[openib-general] on calling rdma_disconnect from non sleepablecontext

[Caitlin Bestler]

openib-general-bounces at openib.org wrote:
>> I have now instrumented iser code to always call rdma_disconnect /
>> rdma_destory_qp etc from sleepable context. Before doing so i was
>> getting this oops few times. My interpertation was that
>> cma_modify_qp_err is not supposted to get called when in_atomic is
>> true, am i correct?
> 
> It looks like this from the backtrace, but this is dependent on the
> lower level driver. 
> 

ULP code such as iSER should not have to read driver-specific
code to determine if their use of the verbs is correct.

Either use of cma_modify_qp_err is legal while in_atomic or
it is not. If it is legal, the fact that a given driver oops
is a bug in that driver. If is not legal then the fact that
a given driver does not cause an oops does not make it legal.

Generally, I do not believe that QP state modifying calls should
be legal in restricted contexts. Whenever there is a requirement
that this be supported it should be explicitly documented because
it creates a special requirement for those implementing drivers.