| Message ID | 1617216194-12890-2-git-send-email-haakon.bugge@oracle.com (mailing list archive) |
|---|---|
| State | Awaiting Upstream |
| Headers | show |
| Series | Introduce rdma_set_min_rnr_timer() and use it in RDS | expand |
| Context | Check | Description |
|---|---|---|
| netdev/cover_letter | success | Link |
| netdev/fixes_present | success | Link |
| netdev/patch_count | success | Link |
| netdev/tree_selection | success | Guessed tree name to be net-next |
| netdev/subject_prefix | success | Link |
| netdev/cc_maintainers | warning | 5 maintainers not CCed: jgg@ziepe.ca maorg@mellanox.com avihaih@nvidia.com markz@mellanox.com leon@kernel.org |
| netdev/source_inline | success | Was 0 now: 0 |
| netdev/verify_signedoff | success | Link |
| netdev/module_param | success | Was 0 now: 0 |
| netdev/build_32bit | success | Errors and warnings before: 180 this patch: 180 |
| netdev/kdoc | success | Errors and warnings before: 20 this patch: 20 |
| netdev/verify_fixes | success | Link |
| netdev/checkpatch | success | total: 0 errors, 0 warnings, 0 checks, 78 lines checked |
| netdev/build_allmodconfig_warn | success | Errors and warnings before: 309 this patch: 309 |
| netdev/header_inline | success | Link |
On Wed, Mar 31, 2021 at 08:43:13PM +0200, Håkon Bugge wrote: > Introduce the ability for kernel ULPs to adjust the minimum RNR Retry > timer. The INIT -> RTR transition executed by RDMA CM will be used for > this adjustment. This avoids an additional ib_modify_qp() call. > > rdma_set_min_rnr_timer() must be called before the call to > rdma_connect() on the active side and before the call to rdma_accept() > on the passive side. > > The default value of RNR Retry timer is zero, which translates to 655 > ms. When the receiver is not ready to accept a send messages, it > encodes the RNR Retry timer value in the NAK. The requestor will then > wait at least the specified time value before retrying the send. > > The 5-bit value to be supplied to the rdma_set_min_rnr_timer() is > documented in IBTA Table 45: "Encoding for RNR NAK Timer Field". > > Signed-off-by: Håkon Bugge <haakon.bugge@oracle.com> > Acked-by: Jason Gunthorpe <jgg@nvidia.com> > --- > drivers/infiniband/core/cma.c | 41 ++++++++++++++++++++++++++++++++++++++ > drivers/infiniband/core/cma_priv.h | 2 ++ > include/rdma/rdma_cm.h | 2 ++ > 3 files changed, 45 insertions(+) > > diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c > index 9409651..5ce097d 100644 > --- a/drivers/infiniband/core/cma.c > +++ b/drivers/infiniband/core/cma.c > @@ -852,6 +852,7 @@ static void cma_id_put(struct rdma_id_private *id_priv) > id_priv->id.qp_type = qp_type; > id_priv->tos_set = false; > id_priv->timeout_set = false; > + id_priv->min_rnr_timer_set = false; > id_priv->gid_type = IB_GID_TYPE_IB; > spin_lock_init(&id_priv->lock); > mutex_init(&id_priv->qp_mutex); > @@ -1141,6 +1142,9 @@ int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr, > if ((*qp_attr_mask & IB_QP_TIMEOUT) && id_priv->timeout_set) > qp_attr->timeout = id_priv->timeout; > > + if ((*qp_attr_mask & IB_QP_MIN_RNR_TIMER) && id_priv->min_rnr_timer_set) > + qp_attr->min_rnr_timer = id_priv->min_rnr_timer; > + > return ret; > } > EXPORT_SYMBOL(rdma_init_qp_attr); > @@ -2615,6 +2619,43 @@ int rdma_set_ack_timeout(struct rdma_cm_id *id, u8 timeout) > } > EXPORT_SYMBOL(rdma_set_ack_timeout); > > +/** > + * rdma_set_min_rnr_timer() - Set the minimum RNR Retry timer of the > + * QP associated with a connection identifier. > + * @id: Communication identifier to associated with service type. > + * @min_rnr_timer: 5-bit value encoded as Table 45: "Encoding for RNR NAK > + * Timer Field" in the IBTA specification. > + * > + * This function should be called before rdma_connect() on active > + * side, and on passive side before rdma_accept(). The timer value > + * will be associated with the local QP. When it receives a send it is > + * not read to handle, typically if the receive queue is empty, an RNR > + * Retry NAK is returned to the requester