Re: State of devfs in 2.6?

Re: State of devfs in 2.6?

[Pascal Schmidt]

On Tue, 09 Dec 2003 11:20:06 +0100, you wrote in linux.kernel:

> # find /dev | wc -l
>     326

40k on ext2 (128 byte inodes). I'm betting devfs is more than 40k of code.
Plus devfs uses more memory than a filesystem-backed /dev, don't you think?

-- 
Ciao,
Pascal

Re: State of devfs in 2.6?

[jw schultz]

On Tue, Dec 09, 2003 at 09:42:54PM -0800, Greg KH wrote:
> On Tue, Dec 09, 2003 at 12:10:47PM -0500, Mark Mielke wrote:
> > On Tue, Dec 09, 2003 at 09:21:56AM +0100, Holger Schurig wrote:
> > > Devfs for embedded devices is just great. It's all in the kernel, no
> > > external process to run (I use my embedded stuff without devfsd). I'm using
> > > it for about one year with various kernels.
> > 
> > I don't see why 'all in the kernel' is the best approach, embedded or
> > otherwise. I believe udev is being written to execute with a minimal
> > runtime environment. No glibc, or other such beasts.
> 
> Exactly.  The current bk tree version of udev (which has more features
> than the 008 release) is weighing in at 49Kb, linked statically, no
> external dependancies.

And if we are talking about the smaller embedded systems it
would be good to remember that their /dev doesn't really
change much so devfs and udev could be left out entirely by
having a small staticly populated /dev in the
initramfs/initrd image, or whatever is used for /.

-- 
________________________________________________________________
	J.W. Schultz            Pegasystems Technologies
	email address:		jw@pegasys.ws

		Remember Cernan and Schmitt

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 12:10:47PM -0500, Mark Mielke wrote:
> On Tue, Dec 09, 2003 at 09:21:56AM +0100, Holger Schurig wrote:
> > Devfs for embedded devices is just great. It's all in the kernel, no
> > external process to run (I use my embedded stuff without devfsd). I'm using
> > it for about one year with various kernels.
> 
> I don't see why 'all in the kernel' is the best approach, embedded or
> otherwise. I believe udev is being written to execute with a minimal
> runtime environment. No glibc, or other such beasts.

Exactly.  The current bk tree version of udev (which has more features
than the 008 release) is weighing in at 49Kb, linked statically, no
external dependancies.

greg k-h

Re: State of devfs in 2.6?

[Bob]

Skip to Bottom line: Does devfs do devpts on 2.6? Didn't for me.

Clemens Schwaighofer wrote:

> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Greg KH wrote:
>
> |
> | I don't think that all 4 users of devfs on 2.6 are all that vocal :)
> | Either way, I haven't been paying attention, as I really don't care.
>
> well ... I think there are more devfs users out there. eg, all the
> Gentoo freaks (me included) are sort of forced into devfs. If they want
> or not. And they will stick to it, until sysfs/udev/hotplug/foobarfs is
> so solid it can replace devfs. 

I'm using devfs on 2.6.0-test11. I have never run into the
impossible to fix problems. I run devfs with very little
old compatible naming. I edited /etc/devfs/compat_symlinks
to give me /dev/sd* /dev/hd* and weirdly ln -s /dev/dsp2 /dev/dsp
since all sound apps only seem to look for /dev/dsp which is only
a dummy with nforce2.

In 2.6 devfs would not do devpts for X so I put devpts back in the
kernel and instantly got on with my work. I'm not sure if that means
devfs can't do devpts in 2.6. I don't care.

I see no devfs devices for a 32-bit cardbus pcmcia controller pci card
which is id'd by yenta and cardmgr(pcmcia-cs v3.2.5), so I am switching
to sysfs udev hotplug to see if that helps. Actually I can mount sysfs
to /sys now and look around first but proc and var agree there is nada
but there are two cards in slots.

In the cdrecord thread Linus said some things about target/bus/lun
naming and I must admit that it is nice to get back where ls /dev
shows what drives there are without having to tree out into
/dev/scsi/host/bus/target/lun/* to see how drives are recognized.
That can be improved in most configurations by using compat
symlinks instructions for /dev/sd* in /etc/devfs/compat_symlinks
but as Linus explained iscsi handled better by sysfs udev /dev/s??
and not 0,0,0 targbuslun cdrecord-style.

