Re: [PATCH 1/5] driver core: make struct class.dev_uevent() take a const *

From: Greg Kroah-Hartman
Date: Mon Nov 28 2022 - 12:25:39 EST


On Sun, Nov 27, 2022 at 06:38:39PM -0800, Bart Van Assche wrote:
> On 11/27/22 05:45, Greg Kroah-Hartman wrote:
> > On Fri, Nov 25, 2022 at 03:51:11PM -0800, Bart Van Assche wrote:
> > > On 11/23/22 04:25, Greg Kroah-Hartman wrote:
> > > > diff --git a/include/linux/mISDNif.h b/include/linux/mISDNif.h
> > > > index 7dd1f01ec4f9..7aab4a769736 100644
> > > > --- a/include/linux/mISDNif.h
> > > > +++ b/include/linux/mISDNif.h
> > > > @@ -586,7 +586,7 @@ extern struct mISDNclock *mISDN_register_clock(char *, int, clockctl_func_t *,
> > > > void *);
> > > > extern void mISDN_unregister_clock(struct mISDNclock *);
> > > > -static inline struct mISDNdevice *dev_to_mISDN(struct device *dev)
> > > > +static inline struct mISDNdevice *dev_to_mISDN(const struct device *dev)
> > > > {
> > > > if (dev)
> > > > return dev_get_drvdata(dev);
> > >
> > > Why does the dev_to_mISDN() function drop constness? I haven't found an
> > > explanation for this in the cover letter.
> >
> > I agree, this is going to be fixed up, see the thread starting here:
> > https://lore.kernel.org/r/Y34+V2bCDdqujBDk@xxxxxxxxx
> >
> > I'll work on making a const / non const version for these so that we
> > don't loose the marking.
> >
> > Oh wait, no, this function is fine, it's not modifying the device
> > structure at all, and only returning the pointer in the private data
> > stored in the device. There is no loss of const-ness here.
>
> Hi Greg,
>
> This is what I found in include/linux/mISDNif.h:
>
> struct mISDNdevice {
> struct mISDNchannel D;
> u_int id;
> u_int Dprotocols;
> u_int Bprotocols;
> u_int nrbchan;
> u_char channelmap[MISDN_CHMAP_SIZE];
> struct list_head bchannels;
> struct mISDNchannel *teimgr;
> struct device dev;
> };
>
> As one can see 'dev' is a member of struct mISDNdevice. I still think that
> dev_to_mISDN() drops constness. Did I perhaps overlook something?

I think you are missing that dev_to_mISDN() is doing something different
than most dev_to_FOO() functions do (i.e. there is no container_of()
call here at all):

static inline struct mISDNdevice *dev_to_mISDN(struct device *dev)
{
if (dev)
return dev_get_drvdata(dev);
else
return NULL;
}

See, no pointer mess or anything else here, all that happens is the
driver data pointer in struct device is returned.

If this was a "normal" dev_to_FOO() function, then yes, the const-ness
of the pointer would be called into question as the thread I linked to
discusses.

thanks,

greg k-h