diff options
author | Michael J. Ruhl <michael.j.ruhl@intel.com> | 2018-09-10 19:49:27 +0300 |
---|---|---|
committer | Jason Gunthorpe <jgg@mellanox.com> | 2018-09-11 18:55:02 +0300 |
commit | 0b79b27748cbec221e1ceabf63578198602bf01d (patch) | |
tree | a62ba4181d233bc314b46b3d12f62fe68e34d5b4 /drivers/infiniband/sw | |
parent | 3e5d60bcc8a42bfd0c888a0cf52a5a7e8398677d (diff) | |
download | linux-0b79b27748cbec221e1ceabf63578198602bf01d.tar.xz |
IB/{hfi1, qib, rdmavt}: Schedule multi RC/UC packets instead of posting
The post_send() path determines if it should post directly or, schedule
the post for later. The current logic is:
if the swqe ring is empty or (for hfi1) wqe->length <= piothreshold
post the send
else
schedule
This can allow large requests to call the send engine directly. Large
requests can potentially produce a large number of packets prior to
returning to the caller, blocking the caller from posting more requests,
and allowing better parallel processing.
Allow the driver(s) more say in this logic (pass call_send to the driver,
rather than examining a return value).
Update hfi1/qib logic to schedule the send engine if an RC or UC message
is larger than the QP MTU size.
Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Jason Gunthorpe <jgg@mellanox.com>
Diffstat (limited to 'drivers/infiniband/sw')
-rw-r--r-- | drivers/infiniband/sw/rdmavt/qp.c | 14 |
1 files changed, 8 insertions, 6 deletions
diff --git a/drivers/infiniband/sw/rdmavt/qp.c b/drivers/infiniband/sw/rdmavt/qp.c index 5ce403c6cddb..a9b7d7ff32ee 100644 --- a/drivers/infiniband/sw/rdmavt/qp.c +++ b/drivers/infiniband/sw/rdmavt/qp.c @@ -1718,7 +1718,7 @@ static inline int rvt_qp_is_avail( */ static int rvt_post_one_wr(struct rvt_qp *qp, const struct ib_send_wr *wr, - int *call_send) + bool *call_send) { struct rvt_swqe *wqe; u32 next; @@ -1825,11 +1825,9 @@ static int rvt_post_one_wr(struct rvt_qp *qp, /* general part of wqe valid - allow for driver checks */ if (rdi->driver_f.check_send_wqe) { - ret = rdi->driver_f.check_send_wqe(qp, wqe); + ret = rdi->driver_f.check_send_wqe(qp, wqe, call_send); if (ret < 0) goto bail_inval_free; - if (ret) - *call_send = ret; } log_pmtu = qp->log_pmtu; @@ -1897,7 +1895,7 @@ int rvt_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, struct rvt_qp *qp = ibqp_to_rvtqp(ibqp); struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device); unsigned long flags = 0; - int call_send; + bool call_send; unsigned nreq = 0; int err = 0; @@ -1930,7 +1928,11 @@ int rvt_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr, bail: spin_unlock_irqrestore(&qp->s_hlock, flags); if (nreq) { - if (call_send) + /* + * Only call do_send if there is exactly one packet, and the + * driver said it was ok. + */ + if (nreq == 1 && call_send) rdi->driver_f.do_send(qp); else rdi->driver_f.schedule_send_no_lock(qp); |