RDMA/rxe: Misc rxe_pool cleanups
Subsystems (2)
v3 cover RDMA/rxe: Misc rxe_pool cleanups
[v3]
Fixed a spelling error suggested by [email protected] Changed the naming of APIs to xxx_locked suggested by [email protected]
[v2]
This series of patches corrects a bug introduced in rxe_pool.c by a recent commit and then addresses several issues raised during discussion of the bug and the proposed fix.
The first patch fixes a real bug but the other five are stylistic and cleanup changes.
Taken together these changes also improve the performance of ib_write_bw (in software loopback) by about 6% by reducing overhead in rxe_pool.c compared to current head of tree.
Bob Pearson (6):
RDMA/rxe: Fix bug in rxe_alloc
RDMA/rxe: Fix misleading comments and names RDMA/rxe: Remove RXE_POOL_ATOMIC
RDMA/rxe: Remove references to ib_device and pool RDMA/rxe: Remove unneeded pool->state RDMA/rxe: Replace missing rxe_pool_get_index_locked
drivers/infiniband/sw/rxe/rxe_mcast.c | 8 +- drivers/infiniband/sw/rxe/rxe_pool.c | 132 +++++++++----------------- drivers/infiniband/sw/rxe/rxe_pool.h | 63 ++++++------ 3 files changed, 76 insertions(+), 127 deletions(-)
Signature · 1 line
2.27.0
PATCH v3 1/6 RDMA/rxe: Fix bug in rxe_alloc
[v1]
A recent patch which added an 'unlocked' version of rxe_alloc introduced a bug causing kzalloc(..., GFP_KERNEL) to be called while holding a spin lock. This patch corrects that error.
rxe_alloc_nl() should always be called while holding the pool->pool_lock so the 2nd argument to kzalloc there should be GFP_ATOMIC.
rxe_alloc() prior to the change only locked the code around checking that pool->state is RXE_POOL_STATE_VALID to avoid races between working threads and a thread shutting down the rxe driver. This patch reverts rxe_alloc() to this behavior so the lock is not held when kzalloc() is called.
--- drivers/infiniband/sw/rxe/rxe_pool.c | 41 ++++++++++++++++++++++++---- 1 file changed, 35 insertions(+), 6 deletions(-)
drivers/infiniband/sw/rxe/rxe_pool.c +35 -6
@@ -343,8 +343,6 @@ void *rxe_alloc_nl(struct rxe_pool *pool)struct rxe_pool_entry *elem;u8 *obj; -might_sleep_if (!(pool->flags &RXE_POOL_ATOMIC )); -if (pool->state !=RXE_POOL_STATE_VALID )return NULL ;
@@ -356,8 +354,7 @@ void *rxe_alloc_nl(struct rxe_pool *pool)if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt; - obj =kzalloc (info->size, (pool->flags &RXE_POOL_ATOMIC ) ? -GFP_ATOMIC :GFP_KERNEL ); + obj =kzalloc (info->size,GFP_ATOMIC );if (!obj) goto out_cnt;
@@ -378,14 +375,46 @@ void *rxe_alloc_nl(struct rxe_pool *pool)void *rxe_alloc (struct rxe_pool *pool) { -u8 *obj;unsigned long flags; +struct rxe_type_info *info = &rxe_type_info[pool->type]; +struct rxe_pool_entry *elem; +u8 *obj; + +might_sleep_if (!(pool->flags &RXE_POOL_ATOMIC ));read_lock_irqsave (&pool->pool_lock, flags); - obj =rxe_alloc_nl (pool); +if (pool->state !=RXE_POOL_STATE_VALID ) { +read_unlock_irqrestore (&pool->pool_lock, flags); +return NULL ; + } + +kref_get (&pool->ref_cnt);read_unlock_irqrestore (&pool->pool_lock, flags); +if (!ib_device_try_get (&pool->rxe->ib_dev)) + goto out_put_pool; + +if (atomic_inc_return (&pool->num_elem) > pool->max_elem) + goto out_cnt; + + obj =kzalloc (info->size, (pool->flags &RXE_POOL_ATOMIC ) ? +GFP_ATOMIC :GFP_KERNEL ); +if (!obj) + goto out_cnt; + + elem = (struct rxe_pool_entry *)(obj + info->elem_offset); + + elem->pool = pool; +kref_init (&elem->ref_cnt); +return obj; + +out_cnt: +atomic_dec (&pool->num_elem); +ib_device_put (&pool->rxe->ib_dev); +out_put_pool: +rxe_pool_put (pool); +return NULL ; }int __rxe_add_to_pool (struct rxe_pool *pool,struct rxe_pool_entry *elem)
Signature · 1 line
2.27.0
PATCH v3 2/6 RDMA/rxe: Fix misleading comments and names
In reply to Bob Pearson[v3]
Was rxe_xxx__() changed to rxe_xxx_locked()
The names and comments of the 'unlocked' pool APIs are very misleading and not what was intended. This patch replaces 'rxe_xxx_nl' with 'rxe_xxx_locked' with comments indicating that the caller is expected to hold the rxe pool lock.
