diff mbox series

[rdma-next,3/3] IB/mlx5: Introduce a new mini-CQE format

Message ID 20180527104234.17261-4-leon@kernel.org
State Not Applicable, archived
Delegated to: David Miller
Headers show
Series Introduce new mlx5 CQE format | expand

Commit Message

Leon Romanovsky May 27, 2018, 10:42 a.m. UTC 
From: Yonatan Cohen <yonatanc@mellanox.com>

The new mini-CQE format includes the stride index, byte count and
packet checksum.
Stride index is needed for striding WQ feature.
This patch exposes this capability and enables its setting
via mlx5 UHW data as part of query device and cq creation.

Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
Reviewed-by: Guy Levi <guyle@mellanox.com>
Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
---
 drivers/infiniband/hw/mlx5/cq.c   | 42 +++++++++++++++++++++++++++++----------
 drivers/infiniband/hw/mlx5/main.c |  4 ++++
 include/uapi/rdma/mlx5-abi.h      |  2 +-
 3 files changed, 37 insertions(+), 11 deletions(-)

Comments

Jason Gunthorpe May 28, 2018, 4:11 p.m. UTC | #1 
On Sun, May 27, 2018 at 01:42:34PM +0300, Leon Romanovsky wrote:
> From: Yonatan Cohen <yonatanc@mellanox.com>
> 
> The new mini-CQE format includes the stride index, byte count and
> packet checksum.
> Stride index is needed for striding WQ feature.
> This patch exposes this capability and enables its setting
> via mlx5 UHW data as part of query device and cq creation.
> 
> Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
> Reviewed-by: Guy Levi <guyle@mellanox.com>
> Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
>  drivers/infiniband/hw/mlx5/cq.c   | 42 +++++++++++++++++++++++++++++----------
>  drivers/infiniband/hw/mlx5/main.c |  4 ++++
>  include/uapi/rdma/mlx5-abi.h      |  2 +-
>  3 files changed, 37 insertions(+), 11 deletions(-)
> 
> diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
> index 7b4ce1a19de0..ad39d64b8108 100644
> +++ b/drivers/infiniband/hw/mlx5/cq.c
> @@ -751,6 +751,28 @@ static int alloc_cq_frag_buf(struct mlx5_ib_dev *dev,
>  	return 0;
>  }
>  
> +enum {
> +	MLX5_CQE_RES_FORMAT_HASH = 0,
> +	MLX5_CQE_RES_FORMAT_CSUM = 1,
> +	MLX5_CQE_RES_FORMAT_CSUM_STRIDX = 3,
> +};

What is this??

> +static int mini_cqe_res_format_to_hw(struct mlx5_ib_dev *dev, u8 format)
> +{
> +	switch (format) {
> +	case MLX5_IB_CQE_RES_FORMAT_HASH:
> +		return MLX5_CQE_RES_FORMAT_HASH;

Used here..

> +		mini_cqe_format =
> +			mini_cqe_res_format_to_hw(dev,
> +						  ucmd.cqe_comp_res_format);

And format comes from a ucmd, so that enum is upai.

Put it in the right place and put the right comment beside
struct mlx5_ib_create_cq's cqe_comp_res_format..

And what is wrong with the user space patches? Where is the update to
enum mlx5dv_cqe_comp_res_format ? And why is this wrong?

struct mlx5dv_cq_init_attr {
        uint64_t comp_mask; /* Use enum mlx5dv_cq_init_attr_mask */
        uint8_t cqe_comp_res_format; /* Use enum mlx5dv_cqe_comp_res_format */
                                                 ^^^^^^^^^^^^^^^^^^^^^^^^^^

No, it isn't, and there isn't even an enum for it. Are you sure this is
designed right? Looks pretty wrong to me.

Fix it all please, and you need to arrange things to share the uapi
header with dv just like verbs is doing.

No more of this lax attitude toward uapi!

