[v5,for-next,0/9] RDMA/hns: Use new interfaces to write/read fields

Message

Weihang Li June 21, 2021, 8 a.m. UTC

hr_reg_*() is simpler than roce_set_*(), and the field/bit can be generated
automatically and accurately. This series first fixes two issues on
hr_reg_*(), then introduces hr_reg_write_bool() for 1 bit-width field, does
the replacement at last.

Changes since v4:
* Change the way to fix the sparse warning, we introduce
  hr_reg_write_bool() for 1 bit-width field.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/

Changes since v3:
* Use "val ? 1 : 0" instead of "!!val" for hr_reg_write to avoid sparse
  warnings.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/

Changes since v2:
* Add a patch to solve the gcc warnings about PREP_FIELD() by adding a
  check for mtu. Therefore only the parts which fix the sparse warning is
  reserved in #1.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/

Changes since v1:
* Add a patch to fix gcc warnings about PREP_FIELD().
* Fix a typo in #5.
* Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/

Lang Cheng (3):
  RDMA/hns: Add hr_reg_write_bool()
  RDMA/hns: Use new interface to modify QP context
  RDMA/hns: Use new interface to get CQE fields

Weihang Li (2):
  RDMA/hns: Fix sparse warnings when calling hr_reg_write()
  RDMA/hns: Add a check to ensure integer mtu is positive

Xi Wang (1):
  RDMA/hns: Clean SRQC structure definition

Yixing Liu (3):
  RDMA/hns: Use new interface to write CQ context.
  RDMA/hns: Use new interface to write FRMR fields
  RDMA/hns: Use new interface to write DB related fields

 drivers/infiniband/hw/hns/hns_roce_common.h |    8 +
 drivers/infiniband/hw/hns/hns_roce_hw_v2.c  | 1124 ++++++++++-----------------
 drivers/infiniband/hw/hns/hns_roce_hw_v2.h  |  862 +++++++-------------
 3 files changed, 679 insertions(+), 1315 deletions(-)

Comments

Jason Gunthorpe June 21, 2021, 6:37 p.m. UTC | #1

On Mon, Jun 21, 2021 at 04:00:34PM +0800, Weihang Li wrote:
> hr_reg_*() is simpler than roce_set_*(), and the field/bit can be generated
> automatically and accurately. This series first fixes two issues on
> hr_reg_*(), then introduces hr_reg_write_bool() for 1 bit-width field, does
> the replacement at last.
> 
> Changes since v4:
> * Change the way to fix the sparse warning, we introduce
>   hr_reg_write_bool() for 1 bit-width field.
> * Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/
> 
> Changes since v3:
> * Use "val ? 1 : 0" instead of "!!val" for hr_reg_write to avoid sparse
>   warnings.
> * Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/
> 
> Changes since v2:
> * Add a patch to solve the gcc warnings about PREP_FIELD() by adding a
>   check for mtu. Therefore only the parts which fix the sparse warning is
>   reserved in #1.
> * Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/
> 
> Changes since v1:
> * Add a patch to fix gcc warnings about PREP_FIELD().
> * Fix a typo in #5.
> * Link: https://patchwork.kernel.org/project/linux-rdma/cover/[e-mail]/
> 
> Lang Cheng (3):
>   RDMA/hns: Add hr_reg_write_bool()
>   RDMA/hns: Use new interface to modify QP context
>   RDMA/hns: Use new interface to get CQE fields
> 
> Weihang Li (2):
>   RDMA/hns: Fix sparse warnings when calling hr_reg_write()
>   RDMA/hns: Add a check to ensure integer mtu is positive
> 
> Xi Wang (1):
>   RDMA/hns: Clean SRQC structure definition
> 
> Yixing Liu (3):
>   RDMA/hns: Use new interface to write CQ context.
>   RDMA/hns: Use new interface to write FRMR fields
>   RDMA/hns: Use new interface to write DB related fields

Applied to for-next, thanks

Jason