RDMA/mlx5: Switch MR cache to use RB-tree
Subsystems (3)
cover RDMA/mlx5: Switch MR cache to use RB-tree
This series moves the MR cache to use RB tree to store the entries of the cache. By doing so, enabling more flexibility when managing the cache entries.
The MR cache will now cache mkeys returned by the user even if they are not from one of the predefined pools, by that allowing restarting applications to reuse the their released mkey and improve restart times.
v0->v1:
- Fix rb tree search from memcmp to dedicated cmp function
- Rewording of some commit messages
Aharon Landau (8):
RDMA/mlx5: Don't keep umrable 'page_shift' in cache entries RDMA/mlx5: Generalize mlx5_cache_cache_mr() to fit all cacheable mkeys RDMA/mlx5: Remove explicit ODP cache entry
RDMA/mlx5: Allow rereg all the mkeys that can load pas with UMR RDMA/mlx5: Introduce mlx5r_cache_rb_key
RDMA/mlx5: Change the cache structure to an RB-tree RDMA/mlx5: Cache all user cacheable mkeys on dereg MR flow RDMA/mlx5: Add work to remove temporary entries from the cache
drivers/infiniband/hw/mlx5/mlx5_ib.h | 42 +- drivers/infiniband/hw/mlx5/mr.c | 574 ++++++++++++++++++++------- drivers/infiniband/hw/mlx5/odp.c | 34 +- include/linux/mlx5/driver.h | 1 - 4 files changed, 465 insertions(+), 186 deletions(-)
Signature · 1 line
2.17.2
PATCH 1/8 RDMA/mlx5: Don't keep umrable 'page_shift' in cache entries
From: Aharon Landau <[email protected]>
mkc.log_page_size can be changed using UMR. Therefore, don't treat it as a cache entry property.
Removing it from struct mlx5_cache_ent.
All cache mkeys will be created with default PAGE_SHIFT, and updated with the needed page_shift using UMR when passing them to a user.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 1 - drivers/infiniband/hw/mlx5/mr.c | 3 +-- drivers/infiniband/hw/mlx5/odp.c | 2 -- 3 files changed, 1 insertion(+), 5 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +0 -1
@@ -735,7 +735,6 @@ struct mlx5_cache_ent {char name[4 ];u32 order;u32 access_mode; -u32 page;unsigned int ndescs; u8 disabled:1;
drivers/infiniband/hw/mlx5/mr.c +1 -2
@@ -297,7 +297,7 @@ static void set_cache_mkc(struct mlx5_cache_ent *ent, void *mkc)MLX5_SET (mkc, mkc, translations_octword_size,get_mkc_octo_size (ent->access_mode, ent->ndescs)); -MLX5_SET (mkc, mkc, log_page_size, ent->page); +MLX5_SET (mkc, mkc, log_page_size,PAGE_SHIFT ); }/* Asynchronously schedule new MRs to be populated in the cache. */
@@ -765,7 +765,6 @@ int mlx5_mkey_cache_init(struct mlx5_ib_dev *dev)if (ent->order >mkey_cache_max_order (dev))continue ; - ent->page =PAGE_SHIFT ; ent->ndescs =1 << ent->order; ent->access_mode =MLX5_MKC_ACCESS_MODE_MTT ; if ((dev->mdev->profile.mask & MLX5_PROF_MASK_MR_CACHE) &&
drivers/infiniband/hw/mlx5/odp.c +0 -2
@@ -1595,14 +1595,12 @@ void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent)switch (ent->order -2 ) {case MLX5_IMR_MTT_CACHE_ENTRY : - ent->page =PAGE_SHIFT ; ent->ndescs =MLX5_IMR_MTT_ENTRIES ; ent->access_mode =MLX5_MKC_ACCESS_MODE_MTT ; ent->limit =0 ;break ;case MLX5_IMR_KSM_CACHE_ENTRY : - ent->page =MLX5_KSM_PAGE_SHIFT ; ent->ndescs = mlx5_imr_ksm_entries; ent->access_mode =MLX5_MKC_ACCESS_MODE_KSM ; ent->limit =0 ;
Signature · 1 line
2.17.2
PATCH 2/8 RDMA/mlx5: Generalize mlx5_cache_cache_mr() to fit all cacheable mkeys
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
All the mkeys that can use UMR, can be cached.
