debugfs in the namespace

debugfs in the namespace

[Pete Zaitcev]

Hi Greg,

what is the canonic place to mount debugfs: /debug, /debugfs, or anything
else? The reason I'm asking is that USBMon has to find it somewhere and
I'd really hate to see it varying from distro to distro.

Thanks,
-- Pete

Re: debugfs in the namespace

[Greg KH]

On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> Hi Greg,
> 
> what is the canonic place to mount debugfs: /debug, /debugfs, or anything
> else? The reason I'm asking is that USBMon has to find it somewhere and
> I'd really hate to see it varying from distro to distro.

Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)

Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?

What do people want?  I guess it's time to write up a LSB proposal :(

thanks,

greg k-h

Re: debugfs in the namespace

[Pete Zaitcev]

On Thu, 16 Dec 2004 11:08:35 -0800, Greg KH <greg@kroah.com> wrote:

> On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> > what is the canonic place to mount debugfs: /debug, /debugfs, or anything
> > else? The reason I'm asking is that USBMon has to find it somewhere and
> > I'd really hate to see it varying from distro to distro.
> 
> Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)

Oh, that's right: usr and creat. How could I forget.

> Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?
> 
> What do people want?  I guess it's time to write up a LSB proposal :(

I use /debug but it's not too late to change. Fedora does not ship it yet,
so I don't think we have an institutional opinion about it.

Personally, I'm against the doubles to prevent issues with the mounting
order on boot, but that's rather weak. The /dev can be specially managed
and I'm concerned with people running find(1) on it. The /proc sounds
better, but mounting anything under /proc requires a kernel component
to create a directory, does it not?

[root@lembas root]# mount -t debugfs any /proc/debug
mount: mount point /proc/debug does not exist
[root@lembas root]# mkdir /proc/debug
mkdir: cannot create directory `/proc/debug': No such file or directory
[root@lembas root]# 

-- Pete

Re: debugfs in the namespace

[Greg KH]

On Thu, Dec 16, 2004 at 11:33:57AM -0800, Pete Zaitcev wrote:
> On Thu, 16 Dec 2004 11:08:35 -0800, Greg KH <greg@kroah.com> wrote:
> 
> > On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> > > what is the canonic place to mount debugfs: /debug, /debugfs, or anything
> > > else? The reason I'm asking is that USBMon has to find it somewhere and
> > > I'd really hate to see it varying from distro to distro.
> > 
> > Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)
> 
> Oh, that's right: usr and creat. How could I forget.
> 
> > Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?
> > 
> > What do people want?  I guess it's time to write up a LSB proposal :(
> 
> I use /debug but it's not too late to change. Fedora does not ship it yet,
> so I don't think we have an institutional opinion about it.
> 
> Personally, I'm against the doubles to prevent issues with the mounting
> order on boot, but that's rather weak. The /dev can be specially managed
> and I'm concerned with people running find(1) on it. The /proc sounds
> better, but mounting anything under /proc requires a kernel component
> to create a directory, does it not?

Yes it does, but debugfs could create the mount point, if people agree
that this is a good place to put it (like usbfs does.)

Personally, I don't want to put it there, but that's just because I hate
proc stuff :)

So, /debug sounds good to me.  Any objections?

thanks,

greg k-h

Re: debugfs in the namespace

[Jan Engelhardt]

>> Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?
>> What do people want?  I guess it's time to write up a LSB proposal :(
>
>The /proc sounds
>better, but mounting anything under /proc requires a kernel component
>to create a directory, does it not?

And in case you debug procfs itself, you usually should not or even can't 
mount it into /proc, e.g. /proc is currently being initialised, but debugfs 
values already spit out.

