[PATCH 0/8] staging: unisys: visorchipset proc fixes

Greg KH gregkh at linuxfoundation.org
Sat Jul 19 19:55:36 UTC 2014


On Sat, Jul 19, 2014 at 02:26:48PM -0400, Ben Romer wrote:
> Greg KH <gregkh at linuxfoundation.org> wrote:
> 
> > What happened to the idea of only creating the sysfs files _if_ it is
> > needed?  You are always creating these files, and then can return
> > -ENODEV if the device really isn't there, that's not what you should do
> > for a sysfs file.  If the file is present, it should return data, not
> > return an error.  If the device isn't there, just don't create the file.
> 
> Greg,
> 
> I submitted a set of patches before this set that does just that. I moved the
> controlvm channel function into visorchipset_main.c and removed the old files,
> and made it so that if the channel is not present the module wouldn't load. I
> also removed all the code that returns ENODEV, except for the module init
> function, where it gets returned if there's no controlvm channel present.
> 
> I could change that to some other error, or let the module load and then not
> create files if the channel isn't present, if you'd prefer that? But if the
> module doesn't load, the files in sys don't get created, so I thought that
> would be a good solution.
> 
> The commit numbers were 524b0b6 for the controlvm channel function, and
> 8a1182e for the extraneous checks and ENODEV errors being removed.

See my comments in your patch 1/8 for the specifics as to why I didn't
know that you have done this work (remember, I deal with hundreds of
patches a week and have no short term memory...)

greg k-h


More information about the devel mailing list