[openib-general] Re: [PATCH] RFC Verbs: add support for transport specific verbs

[Sean Hefty]

Michael S. Tsirkin wrote:
> Okay, but lets try to avoid adding runtime overhead.

This shouldn't add any runtime overhead than what's already there.  The only 
difference is the notation used to get to the process_mad function.

> I dont see how - the union size will likely change anyway.

This could be avoided by making the union reference pointers to structures, 
rather than the structures themselves.  I didn't go this route because I didn't 
want to update where the ib_device structures were allocated.  This would have 
an impact on runtime performance, but the cost is only an additional memory read.

>>>My point is, I have to test whether the function is implemented anyway,
>>>so why add two checks: one for device type, another for function
>>>implementation? Its complicated and inefficient.

Multiple checks are already done today.  For example, the MAD code checks device 
type before trying to use a device, then checks again to see if a function is 
implemented.  Similarly, the CMA must also check device type.

> Then at least lets make it a structure, not a union.
> This way a single test is sufficient to figure out whether
> a specific function is supported.

I don't follow the advantages of embedding a structure inside ib_device.

- Sean