debugfs directory structure

debugfs directory structure

[Roland Dreier]

Hi Greg,

Now that debugfs is merged into Linus's tree, I'm looking at using it
to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).  Is
there any guidance on what the structure of debugfs should look like?
Right now I'm planning on putting all the debug info files under an
ipoib/ top level directory.  Does that sound reasonable?

Thanks,
  Roland

Re: debugfs directory structure

[Greg KH]

On Wed, Jan 12, 2005 at 12:50:51PM -0800, Roland Dreier wrote:
> Hi Greg,
> 
> Now that debugfs is merged into Linus's tree, I'm looking at using it
> to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).

Great!

> Is
> there any guidance on what the structure of debugfs should look like?
> Right now I'm planning on putting all the debug info files under an
> ipoib/ top level directory.  Does that sound reasonable?

Anarchy rules in debugfs.  Do what you want.  If you stomp over someone
else's stuff, I expect complaints and maybe someone will have to
arbitrate, but odds are that will ever happen is pretty slim.

So yes, ipoib/ in the top level sounds just fine.

thanks,

greg k-h

Re: debugfs directory structure

[Dave Jones]

On Wed, Jan 12, 2005 at 12:50:51PM -0800, Roland Dreier wrote:
 > Hi Greg,
 > 
 > Now that debugfs is merged into Linus's tree, I'm looking at using it
 > to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).  Is
 > there any guidance on what the structure of debugfs should look like?
 > Right now I'm planning on putting all the debug info files under an
 > ipoib/ top level directory.  Does that sound reasonable?

How about mirroring the toplevel kernel source structure ?

Ie, you'd make drivers/infiniband/ulp/ipoib ?

It could get ugly quickly without some structure at least to
the toplevel dir.

		Dave

Re: debugfs directory structure

[Greg KH]

On Wed, Jan 12, 2005 at 04:09:45PM -0500, Dave Jones wrote:
> On Wed, Jan 12, 2005 at 12:50:51PM -0800, Roland Dreier wrote:
>  > Hi Greg,
>  > 
>  > Now that debugfs is merged into Linus's tree, I'm looking at using it
>  > to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).  Is
>  > there any guidance on what the structure of debugfs should look like?
>  > Right now I'm planning on putting all the debug info files under an
>  > ipoib/ top level directory.  Does that sound reasonable?
> 
> How about mirroring the toplevel kernel source structure ?
> 
> Ie, you'd make drivers/infiniband/ulp/ipoib ?

But who would be in charge of createing the "drivers/" subdirectory?

debugfs can't handle "/" in a directory name, like procfs does.

> It could get ugly quickly without some structure at least to
> the toplevel dir.

I say ipoib/ is fine, remember, this is for debugging stuff, it will
quickly get ugly anyway :)

thanks,

greg k-h

Re: debugfs directory structure

[Dave Jones]

On Wed, Jan 12, 2005 at 01:41:08PM -0800, Greg KH wrote:
 > On Wed, Jan 12, 2005 at 04:09:45PM -0500, Dave Jones wrote:
 > > On Wed, Jan 12, 2005 at 12:50:51PM -0800, Roland Dreier wrote:
 > >  > Hi Greg,
 > >  > 
 > >  > Now that debugfs is merged into Linus's tree, I'm looking at using it
 > >  > to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).  Is
 > >  > there any guidance on what the structure of debugfs should look like?
 > >  > Right now I'm planning on putting all the debug info files under an
 > >  > ipoib/ top level directory.  Does that sound reasonable?
 > > 
 > > How about mirroring the toplevel kernel source structure ?
 > > 
 > > Ie, you'd make drivers/infiniband/ulp/ipoib ?
 > 
 > But who would be in charge of createing the "drivers/" subdirectory?
 > debugfs can't handle "/" in a directory name, like procfs does.

maybe it should ?

 > > It could get ugly quickly without some structure at least to
 > > the toplevel dir.
 > I say ipoib/ is fine, remember, this is for debugging stuff, it will
 > quickly get ugly anyway :)

with no heirarchy, what happens when two drivers want to make
the same directory / filenames ?

		Dave

Re: debugfs directory structure

[Greg KH]

On Wed, Jan 12, 2005 at 05:01:43PM -0500, Dave Jones wrote:
> On Wed, Jan 12, 2005 at 01:41:08PM -0800, Greg KH wrote:
>  > On Wed, Jan 12, 2005 at 04:09:45PM -0500, Dave Jones wrote:
>  > > On Wed, Jan 12, 2005 at 12:50:51PM -0800, Roland Dreier wrote:
>  > >  > Hi Greg,
>  > >  > 
>  > >  > Now that debugfs is merged into Linus's tree, I'm looking at using it
>  > >  > to replace the IPoIB debugging pseudo-filesystem (ipoib_debugfs).  Is
>  > >  > there any guidance on what the structure of debugfs should look like?
>  > >  > Right now I'm planning on putting all the debug info files under an
>  > >  > ipoib/ top level directory.  Does that sound reasonable?
>  > > 
>  > > How about mirroring the toplevel kernel source structure ?
>  > > 
>  > > Ie, you'd make drivers/infiniband/ulp/ipoib ?
>  > 
>  > But who would be in charge of createing the "drivers/" subdirectory?
>  > debugfs can't handle "/" in a directory name, like procfs does.
> 
> maybe it should ?

Right now debugfs is dentry based, not string based.  If someone wants
to send me patches to change it, I'll reconsider it :)

>  > > It could get ugly quickly without some structure at least to
>  > > the toplevel dir.
>  > I say ipoib/ is fine, remember, this is for debugging stuff, it will
>  > quickly get ugly anyway :)
> 
> with no heirarchy, what happens when two drivers want to make
> the same directory / filenames ?

The second call will fail.  Code should always check return values,
right?

thanks,

greg k-h