Linux-RDMA Archive on lore.kernel.org
 help / color / Atom feed
From: Leon Romanovsky <leon@kernel.org>
To: Doug Ledford <dledford@redhat.com>, Jason Gunthorpe <jgg@nvidia.com>
Cc: linux-rdma@vger.kernel.org,
	"majd@mellanox.com" <majd@mellanox.com>,
	Matan Barak <matanb@mellanox.com>,
	Sagi Grimberg <sagi@grimberg.me>,
	Yishai Hadas <yishaih@mellanox.com>
Subject: [PATCH rdma-next 3/6] RDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for WQ
Date: Mon, 26 Oct 2020 15:26:32 +0200
Message-ID: <20201026132635.1337663-4-leon@kernel.org> (raw)
In-Reply-To: <20201026132635.1337663-1-leon@kernel.org>

From: Jason Gunthorpe <jgg@nvidia.com>

This fixes a subtle bug, the WQ mailbox has only 5 bits to describe the
page_offset, while mlx5_ib_get_buf_offset() is hard wired to only work
with 6 bit page_offsets.

Thus it did not properly reject badly aligned buffers.

YISHAI: WTF? Why does this PRM command only have 5 bits? We must force 4k
alignment for WQ umems in the userspace?

Fixes: 79b20a6c3014 ("IB/mlx5: Add receive Work Queue verbs")
Fixes: 0fb2ed66a14c ("IB/mlx5: Add create and destroy functionality for Raw Packet QP")
Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
---
 drivers/infiniband/hw/mlx5/qp.c | 53 +++++++++++++++++++--------------
 1 file changed, 31 insertions(+), 22 deletions(-)

diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
index 042177f33252..143fd5eae304 100644
--- a/drivers/infiniband/hw/mlx5/qp.c
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -833,7 +833,7 @@ static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 {
 	struct mlx5_ib_ucontext *ucontext = rdma_udata_to_drv_context(
 		udata, struct mlx5_ib_ucontext, ibucontext);
-	int page_shift = 0;
+	unsigned long page_size = 0;
 	u32 offset = 0;
 	int err;
 
@@ -847,24 +847,25 @@ static int create_user_rq(struct mlx5_ib_dev *dev, struct ib_pd *pd,
 		return err;
 	}
 
-	mlx5_ib_cont_pages(rwq->umem, ucmd->buf_addr, 0, &page_shift);
-	err = mlx5_ib_get_buf_offset(ucmd->buf_addr, page_shift,
-				     &rwq->rq_page_offset);
-	if (err) {
+	page_size = mlx5_umem_find_best_quantized_pgoff(
+		rwq->umem, wq, log_wq_pg_sz, MLX5_ADAPTER_PAGE_SHIFT,
+		page_offset, 64, &rwq->rq_page_offset);
+	if (!page_size) {
 		mlx5_ib_warn(dev, "bad offset\n");
+		err = -EINVAL;
 		goto err_umem;
 	}
 
-	rwq->rq_num_pas = ib_umem_num_dma_blocks(rwq->umem, 1UL << page_shift);
-	rwq->page_shift = page_shift;
-	rwq->log_page_size =  page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	rwq->rq_num_pas = ib_umem_num_dma_blocks(rwq->umem, page_size);
+	rwq->page_shift = order_base_2(page_size);
+	rwq->log_page_size =  rwq->page_shift - MLX5_ADAPTER_PAGE_SHIFT;
 	rwq->wq_sig = !!(ucmd->flags & MLX5_WQ_FLAG_SIGNATURE);
 
 	mlx5_ib_dbg(
 		dev,
-		"addr 0x%llx, size %zd, npages %zu, page_shift %d, ncont %d, offset %d\n",
+		"addr 0x%llx, size %zd, npages %zu, page_size %ld, ncont %d, offset %d\n",
 		(unsigned long long)ucmd->buf_addr, rwq->buf_size,
-		ib_umem_num_pages(rwq->umem), page_shift, rwq->rq_num_pas,
+		ib_umem_num_pages(rwq->umem), page_size, rwq->rq_num_pas,
 		offset);
 
