On Mon, 31 Jan 2022, Florian Fainelli wrote:
> On 1/31/2022 10:35 AM, Saeed Mahameed wrote:
>> On 31 Jan 19:30, Geert Uytterhoeven wrote:
>>> On Mon, Jan 31, 2022 at 6:59 PM Stephen Hemminger
>>> <[email protected]> wrote:
>>>> On Mon, 31 Jan 2022 09:24:50 -0800
>>>> Saeed Mahameed <[email protected]> wrote:
>>>>
>>>> > From: Saeed Mahameed <[email protected]>
>>>> >
>>>> > NET_VENDOR_XYZ were defaulted to 'y' for no technical reason.
>>>> >
>>>> > Since all drivers belonging to a vendor are supposed to default to 'n',
>>>> > defaulting all vendors to 'n' shouldn't be an issue, and aligns well
>>>> > with the 'no new drivers' by default mentality.
>>>> >
>>>> > Signed-off-by: Saeed Mahameed <[email protected]>
>>>>
>>>> This was done back when vendors were introduced in the network drivers
>>>> tree.
>>>> The default of Y allowed older configurations to just work.
>>>
>>> And changing the defaults means all defconfigs must be updated first,
>>> else the user's configs will end up without drivers needed.
>>>
>>
>> As I understand correctly, at least for most common net drivers, having
>> NET_VENDOR_XYZ=y doesn't actually build anything, we have flags per
>> module for each vendor and those are defaulted to N.
>
> Right, but once you start hiding NET_VENDOR_DRIVER_XYZ under a NET_VENDOR_XYZ
> Kconfig symbol dependency, if NET_VENDOR_XYZ is not set to Y, then you have
> no way to select NET_VENDOR_DRIVER_XYZ and so your old defconfig breaks.
>
>>
>>>> So there was a reason, not sure if it matters anymore.
>>>> But it seems like useless repainting to change it now.
>>>
>>> It might make sense to tune some of the defaults (i.e. change to
>>> "default y if ARCH_*") for drivers with clear platform dependencies.
>>>
>>
>> either set hard default to 'n' or just keep it as is, anything else is just
>> more confusion.
>
> Maybe the rule should go like this: any new driver vendor defaults to n, and
> existing ones remain set to y, until we deprecate doing that and switching
> them all off to n by 5.18?
Forgive my ignorance, but isn't it a regression if things quit working
even if it's just a configuration change?
From a user perspective I like having everything turned on initially so
it just works. Pruning things down is a lot easier than trying to figure
out what all to turn on. Especially in graphics.
--
Hisashi T Fujinaka - [email protected]
On Mon, Jan 31, 2022 at 10:55:14AM -0800, Hisashi T Fujinaka wrote:
> On Mon, 31 Jan 2022, Florian Fainelli wrote:
>
> > On 1/31/2022 10:35 AM, Saeed Mahameed wrote:
> > > On 31 Jan 19:30, Geert Uytterhoeven wrote:
> > > > On Mon, Jan 31, 2022 at 6:59 PM Stephen Hemminger
> > > > <[email protected]> wrote:
> > > > > On Mon, 31 Jan 2022 09:24:50 -0800
> > > > > Saeed Mahameed <[email protected]> wrote:
> > > > >
> > > > > > From: Saeed Mahameed <[email protected]>
> > > > > >
> > > > > > NET_VENDOR_XYZ were defaulted to 'y' for no technical reason.
> > > > > >
> > > > > > Since all drivers belonging to a vendor are supposed to default to 'n',
> > > > > > defaulting all vendors to 'n' shouldn't be an issue, and aligns well
> > > > > > with the 'no new drivers' by default mentality.
> > > > > >
> > > > > > Signed-off-by: Saeed Mahameed <[email protected]>
> > > > >
> > > > > This was done back when vendors were introduced in the
> > > > > network drivers tree.
> > > > > The default of Y allowed older configurations to just work.
> > > >
> > > > And changing the defaults means all defconfigs must be updated first,
> > > > else the user's configs will end up without drivers needed.
> > > >
> > >
> > > As I understand correctly, at least for most common net drivers,
> > > having NET_VENDOR_XYZ=y doesn't actually build anything, we have
> > > flags per
> > > module for each vendor and those are defaulted to N.
> >
> > Right, but once you start hiding NET_VENDOR_DRIVER_XYZ under a
> > NET_VENDOR_XYZ Kconfig symbol dependency, if NET_VENDOR_XYZ is not set
> > to Y, then you have no way to select NET_VENDOR_DRIVER_XYZ and so your
> > old defconfig breaks.
> >
> > >
> > > > > So there was a reason, not sure if it matters anymore.
> > > > > But it seems like useless repainting to change it now.
> > > >
> > > > It might make sense to tune some of the defaults (i.e. change to
> > > > "default y if ARCH_*") for drivers with clear platform dependencies.
> > > >
> > >
> > > either set hard default to 'n' or just keep it as is, anything else is just
> > > more confusion.
> >
> > Maybe the rule should go like this: any new driver vendor defaults to n,
> > and existing ones remain set to y, until we deprecate doing that and
> > switching them all off to n by 5.18?
>
> Forgive my ignorance, but isn't it a regression if things quit working
> even if it's just a configuration change?
No, kernel configs never were declared as ABI as "regular" users are not
supposed to touch it. They use something provided by the distro.
>
> From a user perspective I like having everything turned on initially so
> it just works. Pruning things down is a lot easier than trying to figure
> out what all to turn on. Especially in graphics.
I have completely opposite view here and prefer to have minimal config
for my CI, and for my working machines as well.
Thanks
>
> --
> Hisashi T Fujinaka - [email protected]
On Tue, Feb 01, 2022 at 10:58:01AM +0200, Leon Romanovsky wrote:
> No, kernel configs never were declared as ABI as "regular" users are not
> supposed to touch it. They use something provided by the distro.
+1