[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Y3e1EvHE4uk2LLLv@nvidia.com>
Date: Fri, 18 Nov 2022 12:38:42 -0400
From: Jason Gunthorpe <jgg@...dia.com>
To: Li Zhijian <lizhijian@...itsu.com>
Cc: Zhu Yanjun <zyjzyj2000@...il.com>,
Leon Romanovsky <leon@...nel.org>, linux-rdma@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [for-next PATCH 1/5] RDMA/rxe: Remove rxe_phys_buf.size
On Fri, Nov 11, 2022 at 04:30:26AM +0000, Li Zhijian wrote:
> Every rxe_phys_buf used by either IB_MR_TYPE_MEM_REG or IB_MR_TYPE_USER
> has the same size, which should be same with ibmr->page_size. So we can
> use ibmr->page_size correspondingly.
ibmr->page_size is really only supposed to be used by MRs that are
going to be used with FMR. It is some protocol to pass information to
the IB_WR_REG_MR op
The whole way rxe stores the MRs really could stand to be modernized,
just replace the whole mr->map and everything under it with a simple
xarray of page pointers.
You are right though, the page_size of the logical page array should
be global to the mr, not stored in every entry. It isn't a sgl, it is
a dynamic array. Just don't abuse ibmr->page_size for it..
Jason
Powered by blists - more mailing lists