/.debugfs/ as my choice, to not pollute/fill the root directory that much :)



Jan Engelhardt
-- 
ENOSPC

Re: debugfs in the namespace

[Mike Waychison]

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

Greg KH wrote:
> On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> 
>>Hi Greg,
>>
>>what is the canonic place to mount debugfs: /debug, /debugfs, or anything
>>else? The reason I'm asking is that USBMon has to find it somewhere and
>>I'd really hate to see it varying from distro to distro.
> 
> 
> Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)
> 
> Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?
> 
> What do people want?  I guess it's time to write up a LSB proposal :(
> 

I thought debugfs was meant for just debugging.  As there is no plans
for standardizing its namespace, why are we allowing ourselves to rely
on it being mounted at all?

AFAICT, there should be no excuse for userspace to actually rely on any
of the data within debugfs.  Otherwise we end up with yet another
filesystem whose role is: Chaotic hodgepodge of magic files created by
drivers that couldn't bother to be well-organized.

Please, let's not make debugfs part of userspace.  Keep it for what it
is, debugging purposes only.


- --
Mike Waychison
Sun Microsystems, Inc.
1 (650) 352-5299 voice
1 (416) 202-8336 voice

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
NOTICE:  The opinions expressed in this email are held by me,
and may not represent the views of Sun Microsystems, Inc.
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.5 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFBwgNWdQs4kOxk3/MRAslqAJwPnra30/EBuZxuXkdpo67SZJXJUQCaAtC1
OeMH0Xiww/8xV9tIfqyzmE4=
=Aa8I
-----END PGP SIGNATURE-----

Re: debugfs in the namespace

[Greg KH]

On Thu, Dec 16, 2004 at 04:51:18PM -0500, Mike Waychison wrote:
> I thought debugfs was meant for just debugging.  As there is no plans
> for standardizing its namespace, why are we allowing ourselves to rely
> on it being mounted at all?
> 
> AFAICT, there should be no excuse for userspace to actually rely on any
> of the data within debugfs.  Otherwise we end up with yet another
> filesystem whose role is: Chaotic hodgepodge of magic files created by
> drivers that couldn't bother to be well-organized.
> 
> Please, let's not make debugfs part of userspace.  Keep it for what it
> is, debugging purposes only.

I'm not saying we will ever make it "required" at all.  It's just that
people are going to want to mount the thing, and are already asking me
where we should mount it at.  If you pick a different place than me,
fine, I don't mind.  It's the user who is asked to report some info that
happens to be in debugfs that is going to want to know where to put it,
as they have no idea even what it is.  Distros are going to ask what to
put in their fstabs for where to mount the thing too.

So, let's pick a place and be done with it.

I like /dbg (3 characters total to get to, which is shorter than /debug
which takes at least 4, 3 chars and a tab).  Pete likes /debug.  Jan
Engelhardt want to hide the thing from people at /.debugfs.

Hm, what about /.debug ?  That's a compromise that I can live with (even
less key strokes to get to...)

Or is their some restriction on putting hidden directories in the root
filesystem as specified by the LSB?

So, /.debug sound acceptable?

thanks,

greg k-h

Re: debugfs in the namespace

[Pete Zaitcev]

On Thu, 16 Dec 2004 14:18:43 -0800, Greg KH <greg@kroah.com> wrote:

> Hm, what about /.debug ?  That's a compromise that I can live with (even
> less key strokes to get to...)

No way, Jan is out of his mind, adding obfuscations like that. Anything
but that. I didn't even bother to reply, because it never occurred to me
that you'd fall for something so retarded.

Otherwise, /dbg sounds good.

Mike's objections sound philosophically congenial to me. What I'm trying
to have here is to support an equivalent of tcpdump, which some may consider
a core function rather than a debugging function. Of course, I could easily
say "this is for debugging only" and thus deflect Mike, but this is not
about winning, and actually I have no investment in any approach. For me the
/sys is obviously out because of the "one file one value" doctrine. The /proc
sounds attractive, but programming procfs is such a bother. If we had a debugfs
style API to procfs, that would be the winner from the standpoint of this
application. Failing that, I guess, it's /dbg.