Jason
Yishai Hadas May 28, 2018, 4:52 p.m. UTC | #2 
On 5/28/2018 7:11 PM, Jason Gunthorpe wrote:
> On Sun, May 27, 2018 at 01:42:34PM +0300, Leon Romanovsky wrote:
>> From: Yonatan Cohen <yonatanc@mellanox.com>
>>
>> The new mini-CQE format includes the stride index, byte count and
>> packet checksum.
>> Stride index is needed for striding WQ feature.
>> This patch exposes this capability and enables its setting
>> via mlx5 UHW data as part of query device and cq creation.
>>
>> Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
>> Reviewed-by: Guy Levi <guyle@mellanox.com>
>> Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
>> Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
>>   drivers/infiniband/hw/mlx5/cq.c   | 42 +++++++++++++++++++++++++++++----------
>>   drivers/infiniband/hw/mlx5/main.c |  4 ++++
>>   include/uapi/rdma/mlx5-abi.h      |  2 +-
>>   3 files changed, 37 insertions(+), 11 deletions(-)
>>
>> diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
>> index 7b4ce1a19de0..ad39d64b8108 100644
>> +++ b/drivers/infiniband/hw/mlx5/cq.c
>> @@ -751,6 +751,28 @@ static int alloc_cq_frag_buf(struct mlx5_ib_dev *dev,
>>   	return 0;
>>   }
>>   
>> +enum {
>> +	MLX5_CQE_RES_FORMAT_HASH = 0,
>> +	MLX5_CQE_RES_FORMAT_CSUM = 1,
>> +	MLX5_CQE_RES_FORMAT_CSUM_STRIDX = 3,
>> +};
> 
> What is this??

Those are mlx5 device values not uapi.

>> +static int mini_cqe_res_format_to_hw(struct mlx5_ib_dev *dev, u8 format)
>> +{
>> +	switch (format) {
>> +	case MLX5_IB_CQE_RES_FORMAT_HASH:
>> +		return MLX5_CQE_RES_FORMAT_HASH;
> 
> Used here..

This is some conversion between the uapi to the device values.

>> +		mini_cqe_format =
>> +			mini_cqe_res_format_to_hw(dev,
>> +						  ucmd.cqe_comp_res_format);
> 
> And format comes from a ucmd, so that enum is upai.

Correct, see mlx5-abi.h as part of this patch.

> Put it in the right place and put the right comment beside
> struct mlx5_ib_create_cq's cqe_comp_res_format..
> 
> And what is wrong with the user space patches? Where is the update to
> enum mlx5dv_cqe_comp_res_format ? And why is this wrong?
> 

See the first patch from below PR [1], it brings the new enum value to 
the user area as part of kernel-headers/rdma/mlx5-abi.h.

[1] https://github.com/linux-rdma/rdma-core/pull/337

> struct mlx5dv_cq_init_attr {
>          uint64_t comp_mask; /* Use enum mlx5dv_cq_init_attr_mask */
>          uint8_t cqe_comp_res_format; /* Use enum mlx5dv_cqe_comp_res_format */
>                                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^
> 

