Re: Fwd: [PATCH] [-mm] ACPI: export ACPI events via netlink

[Patrick McHardy <kaber@trash.net>]

jamal wrote:
> On Fri, 2007-29-06 at 14:48 +0200, Patrick McHardy wrote:
> 
>>Johannes Berg wrote:
> 
> 
>>>Hmm, another thought: since we have 32 bits for group numbers and 16
>>>bits for families we could just reserve 16 bits for groups within each
>>>family. Or do we get trouble with that because in some place there's a
>>>group bitmap or such?
>>
>>
>>Yes, af_netlink has a bitmap per socket that is subscribed to any group.
> 
> 
> 
> I think this is the challenge. The groups belong to a global namespace.
> i.e when you do a socket bind to group - it is unique regardless of the
> family.
> Our philosophy in genetlink is to have dynamic resources allocated and
> released - remember the real reason we even have this is because we were
> running out of numbers ;->


That was more of a rumour :) We have 2^32-1 groups and I think 256
families, of which about 20 are used.

> So while the static allocation of 16 bits per group will work (famous
> last words "noone will ever need more than 640K of RAM";->) it will be
> cleaner imo to allow dynamic allocation/release.
> Maybe a mix (of a few static and mostly dynamic) as Patrick says - but
> that would mean more coding for you ;-> Actually i like the idea of at
> least your ID being your static mcast group and the rest are in the
> dynamic pool (Hey, thanks Patrick;->). This means the first 2^16 are
> static/reserved and if you want more groups, you register for them.


I wouldn't reserve any, just hand them out as needed. Otherwise we'll
run into problems with the af_netlink bitmaps.