Re: PROPOSAL: /proc/dev

Kevin Lentin (kevinl@cs.monash.edu.au)
Sun, 4 Jan 1998 12:09:12 +1100


On Fri, Jan 02, 1998 at 07:14:22PM +0000, Linus Torvalds wrote:
> But the good thing with a devfs would be that the /dev directory would
> only be populated with devices that we actually have - so it wouldn't be
> nearly as large as a "real" /dev.

On the subject of resource consumption. We've heard a lot about having to
hold inodes (or cut down 64 byte structures) for all the devices but what
about all the static data that describes them? For floppies, for instance,
the driver might have to contain a list that contains all of these...

/dev/fd0 /dev/fd0h1600 /dev/fd0u1120 /dev/fd0u1840 /dev/fd0u800
/dev/fd0CompaQ /dev/fd0h360 /dev/fd0u1440 /dev/fd0u1920 /dev/fd0u820
/dev/fd0d360 /dev/fd0h410 /dev/fd0u1600 /dev/fd0u2880 /dev/fd0u830
/dev/fd0h1200 /dev/fd0h420 /dev/fd0u1680 /dev/fd0u3200
/dev/fd0h1440 /dev/fd0h720 /dev/fd0u1722 /dev/fd0u3520
/dev/fd0h1476 /dev/fd0h880 /dev/fd0u1743 /dev/fd0u360
/dev/fd0h1494 /dev/fd0u1040 /dev/fd0u1760 /dev/fd0u3840

That's about 420 bytes. Is that significant to us when you add all the
rest of the structure that surrounds it (including string padding,etc). Or
would a floppy driver prefer to actually generate that list based on the
individual characteristics that the names describe (ie use a set of a few
nested for loops to generate names).

Or am I quibling over tiny little bits of memory? (we put up with that huge
PCI database as it is - although I see plans to move that into userspace
which is good).

-- 
[======================================================================]
[     Kevin Lentin               Email: K.Lentin@cs.monash.edu.au      ]
[   finger kevinl@fangorn.cs.monash.edu.au for PGP public key block.   ]
[  KeyId: 06808EED    FingerPrint: 6024308DE1F84314  811B511DBA6FD596  ]
[======================================================================]