Re: [PATCH v2 23/23] PCI: aardvark: Make main irq_chip structure a static driver structure

From: Marc Zyngier <maz@kernel.org>
To: "Pali Rohár" <pali@kernel.org>
Cc: "Marek Behún" <kabel@kernel.org>,
	"Lorenzo Pieralisi" <lorenzo.pieralisi@arm.com>,
	"Bjorn Helgaas" <helgaas@kernel.org>,
	linux-pci@vger.kernel.org, linux-arm-kernel@lists.infradead.org
Subject: Re: [PATCH v2 23/23] PCI: aardvark: Make main irq_chip structure a static driver structure
Date: Mon, 10 Jan 2022 14:44:17 +0000	[thread overview]
Message-ID: <87mtk3tzum.wl-maz@kernel.org> (raw)
In-Reply-To: <20220110105324.jud6bzdtmoiuvyas@pali>

On Mon, 10 Jan 2022 10:53:24 +0000,
Pali Rohár <pali@kernel.org> wrote:
> 
> On Monday 10 January 2022 09:28:39 Marc Zyngier wrote:
> > On 2022-01-10 01:50, Marek Behún wrote:
> > > Marc Zyngier says [1] that we should use struct irq_chip as a global
> > > static struct in the driver. Even though the structure currently
> > > contains a dynamic member (parent_device), Marc says [2] that he plans
> > > to kill it and make the structure completely static.
> > > 
> > > We have already converted others irq_chip structures in this driver in
> > > this way, but we omitted this one because the .name member is
> > > dynamically created from device's name, and the name is displayed in
> > > sysfs, so changing it would break sysfs ABI.
> > > 
> > > The rationale for changing the name (to "advk-INT") in spite of sysfs
> > > ABI, and thus allowing to convert to a static structure, is that after
> > > the other changes we made in this series, the IRQ chip is basically
> > > something different: it no logner generates ERR and PME interrupts (they
> > > are generated by emulated bridge's rp_irq_chip).
> > 
> > There is no 'is spite of the ABI'. If you don't understand why
> > we don't break the ABI, you have an even bigger problem.
> > 
> > So NAK to this patch, now and forever. Any change to the structure to
> > make it read-only must allow the preservation of the existing names
> > when they are generated by the driver.
> 
> Marc, you already presented that you do not like Armada 3720 platform
> and that you do not care about it.

What I like or not is irrelevant here. What I ask for is that
userspace ABIs are not broken.

> But please do not slowdown development for this platform.

That's quite an accusation.

> Arguments about ABIs, breaking it and similar are not relevant here as
> this current kernel implementation is broken. And has to be replaced by
> a working one. We are doing on it for more than year.
>
> It really does not make sense to try doing some backward compatibility
> with something which is broken by design and does not work. It just take
> lot of time without any value.
> 
> We really need to more forward and fix driver as in current state is
> PCIe on Armada 3720 unusable.

This patch doesn't fix anything. It has the potential to break
userspace, and I'm not having any of it. You may not care about
backward compatibility, but this is thankfully *not* your pet
playground.

You can claim that I am doing a bad job. In which case, feel free to
submit a patch removing me from the MAINTAINER file, and we can have
that discussion.

In the meantime, I will continue to oppose these kind of patches that
pretend to 'fix' things without adding any value.

	M.

-- 
Without deviation from the norm, progress is not possible.