在 2019/1/22 11:57, Jason Gunthorpe 写道:
> On Tue, Jan 22, 2019 at 11:36:02AM +0800, oulijun wrote:
>> 在 2019/1/22 10:23, Jason Gunthorpe 写道:
>>> On Tue, Jan 22, 2019 at 10:23:34AM +0800, Lijun Ou wrote:
>>>> The hns_roce_ib_create_srq_resp is used to interact with the
>>>> user for data. As a result, it is added to apply with matching
>>>> kernel headers.
>>>>
>>>> Signed-off-by: Lijun Ou <oulijun@xxxxxxxxxx>
>>>> include/uapi/rdma/hns-abi.h | 5 +++++
>>>> 1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/include/uapi/rdma/hns-abi.h b/include/uapi/rdma/hns-abi.h
>>>> index ef3c7ec..eb76b38 100644
>>>> +++ b/include/uapi/rdma/hns-abi.h
>>>> @@ -52,6 +52,11 @@ struct hns_roce_ib_create_srq {
>>>> __aligned_u64 que_addr;
>>>> };
>>>>
>>>> +struct hns_roce_ib_create_srq_resp {
>>>> + __u32 srqn;
>>>> + __u32 reserved;
>>>> +};
>>>> +
>>> This makes no sense, how can uapi headers be added without adding a
>>> corresponding kernel code!?!?! What fills in srqn?
>>>
>>> Jason
>> The srqn will be obtained from hns_roce_bitmap_alloc in the kernel mode. When create srq, it need to
>> transfer srqn to user by ib_copy_to_udata function.
> The srqn is part of the ib_uverbs_create_srq_resp, it has nothing to
> do with hns_roce_ib_create_srq_resp.
>
>> the ofed will report srqn by ib_uverbs_create_srq_resp,
>> nextly, the user will use it by resp.srqn when run hns_roce_u_create_srq function.
>> cmdof hns_roce_create_srq need be declared by DECLARE_DRV_CMD.
> What?
>
>> for example,
>> DECLARE_DRV_CMD(hns_roce_create_srq,
>> IB_USER_VERBS_CMD_CREATE_SRQ, hns_roce_ib_create_srq,
>> hns_roce_ib_create_srq_resp);
> Are you sure this shouldn't be
>
> DECLARE_DRV_CMD(hns_roce_create_srq,
> IB_USER_VERBS_CMD_CREATE_SRQ, hns_roce_ib_create_srq,
> empty);
>
> ?
>
> Jason
Yes. It will build fail and prompt hns_roce_create_srq_resp has no member named srqn information.
