diff options
author | Sean Hefty <shefty@nvidia.com> | 2024-11-13 14:12:55 +0300 |
---|---|---|
committer | Leon Romanovsky <leon@kernel.org> | 2024-11-17 12:51:43 +0300 |
commit | 1e5159219076ddb2e44338c667c83fd1bd43dfef (patch) | |
tree | 9a6acfd6fa21b738e7383ef20dfb3ec898e3b2ae /drivers/infiniband | |
parent | 0492458750c9fbd69cfc7baddd3ddcac77f2a0c8 (diff) | |
download | linux-1e5159219076ddb2e44338c667c83fd1bd43dfef.tar.xz |
IB/cm: Do not hold reference on cm_id unless needed
Typically, when the CM sends a MAD it bumps a reference count
on the associated cm_id. There are some exceptions, such
as when the MAD is a direct response to a receive MAD. For
example, the CM may generate an MRA in response to a duplicate
REQ. But, in general, if a MAD may be sent as a result of
the user invoking an API call (e.g. ib_send_cm_rep(),
ib_send_cm_rtu(), etc.), a reference is taken on the cm_id.
This reference is necessary if the MAD requires a response.
The reference allows routing a response MAD back to the
cm_id, or, if no response is received, allows updating the
cm_id state to reflect the failure.
For MADs which do not generate a response from the
target, however, there's no need to hold a reference on the cm_id.
Such MADs will not be retried by the MAD layer and their
completions do not change the state of the cm_id.
There are 2 internal calls used to allocate MADs which take
a reference on the cm_id: cm_alloc_msg() and cm_alloc_priv_msg().
The latter calls the former. It turns out that all other places
where cm_alloc_msg() is called are for MADs that do not generate
a response from the target: sending an RTU, DREP, REJ, MRA, or
SIDR REP. In all of these cases, there's no need to hold a
reference on the cm_id.
The benefit of dropping unneeded references is that it allows
destruction of the cm_id to proceed immediately. Currently,
the cm_destroy_id() call blocks as long as there's a reference
held on the cm_id. Worse, is that cm_destroy_id() will send
MADs, which it then needs to complete. Sending the MADs is
beneficial, as they notify the peer that a connection is
being destroyed. However, since the MADs hold a reference
on the cm_id, they block destruction and cannot be retried.
Move cm_id referencing from cm_alloc_msg() to cm_alloc_priv_msg().
The latter should hold a reference on the cm_id in all cases but
one, which will be handled in a separate patch. cm_alloc_priv_msg()
is used when sending a REQ, REP, DREQ, and SIDR REQ, all of which
require a response.
Also, merge common code into cm_alloc_priv_msg() and combine the
freeing of all messages which do not need a response.
Signed-off-by: Sean Hefty <shefty@nvidia.com>
Signed-off-by: Or Har-Toov <ohartoov@nvidia.com>
Signed-off-by: Vlad Dumitrescu <vdumitrescu@nvidia.com>
Link: https://patch.msgid.link/1f0f96acace72790ecf89087fc765dead960189e.1731495873.git.leon@kernel.org
Signed-off-by: Leon Romanovsky <leon@kernel.org>
Diffstat (limited to 'drivers/infiniband')
-rw-r--r-- | drivers/infiniband/core/cm.c | 66 |
1 files changed, 24 insertions, 42 deletions
diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c index 99246e49dd3a..2517bfebcfd5 100644 --- a/drivers/infiniband/core/cm.c +++ b/drivers/infiniband/core/cm.c @@ -309,12 +309,7 @@ static struct ib_mad_send_buf *cm_alloc_msg(struct cm_id_private *cm_id_priv) goto out; } - /* Timeout set by caller if response is expected. */ m->ah = ah; - m->retries = cm_id_priv->max_cm_retries; - - refcount_inc(&cm_id_priv->refcount); - m->context[0] = cm_id_priv; out: spin_unlock(&cm_id_priv->av.port->cm_dev->mad_agent_lock); @@ -323,16 +318,13 @@ out: static void cm_free_msg(struct ib_mad_send_buf *msg) { - struct cm_id_private *cm_id_priv = msg->context[0]; - if (msg->ah) rdma_destroy_ah(msg->ah, 0); - cm_deref_id(cm_id_priv); ib_free_send_mad(msg); } static struct ib_mad_send_buf * -cm_alloc_priv_msg(struct cm_id_private *cm_id_priv) +cm_alloc_priv_msg(struct cm_id_private *cm_id_priv, enum ib_cm_state state) { struct ib_mad_send_buf *msg; @@ -341,7 +333,15 @@ cm_alloc_priv_msg(struct cm_id_private *cm_id_priv) msg = cm_alloc_msg(cm_id_priv); if (IS_ERR(msg)) return msg; + cm_id_priv->msg = msg; + refcount_inc(&cm_id_priv->refcount); + msg->context[0] = cm_id_priv; + msg->context[1] = (void *) (unsigned long) state; + + msg->retries = cm_id_priv->max_cm_retries; + msg->timeout_ms = cm_id_priv->timeout_ms; + return msg; } @@ -413,13 +413,6 @@ static int cm_alloc_response_msg(struct cm_port *port, return 0; } -static void cm_free_response_msg(struct ib_mad_send_buf *msg) -{ - if (msg->ah) - rdma_destroy_ah(msg->ah, 0); - ib_free_send_mad(msg); -} - static void *cm_copy_private_data(const void *private_data, u8 private_data_len) { void *data; @@ -1567,7 +1560,7 @@ int ib_send_cm_req(struct ib_cm_id *cm_id, if (param->alternate_path) cm_move_av_from_path(&cm_id_priv->alt_av, &alt_av); - msg = cm_alloc_priv_msg(cm_id_priv); + msg = cm_alloc_priv_msg(cm_id_priv, IB_CM_REQ_SENT); if (IS_ERR(msg)) { ret = PTR_ERR(msg); goto out_unlock; @@ -1576,8 +1569,6 @@ int ib_send_cm_req(struct ib_cm_id *cm_id, req_msg = (struct cm_req_msg *)msg->mad; cm_format_req(req_msg, cm_id_priv, param); cm_id_priv->tid = req_msg->hdr.tid; - msg->timeout_ms = cm_id_priv->timeout_ms; - msg->context[1] = (void *)(unsigned long)IB_CM_REQ_SENT; cm_id_priv->local_qpn = cpu_to_be32(IBA_GET(CM_REQ_LOCAL_QPN, req_msg)); cm_id_priv->rq_psn = cpu_to_be32(IBA_GET(CM_REQ_STARTING_PSN, req_msg)); @@ -1634,7 +1625,7 @@ static int cm_issue_rej(struct cm_port *port, IBA_GET(CM_REJ_REMOTE_COMM_ID, rcv_msg)); ret = ib_post_send_mad(msg, NULL); if (ret) - cm_free_response_msg(msg); + cm_free_msg(msg); return ret; } @@ -1990,7 +1981,7 @@ static void cm_dup_req_handler(struct cm_work *work, return; unlock: spin_unlock_irq(&cm_id_priv->lock); -free: cm_free_response_msg(msg); +free: cm_free_msg(msg); } static struct cm_id_private *cm_match_req(struct cm_work *work, @@ -2304,7 +2295,7 @@ int ib_send_cm_rep(struct ib_cm_id *cm_id, goto out; } - msg = cm_alloc_priv_msg(cm_id_priv); + msg = cm_alloc_priv_msg(cm_id_priv, IB_CM_REP_SENT); if (IS_ERR(msg)) { ret = PTR_ERR(msg); goto out; @@ -2312,8 +2303,6 @@ int ib_send_cm_rep(struct ib_cm_id *cm_id, rep_msg = (struct cm_rep_msg *) msg->mad; cm_format_rep(rep_msg, cm_id_priv, param); - msg->timeout_ms = cm_id_priv->timeout_ms; - msg->context[1] = (void *) (unsigned long) IB_CM_REP_SENT; trace_icm_send_rep(cm_id); ret = ib_post_send_mad(msg, NULL); @@ -2479,7 +2468,7 @@ static void cm_dup_rep_handler(struct cm_work *work) goto deref; unlock: spin_unlock_irq(&cm_id_priv->lock); -free: cm_free_response_msg(msg); +free: cm_free_msg(msg); deref: