On Mon, Sep 27, 2021 at 10:53:09AM -0700, David E. Box wrote:
> On Mon, 2021-09-27 at 06:03 +0200, Greg KH wrote:
> > On Sun, Sep 26, 2021 at 06:15:16PM -0700, David E. Box wrote:
> > > > > +static int sdsi_remove(struct platform_device *pdev)
> > > > > +{
> > > > > + struct sdsi_priv *priv = platform_get_drvdata(pdev);
> > > > > +
> > > > > + priv->dev_present = false;
> > > > > + sysfs_remove_bin_file(&priv->pdev->dev.kobj, &priv->registers_bin_attr);
> > > > > + misc_deregister(&priv->miscdev);
> > > > > + kref_put(&priv->kref, sdsi_priv_release);
> > > >
> > > > Why do you need a kref for a structure that already can be controlled by
> > > > a different lifetime rule?
> > >
> > > Which rule am I missing? This kref allows the structure to remain in case the device is removed
> > > while the file is open.
> >
> > This device is on a hardware bus that allows removal?
>
> Well the device can be unbound. A test case covers this.
Great, where are these tests? Why not add them to the kernel tree
itself in the proper location?
And in the real-world, who would ever unbind this?
> > Anyway, you now are dealing with lifetime rules of 3 structures all at
> > once, and the interactions between them is not very obvious. It would
> > probably be simpler just to stick with 2, right? You really only care
> > about the misc structure here.
>
> In the case that the device is unbound, both the pdev and miscdev go away. Something has to outlive
> them in order to handle any open files still trying to use the ioctl.
I do not think that the miscdev goes away if the file handle is still
open, right?
thanks,
greg k-h