--- drivers/infiniband/sw/rxe/rxe_mcast.c | 8 ++-- drivers/infiniband/sw/rxe/rxe_pool.c | 22 +++++------ drivers/infiniband/sw/rxe/rxe_pool.h | 55 +++++++++++++-------------- 3 files changed, 42 insertions(+), 43 deletions(-)
drivers/infiniband/sw/rxe/rxe_mcast.c +4 -4
@@ -15,18 +15,18 @@ static struct rxe_mc_grp *create_grp(struct rxe_dev *rxe,int err;struct rxe_mc_grp *grp; - grp =rxe_alloc_nl (&rxe->mc_grp_pool); + grp =rxe_alloc_locked (&rxe->mc_grp_pool);if (!grp)return ERR_PTR (-ENOMEM );INIT_LIST_HEAD (&grp->qp_list);spin_lock_init (&grp->mcg_lock); grp->rxe = rxe; -rxe_add_key_nl (grp, mgid); +rxe_add_key_locked (grp, mgid); err =rxe_mcast_add (rxe, mgid);if (unlikely (err)) { -rxe_drop_key_nl (grp); +rxe_drop_key_locked (grp);rxe_drop_ref (grp);return ERR_PTR (err); }
@@ -47,7 +47,7 @@ int rxe_mcast_get_grp(struct rxe_dev *rxe, union ib_gid *mgid,write_lock_irqsave (&pool->pool_lock, flags); - grp =rxe_pool_get_key_nl (pool, mgid); + grp =rxe_pool_get_key_locked (pool, mgid);if (grp) goto done;
drivers/infiniband/sw/rxe/rxe_pool.c +11 -11
@@ -266,7 +266,7 @@ static void insert_key(struct rxe_pool *pool, struct rxe_pool_entry *new)return ; } -void __rxe_add_key_nl (struct rxe_pool_entry *elem,void *key) +void __rxe_add_key_locked (struct rxe_pool_entry *elem,void *key) {struct rxe_pool *pool = elem->pool;
@@ -280,11 +280,11 @@ void __rxe_add_key(struct rxe_pool_entry *elem, void *key)unsigned long flags;write_lock_irqsave (&pool->pool_lock, flags); -__rxe_add_key_nl (elem, key); +__rxe_add_key_locked (elem, key);write_unlock_irqrestore (&pool->pool_lock, flags); } -void __rxe_drop_key_nl (struct rxe_pool_entry *elem) +void __rxe_drop_key_locked (struct rxe_pool_entry *elem) {struct rxe_pool *pool = elem->pool;
@@ -297,11 +297,11 @@ void __rxe_drop_key(struct rxe_pool_entry *elem)unsigned long flags;write_lock_irqsave (&pool->pool_lock, flags); -__rxe_drop_key_nl (elem); +__rxe_drop_key_locked (elem);write_unlock_irqrestore (&pool->pool_lock, flags); } -void __rxe_add_index_nl (struct rxe_pool_entry *elem) +void __rxe_add_index_locked (struct rxe_pool_entry *elem) {struct rxe_pool *pool = elem->pool;
@@ -315,11 +315,11 @@ void __rxe_add_index(struct rxe_pool_entry *elem)unsigned long flags;write_lock_irqsave (&pool->pool_lock, flags); -__rxe_add_index_nl (elem); +__rxe_add_index_locked (elem);write_unlock_irqrestore (&pool->pool_lock, flags); } -void __rxe_drop_index_nl (struct rxe_pool_entry *elem) +void __rxe_drop_index_locked (struct rxe_pool_entry *elem) {struct rxe_pool *pool = elem->pool;
@@ -333,11 +333,11 @@ void __rxe_drop_index(struct rxe_pool_entry *elem)unsigned long flags;write_lock_irqsave (&pool->pool_lock, flags); -__rxe_drop_index_nl (elem); +__rxe_drop_index_locked (elem);write_unlock_irqrestore (&pool->pool_lock, flags); } -void *rxe_alloc_nl (struct rxe_pool *pool) +void *rxe_alloc_locked (struct rxe_pool *pool) {struct rxe_type_info *info = &rxe_type_info[pool->type];struct rxe_pool_entry *elem;
