lkml.org 
[lkml]   [1998]   [Aug]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: DEVFSv50 and /dev/fb? (or /dev/fb/? ???)
Terry L. Ridder writes:
> Jurgen Botz wrote:
> >
> > "Anthony Barbachan" wrote:
> > > /dev/sd{a,b,c,...} is definately cleaner and simplier than
> > > /dev/dsk/c0t0u0d0s0 (or whatever?!?!?!?). And EIDE devices definately do
> >
> > As a system administrator with over a decade experience, I must say that
> > I disagree completely. The old naming scheme is problematic and ugly and
> > the DEVFS one is practical and elegant. MHO.
> >
> > (I rather suspect that Anthony and some other detractors have never
> > administered systems with multiple SCSI busses and more than a few
> > drives.)
>
> I can only answer for myself, and what you suspect, at least in my case,
> is incorrect. I have been involved with UNIX for over 20 years, from
> working
> on the original AT&T version 6 source code, programming applications,
> SysAdmin,
> etc. I have administered HP file servers, Sun file servers, etc, with
> multiple SCSI busses, and hundreds of hard drives. I did not like
> Solaris
> then and still do not.
>
> By contrast I have also administered large scale AppleShare file servers
> with multiple SCSI controllers and nearly 64 drives.
> Volume naming is nice. Does not matter care what SCSI controller it is
> on,
> what SCSI ID it has, move it to a new fileserver and it comes up with
> the
> same name as before.
>
> >
> > Personally I really like DEVFS. It solves some real problems and it does
> > it in a scaleable, forward-looking manner. Auto-generation seems like
> > a quick-and-dirty hack by comparison. The counter-arguments I've seen
> > seemed to mostly refer to vague aesthetic issues. I think the aesthetics
> > of this kind of thing flow from its functionality, and by that DEVFS is
> > beautiful.
> >
>
> As others have pointed out.
>
> To quote H. Peter Arvin:
>
> <Begin Quote>
> Auto device generation can be a security hole, and can be done in user
> space without hogging tons of kernel memory.
> <End Quote>
>
> To quote H. Peter Arvin again:
>
> <Begin Quote>
> Counter question: how much kernel memory does it take to
> keep a million devices with all their info (atime, mtime,
> ctime, permissions, ownership all included!)
> <End Quote>
>
> Note no one from the dev_fs side has answered this question yet.

I guess you missed my response:
Answer: way too much! That's why a scsidev or devfs system is
essential.

HPA and I then took our discussion offline. For the public benefit,
the typical RAM consumption of devfs is a few pages. The answer I had
given public was a reference to the fact that with devfs (or scsidev)
you don't need millions of inodes.

> dev_fs uses too much of kernel memory and by doing so inflicts a
> performance hit.

It uses a measly few pages. I don't think this is "too much". As I
also discussed with HPA (quiting myself again):
Yes, having a few extra unswappable pages can potentially hurt
performance. On the other hand, devfs also can improve performance
(both because of less device nodes and a more direct link between
device nodes and device drivers). Which effect is more significant is
unknown at this point.

> Using either tar, or a C program to save and restore permissions,
> user/group ownership, modtimes, etc is a hack.

If needed, devfs can have real persistence to a block device.

> To quote Theodore:
>
> <Begin Quote>
> As far as searching a list when we open a major number, again this is a
> extremely flawed and weak argument. First of all, the vast majority of
> systems out there will only have less than 16 major devices. A typical
> system has less than 10 major devices. (cat /proc/devices and see!) So
> searching the list is simply not a problem. If searching the list were
> an issue, there are plenty of ways of solving this problem internal to
> the kernel, without needing to make any user-visible changes --- such
> using hash table.
> <End Quote>

And quoting my response:
I think that the extra layer between device nodes and device drivers
is an ugly hack. I see the extra level of indirection as unnecessary
and adding some (small, but avoidable) performance overhead. I also
see it as a conceptual and administrative overhead. We now have device
information kept in two places: in the source of each driver and in
devices.txt, and has to be synchronised manually. Devfs avoids that
entirely by keeping it in one place: in the driver.
These are not "killer argument" for it, they're just some (small)
reasons in a long list. As I've said in the FAQ, IMHO the totality of
these reasons does show that devfs is a good idea.

Regards,

Richard....
-
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.altern.org/andrebalsa/doc/lkml-faq.html

\
 
 \ /
  Last update: 2005-03-22 13:43    [from the cache]
©2003-2011 Jasper Spaans