I like devfs and haven't ever had an error burning cd's with ide-scsi
but I'm switching to udev and ide-cd to suck up to Linus before
I start plaguing him about my pcmcia controller and yenta not
working together--udev might work better with hotplug than
devfs so I should try that first.

It seems dubious that Gooch isn't maintaining devfs and syfs
could be moving away from compatibility with devfs, and
there are proven impossibilities about the devfs way. Devfs
is a lot less broken than windows, though. Devfs might
be good enough to Walmart-Microsoft pretty good alright
rip the guts out of userland, but that's not good enough for the
os that runs the internet.

-Bob

Re: State of devfs in 2.6?

[Clemens Schwaighofer]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Greg KH wrote:

|
| I don't think that all 4 users of devfs on 2.6 are all that vocal :)
| Either way, I haven't been paying attention, as I really don't care.

well ... I think there are more devfs users out there. eg, all the
Gentoo freaks (me included) are sort of forced into devfs. If they want
or not. And they will stick to it, until sysfs/udev/hotplug/foobarfs is
so solid it can replace devfs.


- --
Clemens Schwaighofer - IT Engineer & System Administration
==========================================================
Tequila Japan, 6-17-2 Ginza Chuo-ku, Tokyo 104-8167, JAPAN
Tel: +81-(0)3-3545-7703            Fax: +81-(0)3-3545-7343
http://www.tequila.jp
==========================================================
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org

iD8DBQE/1oHQjBz/yQjBxz8RAuumAJ9y5mXjaMSSOA0D3HL9g0pz0wSYkACfW0NC
OqFAQbBoSAdGJwciCJuPsck=
=MIoV
-----END PGP SIGNATURE-----

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 05:47:39PM +0100, Eduard Bloch wrote:
> #include <hallo.h>
> * Greg KH [Tue, Dec 09 2003, 08:27:47AM]:
> 
> > Like Matthew stated, either use the udev rc startup script, or put udev
> > into your initramfs image to catch all of the early boot messages.
> > Doing the initramfs method is still very tough to do right now, but
> > people have reported success that way.  I still recommend just using the
> > init.d script for now.
> 
> Wouln't it be less error-prone to introduce a kind of queing for the
> hotplug program so kernel puts all the registered devices in a list and
> the list is submitted in one pass when udev asks for it?

Heh, no.  It's much easier to just run out of early userspace.

But that's just my opinion, if you think differently, I'd be very glad
to see the code for your solution. :)

thanks,

greg k-h

Re: State of devfs in 2.6?

[Svetoslav Slavtchev]

Hi sorry for using this kind of reply,
but i'm not subscribed to lkml
/* it would be nice if you CC me :-) */

------ original e-mail -----------
On Mon, Dec 08, 2003 at 11:26:17PM -0600, Rob Landley wrote:
> 
> Is there a big rollup patch against that adds all the sys/*/dev entries
for 
> people who want to try udev?

After I get finished catching up on USB patches that people sent me for
the last month, I'll generate this and post it to lkml.

thanks,

greg k-h
-------end original e-mail -----

Greg could you please post them splited,
i'm having some issues with the one that adds
misc devices -- oopses in early boot as described in
http://marc.theaimsgroup.com/?l=linux-kernel&m=107097871212256&w=2

i managed to find 4 of the patches, but i'm still missing
the ones for sound, input, parport

the ones i found are available from the mandrake ml :-)
eg http://marc.theaimsgroup.com/?l=mandrake-cooker&m=107099351100451&w=2
( mem, vcs, fb , misc devices support)

best,

svetljo

-- 
+++ GMX - die erste Adresse für Mail, Message, More +++
Neu: Preissenkung für MMS und FreeMMS! http://www.gmx.net

Re: State of devfs in 2.6?

[Greg KH]

On Mon, Dec 08, 2003 at 11:26:17PM -0600, Rob Landley wrote:
> Is there a big rollup patch against that adds all the sys/*/dev entries for 
> people who want to try udev?

Oh, and you can try udev today with no kernel patches needed.  All block
devices and tty devices and i2c dev devices and usb major devices and
v4l devices work just fine with udev on 2.6.0-test11.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Greg KH]

On Mon, Dec 08, 2003 at 11:26:17PM -0600, Rob Landley wrote:
> 
> Is there a big rollup patch against that adds all the sys/*/dev entries for 
> people who want to try udev?

After I get finished catching up on USB patches that people sent me for
the last month, I'll generate this and post it to lkml.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Mark Mielke]

On Tue, Dec 09, 2003 at 09:21:56AM +0100, Holger Schurig wrote:
> Devfs for embedded devices is just great. It's all in the kernel, no
> external process to run (I use my embedded stuff without devfsd). I'm using
> it for about one year with various kernels.