-- Pete

Re: debugfs in the namespace

[Greg KH]

On Thu, Dec 16, 2004 at 02:45:31PM -0800, Pete Zaitcev wrote:
> On Thu, 16 Dec 2004 14:18:43 -0800, Greg KH <greg@kroah.com> wrote:
> 
> > Hm, what about /.debug ?  That's a compromise that I can live with (even
> > less key strokes to get to...)
> 
> No way, Jan is out of his mind, adding obfuscations like that. Anything
> but that. I didn't even bother to reply, because it never occurred to me
> that you'd fall for something so retarded.

Bah, fine :)

> Otherwise, /dbg sounds good.

Ok, I can live with that.

> Mike's objections sound philosophically congenial to me. What I'm trying
> to have here is to support an equivalent of tcpdump, which some may consider
> a core function rather than a debugging function. Of course, I could easily
> say "this is for debugging only" and thus deflect Mike, but this is not
> about winning, and actually I have no investment in any approach. For me the
> /sys is obviously out because of the "one file one value" doctrine. The /proc
> sounds attractive, but programming procfs is such a bother. If we had a debugfs
> style API to procfs, that would be the winner from the standpoint of this
> application. Failing that, I guess, it's /dbg.

Yes, usb data dumping could go in /dbg, it makes sense.

thanks,

greg k-h

Re: debugfs in the namespace

[Bernd Eckenfels]

In article <20041216144531.3a8d988c@lembas.zaitcev.lan> you wrote:
> No way, Jan is out of his mind, adding obfuscations like that.
...
> Otherwise, /dbg sounds good.

I dont think that a root level directory, especially with an unreadable name
is a good idea. Why dont we at least try to keep the  namespace clean?

Greetings
Bernd

Re: debugfs in the namespace

[Pedro Venda (SYSADM)]

Greg KH wrote:
> On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> 
>>Hi Greg,
>>
>>what is the canonic place to mount debugfs: /debug, /debugfs, or anything
>>else? The reason I'm asking is that USBMon has to find it somewhere and
>>I'd really hate to see it varying from distro to distro.
> 
> 
> Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)
> 
> Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?

well, since there is already a /sys, /proc... /debug wouldn't be that bad. 
perhaps the /.debug is better to keep it simple and away from people's eyes.

regards,
pedro venda.

-- 

Pedro João Lopes Venda
email: pjvenda@rnl.ist.utl.pt
http://maxwell.rnl.ist.utl.pt

Equipa de Administração de Sistemas
Rede das Novas Licenciaturas (RNL)
Instituto Superior Técnico
http://www.rnl.ist.utl.pt
http://mega.ist.utl.pt

Re: debugfs in the namespace

[Grzegorz Kulewski]

On Thu, 16 Dec 2004, Greg KH wrote:

> On Thu, Dec 16, 2004 at 02:45:31PM -0800, Pete Zaitcev wrote:
>> On Thu, 16 Dec 2004 14:18:43 -0800, Greg KH <greg@kroah.com> wrote:
>>
>>> Hm, what about /.debug ?  That's a compromise that I can live with (even
>>> less key strokes to get to...)
>>
>> No way, Jan is out of his mind, adding obfuscations like that. Anything
>> but that. I didn't even bother to reply, because it never occurred to me
>> that you'd fall for something so retarded.
>
> Bah, fine :)
>
>> Otherwise, /dbg sounds good.
>
> Ok, I can live with that.

I agree that anything like /.* is broken and strange... But this is only 
my little opinion. :-)

But why creating dir in /proc - like /proc/debug is bad? Its advantages:
- it does not pollute namespace,
- it can be created by kernel at startup even on systems where debugfs 
will not be used (why not?),
- /proc is mounted in all configurations and often it is the first thing 
that startscripts do,
- if somebody really needs to debug proc using debugfs he can always mount 
it as /debug temporaily.


