Re: [PATCH 05/11] net: core: provide devm_register_netdev()

[Bartosz Golaszewski]

pt., 8 maj 2020 o 07:54 Heiner Kallweit <hkallweit1@xxxxxxxxx> napisaÅ(a):
>
> On 08.05.2020 00:56, Jakub Kicinski wrote:
> > On Thu, 7 May 2020 19:03:44 +0200 Bartosz Golaszewski wrote:
> >>> To implement Edwin's suggestion? Makes sense, but I'm no expert, let's
> >>> also CC Heiner since he was asking about it last time.
> >>
> >> Yes, because taking the last bit of priv_flags from net_device seems
> >> to be more controversial but if net maintainers are fine with that I
> >> can simply go with the current approach.
> >
> > From my perspective what Edwin suggests makes sense. Apart from
> > little use for the bit after probe, it also seems cleaner for devres
> > to be able to recognize managed objects based on its own state.
> >
> What I was saying is that we should catch the case that a driver
> author uses a device-managed register() w/o doing the same for the
> alloc(). A core function should not assume that driver authors do
> sane things only.
> I don't have a strong preference how it should be done.
> Considering what is being discussed, have a look at get_pci_dr() and
> find_pci_dr(), they deal with managing which parts of the PCI
> subsystem are device-managed.

Yes, I have - that's why I asked if anyone objects to me moving all
networking devres functions into their own source file. The reason for
that being: devres_find() needs to know the address of the release
function, meanwhile devm_register_netdev() would have to go into
net/core, while devm_alloc_etherdev() lives in net/ethernet.

Bart