I don't see why 'all in the kernel' is the best approach, embedded or
otherwise. I believe udev is being written to execute with a minimal
runtime environment. No glibc, or other such beasts.

> * space. devfs doesn't eat space like the MAKEDEV approach.

Can you prove this? tmpfs doesn't seem to take up much space, and given
that only devices that exist will require data structures in both cases,
it seems to me that the issue is a little irrelevant in either case.

> * simplicity: I run my system without devfsd and without an initial ramdisk.
> All needed modules are simply compiled into the kernel.

Isn't this an argument for udev?

> * No need for overcomplification, e.g a process that has to be started
> before userspace touches /dev, a specially compiled uclibc-based proggy in
> an initrd

Many of us believe that devfs is 'over-complification'.

> So, when /dev is accessed by userspace, all is there and well.

True in either case...

mark

-- 
mark@mielke.cc/markm@ncf.ca/markm@nortelnetworks.com __________________________
.  .  _  ._  . .   .__    .  . ._. .__ .   . . .__  | Neighbourhood Coder
|\/| |_| |_| |/    |_     |\/|  |  |_  |   |/  |_   | 
|  | | | | \ | \   |__ .  |  | .|. |__ |__ | \ |__  | Ottawa, Ontario, Canada

  One ring to rule them all, one ring to find them, one ring to bring them all
                       and in the darkness bind them...

                           http://mark.mielke.cc/

Re: State of devfs in 2.6?

[Eduard Bloch]

#include <hallo.h>
* Greg KH [Tue, Dec 09 2003, 08:27:47AM]:

> Like Matthew stated, either use the udev rc startup script, or put udev
> into your initramfs image to catch all of the early boot messages.
> Doing the initramfs method is still very tough to do right now, but
> people have reported success that way.  I still recommend just using the
> init.d script for now.

Wouln't it be less error-prone to introduce a kind of queing for the
hotplug program so kernel puts all the registered devices in a list and
the list is submitted in one pass when udev asks for it?

MfG,
Eduard.

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 10:59:09AM +0100, Jan Dittmer wrote:
> Btw. I still haven't figured out, how to use udev properly. I just get
> the nodes of devices I plugin after boot and of the modules I load after
> boot. IDE et all aren't showing up. How early do I need to load udev or
> has my kernel to be all modular for it to work properly?

Like Matthew stated, either use the udev rc startup script, or put udev
into your initramfs image to catch all of the early boot messages.
Doing the initramfs method is still very tough to do right now, but
people have reported success that way.  I still recommend just using the
init.d script for now.

Oh, and if you do have any udev problems, please post them to the
linux-hotplug-devel mailing list.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Matthew Reppert]

On Tue, 2003-12-09 at 03:59, Jan Dittmer wrote:
>
> Btw. I still haven't figured out, how to use udev properly. I just get
> the nodes of devices I plugin after boot and of the modules I load after
> boot. IDE et all aren't showing up. How early do I need to load udev or
> has my kernel to be all modular for it to work properly?

Since, I believe, version 006, udev has shipped with an init script
contributed by rml that will create device nodes for devices present
at system boot. You should be able to just make sure that that runs
during your boot sequence and be fine. (I just ran this script on
my system, and made sure it proper nodes for all my IDE drives and
the partitions contained thereon.)

Matt

Re: State of devfs in 2.6?

[Paul P Komkoff Jr]

Replying to Andrew Walrond:
> My initial query has thrown up lots of interesting debate :)
> 
> I, like most people I suspect, love the concept of a complete auto-populated 
> dev directory, and not having to MAKEDEV.
> 
> devfs provided this, but like most people who read LKML, I stopped using it 
> when it's problems were discussed.
> 
> I really hope udev lives up to its promise, unlike devfs. Manually creating /
> dev just annoys me for no apparent reason other than it's plain inelegance I 
> suppose.

The one of main benefits I considered when focusing my attention on
devfs-like approach is space consumption. Somewhat-populated dev
subdirectory (looking at fedora 1) have about 7k items inside, each of
them eating its inode and (depends on underlying fs) a block.

I agree that previous implementation may be racy, domb, gooched,
whatever.
but
is it sane that for system to function correcly I should carry over
a whole bunch of directory entries, when I actually have all
information about it in kernel, somewhere buried under major-minor
declarations. 