with the min_rnr_timer > + * encoded. The requester will then wait at least the time specified > + * in the NAK before retrying. The default is zero, which translates > + * to a minimum RNR Timer value of 655 ms. > + * > + * Return: 0 for success > + */ > +int rdma_set_min_rnr_timer(struct rdma_cm_id *id, u8 min_rnr_timer) > +{ > + struct rdma_id_private *id_priv; > + > + /* It is a five-bit value */ > + if (min_rnr_timer & 0xe0) > + return -EINVAL; > + > + if (id->qp_type != IB_QPT_RC && id->qp_type != IB_QPT_XRC_TGT) > + return -EINVAL; This is in-kernel API and safe to use WARN_ON() instead of returning error which RDS is not checking anyway. Thanks
diff --git a/drivers/infiniband/core/cma.c b/drivers/infiniband/core/cma.c index 9409651..5ce097d 100644 --- a/drivers/infiniband/core/cma.c +++ b/drivers/infiniband/core/cma.c @@ -852,6 +852,7 @@ static void cma_id_put(struct rdma_id_private *id_priv) id_priv->id.qp_type = qp_type; id_priv->tos_set = false; id_priv->timeout_set = false; + id_priv->min_rnr_timer_set = false; id_priv->gid_type = IB_GID_TYPE_IB; spin_lock_init(&id_priv->lock); mutex_init(&id_priv->qp_mutex); @@ -1141,6 +1142,9 @@ int rdma_init_qp_attr(struct rdma_cm_id *id, struct ib_qp_attr *qp_attr, if ((*qp_attr_mask & IB_QP_TIMEOUT) && id_priv->timeout_set) qp_attr->timeout = id_priv->timeout; + if ((*qp_attr_mask & IB_QP_MIN_RNR_TIMER) && id_priv->min_rnr_timer_set) + qp_attr->min_rnr_timer = id_priv->min_rnr_timer; + return ret; } EXPORT_SYMBOL(rdma_init_qp_attr); @@ -2615,6 +2619,43 @@ int rdma_set_ack_timeout(struct rdma_cm_id *id, u8 timeout) } EXPORT_SYMBOL(rdma_set_ack_timeout); +/** + * rdma_set_min_rnr_timer() - Set the minimum RNR Retry timer of the + * QP associated with a connection identifier. + * @id: Communication identifier to associated with service type. + * @min_rnr_timer: 5-bit value encoded as Table 45: "Encoding for RNR NAK + * Timer Field" in the IBTA specification. + * + * This function should be called before rdma_connect() on active + * side, and on passive side before rdma_accept(). The timer value + * will be associated with the local QP. When it receives a send it is + * not read to handle, typically if the receive queue is empty, an RNR + * Retry NAK is returned to the requester with the min_rnr_timer + * encoded. The requester will then wait at least the time specified + * in the NAK before retrying. The default is zero, which translates + * to a minimum RNR Timer value of 655 ms. + * + * Return: 0 for success + */ +int rdma_set_min_rnr_timer(struct rdma_cm_id *id, u8 min_rnr_timer) +{ + struct rdma_id_private *id_priv; + + /* It is a five-bit value */ + if (min_rnr_timer & 0xe0) + return -EINVAL; + + if (id->qp_type != IB_QPT_RC && id->qp_type != IB_QPT_XRC_TGT) + return -EINVAL; + + id_priv = container_of(id, struct rdma_id_private, id); + id_priv->min_rnr_timer = min_rnr_timer; + id_priv->min_rnr_timer_set = true; + + return 0; +} +EXPORT_SYMBOL(rdma_set_min_rnr_timer); + static void cma_query_handler(int status, struct sa_path_rec *path_rec, void *context) { diff --git a/drivers/infiniband/core/cma_priv.h b/drivers/infiniband/core/cma_priv.h index caece96..bf83d32 100644 --- a/drivers/infiniband/core/cma_priv.h +++ b/drivers/infiniband/core/cma_priv.h @@ -86,9 +86,11 @@ struct rdma_id_private { u8 tos; u8 tos_set:1; u8 timeout_set:1; + u8 min_rnr_timer_set:1; u8 reuseaddr; u8 afonly; u8 timeout; + u8 min_rnr_timer; enum ib_gid_type gid_type; /* diff --git a/include/rdma/rdma_cm.h b/include/rdma/rdma_cm.h index 32a67af..8b0f66e 100644 --- a/include/rdma/rdma_cm.h +++ b/include/rdma/rdma_cm.h @@ -331,6 +331,8 @@ int rdma_join_multicast(struct rdma_cm_id *id, struct sockaddr *addr, int rdma_set_afonly(struct rdma_cm_id *id, int afonly); int rdma_set_ack_timeout(struct rdma_cm_id *id, u8 timeout); + +int rdma_set_min_rnr_timer(struct rdma_cm_id *id, u8 min_rnr_timer); /** * rdma_get_service_id - Return the IB service ID for a specified address. * @id: Communication identifier associated with the address.