Re: [RFC PATCH v2 31/31] kvx: Add IPI driver

From: "Arnd Bergmann" <arnd@arndb.de>
To: "Yann Sionneau" <ysionneau@kalrayinc.com>,
	"Krzysztof Kozlowski" <krzysztof.kozlowski@linaro.org>,
	"Yann Sionneau" <ysionneau@kalray.eu>,
	"Jonathan Corbet" <corbet@lwn.net>,
	"Thomas Gleixner" <tglx@linutronix.de>,
	"Marc Zyngier" <maz@kernel.org>,
	"Rob Herring" <robh+dt@kernel.org>,
	"Krzysztof Kozlowski" <krzysztof.kozlowski+dt@linaro.org>,
	"Will Deacon" <will@kernel.org>,
	"Peter Zijlstra" <peterz@infradead.org>,
	"Boqun Feng" <boqun.feng@gmail.com>,
	"Mark Rutland" <mark.rutland@arm.com>,
	"Eric W. Biederman" <ebiederm@xmission.com>,
	"Kees Cook" <keescook@chromium.org>,
	"Oleg Nesterov" <oleg@redhat.com>,
	"Ingo Molnar" <mingo@redhat.com>,
	"Waiman Long" <longman@redhat.com>,
	"Aneesh Kumar" <aneesh.kumar@linux.ibm.com>,
	"Andrew Morton" <akpm@linux-foundation.org>,
	"Nicholas Piggin" <npiggin@gmail.com>,
	"Paul Moore" <paul@paul-moore.com>,
	"Eric Paris" <eparis@redhat.com>,
	"Christian Brauner" <brauner@kernel.org>,
	"Paul Walmsley" <paul.walmsley@sifive.com>,
	"Palmer Dabbelt" <palmer@dabbelt.com>,
	"Albert Ou" <aou@eecs.berkeley.edu>,
	"Jules Maselbas" <jmaselbas@kalray.eu>,
	"Guillaume Thouvenin" <gthouvenin@kalray.eu>,
	"Clement Leger" <clement@clement-leger.fr>,
	"Vincent Chardon" <vincent.chardon@elsys-design.com>,
	"Marc Poulhiès" <dkm@kataplop.net>,
	"Julian Vetter" <jvetter@kalray.eu>,
	"Samuel Jones" <sjones@kalray.eu>,
	"Ashley Lesdalons" <alesdalons@kalray.eu>,
	"Thomas Costis" <tcostis@kalray.eu>,
	"Marius Gligor" <mgligor@kalray.eu>,
	"Jonathan Borne" <jborne@kalray.eu>,
	"Julien Villette" <jvillette@kalray.eu>,
	"Luc Michel" <lmichel@kalray.eu>,
	"Louis Morhet" <lmorhet@kalray.eu>,
	"Julien Hascoet" <jhascoet@kalray.eu>,
	"Jean-Christophe Pince" <jcpince@gmail.com>,
	"Guillaume Missonnier" <gmissonnier@kalray.eu>,
	"Alex Michon" <amichon@kalray.eu>,
	"Huacai Chen" <chenhuacai@kernel.org>,
	"WANG Xuerui" <git@xen0n.name>,
	"Shaokun Zhang" <zhangshaokun@hisilicon.com>,
	"John Garry" <john.garry@huawei.com>,
	"Guangbin Huang" <huangguangbin2@huawei.com>,
	"Bharat Bhushan" <bbhushan2@marvell.com>,
	"Bibo Mao" <maobibo@loongson.cn>,
	"Atish Patra" <atishp@atishpatra.org>,
	"Jason A . Donenfeld" <Jason@zx2c4.com>,
	"Qi Liu" <liuqi115@huawei.com>,
	"Jiaxun Yang" <jiaxun.yang@flygoat.com>,
	"Catalin Marinas" <catalin.marinas@arm.com>,
	"Mark Brown" <broonie@kernel.org>,
	"Janosch Frank" <frankja@linux.ibm.com>,
	"Alexey Dobriyan" <adobriyan@gmail.com>,
	"Julian Vetter" <jvetter@kalrayinc.com>,
	jmaselbas@zdiv.net
Cc: "Benjamin Mugnier" <mugnier.benjamin@gmail.com>,
	linux-doc@vger.kernel.org, linux-kernel@vger.kernel.org,
	devicetree@vger.kernel.org, linux-mm@kvack.org,
	Linux-Arch <linux-arch@vger.kernel.org>,
	linux-audit@redhat.com, linux-riscv@lists.infradead.org,
	bpf@vger.kernel.org
Subject: Re: [RFC PATCH v2 31/31] kvx: Add IPI driver
Date: Wed, 31 Jan 2024 11:28:57 +0100	[thread overview]
Message-ID: <1036060c-25dd-439b-b081-893d0cb000f6@app.fastmail.com> (raw)
In-Reply-To: <269edff0-d989-4ac8-b0c3-bce31283806b@kalrayinc.com>