That is, udev backed on tmpfs approach are almost
solving our problem. But not completely. Module autoloading is useful,
actually, was useful in conjunction with module unloading - if
unloading support is poor autoloading is almost useless ...

-- 
Paul P 'Stingray' Komkoff Jr // http://stingr.net/key <- my pgp key
 This message represents the official view of the voices in my head

Re: State of devfs in 2.6?

[Måns Rullgård]

Andrew Walrond <andrew@walrond.org> writes:

> My initial query has thrown up lots of interesting debate :)
>
> I, like most people I suspect, love the concept of a complete auto-populated 
> dev directory, and not having to MAKEDEV.

So do I.

> devfs provided this, but like most people who read LKML, I stopped using it 
> when it's problems were discussed.
>
> I really hope udev lives up to its promise, unlike devfs. Manually creating /
> dev just annoys me for no apparent reason other than it's plain inelegance I 
> suppose.

Does anyone else remember all the talk about the supreme elegance of
devfs back when it was new?

-- 
Måns Rullgård
mru@kth.se

Re: State of devfs in 2.6?

[Andrew Walrond]

My initial query has thrown up lots of interesting debate :)

I, like most people I suspect, love the concept of a complete auto-populated 
dev directory, and not having to MAKEDEV.

devfs provided this, but like most people who read LKML, I stopped using it 
when it's problems were discussed.

I really hope udev lives up to its promise, unlike devfs. Manually creating /
dev just annoys me for no apparent reason other than it's plain inelegance I 
suppose.

Andrew Walrond

Re: State of devfs in 2.6?

[Holger Schurig]

>> * space. devfs doesn't eat space like the MAKEDEV approach.
> 
> Um, devfs does actually use a non-zero amount of code...

Yeh, it uses code. But if you look into a MAKEDEV bases file system, you see
hundreds of device files. Which uses inode and directory space on the
medium, may it be e2fs or jffs2.

I did not measure if the code for devfs is smaller as the code+config files
for udev. But I didn't make a statement about this.

> For a typical embedded device with about 5 entries in /dev I wouldn't
> be surprised if devfs used a lot _more_ space...

# find /dev | wc -l
    326

"Embedded" is not automatically a washing maschine with only 5 entries, it
can be a full blown Linux computer with 400 MHz, Framebuffer, USB Host, USB
Client, 7 serial ports for GSM, Barcode Scanner, Whatever, and so on. Like
our device.

However, even on such hardware-rich devices you usually have a constraint on
Flash Memory size. So having things small & simple is nice. That makes room
for the user applications & data.

-- 
Try Linux 2.6 from BitKeeper for PXA2x0 CPUs at
http://www.mn-logistik.de/unsupported/linux-2.6/

Re: State of devfs in 2.6?

[Jan Dittmer]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Greg KH wrote:
| On Tue, Dec 09, 2003 at 08:02:19AM +0100, Andreas Jellinghaus wrote:
|
|>>Regardless of the state of udev, devfs has insolvable problems and you
|>>should not use it.  End of story.
|>
|>how many bug reports did you see in the last three months of people
|>having problems with devfs?
|
|
| I don't think that all 4 users of devfs on 2.6 are all that vocal :)
| Either way, I haven't been paying attention, as I really don't care.
|

FWIW, I've been using devfs from the beginning of 2.4 and with 2.5/2.6
with Debian and never had a problem (knock on wood). I really like the
way of having device nodes only for present devices.
Btw. I still haven't figured out, how to use udev properly. I just get
the nodes of devices I plugin after boot and of the modules I load after
boot. IDE et all aren't showing up. How early do I need to load udev or
has my kernel to be all modular for it to work properly?

Thanks,

Jan
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQE/1ZztLqMJRclVKIYRAlgWAJ0cFbRv2QbWrQbFNWACwHrp/opQiQCfZJVD
UjI7PkOYwt4auQb1qTRtwx8=
=40wq
-----END PGP SIGNATURE-----

Re: State of devfs in 2.6?

[Andreas Jellinghaus]

Greg, 

what about adding those patches to the hotplug utils
directory, so people can find them easily and test
sysfs+udev on empty /dev directories?

Regards, Andreas

Re: State of devfs in 2.6?

[Miles Bader]

Holger Schurig <h.schurig@mn-logistik.de> writes:
> Devfs for embedded devices is just great.
> 
> * space. devfs doesn't eat space like the MAKEDEV approach.

Um, devfs does actually use a non-zero amount of code...