Thanks,

Grzegorz Kulewski

Re: debugfs in the namespace

[Greg KH]

On Fri, Dec 17, 2004 at 12:39:00AM +0100, Grzegorz Kulewski wrote:
> On Thu, 16 Dec 2004, Greg KH wrote:
> 
> >On Thu, Dec 16, 2004 at 02:45:31PM -0800, Pete Zaitcev wrote:
> >>On Thu, 16 Dec 2004 14:18:43 -0800, Greg KH <greg@kroah.com> wrote:
> >>
> >>>Hm, what about /.debug ?  That's a compromise that I can live with (even
> >>>less key strokes to get to...)
> >>
> >>No way, Jan is out of his mind, adding obfuscations like that. Anything
> >>but that. I didn't even bother to reply, because it never occurred to me
> >>that you'd fall for something so retarded.
> >
> >Bah, fine :)
> >
> >>Otherwise, /dbg sounds good.
> >
> >Ok, I can live with that.
> 
> I agree that anything like /.* is broken and strange... But this is only 
> my little opinion. :-)
> 
> But why creating dir in /proc - like /proc/debug is bad? Its advantages:
> - it does not pollute namespace,
> - it can be created by kernel at startup even on systems where debugfs 
> will not be used (why not?),
> - /proc is mounted in all configurations and often it is the first thing 
> that startscripts do,
> - if somebody really needs to debug proc using debugfs he can always mount 
> it as /debug temporaily.

Disadvantage:
 - it puts a non-process type thing into /proc which is what I am
   specifically trying to get away from doing.

Only process related things _should_ be in /proc.  Now if I can ever
fully achieve that goal in my lifetime is something that is left to be
seen...

thanks,

greg k-h

Re: debugfs in the namespace

[Grzegorz Kulewski]

On Thu, 16 Dec 2004, Greg KH wrote:
> Disadvantage:
> - it puts a non-process type thing into /proc which is what I am
>   specifically trying to get away from doing.
>
> Only process related things _should_ be in /proc.  Now if I can ever
> fully achieve that goal in my lifetime is something that is left to be
> seen...

Ok, probably - but who said this? IIRC there is no standard describing 
what should be in proc and what should not. I do not think that after so 
many years of storing everyting in /proc there is any chance that you will 
remove all not [0-9]* dirs and files from /proc before the sun will stop 
lighting... :-) Many, really _many_, binary only apps are using proc for 
999999 different (often stupid) things - how you will change that? There 
is way too late for such change, in my opinion.

And polluting / with proc, sys, dev, selinux, debug and who knows what 
else is at least equally bad.


Thanks,

Grzegorz Kulewski

Re: debugfs in the namespace

[Greg KH]

On Fri, Dec 17, 2004 at 01:08:05AM +0100, Grzegorz Kulewski wrote:
> On Thu, 16 Dec 2004, Greg KH wrote:
> >Disadvantage:
> >- it puts a non-process type thing into /proc which is what I am
> >  specifically trying to get away from doing.
> >
> >Only process related things _should_ be in /proc.  Now if I can ever
> >fully achieve that goal in my lifetime is something that is left to be
> >seen...
> 
> Ok, probably - but who said this?

Well, if you can't find it written down anymore, here, I'll give you a
quote:
	"/proc should only be for process related things."
			- Greg Kroah-Hartman, December 16, 2004

> IIRC there is no standard describing what should be in proc and what
> should not.

We have one now, see above :)

> I do not think that after so many years of storing everyting in /proc
> there is any chance that you will remove all not [0-9]* dirs and files
> from /proc before the sun will stop lighting... :-)

Hey, everyone needs a windmill to chase after to define their life's
purpose...

> Many, really _many_, binary only apps are using proc for 999999
> different (often stupid) things

Yes, I'm well aware of the /proc/cpu abusing that is done by many binary
programs (oracle, jvm, etc.)  But it's the other things that definitely
don't belong in there (like the /proc/drivers/ tree stuff) that should
get moved out over time.  What do you think I've been doing over the
past 3 years with sysfs...

> - how you will change that? There is way too late for such change, in
> my opinion.

Slowly, over time, with creating good, standardised things like sysfs
(which has order to it), and fun, anarchistic things like debugfs.
Between the combination of the two, we'll get closer to an ideal, which
is better than what we have today.

