Re: possible SCSI device numbering solution

Albert Cahalan (albert@ccs.neu.edu)
Tue, 25 Jun 1996 19:22:38 -0400 (EDT)

From: koenig@tat.physik.uni-tuebingen.de (Harald Koenig)
>>
>> But anyhoot, 16/48 bit majors/minors seems reasonable to me.
>> Of course, this is a blind guess - we really should discuss how
>> majors/minors will be assigned and used first. We could easily gobble up
>> 128-bits with a poor system, or, improve the current one and stick with
>> 16-bits.
>
> any idea how the "ls -l" output should look like for larger dev_t ?
>
> crw-r----- 1 root kmem 1, 2 Aug 29 1992 /dev/kmem
> brw-rw---- 1 root root 65535, 281474976710655 Apr 1 2001 /dev/last_dev_16_48
> brw-rw---- 1 root root 4294967295, 4294967295 Apr 1 2001 /dev/last_dev_32_32
> brw-rw---- 1 root disk 8, 0 May 4 1994 /dev/sda
>
> doesn't look too nice ;-)

------------------ For 64-bit device numbers -------------------

Here's the old way:
brw-rw---- 1 root disk 8, 0 May 4 1994 /dev/sda

Hex won't quite fit:
brw-rw---- 1 root disk 2847a3b5f9e310f9 May 4 1994 /dev/sda

Use base-64 encoding:
brw-rw---- 1 root disk 3Aib31fzeTc May 4 1994 /dev/sda

------------------- For something that fits --------------------

If we only use 48 bits (dev_t is padded to 64-bit) then hex will
fit in the display correctly:
brw-rw---- 1 root disk a3b5f9e310f9 May 4 1994 /dev/sda

Break up device numbers on nibble boundries for readability.

controller:4
bus:4
device:8
lun:8
partition:4

Hmmm, that is only 28 bits. Did I miss something? If SCSI, IDE,
and all the weird stuff gets unified naming, then 36 bits are left
over! Even with 8-bit partition numbers, 32 bits is enough.