[openib-general] Re: InfiniBand MAD device number request

Torben Mathiasen torben.mathiasen at hp.com
Fri Jan 14 02:16:50 PST 2005


On Thu, Jan 13 2005, Roland Dreier wrote:
>     Torben> Allright, I begun looking into using the 2.6 specific
>     Torben> major/minor allocations.  2.6 allows 20 bit worth of
>     Torben> minors per major. I'll let you know once I know more.
> 
> It seems that using a large major (ie > 255) is a good solution, since
> we have no backwards compatibility to worry about.  So I would propose
> the following if it seems acceptable to you: reserve something like a
> 2K or 4K chunk of minors in the large major space for IB (eg
> 256:0...256:4095 -- of course the major isn't important if you want to
> start assigning somewhere else) and assign the first 256 minors from
> this range as /dev/infiniband/umad0 ... /dev/infiniband/umad255.
>

Yeah, I verified that we _can_ start the new 2.6 allocations, so you guys are
going to be some of the first.

But, initially we were talking 32 or 64 minors, now its 2K or 4K? If you really
need that many, no problem, but try to think about if its really necessary.
Thats a LOT of minors.

Thanks,
Torben



More information about the general mailing list