 	err = mlx5_ib_db_map_user(ucontext, udata, ucmd->db_addr, &rwq->db);
@@ -1209,17 +1210,24 @@ static int create_raw_packet_qp_sq(struct mlx5_ib_dev *dev,
 	void *wq;
 	int inlen;
 	int err;
-	int page_shift = 0;
-	u32 offset = 0;
-
-	err = mlx5_ib_umem_get(dev, udata, ubuffer->buf_addr, ubuffer->buf_size,
-			       &sq->ubuffer.umem, &page_shift, &offset);
-	if (err)
-		return err;
+	unsigned int page_offset_quantized;
+	unsigned long page_size;
+
+	sq->ubuffer.umem = ib_umem_get(&dev->ib_dev, ubuffer->buf_addr,
+				       ubuffer->buf_size, 0);
+	if (IS_ERR(sq->ubuffer.umem))
+		return PTR_ERR(sq->ubuffer.umem);
+	page_size = mlx5_umem_find_best_quantized_pgoff(
+		ubuffer->umem, wq, log_wq_pg_sz, MLX5_ADAPTER_PAGE_SHIFT,
+		page_offset, 64, &page_offset_quantized);
+	if (!page_size) {
+		err = -EINVAL;
+		goto err_umem;
+	}
 
 	inlen = MLX5_ST_SZ_BYTES(create_sq_in) +
-		sizeof(u64) * ib_umem_num_dma_blocks(sq->ubuffer.umem,
-						     1UL << page_shift);
+		sizeof(u64) *
+			ib_umem_num_dma_blocks(sq->ubuffer.umem, page_size);
 	in = kvzalloc(inlen, GFP_KERNEL);
 	if (!in) {
 		err = -ENOMEM;
@@ -1247,11 +1255,12 @@ static int create_raw_packet_qp_sq(struct mlx5_ib_dev *dev,
 	MLX5_SET64(wq, wq, dbr_addr, MLX5_GET64(qpc, qpc, dbr_addr));
 	MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB));
 	MLX5_SET(wq, wq, log_wq_sz, MLX5_GET(qpc, qpc, log_sq_size));
-	MLX5_SET(wq, wq, log_wq_pg_sz,  page_shift - MLX5_ADAPTER_PAGE_SHIFT);
-	MLX5_SET(wq, wq, page_offset, offset);
+	MLX5_SET(wq, wq, log_wq_pg_sz,
+		 order_base_2(page_size) - MLX5_ADAPTER_PAGE_SHIFT);
+	MLX5_SET(wq, wq, page_offset, page_offset_quantized);
 
 	pas = (__be64 *)MLX5_ADDR_OF(wq, wq, pas);
-	mlx5_ib_populate_pas(sq->ubuffer.umem, 1UL << page_shift, pas, 0);
+	mlx5_ib_populate_pas(sq->ubuffer.umem, page_size, pas, 0);
 
 	err = mlx5_core_create_sq_tracked(dev, in, inlen, &sq->base.mqp);
 
-- 
2.26.2


  parent reply index

Thread overview: 9+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-10-26 13:26 [PATCH rdma-next 0/6] Use ib_umem_find_best_pgsz() for all umems Leon Romanovsky
2020-10-26 13:26 ` [PATCH rdma-next 1/6] RDMA/mlx5: Use ib_umem_find_best_pgsz() for devx Leon Romanovsky
2020-10-26 13:26 ` [PATCH rdma-next 2/6] RDMA/mlx5: Use ib_umem_find_best_pgoff() for SRQ Leon Romanovsky
2020-10-26 13:26 ` Leon Romanovsky [this message]
2020-10-26 14:42   ` [PATCH rdma-next 3/6] RDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for WQ Gal Pressman
2020-10-27  6:03     ` Leon Romanovsky
2020-10-26 13:26 ` [PATCH rdma-next 4/6] RDMA/mlx5: Use mlx5_umem_find_best_quantized_pgoff() for QP Leon Romanovsky
2020-10-26 13:26 ` [PATCH rdma-next 5/6] RDMA/mlx5: mlx5_umem_find_best_quantized_pgoff() for CQ Leon Romanovsky
2020-10-26 13:26 ` [PATCH rdma-next 6/6] RDMA/mlx5: Lower setting the umem's PAS for SRQ 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=20201026132635.1337663-4-leon@kernel.org \
    --to=leon@kernel.org \
    --cc=dledford@redhat.com \
    --cc=jgg@nvidia.com \
    --cc=linux-rdma@vger.kernel.org \
    --cc=majd@mellanox.com \
    --cc=matanb@mellanox.com \
    --cc=sagi@grimberg.me \
    --cc=yishaih@mellanox.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

Linux-RDMA Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-rdma/0 linux-rdma/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-rdma linux-rdma/ https://lore.kernel.org/linux-rdma \
		linux-rdma@vger.kernel.org
	public-inbox-index linux-rdma

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-rdma


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git