@@ -507,7 +507,7 @@ void *rxe_pool_get_index(struct rxe_pool *pool, u32 index)return obj; } -void *rxe_pool_get_key_nl (struct rxe_pool *pool,void *key) +void *rxe_pool_get_key_locked (struct rxe_pool *pool,void *key) {struct rxe_type_info *info = &rxe_type_info[pool->type];struct rb_node *node;
@@ -551,7 +551,7 @@ void *rxe_pool_get_key(struct rxe_pool *pool, void *key)unsigned long flags;read_lock_irqsave (&pool->pool_lock, flags); - obj =rxe_pool_get_key_nl (pool, key); + obj =rxe_pool_get_key_locked (pool, key);read_unlock_irqrestore (&pool->pool_lock, flags);return obj;
drivers/infiniband/sw/rxe/rxe_pool.h +27 -28
98 lines not on this page · raw message
Signature · 1 line
2.27.0
PATCH v3 3/6 RDMA/rxe: Remove RXE_POOL_ATOMIC
In reply to Bob Pearson[v2]rxe_alloc() used the RXE_POOL_ATOMIC flag in rxe_type_info to select GFP_ATOMIC in calls to kzalloc(). This was intended to handle cases where an object could be created in interrupt context. This no longer occurs since allocating those objects has moved into the core so this flag is not necessary. An incorrect use of this flag was still present for rxe_mc_elem objects and is removed.
--- drivers/infiniband/sw/rxe/rxe_pool.c | 6 +----- drivers/infiniband/sw/rxe/rxe_pool.h | 1 - 2 files changed, 1 insertion(+), 6 deletions(-)
drivers/infiniband/sw/rxe/rxe_pool.c +1 -5
@@ -84,7 +84,6 @@ struct rxe_type_info rxe_type_info[RXE_NUM_TYPES] = { .name = "rxe-mc_elem", .size = sizeof(struct rxe_mc_elem), .elem_offset = offsetof(struct rxe_mc_elem, pelem), - .flags = RXE_POOL_ATOMIC, }, };
@@ -380,8 +379,6 @@ void *rxe_alloc(struct rxe_pool *pool)struct rxe_pool_entry *elem;u8 *obj; -might_sleep_if (!(pool->flags &RXE_POOL_ATOMIC )); -read_lock_irqsave (&pool->pool_lock, flags);if (pool->state !=RXE_POOL_STATE_VALID ) {read_unlock_irqrestore (&pool->pool_lock, flags);
@@ -397,8 +394,7 @@ void *rxe_alloc(struct rxe_pool *pool)if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt; - obj =kzalloc (info->size, (pool->flags &RXE_POOL_ATOMIC ) ? -GFP_ATOMIC :GFP_KERNEL ); + obj =kzalloc (info->size,GFP_KERNEL );if (!obj) goto out_cnt;
drivers/infiniband/sw/rxe/rxe_pool.h +0 -1
@@ -11,7 +11,6 @@#define RXE_POOL_CACHE_FLAGS (0)enum rxe_pool_flags { -RXE_POOL_ATOMIC =BIT (0 ),RXE_POOL_INDEX =BIT (1 ),RXE_POOL_KEY =BIT (2 ),RXE_POOL_NO_ALLOC =BIT (4 ),
Signature · 1 line
2.27.0
PATCH v3 4/6 RDMA/rxe: Remove references to ib_device and pool
In reply to Bob Pearson[v2]rxe_pool.c takes references to the pool and ib_device structs for each object allocated and also keeps an atomic num_elem count in each pool. This is more work than is needed. Pool allocation is only called from verbs APIs which already have references to ib_device and pools are only diasbled when the driver is removed so no protection of the pool addresses are needed. The elem count is used to warn if elements are still present in a pool when it is cleaned up which is useful.
