On Thu, Apr 05, 2018 at 01:15:09PM -0600, Jason Gunthorpe wrote: > On Thu, Apr 05, 2018 at 12:02:04PM -0600, Jason Gunthorpe wrote: > > On Thu, Apr 05, 2018 at 06:53:21PM +0300, Leon Romanovsky wrote: > > > From: Leon Romanovsky <leonro@xxxxxxxxxxxx> > > > > > > Changelog; > > > v1->v2: > > > * Slightly changed allocation algorithm > > > * Removed uverbs_dealloc_dm() > > > * Fixed error-handling paths > > > * Used __aligned_u64 > > > v0->v1: The main change that Ariel moved almost all alloc/dealloc code > > > from netdev part of mlx5 driver to ib part. > > > > > Hi > > > > > > The feature itself is implemented on top of new ioctl() infrastructure, > > > the patches itself are stacked on previously sent IPsec series. In > > > addition to being tied to IPsec, this series will have extremely minor > > > conflict with Dave's netdev tree. > > > > > > +++ b/include/linux/mlx5/device.h > > > @@@ -1016,7 -1018,7 +1021,8 @@@ enum mlx5_cap_type > > > MLX5_CAP_RESERVED, > > > MLX5_CAP_VECTOR_CALC, > > > MLX5_CAP_QOS, > > > + MLX5_CAP_DEBUG, > > > + MLX5_CAP_RESERVED_14, > > > + MLX5_CAP_DEV_MEM, > > > /* NUM OF CAP Types */ > > > MLX5_CAP_NUM > > > }; > > > > This looks like it is refering to firmware structures so order matters > > here? If there are firmware bits they should have explicit numbers, IMHO. > > > > So after merge resolution we must have > > > > MLX5_CAP_DEV_MEM = 15 > > MLX5_CAP_DEBUG = 13 > > > > ? > > I changed this series to include MLX5_CAP_DEBUG as the 13th element > replacing what was called here MLx5_CAP_RESERVED_13. This makes the > conflict resolution and history more intuitive This is FW structures, it is the op_mod field from "Table 141 - QUERY_HCA_CAP Input Structure Field Descriptions" from IB supplementary part of Eth PRM [1]. Thanks [1] http://www.mellanox.com/related-docs/user_manuals/Ethernet_Adapters_Programming_Manual.pdf > > Jason
Attachment:
signature.asc
Description: PGP signature