The user space uses the DV prefix (e.g. 
MLX5DV_CQE_RES_FORMAT_CSUM_STRIDX), no change from previous flags around 
enum mlx5dv_cqe_comp_res_format.
Jason Gunthorpe May 28, 2018, 5 p.m. UTC | #3 
On Mon, May 28, 2018 at 07:52:03PM +0300, Yishai Hadas wrote:
> On 5/28/2018 7:11 PM, Jason Gunthorpe wrote:
> >On Sun, May 27, 2018 at 01:42:34PM +0300, Leon Romanovsky wrote:
> >>From: Yonatan Cohen <yonatanc@mellanox.com>
> >>
> >>The new mini-CQE format includes the stride index, byte count and
> >>packet checksum.
> >>Stride index is needed for striding WQ feature.
> >>This patch exposes this capability and enables its setting
> >>via mlx5 UHW data as part of query device and cq creation.
> >>
> >>Reviewed-by: Yishai Hadas <yishaih@mellanox.com>
> >>Reviewed-by: Guy Levi <guyle@mellanox.com>
> >>Signed-off-by: Yonatan Cohen <yonatanc@mellanox.com>
> >>Signed-off-by: Leon Romanovsky <leonro@mellanox.com>
> >>  drivers/infiniband/hw/mlx5/cq.c   | 42 +++++++++++++++++++++++++++++----------
> >>  drivers/infiniband/hw/mlx5/main.c |  4 ++++
> >>  include/uapi/rdma/mlx5-abi.h      |  2 +-
> >>  3 files changed, 37 insertions(+), 11 deletions(-)
> >>
> >>diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
> >>index 7b4ce1a19de0..ad39d64b8108 100644
> >>+++ b/drivers/infiniband/hw/mlx5/cq.c
> >>@@ -751,6 +751,28 @@ static int alloc_cq_frag_buf(struct mlx5_ib_dev *dev,
> >>  	return 0;
> >>  }
> >>+enum {
> >>+	MLX5_CQE_RES_FORMAT_HASH = 0,
> >>+	MLX5_CQE_RES_FORMAT_CSUM = 1,
> >>+	MLX5_CQE_RES_FORMAT_CSUM_STRIDX = 3,
> >>+};
> >
> >What is this??
> 
> Those are mlx5 device values not uapi.
> 
> >>+static int mini_cqe_res_format_to_hw(struct mlx5_ib_dev *dev, u8 format)
> >>+{
> >>+	switch (format) {
> >>+	case MLX5_IB_CQE_RES_FORMAT_HASH:
> >>+		return MLX5_CQE_RES_FORMAT_HASH;
> >
> >Used here..
> 
> This is some conversion between the uapi to the device values.
> 
> >>+		mini_cqe_format =
> >>+			mini_cqe_res_format_to_hw(dev,
> >>+						  ucmd.cqe_comp_res_format);
> >
> >And format comes from a ucmd, so that enum is upai.
> 
> Correct, see mlx5-abi.h as part of this patch.
> 
> >Put it in the right place and put the right comment beside
> >struct mlx5_ib_create_cq's cqe_comp_res_format..
> >
> >And what is wrong with the user space patches? Where is the update to
> >enum mlx5dv_cqe_comp_res_format ? And why is this wrong?
> >
> 
> See the first patch from below PR [1], it brings the new enum value to the
> user area as part of kernel-headers/rdma/mlx5-abi.h.
> 
> [1] https://github.com/linux-rdma/rdma-core/pull/337
> 
> >struct mlx5dv_cq_init_attr {
> >         uint64_t comp_mask; /* Use enum mlx5dv_cq_init_attr_mask */
> >         uint8_t cqe_comp_res_format; /* Use enum mlx5dv_cqe_comp_res_format */
> >                                                  ^^^^^^^^^^^^^^^^^^^^^^^^^^
> >
> 
> The user space uses the DV prefix (e.g. MLX5DV_CQE_RES_FORMAT_CSUM_STRIDX),
> no change from previous flags around enum mlx5dv_cqe_comp_res_format.

This still needs eventual cleaning up in the verbs_abi.h way.

But OK, this is not what I thought.

Jason
diff mbox series

Patch

diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
index 7b4ce1a19de0..ad39d64b8108 100644
--- a/drivers/infiniband/hw/mlx5/cq.c
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -751,6 +751,28 @@  static int alloc_cq_frag_buf(struct mlx5_ib_dev *dev,
 	return 0;
 }
 
