On 2020/11/17 0:09, Jason Gunthorpe wrote:
> On Sat, Nov 14, 2020 at 03:46:58AM +0000, liweihang wrote:
>> On 2020/11/13 2:33, Jason Gunthorpe wrote:
>>> On Fri, Oct 30, 2020 at 07:39:31PM +0800, Weihang Li wrote:
>>>> From: Jiaran Zhang <zhangjiaran@xxxxxxxxxx>
>>>>
>>>> According to the RoCE v1 specification, the sl (service level) 0-7 are
>>>> mapped directly to priorities 0-7 respectively, sl 8-15 are reserved. The
>>>> driver should verify whether the value of sl is larger than 7, if so, an
>>>> exception should be returned.
>>>>
>>>> Fixes: d6a3627e311c ("RDMA/hns: Optimize wqe buffer set flow for post send")
>>>> Signed-off-by: Jiaran Zhang <zhangjiaran@xxxxxxxxxx>
>>>> Signed-off-by: Weihang Li <liweihang@xxxxxxxxxx>
>>>> drivers/infiniband/hw/hns/hns_roce_hw_v2.c | 10 +++++++++-
>>>> 1 file changed, 9 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/infiniband/hw/hns/hns_roce_hw_v2.c b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> index 7a1d30f..69386a5 100644
>>>> +++ b/drivers/infiniband/hw/hns/hns_roce_hw_v2.c
>>>> @@ -427,9 +427,10 @@ static inline int set_ud_wqe(struct hns_roce_qp *qp,
>>>> void *wqe, unsigned int *sge_idx,
>>>> unsigned int owner_bit)
>>>> {
>>>> - struct hns_roce_dev *hr_dev = to_hr_dev(qp->ibqp.device);
>>>> struct hns_roce_ah *ah = to_hr_ah(ud_wr(wr)->ah);
>>>> struct hns_roce_v2_ud_send_wqe *ud_sq_wqe = wqe;
>>>> + struct ib_device *ib_dev = qp->ibqp.device;
>>>> + struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
>>>> unsigned int curr_idx = *sge_idx;
>>>> int valid_num_sge;
>>>> u32 msg_len = 0;
>>>> @@ -489,6 +490,13 @@ static inline int set_ud_wqe(struct hns_roce_qp *qp,
>>>> V2_UD_SEND_WQE_BYTE_36_TCLASS_S, ah->av.tclass);
>>>> roce_set_field(ud_sq_wqe->byte_40, V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_M,
>>>> V2_UD_SEND_WQE_BYTE_40_FLOW_LABEL_S, ah->av.flowlabel);
>>>> +
>>>> + if (unlikely(ah->av.sl > MAX_SERVICE_LEVEL)) {
>>>> + ibdev_err(ib_dev,
>>>> + "failed to fill ud av, ud sl (%d) shouldn't be larger than %d.\n",
>>>> + ah->av.sl, MAX_SERVICE_LEVEL);
>>>> + return -EINVAL;
>>>> + }
>>>
>>> We should not print for things like this, IIRC userspace can cause the
>>> ah's sl to become set out of bounds
>
>
>> In "Annex A 16: RoCE", I found the following description:
>>
>> SL 0-7 are mapped directly to Priorities 0-7, respectively
>>
>> SL 8-15 are reserved.
>>
>> CA16-18: An attempt to use an Address Vector for a RoCE port containing
>> a reserved SL value shall result in the Invalid Address Vector verb result.
>>
>> So what should we do if the user wants to use the reserved sl? Should I just let it
>> do mask with 0x7 when creating AH?
>
> Fail and don't print anything
>
> Jason
>
OK, thank you.
Weihang
