Re: [PATCH v2] tpm: enable PPI for TPM 2.0

[Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>]

On Tue, Apr 21, 2015 at 01:33:38PM -0600, Jason Gunthorpe wrote:
> On Tue, Apr 21, 2015 at 09:30:55PM +0300, Jarkko Sakkinen wrote:
> > Enabled PPI interface to the character device sysfs directory accessible
> > both for 1.x and 2.0 devices.
> > 
> > The ppi group is moved from the platform device directory to the
> > character device directory. In order to retain backwards compatibility
> > with the 1.x devices, a symlink is created to the platform device
> > directory.
> > 
> > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
> 
> > Reviewed-by: Jason Gunthorpe <jason.gunthorpe@obsidianresearch.com>
> 
> Jumping the gun a bit, there :)
> But yes, the TPM bits still look OK to me.
> Ah, don't forget to update Documentation/ABI/testing/sysfs-driver-ppi

Right. Is there a rigid naming convention for these files? I don't see
any obvious pattern and there's no documentation for file names in here:

https://www.kernel.org/doc/Documentation/ABI/README

The legacy TPM 1.x sysfs attributes are in a file named as sysfs-class-tpm.

Maybe we could:

* Rename existing sysfs-class-tpm as sysfs-class-tpm1
* Rename sysfs-driver-ppi as sysfs-class-tpm

What do you think? I don't know exactly what to do here. That is the
reason I haven't updated the documentation yet.

> > diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c
> > index 6acc964..749cea3 100644
> > +++ b/fs/kernfs/dir.c
> > @@ -1225,6 +1225,7 @@ int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name,
> >  	else
> >  		return -ENOENT;
> >  }
> > +EXPORT_SYMBOL_GPL(kernfs_remove_by_name_ns);
> 
> ??
> I don't see this being called

My bad, it is clutter in the patch file. I'll remove it.

> > +/**
> > + * sysfs_link_group_to_kobj - add a symlink to a kobject pointing to a group
> > + * @kobj:		The kobject containing the group.
> > + * @target_kobj:	The target kobject.
> > + * @target_name:	Name of the target group.
> > + */
> > +int sysfs_link_group_to_kobj(struct kobject *kobj, struct kobject *target_kobj,
> > +			     const char *target_name)
> 
> Agree with Guenter, separate patch.
> 
> I liked the signature I suggested earlier:
> 
> sysfs_link_group_to_kobj(struct kobject *from_kobj,
> 			 const char *from_name,
>                          struct kobject *target_kobj,
>                          const char *target_name)
> 
> 'link_group' may be too specific a name, I think this would work to
> establish a link to any file in a sysfs directory? Which is good,
> we'll need that for future TPM patches that relocate the other sysfs
> files..

I think it should.

> Jason.

/Jarkko