> And polluting / with proc, sys, dev, selinux, debug and who knows what 
> else is at least equally bad.

Why?  Each location is defined to have one, specific, well defined thing
there that people can count on (or not count on in the case of /debug.)

thanks,

greg k-h

Re: debugfs in the namespace

[Grzegorz Kulewski]

On Thu, 16 Dec 2004, Greg KH wrote:

> On Fri, Dec 17, 2004 at 01:08:05AM +0100, Grzegorz Kulewski wrote:
>> On Thu, 16 Dec 2004, Greg KH wrote:
>>> Disadvantage:
>>> - it puts a non-process type thing into /proc which is what I am
>>>  specifically trying to get away from doing.
>>>
>>> Only process related things _should_ be in /proc.  Now if I can ever
>>> fully achieve that goal in my lifetime is something that is left to be
>>> seen...
>>
>> Ok, probably - but who said this?
>
> Well, if you can't find it written down anymore, here, I'll give you a
> quote:
> 	"/proc should only be for process related things."
> 			- Greg Kroah-Hartman, December 16, 2004
>
>> IIRC there is no standard describing what should be in proc and what
>> should not.
>
> We have one now, see above :)

Thanks, I needed that! :)


>> I do not think that after so many years of storing everyting in /proc
>> there is any chance that you will remove all not [0-9]* dirs and files
>> from /proc before the sun will stop lighting... :-)
>
> Hey, everyone needs a windmill to chase after to define their life's
> purpose...

Ok, you are kernel and Gentoo developer - you are right by definition. :)


>> And polluting / with proc, sys, dev, selinux, debug and who knows what
>> else is at least equally bad.
>
> Why?  Each location is defined to have one, specific, well defined thing
> there that people can count on (or not count on in the case of /debug.)

Because in short time we will end with / occupying >1 page of console - 
and it will be bad in my opinion. Besides do we really need that many 
fses - each for exporting kernel data to userspace? This is at least 
strange. Why can not /dev, /selinux be merged into /sys (ok, maybe there 
should be symlinks in /dev to devices in right device directory in /sys). 
Why debugfs can not live in /sys - for example in /sys/debug? What is 
wrong with one value per file and debugfs?

And I have also other question: Where can I find some info about using 
/sys (in kernel) and some small note about its implementation and overhead 
(cpu and memory)?


Thanks,

Grzegorz Kulewski

Re: debugfs in the namespace

[Greg KH]

On Fri, Dec 17, 2004 at 02:15:22AM +0100, Grzegorz Kulewski wrote:
> >>And polluting / with proc, sys, dev, selinux, debug and who knows what
> >>else is at least equally bad.
> >
> >Why?  Each location is defined to have one, specific, well defined thing
> >there that people can count on (or not count on in the case of /debug.)
> 
> Because in short time we will end with / occupying >1 page of console - 
> and it will be bad in my opinion. Besides do we really need that many 
> fses - each for exporting kernel data to userspace?

Yes, we do.

> This is at least strange. Why can not /dev, /selinux be merged into
> /sys (ok, maybe there should be symlinks in /dev to devices in right
> device directory in /sys). 

Because all three of them are radically different things.

> And I have also other question: Where can I find some info about using 
> /sys (in kernel)

Documentation/driver-model/* is a good start.  It's a bit out of date,
but better than nothing.  The lwn.net series of articles is also good to
look at.

> and some small note about its implementation and overhead (cpu and
> memory)?

For that you will have to look at the code itself.

Good luck,

greg k-h

Re: debugfs in the namespace

[Phil Lougher]

On Fri, 17 Dec 2004 00:21:34 +0100, Bernd Eckenfels
<ecki-news2004-05@lina.inka.de> wrote:
> In article <20041216144531.3a8d988c@lembas.zaitcev.lan> you wrote:
> > Otherwise, /dbg sounds good.
>
> I dont think that a root level directory, especially with an unreadable name
> is a good idea. Why dont we at least try to keep the  namespace clean?