Generalize the flow so that creation of all mkeys that can be cached will go through to the cache instead of reg_create(). This will allow us to to extend the cache in following patches to support additional non-default cache entries.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 6 +- drivers/infiniband/hw/mlx5/mr.c | 147 +++++++++++++++------------ drivers/infiniband/hw/mlx5/odp.c | 9 +- 3 files changed, 87 insertions(+), 75 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +3 -3
@@ -1313,9 +1313,9 @@ int mlx5_ib_get_cqe_size(struct ib_cq *ibcq);int mlx5_mkey_cache_init (struct mlx5_ib_dev *dev);int mlx5_mkey_cache_cleanup (struct mlx5_ib_dev *dev); -struct mlx5_ib_mr *mlx5_mr_cache_alloc (struct mlx5_ib_dev *dev, -struct mlx5_cache_ent *ent, -int access_flags); +struct mlx5_ib_mr *mlx5_mr_cache_alloc (struct mlx5_ib_dev *dev,u8 access_mode, +unsigned int access_flags, +unsigned int ndescs);int mlx5_ib_check_mr_status (struct ib_mr *ibmr,u32 check_mask,struct ib_mr_status *mr_status);
drivers/infiniband/hw/mlx5/mr.c +80 -67
216 lines not on this page · raw message
drivers/infiniband/hw/mlx5/odp.c +4 -5
@@ -418,8 +418,8 @@ static struct mlx5_ib_mr *implicit_get_child_mr(struct mlx5_ib_mr *imr,if (IS_ERR (odp))return ERR_CAST (odp); - mr =mlx5_mr_cache_alloc (dev, &dev->cache.ent[MLX5_IMR_MTT_CACHE_ENTRY ], - imr->access_flags); + mr =mlx5_mr_cache_alloc (dev,MLX5_MKC_ACCESS_MODE_MTT , + imr->access_flags,MLX5_IMR_MTT_ENTRIES );if (IS_ERR (mr)) {ib_umem_odp_release (odp);return mr;
@@ -493,9 +493,8 @@ struct mlx5_ib_mr *mlx5_ib_alloc_implicit_mr(struct mlx5_ib_pd *pd,if (IS_ERR (umem_odp))return ERR_CAST (umem_odp); - imr =mlx5_mr_cache_alloc (dev, - &dev->cache.ent[MLX5_IMR_KSM_CACHE_ENTRY ], - access_flags); + imr =mlx5_mr_cache_alloc (dev,MLX5_MKC_ACCESS_MODE_KSM , access_flags, + mlx5_imr_ksm_entries);if (IS_ERR (imr)) {ib_umem_odp_release (umem_odp);return imr;
Signature · 1 line
2.17.2
PATCH 3/8 RDMA/mlx5: Remove explicit ODP cache entry
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
Explicit ODP mkey doesn't have unique properties. There is no need to devote to it a special entry. Removing it.
--- drivers/infiniband/hw/mlx5/odp.c | 16 ++-------------- include/linux/mlx5/driver.h | 1 - 2 files changed, 2 insertions(+), 15 deletions(-)
drivers/infiniband/hw/mlx5/odp.c +2 -14
@@ -1591,20 +1591,8 @@ void mlx5_odp_init_mkey_cache_entry(struct mlx5_cache_ent *ent) {if (!(ent->dev->odp_caps.general_caps &IB_ODP_SUPPORT_IMPLICIT ))return ; - -switch (ent->order -2 ) { -case MLX5_IMR_MTT_CACHE_ENTRY : - ent->ndescs =MLX5_IMR_MTT_ENTRIES ; - ent->access_mode =MLX5_MKC_ACCESS_MODE_MTT ; - ent->limit =0 ; -break ; - -case MLX5_IMR_KSM_CACHE_ENTRY : - ent->ndescs = mlx5_imr_ksm_entries; - ent->access_mode =MLX5_MKC_ACCESS_MODE_KSM ; - ent->limit =0 ; -break ; - } + ent->ndescs = mlx5_imr_ksm_entries; + ent->access_mode =MLX5_MKC_ACCESS_MODE_KSM ; }static const struct ib_device_ops mlx5_ib_dev_odp_ops = {
include/linux/mlx5/driver.h +0 -1
@@ -736,7 +736,6 @@ enum {enum {MKEY_CACHE_LAST_STD_ENTRY =20 , -MLX5_IMR_MTT_CACHE_ENTRY ,MLX5_IMR_KSM_CACHE_ENTRY ,MAX_MKEY_CACHE_ENTRIES };
Signature · 1 line
2.17.2
PATCH 4/8 RDMA/mlx5: Allow rereg all the mkeys that can load pas with UMR
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
Keep track of the mkey size of all cacheable mkeys, and by this allow to rereg them.
