All of lore.kernel.org
 help / color / mirror / Atom feed
From: Leon Romanovsky <leon@kernel.org>
To: Doug Ledford <dledford@redhat.com>, Jason Gunthorpe <jgg@nvidia.com>
Cc: Yishai Hadas <yishaih@nvidia.com>,
	linux-rdma@vger.kernel.org, Christoph Hellwig <hch@infradead.org>
Subject: [PATCH rdma-next v2 3/4] RDMA/mlx5: Extend advice MR to support non faulting mode
Date: Tue, 22 Sep 2020 11:21:03 +0300	[thread overview]
Message-ID: <20200922082104.2148873-4-leon@kernel.org> (raw)
In-Reply-To: <20200922082104.2148873-1-leon@kernel.org>

From: Yishai Hadas <yishaih@nvidia.com>

Extend advice MR to support non faulting mode, this improves performance
by eliminating page faults and bring only the existing CPU pages.

Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/mr.c         | 3 ++-
 drivers/infiniband/hw/mlx5/odp.c        | 7 ++++++-
 include/uapi/rdma/ib_user_ioctl_verbs.h | 1 +
 3 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
index 6b0d6109afc6..dea65e511a3e 100644
--- a/drivers/infiniband/hw/mlx5/mr.c
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -1321,7 +1321,8 @@ int mlx5_ib_advise_mr(struct ib_pd *pd,
 		      struct uverbs_attr_bundle *attrs)
 {
 	if (advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH &&
-	    advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE)
+	    advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE &&
+	    advice != IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT)
 		return -EOPNOTSUPP;

 	return mlx5_ib_advise_mr_prefetch(pd, advice, flags,
diff --git a/drivers/infiniband/hw/mlx5/odp.c b/drivers/infiniband/hw/mlx5/odp.c
index 5bd5e19d76a2..28b7227d31bf 100644
--- a/drivers/infiniband/hw/mlx5/odp.c
+++ b/drivers/infiniband/hw/mlx5/odp.c
@@ -665,6 +665,7 @@ void mlx5_ib_fence_odp_mr(struct mlx5_ib_mr *mr)
 }

 #define MLX5_PF_FLAGS_DOWNGRADE BIT(1)
+#define MLX5_PF_FLAGS_SNAPSHOT BIT(2)
 static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,
 			     u64 user_va, size_t bcnt, u32 *bytes_mapped,
 			     u32 flags)
@@ -673,6 +674,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,
 	bool downgrade = flags & MLX5_PF_FLAGS_DOWNGRADE;
 	u64 access_mask;
 	u64 start_idx;
+	bool fault = !(flags & MLX5_PF_FLAGS_SNAPSHOT);

 	page_shift = odp->page_shift;
 	start_idx = (user_va - ib_umem_start(odp)) >> page_shift;
@@ -681,7 +683,7 @@ static int pagefault_real_mr(struct mlx5_ib_mr *mr, struct ib_umem_odp *odp,
 	if (odp->umem.writable && !downgrade)
 		access_mask |= ODP_WRITE_ALLOWED_BIT;

-	np = ib_umem_odp_map_dma_and_lock(odp, user_va, bcnt, access_mask, true);
+	np = ib_umem_odp_map_dma_and_lock(odp, user_va, bcnt, access_mask, fault);
 	if (np < 0)
 		return np;

@@ -1851,6 +1853,9 @@ int mlx5_ib_advise_mr_prefetch(struct ib_pd *pd,
 	if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH)
 		pf_flags |= MLX5_PF_FLAGS_DOWNGRADE;

+	if (advice == IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT)
+		pf_flags |= MLX5_PF_FLAGS_SNAPSHOT;
+
 	if (flags & IB_UVERBS_ADVISE_MR_FLAG_FLUSH)
 		return mlx5_ib_prefetch_sg_list(pd, advice, pf_flags, sg_list,
 						num_sge);
diff --git a/include/uapi/rdma/ib_user_ioctl_verbs.h b/include/uapi/rdma/ib_user_ioctl_verbs.h
index cfea82acfe57..22483799cd07 100644
--- a/include/uapi/rdma/ib_user_ioctl_verbs.h
+++ b/include/uapi/rdma/ib_user_ioctl_verbs.h
@@ -208,6 +208,7 @@ enum ib_uverbs_read_counters_flags {
 enum ib_uverbs_advise_mr_advice {
 	IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH,
 	IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_WRITE,
+	IB_UVERBS_ADVISE_MR_ADVICE_PREFETCH_NO_FAULT,
 };

 enum ib_uverbs_advise_mr_flag {
--
2.26.2


  parent reply	other threads:[~2020-09-22  8:21 UTC|newest]

Thread overview: 19+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-09-22  8:21 [PATCH rdma-next v2 0/4] Improve ODP by using HMM API Leon Romanovsky
2020-09-22  8:21 ` [PATCH rdma-next v2 1/4] IB/core: Improve ODP to use hmm_range_fault() Leon Romanovsky
2020-09-29 17:59   ` Jason Gunthorpe
2020-09-29 18:02     ` Christoph Hellwig
2020-09-29 18:13       ` Jason Gunthorpe
2020-09-29 18:15         ` Christoph Hellwig
2020-09-29 18:27           ` Jason Gunthorpe
2020-09-29 20:20     ` Yishai Hadas
2020-09-29 19:27   ` Jason Gunthorpe
2020-09-29 20:09     ` Yishai Hadas
2020-09-29 20:13       ` Jason Gunthorpe
2020-09-29 20:30         ` Yishai Hadas
2020-09-30  0:37           ` Jason Gunthorpe
2020-09-29 21:34         ` Yishai Hadas
2020-09-30  0:35           ` Jason Gunthorpe
2020-09-30  7:32             ` Yishai Hadas
2020-09-22  8:21 ` [PATCH rdma-next v2 2/4] IB/core: Enable ODP sync without faulting Leon Romanovsky
2020-09-22  8:21 ` Leon Romanovsky [this message]
2020-09-22  8:21 ` [PATCH rdma-next v2 4/4] RDMA/mlx5: Sync device with CPU pages upon ODP MR registration Leon Romanovsky

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200922082104.2148873-4-leon@kernel.org \
    --to=leon@kernel.org \
    --cc=dledford@redhat.com \
    --cc=hch@infradead.org \
    --cc=jgg@nvidia.com \
    --cc=linux-rdma@vger.kernel.org \
    --cc=yishaih@nvidia.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body. 
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.