This patch eliminates the references to the ib_device and pool structs.
[v1]
The previous version only removed the ib_device reference.
--- drivers/infiniband/sw/rxe/rxe_pool.c | 42 ++-------------------------- drivers/infiniband/sw/rxe/rxe_pool.h | 1 - 2 files changed, 2 insertions(+), 41 deletions(-)
drivers/infiniband/sw/rxe/rxe_pool.c +2 -40
@@ -142,8 +142,6 @@ int rxe_pool_init(atomic_set (&pool->num_elem,0 ); -kref_init (&pool->ref_cnt); -rwlock_init (&pool->pool_lock);if (rxe_type_info[type].flags &RXE_POOL_INDEX ) {
@@ -165,19 +163,6 @@ int rxe_pool_init(return err; } -static void rxe_pool_release (struct kref *kref) -{ - struct rxe_pool *pool = container_of(kref, struct rxe_pool, ref_cnt); - - pool->state =RXE_POOL_STATE_INVALID ; -kfree (pool->index.table); -} - -static void rxe_pool_put (struct rxe_pool *pool) -{ -kref_put (&pool->ref_cnt, rxe_pool_release); -} -void rxe_pool_cleanup (struct rxe_pool *pool) {unsigned long flags;
@@ -189,7 +174,8 @@ void rxe_pool_cleanup(struct rxe_pool *pool) pool_name(pool));write_unlock_irqrestore (&pool->pool_lock, flags); -rxe_pool_put (pool); + pool->state =RXE_POOL_STATE_INVALID ; +kfree (pool->index.table); }static u32 alloc_index (struct rxe_pool *pool)
@@ -345,11 +331,6 @@ void *rxe_alloc_locked(struct rxe_pool *pool)if (pool->state !=RXE_POOL_STATE_VALID )return NULL ; -kref_get (&pool->ref_cnt); - -if (!ib_device_try_get (&pool->rxe->ib_dev)) - goto out_put_pool; -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -366,9 +347,6 @@ void *rxe_alloc_locked(struct rxe_pool *pool) out_cnt:atomic_dec (&pool->num_elem); -ib_device_put (&pool->rxe->ib_dev); -out_put_pool: -rxe_pool_put (pool);return NULL ; }
@@ -385,12 +363,8 @@ void *rxe_alloc(struct rxe_pool *pool)return NULL ; } -kref_get (&pool->ref_cnt);read_unlock_irqrestore (&pool->pool_lock, flags); -if (!ib_device_try_get (&pool->rxe->ib_dev)) - goto out_put_pool; -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -407,9 +381,6 @@ void *rxe_alloc(struct rxe_pool *pool) out_cnt:atomic_dec (&pool->num_elem); -ib_device_put (&pool->rxe->ib_dev); -out_put_pool: -rxe_pool_put (pool);return NULL ; }
@@ -422,12 +393,8 @@ int __rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_entry *elem)read_unlock_irqrestore (&pool->pool_lock, flags);return -EINVAL ; } -kref_get (&pool->ref_cnt);read_unlock_irqrestore (&pool->pool_lock, flags); -if (!ib_device_try_get (&pool->rxe->ib_dev)) - goto out_put_pool; -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -438,9 +405,6 @@ int __rxe_add_to_pool(struct rxe_pool *pool, struct rxe_pool_entry *elem) out_cnt:atomic_dec (&pool->num_elem); -ib_device_put (&pool->rxe->ib_dev); -out_put_pool: -rxe_pool_put (pool);return -EINVAL ; }
@@ -461,8 +425,6 @@ void rxe_elem_release(struct kref *kref) }atomic_dec (&pool->num_elem); -ib_device_put (&pool->rxe->ib_dev); -rxe_pool_put (pool); }void *rxe_pool_get_index (struct rxe_pool *pool,u32 index)