--- drivers/infiniband/hw/mlx5/mr.c | 7 ++----- 1 file changed, 2 insertions(+), 5 deletions(-)
drivers/infiniband/hw/mlx5/mr.c +2 -5
@@ -709,6 +709,7 @@ struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, u8 access_mode,kfree (mr);return ERR_PTR (err); } + mr->mmkey.ndescs = ndescs; } mr->mmkey.type =MLX5_MKEY_MR ;init_waitqueue_head (&mr->mmkey.wait);
@@ -1374,9 +1375,6 @@ static bool can_use_umr_rereg_pas(struct mlx5_ib_mr *mr, {struct mlx5_ib_dev *dev =to_mdev (mr->ibmr.device); -/* We only track the allocated sizes of MRs from the cache */ -if (!mr->mmkey.cache_ent) -return false;if (!mlx5r_umr_can_load_pas (dev, new_umem->length))return false;
@@ -1384,8 +1382,7 @@ static bool can_use_umr_rereg_pas(struct mlx5_ib_mr *mr,mlx5_umem_find_best_pgsz (new_umem, mkc, log_page_size,0 , iova);if (WARN_ON (!*page_size))return false; -return (1ULL << mr->mmkey.cache_ent->order) >= -ib_umem_num_dma_blocks (new_umem, *page_size); +return mr->mmkey.ndescs >=ib_umem_num_dma_blocks (new_umem, *page_size); } static int umr_rereg_pas(struct mlx5_ib_mr *mr, struct ib_pd *pd,
Signature · 1 line
2.17.2
PATCH 5/8 RDMA/mlx5: Introduce mlx5r_cache_rb_key
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
In the next patch, I will change the cache structure to an RB tree. Introducing the key of the tree.
The key is all the mkey properties that UMR operations can't modify. Using this key to define the cache entries and to search and create cache mkeys.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 14 +++++- drivers/infiniband/hw/mlx5/mr.c | 73 +++++++++++++++++++--------- drivers/infiniband/hw/mlx5/odp.c | 8 ++- 3 files changed, 69 insertions(+), 26 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +12 -2
@@ -613,6 +613,16 @@ enum mlx5_mkey_type { MLX5_MKEY_INDIRECT_DEVX, }; +struct mlx5r_cache_rb_key { +unsigned int access_mode; +unsigned int access_flags; +/* +* keep ndescs as the last member so entries with about the same ndescs +* will be close in the tree +*/ +unsigned int ndescs; +}; +struct mlx5_ib_mkey {u32 key;enum mlx5_mkey_type type;
@@ -732,10 +742,10 @@ struct mlx5_cache_ent {unsigned long stored;unsigned long reserved; +struct mlx5r_cache_rb_key rb_key; +char name[4 ];u32 order; -u32 access_mode; -unsigned int ndescs; u8 disabled:1; u8 fill_to_high_water:1;
drivers/infiniband/hw/mlx5/mr.c +51 -22
154 lines not on this page · raw message
drivers/infiniband/hw/mlx5/odp.c +6 -2
@@ -1589,10 +1589,14 @@ mlx5_ib_odp_destroy_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)void mlx5_odp_init_mkey_cache_entry (struct mlx5_cache_ent *ent) { +struct mlx5r_cache_rb_key rb_key = { + .access_mode =MLX5_MKC_ACCESS_MODE_KSM , + .ndescs = mlx5_imr_ksm_entries + }; +if (!(ent->dev->odp_caps.general_caps &IB_ODP_SUPPORT_IMPLICIT ))return ; - ent->ndescs = mlx5_imr_ksm_entries; - ent->access_mode =MLX5_MKC_ACCESS_MODE_KSM ; + ent->rb_key = rb_key; }static const struct ib_device_ops mlx5_ib_dev_odp_ops = {
Signature · 1 line
2.17.2
PATCH 6/8 RDMA/mlx5: Change the cache structure to an RB-tree
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
Currently, the cache structure is a static linear array. Therefore, his size is limited to the number of entries in it and is not expandable. The entries are dedicated to mkeys of size 2^x and no access_flags. Mkeys with different properties are not cacheable.
