>>> I suggest, as you say, a 32:32 split (it's simple).
>>
>> 64bit dev_t's are going to cause some real problems. I would suggest
>> you try passing a 64bit dev_t over NFS - especially to non unix
>> clients before you do this.
>>
>> If you go to a 64bit dev_t you will have to do hackish horrors for
>> NFS support even Linux to Linux. Stay with a 32bit dev_t, supporting
>> more device files than we can create 1 a second within our epoch is crazy
>
> Not really... some devices have a legitimate need for a sparse minor
> number space. I'd be interested in how HP/UX deals with this; they've
> had a fairly large dev_t for aeons.
HP/UX was using 8:24 last I heard, with /bin/ls using hex.
The sparse minor argument is mostly garbage now. It made perfect
sense when wide SCSI (15 devices) was the worst case. Now we can
have tree-like systems that need infinite bits.
It is easy to use up 64 bits. Look here:
8 node
4 bus
4 bus
4 slot
8 port (place where a cable connects to a card)
8 port (on hub)
8 port (on hub)
8 port (on hub)
8 port (on hub)
... ...
8 LUN
8 partition (native)
8 partition (nested)
If you have a solution to the above, it can be used to fit everything
into 32 bits. I suggest dynamic numbers with devfs, but that need not
be the only way to solve the problem.
If you have userspace generating /dev entries as needed (to avoid
terabytes of inodes), then you might as well use dynamic numbers.
You already lost the advantage of static number assignments.
OSF/1, Digital UNIX, and Tru64 all :-) use 12:20 device numbers.
That would be nice. It allows static allocation for the easy
cases while not causing NFS trouble and general bloat on ia32.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/