Quoting r. Sean Hefty <[EMAIL PROTECTED]>:
> Subject: common userspace support
> 
> I'm implementing the userspace CMA and noticed that there are a couple of 
> areas 
> where userspace support overlaps.
> 
> For example, both the CMA and IB CM need to copy path records between 
> userspace 
> and the kernel.  They also copy QP attributes, which would also be needed by 
> verbs at some point to support query QP.  In these cases, the data structures 
> passed between userspace and the kernel are the same, as is the code to copy 
> them.
> 
> Does anyone have a preference for how to deal with this issue on both the 
> kernel 
> and userspace sides?
> 
> My thinking is that for the kernel, the kernel structures would be defined in 
> a 
> common header, with functions exported to copy to/from them.  This results in 
> additional dependencies between modules.  (E.g. rdma_ucm would require 
> ib_uverbs 
> and ib_usa modules.  ib_user_verbs.h would define the QP attribute structure 
> and 
> uverbs_?.c would export copy routines.)
> 
> For userspace, we can do something similar, which would build dependencies 
> between the different libraries.
> 
> - Sean

I see what you mean now.
In my opinion, given that cma is going to be used with uverbs anyway,
it shouldnt be a problem to make cma depend on uverbs.

-- 
MST
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to