cm_deref_id(cm_id_priv); } @@ -2683,7 +2672,7 @@ static int cm_send_dreq_locked(struct cm_id_private *cm_id_priv, cm_id_priv->id.lap_state == IB_CM_MRA_LAP_RCVD) ib_cancel_mad(cm_id_priv->msg); - msg = cm_alloc_priv_msg(cm_id_priv); + msg = cm_alloc_priv_msg(cm_id_priv, IB_CM_DREQ_SENT); if (IS_ERR(msg)) { cm_enter_timewait(cm_id_priv); return PTR_ERR(msg); @@ -2691,8 +2680,6 @@ static int cm_send_dreq_locked(struct cm_id_private *cm_id_priv, cm_format_dreq((struct cm_dreq_msg *) msg->mad, cm_id_priv, private_data, private_data_len); - msg->timeout_ms = cm_id_priv->timeout_ms; - msg->context[1] = (void *) (unsigned long) IB_CM_DREQ_SENT; trace_icm_send_dreq(&cm_id_priv->id); ret = ib_post_send_mad(msg, NULL); @@ -2819,7 +2806,7 @@ static int cm_issue_drep(struct cm_port *port, IBA_GET(CM_DREQ_REMOTE_COMM_ID, dreq_msg)); ret = ib_post_send_mad(msg, NULL); if (ret) - cm_free_response_msg(msg); + cm_free_msg(msg); return ret; } @@ -2878,7 +2865,7 @@ static int cm_dreq_handler(struct cm_work *work) if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) || ib_post_send_mad(msg, NULL)) - cm_free_response_msg(msg); + cm_free_msg(msg); goto deref; case IB_CM_DREQ_RCVD: atomic_long_inc(&work->port->counters[CM_RECV_DUPLICATES] @@ -3386,7 +3373,7 @@ static int cm_lap_handler(struct cm_work *work) if (cm_create_response_msg_ah(work->port, work->mad_recv_wc, msg) || ib_post_send_mad(msg, NULL)) - cm_free_response_msg(msg); + cm_free_msg(msg); goto deref; case IB_CM_LAP_RCVD: atomic_long_inc(&work->port->counters[CM_RECV_DUPLICATES] @@ -3525,7 +3512,7 @@ int ib_send_cm_sidr_req(struct ib_cm_id *cm_id, goto out_unlock; } - msg = cm_alloc_priv_msg(cm_id_priv); + msg = cm_alloc_priv_msg(cm_id_priv, IB_CM_SIDR_REQ_SENT); if (IS_ERR(msg)) { ret = PTR_ERR(msg); goto out_unlock; @@ -3533,8 +3520,6 @@ int ib_send_cm_sidr_req(struct ib_cm_id *cm_id, cm_format_sidr_req((struct cm_sidr_req_msg *)msg->mad, cm_id_priv, param); - msg->timeout_ms = cm_id_priv->timeout_ms; - msg->context[1] = (void *)(unsigned long)IB_CM_SIDR_REQ_SENT; trace_icm_send_sidr_req(&cm_id_priv->id); ret = ib_post_send_mad(msg, NULL); @@ -3780,17 +3765,17 @@ out: static void cm_process_send_error(struct cm_id_private *cm_id_priv, struct ib_mad_send_buf *msg, - enum ib_cm_state state, enum ib_wc_status wc_status) { + enum ib_cm_state state = (unsigned long) msg->context[1]; struct ib_cm_event cm_event = {}; int ret; - /* Discard old sends or ones without a response. */ + /* Discard old sends. */ spin_lock_irq(&cm_id_priv->lock); if (msg != cm_id_priv->msg) { spin_unlock_irq(&cm_id_priv->lock); - cm_free_msg(msg); + cm_free_priv_msg(msg); return; } cm_free_priv_msg(msg); @@ -3839,8 +3824,6 @@ static void cm_send_handler(struct ib_mad_agent *mad_agent, { struct ib_mad_send_buf *msg = mad_send_wc->send_buf; struct cm_id_private *cm_id_priv; - enum ib_cm_state state = - (enum ib_cm_state)(unsigned long)msg->context[1]; struct cm_port *port; u16 attr_index; @@ -3861,10 +3844,9 @@ static void cm_send_handler(struct ib_mad_agent *mad_agent, &port->counters[CM_XMIT_RETRIES][attr_index]); if (cm_id_priv) - cm_process_send_error(cm_id_priv, msg, state, - mad_send_wc->status); + cm_process_send_error(cm_id_priv, msg, mad_send_wc->status); else - cm_free_response_msg(msg); + cm_free_msg(msg); } static void cm_work_handler(struct work_struct *_work) |