Re: [PATCH 0/4] treewide: add missing put_device calls

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Fri, Dec 13, 2013 at 01:42:05PM -0700, Bjorn Helgaas wrote:
> [+cc Greg]
> 
> On Fri, Dec 13, 2013 at 12:22 PM, Levente Kurusa <levex@linux.com> wrote:
> > Hi,
> >
> > This is just the beginning of patchset-set that aims to fix possible
> > problems caused by not calling put_device() if device_register() fails.
> >
> > The root cause for the need to call put_device() is that the underlying
> > kobject still has a reference count of 1. Thus, device.release() will not
> > be called and the device will just sit there waiting for a put_device().
> > Adding the put_device() also removes the need for the call to kfree() as most
> > release functions already call kfree() on the container of the device.
> >
> > While these have not been experienced, they are potential issues and thus
> > they need to be fixed. Also, they are a few more files that have the same
> > kind of issue, those will be fixed if these are accepted.
> 
> Thanks for doing this.  This is the sort of mistake that just gets
> copied everywhere, so fixing the examples in the tree will help
> prevent the problem from spreading more.
> 
> I don't know if there's really value in having device_register()
> return an error but rely on the caller to do the put_device().  Are
> there cases where the caller still needs the struct device even if
> device_register() fails?  E.g., could we do something like this
> instead (I know some callers would also require corresponding changes
> to avoid double puts):

Yeah, that might make more sense, but I was trying to not have the
driver core suddenly free memory if something you pass to it goes wrong.
That's a pretty "odd" thing for an api call to do in the kernel, usually
the caller is always responsible for cleaning up for errors happening.

And there's going to be a ton of changes to get this fixed, as you
really need to do it all in one patch, which makes for a bad "flag-day"
of the api.

thanks,

greg k-h