In this patch, we change the cache structure to an RB-tree. By this, we enable caching all user mkeys that can use umr.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 17 +- drivers/infiniband/hw/mlx5/mr.c | 307 +++++++++++++++++++-------- drivers/infiniband/hw/mlx5/odp.c | 9 +- 3 files changed, 237 insertions(+), 96 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +12 -5
@@ -742,10 +742,10 @@ struct mlx5_cache_ent {unsigned long stored;unsigned long reserved; +struct rb_node node;struct mlx5r_cache_rb_key rb_key;char name[4 ]; -u32 order; u8 disabled:1; u8 fill_to_high_water:1;
@@ -776,8 +776,9 @@ struct mlx5r_async_create_mkey {struct mlx5_mkey_cache {struct workqueue_struct *wq; -struct mlx5_cache_ent ent[MAX_MKEY_CACHE_ENTRIES ]; -struct dentry *root; +struct rb_root rb_root; +struct mutex rb_lock; +struct dentry *fs_root;unsigned long last_add; };
@@ -1322,6 +1323,8 @@ void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);int mlx5_ib_get_cqe_size (struct ib_cq *ibcq);int mlx5_mkey_cache_init (struct mlx5_ib_dev *dev);int mlx5_mkey_cache_cleanup (struct mlx5_ib_dev *dev); +struct mlx5_cache_ent *mlx5r_cache_create_ent (struct mlx5_ib_dev *dev, +struct mlx5r_cache_rb_key rb_key); struct mlx5_ib_mr *mlx5_mr_cache_alloc(struct mlx5_ib_dev *dev, u8 access_mode, unsigned int access_flags,
@@ -1349,7 +1352,7 @@ int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq);void mlx5_ib_odp_cleanup_one (struct mlx5_ib_dev *ibdev);int __init mlx5_ib_odp_init (void );void mlx5_ib_odp_cleanup (void ); -void mlx5_odp_init_mkey_cache_entry (struct mlx5_cache_ent *ent); +struct mlx5_cache_ent *mlx5_odp_init_mkey_cache_entry (struct mlx5_ib_dev *dev);void mlx5_odp_populate_xlt (void *xlt,size_t idx,size_t nentries,struct mlx5_ib_mr *mr,int flags);
@@ -1368,7 +1371,11 @@ static inline int mlx5r_odp_create_eq(struct mlx5_ib_dev *dev,static inline void mlx5_ib_odp_cleanup_one (struct mlx5_ib_dev *ibdev) {}static inline int mlx5_ib_odp_init (void ) {return 0 ; }static inline void mlx5_ib_odp_cleanup (void ) {} -static inline void mlx5_odp_init_mkey_cache_entry (struct mlx5_cache_ent *ent) {} +static inline struct mlx5_cache_ent * +mlx5_odp_init_mkey_cache_entry (struct mlx5_ib_dev *dev) +{ +return NULL ; +}static inline void mlx5_odp_populate_xlt (void *xlt,size_t idx,size_t nentries,struct mlx5_ib_mr *mr,int flags) {}
drivers/infiniband/hw/mlx5/mr.c +220 -87
400 lines not on this page · raw message
drivers/infiniband/hw/mlx5/odp.c +5 -4
@@ -1587,16 +1587,17 @@ mlx5_ib_odp_destroy_eq(struct mlx5_ib_dev *dev, struct mlx5_ib_pf_eq *eq)return err; } -void mlx5_odp_init_mkey_cache_entry (struct mlx5_cache_ent *ent) +struct mlx5_cache_ent *mlx5_odp_init_mkey_cache_entry (struct mlx5_ib_dev *dev) {struct mlx5r_cache_rb_key rb_key = { .access_mode =MLX5_MKC_ACCESS_MODE_KSM , .ndescs = mlx5_imr_ksm_entries }; -if (!(ent->dev->odp_caps.general_caps &IB_ODP_SUPPORT_IMPLICIT )) -return ; - ent->rb_key = rb_key; +if (!(dev->odp_caps.general_caps &IB_ODP_SUPPORT_IMPLICIT )) +return NULL ; + +return mlx5r_cache_create_ent (dev, rb_key); }static const struct ib_device_ops mlx5_ib_dev_odp_ops = {
