Re: [PATCH v5 3/4] rpmsg: Move the rpmsg control device from rpmsg_char to rpmsg_ctrl

[Bjorn Andersson <bjorn.andersson@linaro.org>]

On Mon 18 Oct 02:13 PDT 2021, Arnaud POULIQUEN wrote:

> 
> 
> On 10/16/21 6:46 AM, Bjorn Andersson wrote:
> > On Mon 11 Oct 05:46 CDT 2021, Arnaud POULIQUEN wrote:
> > 
> >>
> >>
> >> On 10/9/21 1:35 AM, Bjorn Andersson wrote:
> >>> On Mon 12 Jul 05:37 PDT 2021, Arnaud Pouliquen wrote:
> >>>
> >>>> Create the rpmsg_ctrl.c module and move the code related to the
> >>>> rpmsg_ctrldev device in this new module.
> >>>>
> >>>> Add the dependency between rpmsg_char and rpmsg_ctrl in the
> >>>> kconfig file.
> >>>>
> >>>
> >>> As I said in the cover letter, the only reason I can see for doing this
> >>> refactoring is in relation to the introduction of
> >>> RPMSG_CREATE_DEV_IOCTL. So I would like this patch to go together with
> >>> that patch, together with a good motivation why there's merit to
> >>> creating yet another kernel module (and by bind/unbind can't be used).
> >>>
> >>> Perhaps I'm just missing some good usecase related to this?
> >>
> >>
> >>>
> >>>> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen@foss.st.com>
> >>>> Reviewed-by: Mathieu Poirier <mathieu.poirier@linaro.org>
> >>>> ---
> >>>>  drivers/rpmsg/Kconfig      |   9 ++
> >>>>  drivers/rpmsg/Makefile     |   1 +
> >>>>  drivers/rpmsg/rpmsg_char.c | 170 +----------------------------
> >>>>  drivers/rpmsg/rpmsg_char.h |   2 +
> >>>>  drivers/rpmsg/rpmsg_ctrl.c | 215 +++++++++++++++++++++++++++++++++++++
> >>>>  5 files changed, 229 insertions(+), 168 deletions(-)
> >>>>  create mode 100644 drivers/rpmsg/rpmsg_ctrl.c
> >>>>
> >>> [..]
> >>>> diff --git a/drivers/rpmsg/rpmsg_char.c b/drivers/rpmsg/rpmsg_char.c
> >>> [..]
> >>>> -static int rpmsg_chrdev_probe(struct rpmsg_device *rpdev)
> >>>> -{
> >>> [..]
> >>>> -	dev = &ctrldev->dev;
> >>>> -	device_initialize(dev);
> >>>> -	dev->parent = &rpdev->dev;
> >>>> -	dev->class = rpmsg_class;
> >>> [..]
> >>>> diff --git a/drivers/rpmsg/rpmsg_ctrl.c b/drivers/rpmsg/rpmsg_ctrl.c
> >>> [..]
> >>>> +static int rpmsg_ctrldev_probe(struct rpmsg_device *rpdev)
> >>>> +{
> >>> [..]
> >>>> +	dev = &ctrldev->dev;
> >>>> +	device_initialize(dev);
> >>>> +	dev->parent = &rpdev->dev;
> >>>
> >>> You lost the assignment of dev->class here, which breaks the udev rules
> >>> we use to invoke rpmsgexport to create endpoints and it causes udevadm
> >>> to complain that rpmsg_ctrlN doesn't have a "subsystem".
> >>
> >> We discussed this point with Mathieu, as a first step i kept the class, but that
> >> generated another dependency with the rpmsg_char device while information was
> >> available on the rpmsg bus. The char device and ctrl device should share the
> >> same class. As rpmsg_ctrl is created first it would have to create the class,and
> >> provide an API to rpmsg char
> >>
> > 
> > Perhaps if this is considered a common piece shared between multiple
> > rpmsg modules we can create such class in the rpmsg "core" itself?
> 
> Yes that seems a good alternative
> 
> > 
> >> Please could you details what does means "rpmsg_ctrlN doesn't have a
> >> "subsystem"." What exactly the udev is looking for? could it base it check on
> >> the /dev/rpmsg_ctrl0 or /sys/bus/rpmsg/devices/...?
> >>
> > 
> > If I read the uevent messages correctly they seem to contain a SUBSYTEM=
> > property when the class is provided. But I'm not sure about the reasons
> > for that.
> 
> If it part of the udev requirement, i suppose that it is mandatory, and in this
> case, declare the class in the core make sense.
> 

I don't know if it's a requirement. But I think it's worth keeping the
class around, as it's the only problem I've found with existing users.

> I will send a new patchset that will squash all the remaining patches, taking
> into account your comment.
> 

Thanks,
Bjorn

> Thanks,
> Arnaud
> 
> > 
> > Regards,
> > Bjorn
> > 
> >> Thanks,
> >> Arnaud
> >>
> >>>
> >>> Regards,
> >>> Bjorn
> >>>