For a typical embedded device with about 5 entries in /dev I wouldn't
be surprised if devfs used a lot _more_ space...

-miles
-- 
`...the Soviet Union was sliding in to an economic collapse so comprehensive
 that in the end its factories produced not goods but bads: finished products
 less valuable than the raw materials they were made from.'  [The Economist]

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 08:02:19AM +0100, Andreas Jellinghaus wrote:
> On Tue, 2003-12-09 at 00:34, Greg KH wrote:
> > You have 15 floppy devices connected to your box?  All floppy devices
> > should show up in /sys/block.
> 
> No, 16 devices are normal, sysfs has only one:
> aj@simulacron:~/torrent/j-tv/download$ ls /dev/floppy/
> 0       0u1120  0u1600  0u1722  0u1760  0u1920  0u720  0u820
> 0u1040  0u1440  0u1680  0u1743  0u1840  0u360   0u800  0u830
> aj@simulacron:~/torrent/j-tv/download$ find /sys/block/fd* -name dev
> /sys/block/fd0/dev
> 
> Are those floppy devices obsolete? fdformat was the only app to use
> them anyway, I guess. (Not that I use my floppy, I simply noticed
> the change.)

I don't know if they are obsolete or not.  I've never used them, and
trying to figure out the floppy driver just makes my head hurt.  I'm
sure that if anyone really does use them, and udev doesn't work for
them, I'll hear about it :)

> > Regardless of the state of udev, devfs has insolvable problems and you
> > should not use it.  End of story.
> 
> how many bug reports did you see in the last three months of people
> having problems with devfs?

I don't think that all 4 users of devfs on 2.6 are all that vocal :)
Either way, I haven't been paying attention, as I really don't care.

> I don't doubt the problems in theory, but but I simply haven't seen
> them happening. Most users seem quite happy.

There's no accounting for taste, is there.  Anyway, sure some users
might be happy, but when the kernel vfs maintainer shows that it's
pretty simple to deadlock your kernel, and when the devfs maintainer
disappears from view for over a year, that might be a good indication
that you might want to stop using it.

Besides, there's only one "major" distro still using devfs, and that's
just because they don't know any better.

thanks,

greg k-h

/me prepares for another onslaught of complaints from Gentoo users 

Re: State of devfs in 2.6?

[Holger Schurig]

>> how many bug reports did you see in the last three months of people
>> having problems with devfs? I don't doubt the problems in theory, but
>> but I simply haven't seen them happening. Most users seem quite happy.
>> 
> 
> Actually, I think most users who have problems just disable devfs. Most of
> the people I know have done that. No point in making bug reports about
> something that is unmaintained and deprecated.

No, not really.

Devfs for embedded devices is just great. It's all in the kernel, no
external process to run (I use my embedded stuff without devfsd). I'm using
it for about one year with various kernels.

For me, I see several benefits:

* space. devfs doesn't eat space like the MAKEDEV approach.
* simplicity: I run my system without devfsd and without an initial ramdisk.
All needed modules are simply compiled into the kernel.
* No need for overcomplification, e.g a process that has to be started
before userspace touches /dev, a specially compiled uclibc-based proggy in
an initrd

So, when /dev is accessed by userspace, all is there and well.

-- 
Try Linux 2.6 from BitKeeper for PXA2x0 CPUs at
http://www.mn-logistik.de/unsupported/linux-2.6/

Re: State of devfs in 2.6?

[Vojtech Pavlik]

On Mon, Dec 08, 2003 at 03:34:28PM -0800, Greg KH wrote:

> >  - 5 input/ devices
> 
> Patch for sysfs support for this has been posted by Hanna Linder.  It
> still needs work before being added to the kernel tree.

I missed it. Where can I find it?

-- 
Vojtech Pavlik
SuSE Labs, SuSE CR

Re: State of devfs in 2.6?