Are you suggesting we should rename "etc", "mnt" etc? :-)  I like
"/dbg" it follows the gdb, kgdb naming convention and it was the Unix
way to name things like this.  Though perhaps debugfs should have been
named dbgfs in this case...

I don't like "/.debug", hiding it in way this implies that you don't
think it should be there (and so you've hidden it).  A properly
decided upon mount point shouldn't have these connotations?  If you're
using debugfs I think you should want to have the mount point visible.

Phillip

Re: debugfs in the namespace

[H. Peter Anvin]

Followup to:  <20041216110002.3e0ddf52@lembas.zaitcev.lan>
By author:    Pete Zaitcev <zaitcev@redhat.com>
In newsgroup: linux.dev.kernel
>
> Hi Greg,
> 
> what is the canonic place to mount debugfs: /debug, /debugfs, or anything
> else? The reason I'm asking is that USBMon has to find it somewhere and
> I'd really hate to see it varying from distro to distro.
> 
> Thanks,
> -- Pete
> 

/dev is the normal place where stuff related to the user/kernel
interface should go, so /dev/debug or /dev/debugfs would be the right
answer (and yes, /proc and /sys violate this, but we're stuck with
those.)

	-hpa

Re: debugfs in the namespace

[Jan Engelhardt]

>So, let's pick a place and be done with it.
>
>I like /dbg (3 characters total to get to, which is shorter than /debug
>which takes at least 4, 3 chars and a tab).  Pete likes /debug.  Jan
>Engelhardt want to hide the thing from people at /.debugfs.
>
>Hm, what about /.debug ?  That's a compromise that I can live with (even
>less key strokes to get to...)

Anything is acceptable at least if my dot is there :)

>So, /.debug sound acceptable?

No objections.

>

Jan Engelhardt
-- 
ENOSPC

Re: debugfs in the namespace

[Jan Engelhardt]

> But why creating dir in /proc - like /proc/debug is bad? Its advantages:

Like I said, /proc would need to be mounted first -- which is probably not the 
case right away when you boot with "-b" or "init=/bin/bash"



Jan Engelhardt
-- 
ENOSPC

Re: debugfs in the namespace [u]

[Martin Schlemmer [c]]

[-- Attachment #1: Type: text/plain, Size: 1836 bytes --]

On Thu, 2004-12-16 at 14:18 -0800, Greg KH wrote:
> On Thu, Dec 16, 2004 at 04:51:18PM -0500, Mike Waychison wrote:
> > I thought debugfs was meant for just debugging.  As there is no plans
> > for standardizing its namespace, why are we allowing ourselves to rely
> > on it being mounted at all?
> > 
> > AFAICT, there should be no excuse for userspace to actually rely on any
> > of the data within debugfs.  Otherwise we end up with yet another
> > filesystem whose role is: Chaotic hodgepodge of magic files created by
> > drivers that couldn't bother to be well-organized.
> > 
> > Please, let's not make debugfs part of userspace.  Keep it for what it
> > is, debugging purposes only.
> 
> I'm not saying we will ever make it "required" at all.  It's just that
> people are going to want to mount the thing, and are already asking me
> where we should mount it at.  If you pick a different place than me,
> fine, I don't mind.  It's the user who is asked to report some info that
> happens to be in debugfs that is going to want to know where to put it,
> as they have no idea even what it is.  Distros are going to ask what to
> put in their fstabs for where to mount the thing too.
> 
> So, let's pick a place and be done with it.
> 
> I like /dbg (3 characters total to get to, which is shorter than /debug
> which takes at least 4, 3 chars and a tab).  Pete likes /debug.  Jan
> Engelhardt want to hide the thing from people at /.debugfs.
> 
> Hm, what about /.debug ?  That's a compromise that I can live with (even
> less key strokes to get to...)
> 
> Or is their some restriction on putting hidden directories in the root
> filesystem as specified by the LSB?
> 
> So, /.debug sound acceptable?
> 

Prob silly, but what about /sys/dbg ?  Out of sight ...