Signature · 1 line
2.17.2
PATCH 7/8 RDMA/mlx5: Cache all user cacheable mkeys on dereg MR flow
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
Currently, when dereging an MR, if the mkey doesn't belong to a cache entry, it will be destroyed.
As a result, the restart of applications with many non-cached mkeys is not efficient since all the mkeys are destroyed and then recreated. This process takes a long time (for 100,000 MRs, it is ~20 seconds for dereg and ~28 seconds for re-reg).
To shorten the restart runtime, insert all cacheable mkeys to the cache. If there is no fitting entry to the mkey properties, create a temporary entry that fits it.
After a predetermined timeout, the cache entries will shrink to the initial high limit.
The mkeys will still be in the cache when consuming them again after an application restart. Therefore, the registration will be much faster (for 100,000 MRs, it is ~4 seconds for dereg and ~5 seconds for re-reg).
The temporary cache entries created to store the non-cache mkeys are not exposed through sysfs like the default cache entries.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 2 ++ drivers/infiniband/hw/mlx5/mr.c | 48 +++++++++++++++++++++++----- 2 files changed, 42 insertions(+), 8 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +2 -0
@@ -629,6 +629,8 @@ struct mlx5_ib_mkey {unsigned int ndescs;struct wait_queue_head wait;refcount_t usecount; +/* User Mkey must hold either a cache_key or a cache_ent. */ +struct mlx5r_cache_rb_key rb_key;struct mlx5_cache_ent *cache_ent; };
drivers/infiniband/hw/mlx5/mr.c +40 -8
68 lines not on this page · raw message
Signature · 1 line
2.17.2
PATCH 8/8 RDMA/mlx5: Add work to remove temporary entries from the cache
In reply to Michael GuralnikFrom: Aharon Landau <[email protected]>
The non-cache mkeys are stored in the cache only to shorten restarting application time. Don't store them longer than needed.
Configure cache entries that store non-cache MR as temporary entries. If 30 seconds have passed and no user reclaimed the temporarily cached mkeys, an asynchronous work will destroy the mkeys and the temporary entries.
When allocating an mkey from a temporary entry, don't keep a pointer to the entry, as it might be destroyed.
--- drivers/infiniband/hw/mlx5/mlx5_ib.h | 2 + drivers/infiniband/hw/mlx5/mr.c | 95 ++++++++++++++++++++++++---- 2 files changed, 83 insertions(+), 14 deletions(-)
drivers/infiniband/hw/mlx5/mlx5_ib.h +2 -0
@@ -749,6 +749,7 @@ struct mlx5_cache_ent {char name[4 ]; + u8 is_tmp:1; u8 disabled:1; u8 fill_to_high_water:1;
@@ -782,6 +783,7 @@ struct mlx5_mkey_cache {struct mutex rb_lock;struct dentry *fs_root;unsigned long last_add; +struct delayed_work remove_ent_dwork; };struct mlx5_ib_port_resources {
drivers/infiniband/hw/mlx5/mr.c +81 -14
179 lines not on this page · raw message
Signature · 1 line
2.17.2