[Murray J. Root]

On Tue, Dec 09, 2003 at 08:02:19AM +0100, Andreas Jellinghaus wrote:
> On Tue, 2003-12-09 at 00:34, Greg KH wrote:
> > You have 15 floppy devices connected to your box?  All floppy devices
> > should show up in /sys/block.
> 
> No, 16 devices are normal, sysfs has only one:
> aj@simulacron:~/torrent/j-tv/download$ ls /dev/floppy/
> 0       0u1120  0u1600  0u1722  0u1760  0u1920  0u720  0u820
> 0u1040  0u1440  0u1680  0u1743  0u1840  0u360   0u800  0u830
> aj@simulacron:~/torrent/j-tv/download$ find /sys/block/fd* -name dev
> /sys/block/fd0/dev
> 
> Are those floppy devices obsolete? fdformat was the only app to use
> them anyway, I guess. (Not that I use my floppy, I simply noticed
> the change.)
> 
> > > I wouldn't call udev deprecated, unless a newer kernel has the
> > > essential devices, too.
> > 
> > You mean s/udev/devfs/ right?  :)
> 
> oops, sorry.
> 
> > > and
> > > re-introducing makedev for devices not represented
> > > in sysfs doesn't sound very nice either. So 2.8.* might be a nice time
> > > frame for dropping devfs, or at least give sysfs and udev a few months
> > > to catch up on the issues mentioned.
> > 
> > Regardless of the state of udev, devfs has insolvable problems and you
> > should not use it.  End of story.
> 
> how many bug reports did you see in the last three months of people
> having problems with devfs? I don't doubt the problems in theory, but
> but I simply haven't seen them happening. Most users seem quite happy.
> 

Actually, I think most users who have problems just disable devfs. Most of
the people I know have done that. No point in making bug reports about
something that is unmaintained and deprecated.

-- 
Murray J. Root

Re: State of devfs in 2.6?

[Andreas Jellinghaus]

On Tue, 2003-12-09 at 00:34, Greg KH wrote:
> You have 15 floppy devices connected to your box?  All floppy devices
> should show up in /sys/block.

No, 16 devices are normal, sysfs has only one:
aj@simulacron:~/torrent/j-tv/download$ ls /dev/floppy/
0       0u1120  0u1600  0u1722  0u1760  0u1920  0u720  0u820
0u1040  0u1440  0u1680  0u1743  0u1840  0u360   0u800  0u830
aj@simulacron:~/torrent/j-tv/download$ find /sys/block/fd* -name dev
/sys/block/fd0/dev

Are those floppy devices obsolete? fdformat was the only app to use
them anyway, I guess. (Not that I use my floppy, I simply noticed
the change.)

> > I wouldn't call udev deprecated, unless a newer kernel has the
> > essential devices, too.
> 
> You mean s/udev/devfs/ right?  :)

oops, sorry.

> > and
> > re-introducing makedev for devices not represented
> > in sysfs doesn't sound very nice either. So 2.8.* might be a nice time
> > frame for dropping devfs, or at least give sysfs and udev a few months
> > to catch up on the issues mentioned.
> 
> Regardless of the state of udev, devfs has insolvable problems and you
> should not use it.  End of story.

how many bug reports did you see in the last three months of people
having problems with devfs? I don't doubt the problems in theory, but
but I simply haven't seen them happening. Most users seem quite happy.

Andreas

Re: State of devfs in 2.6?

[Rob Landley]

On Monday 08 December 2003 17:34, Greg KH wrote:

> >  - 5 input/ devices
>
> Patch for sysfs support for this has been posted by Hanna Linder.  It
> still needs work before being added to the kernel tree.
>
> >  - full, kmem, kmsg, mem, null, port, random, urandom, zero
>
> Patch for this has been posted by me to lkml in the past.  It will
> probably go into 2.6.1
>
> >  - printers/0
>
> Hanna Linder is working on a patch for these devices.
>
> >  - 5 misc/ devices
>
> Patch for this has been posted by me to lkml in the past.  It will
> probably go into 2.6.1.
>
> >  - 12 snd/ devices
> >  - 5 sound/ devices
>
> I have a patch here from Leann Ogasawara that adds sysfs support for
> these devices.  I've been lacking time to test it better, but again, it
> will probably make it into 2.6.1.

Is there a big rollup patch against that adds all the sys/*/dev entries for 
people who want to try udev?

Rob

Re: State of devfs in 2.6?

[Rob Landley]

On Monday 08 December 2003 09:59, Andrew Walrond wrote:
> On Monday 08 Dec 2003 3:42 pm, William Lee Irwin III wrote:
> > I would say it's deprecated at the very least. sysfs and udev are
> > supposed to provide equivalent functionality, albeit by a somewhat
> > different mechanism.
>
> Thanks for the pointer.
>
> So how good is the device coverage offered by sysfs/udev ? Do they provide
> a viable/complete MAKEDEV replacement yet?