+enum {
+	MLX5_CQE_RES_FORMAT_HASH = 0,
+	MLX5_CQE_RES_FORMAT_CSUM = 1,
+	MLX5_CQE_RES_FORMAT_CSUM_STRIDX = 3,
+};
+
+static int mini_cqe_res_format_to_hw(struct mlx5_ib_dev *dev, u8 format)
+{
+	switch (format) {
+	case MLX5_IB_CQE_RES_FORMAT_HASH:
+		return MLX5_CQE_RES_FORMAT_HASH;
+	case MLX5_IB_CQE_RES_FORMAT_CSUM:
+		return MLX5_CQE_RES_FORMAT_CSUM;
+	case MLX5_IB_CQE_RES_FORMAT_CSUM_STRIDX:
+		if (MLX5_CAP_GEN(dev->mdev, mini_cqe_resp_stride_index))
+			return MLX5_CQE_RES_FORMAT_CSUM_STRIDX;
+		return -EOPNOTSUPP;
+	default:
+		return -EINVAL;
+	}
+}
+
 static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 			  struct ib_ucontext *context, struct mlx5_ib_cq *cq,
 			  int entries, u32 **cqb,
@@ -816,6 +838,8 @@  static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 	*index = to_mucontext(context)->bfregi.sys_pages[0];
 
 	if (ucmd.cqe_comp_en == 1) {
+		int mini_cqe_format;
+
 		if (!((*cqe_size == 128 &&
 		       MLX5_CAP_GEN(dev->mdev, cqe_compression_128)) ||
 		      (*cqe_size == 64  &&
@@ -826,20 +850,18 @@  static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
 			goto err_cqb;
 		}
 
-		if (unlikely(!ucmd.cqe_comp_res_format ||
-			     !(ucmd.cqe_comp_res_format <
-			       MLX5_IB_CQE_RES_RESERVED) ||
-			     (ucmd.cqe_comp_res_format &
-			      (ucmd.cqe_comp_res_format - 1)))) {
-			err = -EOPNOTSUPP;
-			mlx5_ib_warn(dev, "CQE compression res format %d is not supported!\n",
-				     ucmd.cqe_comp_res_format);
+		mini_cqe_format =
+			mini_cqe_res_format_to_hw(dev,
+						  ucmd.cqe_comp_res_format);
+		if (mini_cqe_format < 0) {
+			err = mini_cqe_format;
+			mlx5_ib_dbg(dev, "CQE compression res format %d error: %d\n",
+				    ucmd.cqe_comp_res_format, err);
 			goto err_cqb;
 		}
 
 		MLX5_SET(cqc, cqc, cqe_comp_en, 1);
-		MLX5_SET(cqc, cqc, mini_cqe_res_format,
-			 ilog2(ucmd.cqe_comp_res_format));
+		MLX5_SET(cqc, cqc, mini_cqe_res_format, mini_cqe_format);
 	}
 
 	if (ucmd.flags & MLX5_IB_CREATE_CQ_FLAGS_CQE_128B_PAD) {
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
index 95e67a85078c..238f1eed714c 100644
--- a/drivers/infiniband/hw/mlx5/main.c
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -992,6 +992,10 @@  static int mlx5_ib_query_device(struct ib_device *ibdev,
 			resp.cqe_comp_caps.supported_format =
 				MLX5_IB_CQE_RES_FORMAT_HASH |
 				MLX5_IB_CQE_RES_FORMAT_CSUM;
+
+			if (MLX5_CAP_GEN(dev->mdev, mini_cqe_resp_stride_index))
+				resp.cqe_comp_caps.supported_format |=
+					MLX5_IB_CQE_RES_FORMAT_CSUM_STRIDX;
 		}
 	}
 
diff --git a/include/uapi/rdma/mlx5-abi.h b/include/uapi/rdma/mlx5-abi.h
index beec971effef..a03b68b3e26c 100644
--- a/include/uapi/rdma/mlx5-abi.h
+++ b/include/uapi/rdma/mlx5-abi.h
@@ -166,7 +166,7 @@  struct mlx5_ib_rss_caps {
 enum mlx5_ib_cqe_comp_res_format {
 	MLX5_IB_CQE_RES_FORMAT_HASH	= 1 << 0,
 	MLX5_IB_CQE_RES_FORMAT_CSUM	= 1 << 1,
-	MLX5_IB_CQE_RES_RESERVED	= 1 << 2,
+	MLX5_IB_CQE_RES_FORMAT_CSUM_STRIDX = 1 << 2,
 };
 
 struct mlx5_ib_cqe_comp_caps {