Re: [PATCH 5/5] nvmem: Add support for write-only instances

[Greg KH]

On Tue, Mar 24, 2020 at 02:24:21PM +0000, Nicholas Johnson wrote:
> On Tue, Mar 24, 2020 at 01:25:46PM +0000, Srinivas Kandagatla wrote:
> > 
> > 
> > On 24/03/2020 12:29, Greg KH wrote:
> > > > But the Idea here is :
> > > > We ended up with providing different options like read-only,root-only to
> > > > nvmem providers combined with read/write callbacks.
> > > > With that, there are some cases which are totally invalid, existing code
> > > > does very minimal check to ensure that before populating with correct
> > > > attributes to sysfs file. One of such case is with thunderbolt provider
> > > > which supports only write callback.
> > > > 
> > > > With this new checks in place these flags and callbacks are correctly
> > > > validated, would result in correct file attributes.
> > > Why this crazy set of different groups?  You can set the mode of a sysfs
> > > file in the callback for when the file is about to be created, that's so
> > > much simpler and is what it is for.  This feels really hacky and almost
> > > impossible to follow:(
> > Thanks for the inputs, That definitely sounds much simpler to deal with.
> > 
> > Am guessing you are referring to is_bin_visible callback?
> > 
> > I will try to clean this up!
> I am still onboard and willing do the work, but we may need to discuss
> to be on the same page with new plans. How do you wish to do this?
> 
> Does this new approach still allow us to abort if we receive an invalid
> configuration? Or do we still need to have something in nvmem_register()
> to abort in invalid case?
> 
> The documentation of is_bin_visible says only read/write permissions are 
> accepted. Does this mean that it will not take read-only or write-only? 
> That is one way of interpreting it.

That's a funny way of interpreting it :)

Please be sane, you pass back the permissions of the file, look at all
of the places in the kernel is it used for examples...

thanks,

greg k-h