My understanding is that udev takes the information exported by sysfs about 
what devices exist in the system, and creates device nodes in /dev (which can 
be a ramfs mount or part of a persistent filesystem, udev itself doesn't 
care).  I'm guessing it traverses sysfs to see what the system's got on 
startup (some variant of "find /sys -name device", perhaps) and then receives 
hotplug events when new devices are added later.  On the whole, this is 
generally cool, hotplug friendly, and small and simple.  _and_ the result 
looks like a recognizable /dev directory, so end-user applications don't have 
to be "devfs aware" (which was a bad sign from day 1 if you ask me).

Unfortunately, sysfs doesn't yet export device node information for everything 
in the system yet.  (There aren't any under /sys/cdev, /sys/devices/legacy, 
or /sys/devices/system, for example).  There are pending patches to add more, 
but they're not considered bug fixes, so Linus won't take them before 2.6.0 
and we'll have to wait until after 2.6.0 for development on this subsystem to 
finish.

Probably somewhere in the 2.6.4 to 2.6.6 timeframe, sysfs will have all the 
device exports udev needs.  (Or at least all the ones anybody's complained 
about yet.)  Until then...  dunno.  Maybe you can use a /dev directory on a 
persistent filesystem that you mknod any extra devices you need into 
yourself?)

Rob

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 01:31:00AM +0100, Sven-Haegar Koch wrote:
> On Mon, 8 Dec 2003, Greg KH wrote:
> 
> > > After ignoring .devfsd we are left with 70 devices missing:
> > >  - 15 floppy devices
> >
> > You have 15 floppy devices connected to your box?  All floppy devices
> > should show up in /sys/block.
> 
> perhaps he means fd0u1440, fd0u1600 and friends
> 
> ls /dev/fd0u*|wc -l  -> 15

Yeah, but are those devices actually connected all at once?  Yeah, I
know the way of talking to the same device in different manners...

Hm, udev can now do symlinks, does it also need to handle multiple
minors for floppy devices?  I'll have to look into how the kernel
handles the different block minors for floppy devices.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Sven-Haegar Koch]

On Mon, 8 Dec 2003, Greg KH wrote:

> > After ignoring .devfsd we are left with 70 devices missing:
> >  - 15 floppy devices
>
> You have 15 floppy devices connected to your box?  All floppy devices
> should show up in /sys/block.

perhaps he means fd0u1440, fd0u1600 and friends

ls /dev/fd0u*|wc -l  -> 15

c'ya
sven

-- 

The Internet treats censorship as a routing problem, and routes around it.
(John Gilmore on http://www.cygnus.com/~gnu/)

Re: State of devfs in 2.6?

[Greg KH]

On Mon, Dec 08, 2003 at 03:59:04PM +0000, Andrew Walrond wrote:
> 
> So how good is the device coverage offered by sysfs/udev ? Do they provide a 
> viable/complete MAKEDEV replacement yet?

It works for me on my laptop :)

You might have different devices and need other things.  If so, please
let me know after trying it out.

greg k-h

Re: State of devfs in 2.6?

[Greg KH]

On Mon, Dec 08, 2003 at 03:36:26PM +0000, Andrew Walrond wrote:
> Whats the general feeling about devfs now? I remember Christoph and others 
> making some nasty remarks about it 6months ago or so, but later noted 
> christoph doing some slashing and burning thereof.
> 
> Is it 'nice' yet? 

The kernel code is nicer, but is is marked OBSOLETE.  It also has
insolvalble race conditions that have been detailed many times here on
lkml in the past.  Please search the archives for more info.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Greg KH]

On Tue, Dec 09, 2003 at 12:04:08AM +0100, Andreas Jellinghaus wrote:
> On Mon, 08 Dec 2003 15:50:45 +0000, William Lee Irwin III wrote:
> > I would say it's deprecated at the very least. sysfs and udev are
> > supposed to provide equivalent functionality, albeit by a somewhat
> > different mechanism.
> 
> huh?
> 
> aj@simulacron:/dev$ find -type c -mount |grep -v pty |wc -l
>     164
> aj@simulacron:/dev$ find -type b |wc -l
>     157
> aj@simulacron:/dev$ find /sys/ -name dev |wc -l
>     250
> 
> After ignoring .devfsd we are left with 70 devices missing:
>  - 15 floppy devices

You have 15 floppy devices connected to your box?  All floppy devices
should show up in /sys/block.

>  - 5 input/ devices

Patch for sysfs support for this has been posted by Hanna Linder.  It
still needs work before being added to the kernel tree.