drivers/infiniband/sw/rxe/rxe_pool.h +0 -1
@@ -68,7 +68,6 @@ struct rxe_pool {struct rxe_dev *rxe;rwlock_t pool_lock;/* protects pool add/del/search */ size_t elem_size; -struct kref ref_cnt;void (*cleanup)(struct rxe_pool_entry *obj);enum rxe_pool_state state;enum rxe_pool_flags flags;
Signature · 1 line
2.27.0
PATCH v3 5/6 RDMA/rxe: Remove unneeded pool->state
In reply to Bob Pearson[v3]
Fixed spelling error.
rxe_pool.c uses the field pool->state to mark a pool as invalid when it is shut down and checks it in several pool APIs to verify that the pool has not been shut down. This is unneeded because the pools are not marked invalid unless the entire driver is being removed at which point no functional APIs should or could be executing. This patch removes this field and associated code.
--- drivers/infiniband/sw/rxe/rxe_pool.c | 38 +--------------------------- drivers/infiniband/sw/rxe/rxe_pool.h | 6 ----- 2 files changed, 1 insertion(+), 43 deletions(-)
drivers/infiniband/sw/rxe/rxe_pool.c +1 -37
@@ -157,24 +157,16 @@ int rxe_pool_init( pool->key.key_size = rxe_type_info[type].key_size; } - pool->state =RXE_POOL_STATE_VALID ; - out:return err; }void rxe_pool_cleanup (struct rxe_pool *pool) { -unsigned long flags; - -write_lock_irqsave (&pool->pool_lock, flags); - pool->state =RXE_POOL_STATE_INVALID ;if (atomic_read (&pool->num_elem) >0 )pr_warn ("%s pool destroyed with unfree'd elem\n" ,pool_name (pool)); -write_unlock_irqrestore (&pool->pool_lock, flags); - pool->state =RXE_POOL_STATE_INVALID ;kfree (pool->index.table); }
@@ -328,9 +320,6 @@ void *rxe_alloc_locked(struct rxe_pool *pool)struct rxe_pool_entry *elem;u8 *obj; -if (pool->state !=RXE_POOL_STATE_VALID ) -return NULL ; -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -352,19 +341,10 @@ void *rxe_alloc_locked(struct rxe_pool *pool)void *rxe_alloc (struct rxe_pool *pool) { -unsigned long flags;struct rxe_type_info *info = &rxe_type_info[pool->type];struct rxe_pool_entry *elem;u8 *obj; -read_lock_irqsave (&pool->pool_lock, flags); -if (pool->state !=RXE_POOL_STATE_VALID ) { -read_unlock_irqrestore (&pool->pool_lock, flags); -return NULL ; - } - -read_unlock_irqrestore (&pool->pool_lock, flags); -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -386,15 +366,6 @@ void *rxe_alloc(struct rxe_pool *pool)int __rxe_add_to_pool (struct rxe_pool *pool,struct rxe_pool_entry *elem) { -unsigned long flags; - -read_lock_irqsave (&pool->pool_lock, flags); -if (pool->state !=RXE_POOL_STATE_VALID ) { -read_unlock_irqrestore (&pool->pool_lock, flags); -return -EINVAL ; - } -read_unlock_irqrestore (&pool->pool_lock, flags); -if (atomic_inc_return (&pool->num_elem) > pool->max_elem) goto out_cnt;
