On 2020/11/27 3:24, Jason Gunthorpe wrote:
> On Thu, Nov 19, 2020 at 06:15:42AM +0000, liweihang wrote:
>> On 2020/11/19 3:11, Jason Gunthorpe wrote:
>>> On Mon, Nov 16, 2020 at 07:33:28PM +0800, Weihang Li wrote:
>>>> @@ -503,7 +581,23 @@ static inline int set_ud_wqe(struct hns_roce_qp *qp,
>>>> if (ret)
>>>> return ret;
>>>>
>>>> - set_extend_sge(qp, wr, &curr_idx, valid_num_sge);
>>>> + if (wr->send_flags & IB_SEND_INLINE) {
>>>> + ret = set_ud_inl(qp, wr, ud_sq_wqe, &curr_idx);
>>>> + if (ret)
>>>> + return ret;
>>>
>>> Why are you implementing this in the kernel? No kernel ULP sets this
>>> flag?
>>
>> Sorry, I don't understand. Some kernel users may set IB_SEND_INLINE
>> when using UD, some may not, we just check this flag to decide how
>> to fill the data into UD SQ WQE here.
>
> I mean if you 'git grep IB_SEND_INLINE' nothing uses it.
>
> This is all dead code.
>
> How did you test it?
>
> Jason
>
Hi Jason,
We have tested it with our own tools. After running 'git grep IB_SEND_INLINE',
I found following drivers refer to this flag:
bnxt_re/cxgb4/i40iw/mlx5/ocrdma/qedr/rxe/siw
So I'm still a bit confused. Do you mean that no kernel ULP uses UD inline or
the IB_SEND_INLINE flag? Should current related codes be removed?
I also found a very old discussion about removing this flag, but there was no
futher information:
https://lists.openfabrics.org/pipermail/general/2007-August/039826.html
I will appreciate it if you can give us more detailed suggestions.
Thanks
Weihang