-- 
Martin Schlemmer


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: debugfs in the namespace

[John Levon]

On Fri, Dec 17, 2004 at 04:06:00AM +0000, H. Peter Anvin wrote:

> /dev is the normal place where stuff related to the user/kernel
> interface should go, so /dev/debug or /dev/debugfs would be the right
> answer (and yes, /proc and /sys violate this, but we're stuck with
> those.)

There is also /dev/oprofile/ as recent 2.6 precedent for this.

john

Re: debugfs in the namespace

[Pavel Machek]

Hi!

> > better, but mounting anything under /proc requires a kernel component
> > to create a directory, does it not?
> 
> Yes it does, but debugfs could create the mount point, if people agree
> that this is a good place to put it (like usbfs does.)
> 
> Personally, I don't want to put it there, but that's just because I hate
> proc stuff :)
> 
> So, /debug sounds good to me.  Any objections?

Yes... /debug is something users may actually use already... Like
having scratch filesystem mount on /debug.

								Pavel
-- 
People were complaining that M$ turns users into beta-testers...
...jr ghea gurz vagb qrirybcref, naq gurl frrz gb yvxr vg gung jnl!

Re: debugfs in the namespace

[Jörn Engel]

On Fri, 17 December 2004 20:08:56 +0100, Pavel Machek wrote:
> > 
> > So, /debug sounds good to me.  Any objections?
> 
> Yes... /debug is something users may actually use already... Like
> having scratch filesystem mount on /debug.

Wouldn't it be possible to parse /proc/mounts and determine the
correct mount point that way?

DEBUGFS = grep debugfs /proc/mounts | cut -f2 -d' '

Jörn

-- 
Premature optimization is the root of all evil.
-- Donald Knuth

Re: debugfs in the namespace

[Matt Mackall]

On Thu, Dec 16, 2004 at 11:08:35AM -0800, Greg KH wrote:
> On Thu, Dec 16, 2004 at 11:00:02AM -0800, Pete Zaitcev wrote:
> > Hi Greg,
> > 
> > what is the canonic place to mount debugfs: /debug, /debugfs, or anything
> > else? The reason I'm asking is that USBMon has to find it somewhere and
> > I'd really hate to see it varying from distro to distro.
> 
> Hm, in my testing I've been putting it in /dbg, but I don't like vowels :)
> 
> Anyway, I don't really know.  /dev/debug/ ?  /proc/debug ?  /debug ?
> 
> What do people want?  I guess it's time to write up a LSB proposal :(

What I'd like is to _not_ have a standard, so that no one is tempted
to depend on it being there. No one but developers should mount it by
default and no standard tools should depend on it for functionality.

If you put it in a standard place, people will start treating it as
ABI and then a year down the road we will have _three_ /proc-style
ABI barf-bags rather than just one.

-- 
Mathematics is the supreme nostalgia of our time.

Re: debugfs in the namespace

[Bodo Eggert]

Greg KH wrote:
> On Thu, Dec 16, 2004 at 04:51:18PM -0500, Mike Waychison wrote:

>> Please, let's not make debugfs part of userspace.  Keep it for what it
>> is, debugging purposes only.
> 
> I'm not saying we will ever make it "required" at all.  It's just that
> people are going to want to mount the thing, and are already asking me
> where we should mount it at.  If you pick a different place than me,
> fine, I don't mind.  It's the user who is asked to report some info that
> happens to be in debugfs that is going to want to know where to put it,
> as they have no idea even what it is.  Distros are going to ask what to
> put in their fstabs for where to mount the thing too.
> 
> So, let's pick a place and be done with it.

/mnt (users) or /var/adm/mount/debugfs (distros, asuming they mount their
temporary stuff there).

I'm asuming you should almost never need to touch devfs-entries, so if
your distro mounts it just in case, it should be out of the way.