[PATCH v4 2/6] IB/mlx4: Eliminate duplicate barriers on weakly-ordered archs
From: jgg@ziepe.ca (Jason Gunthorpe)
Date: 2018-03-20 14:49:04
Also in:
linux-arm-msm, linux-rdma, lkml
On Mon, Mar 19, 2018 at 10:47:44PM -0400, Sinan Kaya wrote:
Code includes wmb() followed by writel(). writel() already has a barrier on some architectures like arm64. This ends up CPU observing two barriers back to back before executing the register write. Since code already has an explicit barrier call, changing writel() to writel_relaxed(). Signed-off-by: Sinan Kaya <redacted> --- drivers/infiniband/hw/mlx4/qp.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
Reviewed-by: Jason Gunthorpe <redacted>
quoted hunk ↗ jump to hunk
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c index f045491..74b27b0 100644 --- a/drivers/infiniband/hw/mlx4/qp.c +++ b/drivers/infiniband/hw/mlx4/qp.c@@ -3880,8 +3880,8 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, */ wmb(); - writel(qp->doorbell_qpn, - to_mdev(ibqp->device)->uar_map + MLX4_SEND_DOORBELL); + writel_relaxed(qp->doorbell_qpn, + to_mdev(ibqp->device)->uar_map + MLX4_SEND_DOORBELL); /* * Make sure doorbells don't leak out of SQ spinlock-- 2.7.4