@@ -437,9 +408,6 @@ void *rxe_pool_get_index(struct rxe_pool *pool, u32 index)read_lock_irqsave (&pool->pool_lock, flags); -if (pool->state !=RXE_POOL_STATE_VALID ) - goto out; - node = pool->index.tree.rb_node;while (node) {
@@ -460,8 +428,8 @@ void *rxe_pool_get_index(struct rxe_pool *pool, u32 index) obj =NULL ; } -out:read_unlock_irqrestore (&pool->pool_lock, flags); +return obj; }
@@ -473,9 +441,6 @@ void *rxe_pool_get_key_locked(struct rxe_pool *pool, void *key)u8 *obj =NULL ;int cmp; -if (pool->state !=RXE_POOL_STATE_VALID ) - goto out; - node = pool->key.tree.rb_node;while (node) {
@@ -499,7 +464,6 @@ void *rxe_pool_get_key_locked(struct rxe_pool *pool, void *key) obj =NULL ; } -out:return obj; }
drivers/infiniband/sw/rxe/rxe_pool.h +0 -6
@@ -46,11 +46,6 @@ struct rxe_type_info {extern struct rxe_type_info rxe_type_info[]; -enum rxe_pool_state { -RXE_POOL_STATE_INVALID , -RXE_POOL_STATE_VALID , -}; -struct rxe_pool_entry {struct rxe_pool *pool;struct kref ref_cnt;
@@ -69,7 +64,6 @@ struct rxe_pool {rwlock_t pool_lock;/* protects pool add/del/search */ size_t elem_size;void (*cleanup)(struct rxe_pool_entry *obj); -enum rxe_pool_state state;enum rxe_pool_flags flags;enum rxe_elem_type type;
Signature · 1 line
2.27.0
PATCH v3 6/6 RDMA/rxe: Replace missing rxe_pool_get_index_locked
In reply to Bob Pearson[v2]
One of the pool APIs for when caller is holding lock was not defined but is declared in rxe_pool.h. This patch adds the definition.
--- drivers/infiniband/sw/rxe/rxe_pool.c | 21 ++++++++++++++------- 1 file changed, 14 insertions(+), 7 deletions(-)
drivers/infiniband/sw/rxe/rxe_pool.c +14 -7
@@ -398,15 +398,12 @@ void rxe_elem_release(struct kref *kref)atomic_dec (&pool->num_elem); } -void *rxe_pool_get_index (struct rxe_pool *pool,u32 index) +void *rxe_pool_get_index_locked (struct rxe_pool *pool,u32 index) {struct rxe_type_info *info = &rxe_type_info[pool->type];struct rb_node *node;struct rxe_pool_entry *elem; -u8 *obj =NULL ; -unsigned long flags; - -read_lock_irqsave (&pool->pool_lock, flags); +u8 *obj; node = pool->index.tree.rb_node;
@@ -428,6 +425,16 @@ void *rxe_pool_get_index(struct rxe_pool *pool, u32 index) obj =NULL ; } +return obj; +} + +void *rxe_pool_get_index (struct rxe_pool *pool,u32 index) +{ +u8 *obj; +unsigned long flags; + +read_lock_irqsave (&pool->pool_lock, flags); + obj =rxe_pool_get_index_locked (pool, index);read_unlock_irqrestore (&pool->pool_lock, flags);return obj;
@@ -438,7 +445,7 @@ void *rxe_pool_get_key_locked(struct rxe_pool *pool, void *key)struct rxe_type_info *info = &rxe_type_info[pool->type];struct rb_node *node;struct rxe_pool_entry *elem; -u8 *obj =NULL ; +u8 *obj;int cmp; node = pool->key.tree.rb_node;
@@ -469,7 +476,7 @@ void *rxe_pool_get_key_locked(struct rxe_pool *pool, void *key)void *rxe_pool_get_key (struct rxe_pool *pool,void *key) { -u8 *obj =NULL ; +u8 *obj;unsigned long flags;read_lock_irqsave (&pool->pool_lock, flags);
Signature · 1 line
2.27.0