>  - full, kmem, kmsg, mem, null, port, random, urandom, zero

Patch for this has been posted by me to lkml in the past.  It will
probably go into 2.6.1

>  - printers/0 

Hanna Linder is working on a patch for these devices.

>  - 5 misc/ devices

Patch for this has been posted by me to lkml in the past.  It will
probably go into 2.6.1.

>  - 12 snd/ devices
>  - 5 sound/ devices

I have a patch here from Leann Ogasawara that adds sysfs support for
these devices.  I've been lacking time to test it better, but again, it
will probably make it into 2.6.1.

>  - 18 vcc/ devices

Hm, good catch.  I wonder why these aren't getting picked up in
/sys/class/tty as they are tty devices.  I thought they used to be
there...

> I wouldn't call udev deprecated, unless a newer kernel has the
> essential devices, too.

You mean s/udev/devfs/ right?  :)

> And is there a udev version that can
> do devfs names? last time I checked only lanana names were supported.

There is a udev config file that was just posted to linux-hotplug-devel
that supports a lot of devfs names.  If there are any missing that you
use, please post a config file for them.

Remember, I don't use devfs, so I really don't care about a udev mapping
for it :)

> Some distributions were quite happy to move from /dev and lanana to
> devfs with better names.

Hm, 2?  And one of them (Mandrake) got smart and went back...

> I doubt everyone will rush to udev with lanana names,

Why not?  It's the standard afterall.  Remember, the devfs users are in
the tiny minority here.

> and
> re-introducing makedev for devices not represented
> in sysfs doesn't sound very nice either. So 2.8.* might be a nice time
> frame for dropping devfs, or at least give sysfs and udev a few months
> to catch up on the issues mentioned.

Regardless of the state of udev, devfs has insolvable problems and you
should not use it.  End of story.

thanks,

greg k-h

Re: State of devfs in 2.6?

[Andreas Jellinghaus]

On Mon, 08 Dec 2003 15:50:45 +0000, William Lee Irwin III wrote:
> I would say it's deprecated at the very least. sysfs and udev are
> supposed to provide equivalent functionality, albeit by a somewhat
> different mechanism.

huh?

aj@simulacron:/dev$ find -type c -mount |grep -v pty |wc -l
    164
aj@simulacron:/dev$ find -type b |wc -l
    157
aj@simulacron:/dev$ find /sys/ -name dev |wc -l
    250

After ignoring .devfsd we are left with 70 devices missing:
 - 15 floppy devices
 - 5 input/ devices
 - full, kmem, kmsg, mem, null, port, random, urandom, zero
 - printers/0 
 - 5 misc/ devices
 - 12 snd/ devices
 - 5 sound/ devices
 - 18 vcc/ devices

I wouldn't call udev deprecated, unless a newer kernel has the
essential devices, too. And is there a udev version that can
do devfs names? last time I checked only lanana names were supported.

Some distributions were quite happy to move from /dev and lanana to
devfs with better names. I doubt everyone will rush to udev with
lanana names, and re-introducing makedev for devices not represented
in sysfs doesn't sound very nice either. So 2.8.* might be a nice time
frame for dropping devfs, or at least give sysfs and udev a few months
to catch up on the issues mentioned.

Andreas

Re: State of devfs in 2.6?

[Andrew Walrond]

On Monday 08 Dec 2003 3:42 pm, William Lee Irwin III wrote:
>
> I would say it's deprecated at the very least. sysfs and udev are
> supposed to provide equivalent functionality, albeit by a somewhat
> different mechanism.
>

Thanks for the pointer. 

So how good is the device coverage offered by sysfs/udev ? Do they provide a 
viable/complete MAKEDEV replacement yet?

Re: State of devfs in 2.6?

[William Lee Irwin III]

On Mon, Dec 08, 2003 at 03:36:26PM +0000, Andrew Walrond wrote:
> Whats the general feeling about devfs now? I remember Christoph and others 
> making some nasty remarks about it 6months ago or so, but later noted 
> christoph doing some slashing and burning thereof.
> Is it 'nice' yet? 
> Andrew Walrond

I would say it's deprecated at the very least. sysfs and udev are
supposed to provide equivalent functionality, albeit by a somewhat
different mechanism.


-- wli

State of devfs in 2.6?

[Andrew Walrond]

Whats the general feeling about devfs now? I remember Christoph and others 
making some nasty remarks about it 6months ago or so, but later noted 
christoph doing some slashing and burning thereof.

Is it 'nice' yet? 

Andrew Walrond