Re: [PATCH 1/2] genirq/msi, platform-msi: Adjust return value of msi_domain_prepare_irqs()

From: Huacai Chen <chenhuacai@gmail.com>
To: Marc Zyngier <maz@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Huacai Chen <chenhuacai@loongson.cn>,
	Bjorn Helgaas <bhelgaas@google.com>,
	linux-kernel@vger.kernel.org, loongson-kernel@lists.loongnix.cn,
	Xuefeng Li <lixuefeng@loongson.cn>,
	Jiaxun Yang <jiaxun.yang@flygoat.com>
Subject: Re: [PATCH 1/2] genirq/msi, platform-msi: Adjust return value of msi_domain_prepare_irqs()
Date: Sun, 28 May 2023 20:07:56 +0800	[thread overview]
Message-ID: <CAAhV-H6KpNhL5VvumvhcAKGOpe-EO0zfzm_xPprP0rTVf18Leg@mail.gmail.com> (raw)
In-Reply-To: <86fs7gdhid.wl-maz@kernel.org>

Hi, Marc,

On Sun, May 28, 2023 at 3:47 PM Marc Zyngier <maz@kernel.org> wrote:
>
> On Sat, 27 May 2023 15:03:29 +0100,
> Thomas Gleixner <tglx@linutronix.de> wrote:
> >
> > On Sat, May 27 2023 at 13:46, Huacai Chen wrote:
> > > Adjust the return value semanteme of msi_domain_prepare_irqs(), which
> > > allows us to modify the input nvec by overriding the msi_domain_ops::
> > > msi_prepare(). This is necessary for the later patch.
> > >
> > > Before:
> > > 0 on success, others on error.
> > >
> > > After:
> > > = 0: Success;
> > >> 0: The modified nvec;
> > > < 0: Error code.
> >
> > This explains what the patch does, but provides zero justification for
> > this nor any analysis why this is correct for the existing use cases.
> >
> > That longsoon MSI domain is a PCI MSI domain. PCI/MSI has already a
> > mechanism to return the actual possible number of vectors if the
> > underlying space is exhausted.
> >
> > Why is that not sufficient for your problem at hand?
>
> I've already made that point, but it seems that the argument is
> falling on deaf ears.
I'm very sorry that I didn't answer your question directly.

>
> Being able to allocate MSIs is not a guarantee, and is always
> opportunistic. If some drivers badly fail because the they don't get
> the number of MSIs they need, then they need fixing.
Yes, I know allocating MSIs is not a guarantee, and most existing
drivers will fallback to use legacy irqs when failed. However, as I
replied in an early mail, we want to do some proactive throttling in
the loongson-pch-msi irqchip driver, rather than consume msi vectors
aggressively. For example, if we have two NICs, we want both of them
to get 32 msi vectors; not one exhaust all available vectors, and the
other fallback to use legacy irq.

I hope I have explained clearly, thanks.

Huacai

>
> I really don't see the point in papering over this at the lowest level
> of the stack.
>
>         M.
>
> --
> Without deviation from the norm, progress is not possible.