Re: [PATCH AUTOSEL 5.4 2/5] gpio: ixp4xx: Make irqchip immutable

[Bartosz Golaszewski <brgl@xxxxxxxx>]

On Thu, Sep 22, 2022 at 11:17 AM Marc Zyngier <maz@xxxxxxxxxx> wrote:
>
> Hi Bartosz,
>
> On Wed, 21 Sep 2022 21:04:27 +0100,
> Bartosz Golaszewski <brgl@xxxxxxxx> wrote:
> >
> > On Wed, Sep 21, 2022 at 6:57 PM Marc Zyngier <maz@xxxxxxxxxx> wrote:
> > >
> > > On 2022-09-21 16:54, Sasha Levin wrote:
> > > > From: Linus Walleij <linus.walleij@xxxxxxxxxx>
> > > >
> > > > [ Upstream commit 94e9bc73d85aa6ecfe249e985ff57abe0ab35f34 ]
> > > >
> > > > This turns the IXP4xx GPIO irqchip into an immutable
> > > > irqchip, a bit different from the standard template due
> > > > to being hierarchical.
> > > >
> > > > Tested on the IXP4xx which uses drivers/ata/pata_ixp4xx_cf.c
> > > > for a rootfs on compact flash with IRQs from this GPIO
> > > > block to the CF ATA controller.
> > > >
> > > > Cc: Marc Zyngier <maz@xxxxxxxxxx>
> > > > Signed-off-by: Linus Walleij <linus.walleij@xxxxxxxxxx>
> > > > Acked-by: Marc Zyngier <maz@xxxxxxxxxx>
> > > > Signed-off-by: Bartosz Golaszewski <brgl@xxxxxxxx>
> > > > Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
> > >
> > > Why? The required dependencies are only in 5,19, and are
> > > definitely NOT a stable candidate...
> > >
> > > This isn't a fix by any stretch of the imagination.
> > >
> >
> > Hi Marc,
> >
> > While I didn't mark it for stable (and it shouldn't go into any branch
> > earlier than 5.19.x), I did send the patches making the irqchips
> > immutable to Linus Torvalds as fixes as they technically do *fix* the
> > warning emitted by gpiolib and make the implementation correct.
> >
> > I think these patches should still be part of the v5.19.x stable branch.
>
> 5.19, sure. All the dependencies are there, and tightening the driver
> implementations is a valuable goal.
>
> However, targeting all the other stable releases (5.4, 5.10, 5.15)
> makes little sense. It won't even compile! Do the dependencies need to
> be backported? I don't think it is worthwhile, as this is a long
> series containing multiple related changes spread all over the tree.
> This would defeat the very purpose of a stable tree.
>

That's what I'm saying. All the conversions to immutable irqchips
should go to v5.19 but nowhere else.

I thought that by saying "This isn't a fix by any stretch of the
imagination." you meant it should go like regular feature patches into
the next merge window only.

Sasha: can you drop this from all branches earlier than v5.19?

Thanks,
Bartosz