Re: [for-next PATCH v6 09/10] RDMA/cm: Make QP FLUSHABLE

["lizhijian@fujitsu.com" <lizhijian@fujitsu.com>]

On 22/11/2022 22:52, Jason Gunthorpe wrote:
> On Wed, Nov 16, 2022 at 04:19:50PM +0800, Li Zhijian wrote:
>> It enables flushable access flag for qp
>>
>> Reviewed-by: Zhu Yanjun <zyjzyj2000@gmail.com>
>> Signed-off-by: Li Zhijian <lizhijian@fujitsu.com>
>> ---
>> V5: new patch, inspired by Bob
>> ---
>>   drivers/infiniband/core/cm.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c
>> index 1f9938a2c475..58837aac980b 100644
>> --- a/drivers/infiniband/core/cm.c
>> +++ b/drivers/infiniband/core/cm.c
>> @@ -4096,7 +4096,8 @@ static int cm_init_qp_init_attr(struct cm_id_private *cm_id_priv,
>>   		qp_attr->qp_access_flags = IB_ACCESS_REMOTE_WRITE;
>>   		if (cm_id_priv->responder_resources)
>>   			qp_attr->qp_access_flags |= IB_ACCESS_REMOTE_READ |
>> -						    IB_ACCESS_REMOTE_ATOMIC;
>> +						    IB_ACCESS_REMOTE_ATOMIC |
>> +						    IB_ACCESS_FLUSHABLE;
> 
> What is the point of this? Nothing checks IB_ACCESS_FLUSHABLE ?

Previous, responder of RXE will check qp_access_flags in check_op_valid():
  256 static enum resp_states check_op_valid(struct rxe_qp *qp, 

  257                                        struct rxe_pkt_info *pkt) 

  258 { 

  259         switch (qp_type(qp)) { 

  260         case IB_QPT_RC: 

  261                 if (((pkt->mask & RXE_READ_MASK) && 

  262                      !(qp->attr.qp_access_flags & 
IB_ACCESS_REMOTE_READ)) || 
 

  263                     ((pkt->mask & RXE_WRITE_MASK) && 

  264                      !(qp->attr.qp_access_flags & 
IB_ACCESS_REMOTE_WRITE)) ||
  265                     ((pkt->mask & RXE_ATOMIC_MASK) && 

  266                      !(qp->attr.qp_access_flags & 
IB_ACCESS_REMOTE_ATOMIC))) {
  267                         return RESPST_ERR_UNSUPPORTED_OPCODE; 

  268                 }

based on this, additional IB_FLUSH_PERSISTENT and IB_ACCESS_FLUSH_GLOBAL 
were added in patch7 since V5 suggested by Bob.
Because of this change, QP should become FLUSHABLE correspondingly.

> 
> Do flush ops require a responder resource?

Yes, i think so. See IBA spec, oA19-9: FLUSH shall consume a single 
responder...


> 
> Why should CM set it unconditionally?
> 

I had ever checked git history log of qp->qp_access_flags, they did as 
it's. So i also think qp_access_flags should accept all new IBA 
abilities unconditionally.

What do you think of this @Jason


Thanks
Zhijian
> Explain in the commit message
> 
> Jason