Re: [PATCH 4/5] PCI: Add quirk for multifunction devices of LS7A

From: Jiaxun Yang <jiaxun.yang@flygoat.com>
To: Huacai Chen <chenhuacai@gmail.com>, Bjorn Helgaas <helgaas@kernel.org>
Cc: Huacai Chen <chenhuacai@loongson.cn>,
	Bjorn Helgaas <bhelgaas@google.com>,
	linux-pci <linux-pci@vger.kernel.org>,
	Jianmin Lv <lvjianmin@loongson.cn>,
	Rob Herring <robh+dt@kernel.org>
Subject: Re: [PATCH 4/5] PCI: Add quirk for multifunction devices of LS7A
Date: Wed, 19 May 2021 11:05:25 +0800	[thread overview]
Message-ID: <933330cb-9d86-2b22-9bed-64becefbe2d1@flygoat.com> (raw)
In-Reply-To: <CAAhV-H7DfZffubU_rxoMXdbk_j9p0LwA5S7wF7qvAPOJn=YEdw@mail.gmail.com>

在 2021/5/19 10:26, Huacai Chen 写道:
> Hi, Bjorn,
>
> On Tue, May 18, 2021 at 9:59 PM Bjorn Helgaas <helgaas@kernel.org> wrote:
>> [+cc Rob, beginning of thread
>> https://lore.kernel.org/r/20210514080025.1828197-5-chenhuacai@loongson.cn]
>>
>> On Sat, May 15, 2021 at 11:52:53AM +0800, Huacai Chen wrote:
>>> On Fri, May 14, 2021 at 10:52 PM Jiaxun Yang <jiaxun.yang@flygoat.com> wrote:
>>>> 在 2021/5/14 16:00, Huacai Chen 写道:
>>>>> From: Jianmin Lv <lvjianmin@loongson.cn>
>>>>>
>>>>> In LS7A, multifunction device use same pci PIN and different
>>>>> irq for different function, so fix it for standard pci PIN
>>>>> usage.
>>>> Hmm, I'm unsure about this change.
>>>> The PCIe port, or PCI-to-PCI bridge on LS7A only have a single
>>>> upstream interrupt specified in DeviceTree, how can this quirk
>>>> work?
>>> LS7A will be shared by MIPS-based Loongson and LoongArch-based
>>> Loongson, LoongArch use ACPI rather than FDT, so this quirk is needed.
>> Can you expand on this a little bit?
>>
>> Which DT binding are you referring to?  Is it in the Linux source
>> tree?
> I referring to arch/mips/boot/dts/loongson/ls7a-pch.dtsi, it is
> already in Linux source tree.
>
>> I think Linux reads Interrupt Pin for both FDT and ACPI systems, and
>> apparently that register contains the same value for all functions of
>> this multi-function device.
>>
>> The quirk will be applied for both FDT and ACPI systems, but it sounds
>> like you're saying this is needed *because* LoongArch uses ACPI.
> Jiaxun said that FDT system doesn't need this quirk, maybe he know more.

For FDT system, the IRQ routine of PCI ports looks like:

                 interrupt-map-mask = <0 0 0 0>;
                 interrupt-map = <0 0 0 0 &pic 41 IRQ_TYPE_LEVEL_HIGH>;

It means they are all going to the same parent IRQ no matter what we read
from pin register. And it's the actual hardware behavior AFAIK.

Thanks.

- Jiaxun

>
> Huacai
>> Bjorn