On Wed, Jan 31, 2024, at 10:52, Yann Sionneau wrote:
> On 22/01/2023 12:54, Krzysztof Kozlowski wrote:
>> On 20/01/2023 15:10, Yann Sionneau wrote:
>>> +
>>> +int __init kvx_ipi_ctrl_probe(irqreturn_t (*ipi_irq_handler)(int, void *))
>>> +{
>>> +	struct device_node *np;
>>> +	int ret;
>>> +	unsigned int ipi_irq;
>>> +	void __iomem *ipi_base;
>>> +
>>> +	np = of_find_compatible_node(NULL, NULL, "kalray,kvx-ipi-ctrl");
>> Nope, big no.
>>
>> Drivers go to drivers, not to arch code. Use proper driver infrastructure.
> Thank you for your review.
>
> It raises questions on our side about how to handle this change.
>
> First let me describe the hardware:
>
> The coolidge ipi controller device handles IPI communication between cpus
> inside a cluster.
>
> Each cpu has 8 of its dedicated irq lines (24 to 31) hard-wired to the ipi.
> The ipi controller has 8 sets of 2 registers:
> - a 17-bit "interrupt" register
> - a 17-bit "mask" register
>
> Each couple of register is dedicated to 1 of the 8 irqlines.
> Each of the 17 bits of interrupt/mask register
> identifies a cpu (cores 0 to 15 + secure_core).
> Writing bit i in interrupt register sends an irq to cpu i, according to the mask
> in mask register.
> Writing in interrupt/mask register couple N targets irq line N of the core.
>
> - Ipi generates an interrupt to the cpu when message is ready.
> - Messages are delivered via Axi.
> - Ipi does not have any interrupt input lines.
>
>
>   +---------------+   irq       axi_w
>   |         |  i  |<--/--- ipi <------
>   | CPU     |  n  |  x8
>   |  core0  |  t  |
>   |         |  c  |  irq          irq         msi
>   |         |  t  |<--/--- apic <----- mbox <-------
>   |         |  l  |  x4
>   +---------------+
>   with intctl = core-irq controller
>    
>
> We analyzed how other Linux ports are handling this situation (IPI) and 
> here are several possible solutions:
>
> 1/ put everything in smp.c like what longarch is doing.
>   * Except that IPI in longarch seems to involve writing to a special 
> purpose CPU register and not doing a memory mapped write like kvx.
>
> 2/ write a device driver in drivers/xxx/ with the content from ipi.c
>   * the probe would just ioremap the reg from DT and register the irq 
> using request_percpu_irq()
>   * it would export a function "kvx_ipi_send()" that would directly be 
> called by smp.c
>   * Question : where would this driver be placed in drivers/ ? 
> drivers/irqchip/ ? Even if this is not per-se an interrupt-controller 
> driver?

This looks like it's close enough to the irqchip driver
that you can just have it in the same file as the 'intctl'
portion.

Top-level irqchip implementations tend to be rather architecture
specific, as does the IPI mechanism. Depending on the register
layout, I think you can have a single devicetree node for
the combination of the core-irq (for managing your
own interrupts) and the ipi (for receiving interrupts from
others), and then have a driver in drivers/irqchip to
deal with both. For the ipi mechanism, trying to abstract
it too much generally makes it too slow, so I would not
go through a nested irqchip or a mailbox driver etc.

I don't know what the 'apic' in your diagram is, so that
would be either a nested irqchip or could be part of
the same driver as well.

     Arnd