memory management codes to common driver Vu Pham
2020-04-07 16:48   ` [dpdk-dev] [PATCH v2 1/4] common/mlx5: refactor MP IPC handling " Vu Pham
2020-04-08  9:05     ` Slava Ovsiienko
2020-04-07 16:48   ` [dpdk-dev] [PATCH v2 2/4] net/mlx5: modify net pmd to use common multi-process APIs Vu Pham
2020-04-07 16:48   ` [dpdk-dev] [PATCH v2 3/4] common/mlx5: refactor memory management codes Vu Pham
2020-04-07 16:48   ` [dpdk-dev] [PATCH v2 4/4] net/mlx5: modify net pmd to use common MR driver Vu Pham
2020-04-07 17:00 ` [dpdk-dev] [PATCH v3 0/4] refactor multi-process IPC and memory management codes to common driver Vu Pham
2020-04-07 17:00   ` [dpdk-dev] [PATCH v3 1/4] common/mlx5: refactor multi-process IPC handling " Vu Pham
2020-04-08  9:05     ` Slava Ovsiienko
2020-04-07 17:00   ` [dpdk-dev] [PATCH v3 2/4] net/mlx5: modify net PMD to use common multi-process APIs Vu Pham
2020-04-08  9:05     ` Slava Ovsiienko
2020-04-07 17:00   ` [dpdk-dev] [PATCH v3 3/4] common/mlx5: refactor memory management codes Vu Pham
2020-04-08  9:04     ` Slava Ovsiienko [this message]
2020-04-07 17:00   ` [dpdk-dev] [PATCH v3 4/4] net/mlx5: modify net PMD to use common MR driver Vu Pham
2020-04-08  9:06     ` Slava Ovsiienko
2020-04-13 21:17 ` [dpdk-dev] [PATCH v4 0/2] refactor multi-process IPC and memory management codes to common driver Vu Pham
2020-04-13 21:17   ` [dpdk-dev] [PATCH v4 1/2] common/mlx5: refactor multi-process IPC handling " Vu Pham
2020-04-14  7:26     ` Slava Ovsiienko
2020-04-13 21:17   ` [dpdk-dev] [PATCH v4 2/2] common/mlx5: refactor memory management codes Vu Pham
2020-04-14  7:27     ` Slava Ovsiienko
2020-04-15  9:30   ` [dpdk-dev] [PATCH v4 0/2] refactor multi-process IPC and memory management codes to common driver Raslan Darawsheh

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=AM4PR05MB32652EE4B9324560E4170210D2C00@AM4PR05MB3265.eurprd05.prod.outlook.com \
    --to=viacheslavo@mellanox.com \
    --cc=dev@dpdk.org \
    --cc=matan@mellanox.com \
    --cc=orika@mellanox.com \
    --cc=rasland@mellanox.com \
    --cc=vuhuong@mellanox.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).