On 2020/11/13 2:35, Jason Gunthorpe wrote:
> On Fri, Oct 30, 2020 at 07:39:34PM +0800, Weihang Li wrote:
>> HIP09 supports service type of Unreliable Datagram, add necessary process
>> to enable this feature.
>>
>> Signed-off-by: Weihang Li <liweihang@xxxxxxxxxx>
>> drivers/infiniband/hw/hns/hns_roce_device.h | 2 ++
>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 8 +++++---
>> drivers/infiniband/hw/hns/hns_roce_qp.c | 3 ++-
>> 3 files changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/infiniband/hw/hns/hns_roce_device.h b/drivers/infiniband/hw/hns/hns_roce_device.h
>> index 9a032d0..23f8fe7 100644
>> +++ b/drivers/infiniband/hw/hns/hns_roce_device.h
>> @@ -222,7 +222,9 @@ enum {
>> HNS_ROCE_CAP_FLAG_FRMR = BIT(8),
>> HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL = BIT(9),
>> HNS_ROCE_CAP_FLAG_ATOMIC = BIT(10),
>> + HNS_ROCE_CAP_FLAG_UD = BIT(11),
>
> Why add this flag if nothing reads it?
Hi Jason,
I checked it to set IB_USER_VERBS_CMD_CREATE_AH and IB_USER_VERBS_CMD_DESTROY_AH
in uverbs_cmd_mask which is not needed recently, I forgot to remove it in this
patch :) Will drop it.
>
>> HNS_ROCE_CAP_FLAG_SDI_MODE = BIT(14),
>> +
>> };
>
> Extra space
OK, will remove it.
>
>
> If I recall properly earlier chips did not have a GID table so could
> not support UD because they could not secure the AH, or something like
> that.
>
> So, I would expect to see that only the new devices support UD, but
> I can't quite see that in this patch??
You are right. I made a mistake, I thought it's enough to add judgment of hardware
version in rdma-core to prevent the HIP08's user from using UD in userspace and I
realize it's meaningless just now...
I think I can add a check about the hardware version when creating a user's UD QP.
Thanks again for the reminder.
Weihang
>
> Jason
>
