-----"Jason Gunthorpe" <jgg@xxxxxxxx> wrote: -----
>To: "Bernard Metzler" <bmt@xxxxxxxxxxxxxx>
>From: "Jason Gunthorpe" <jgg@xxxxxxxx>
>Date: 12/09/2019 05:07PM
>Cc: linux-rdma@xxxxxxxxxxxxxxx, krishna2@xxxxxxxxxxx, leon@xxxxxxxxxx
>Subject: [EXTERNAL] Re: [PATCH for-next] RDMA/siw: Simplify QP
>representation.
>
>On Fri, Nov 29, 2019 at 05:25:09PM +0100, Bernard Metzler wrote:
>> Change siw_qp to contain ib_qp. Use ib_qp's uobject pointer
>> to distinguish kernel level and user level applications.
>> Apply same mechanism for kerne/user level application
>> detection to shared receive queues and completion queues.
>
>Drivers should not touch the uobject. If I recall you can use
>restrack
>to tell if it is kernel or user created
>
'bool res->user' would probably be it, but I stumbled
upon this comment (e.g. in struct ib_qp):
/*
* Implementation details of the RDMA core, don't use in drivers:
*/
struct rdma_restrack_entry res;
So we shall not use restrack information in drivers..?
Shall restrack better export a query such as
'rdma_restrack_is_user(resource)'?
After a quick investigation, current drivers do have
their own solution for the issue:
mlx5, mlx4, mthca, hns, cxgb4, qedr:
tests ib_xx->uobject as I proposed here for siw as well.
bnxt_re, qedr, hfi, i40iw, vmw_pvrdma:
use their own local resource flag ('is_user', 'is_kernel',
whatever), as siw does it until now, and what is not
preferred as well. How shall we proceed?
Thanks,
Bernard.
