Re: [PATCH v5 1/6] PCI: OF: Don't remap iospace on unsupported platform

From: Rob Herring <robh+dt@kernel.org>
To: Jiaxun Yang <jiaxun.yang@flygoat.com>
Cc: PCI <linux-pci@vger.kernel.org>,
	Bjorn Helgaas <bhelgaas@google.com>,
	Thomas Bogendoerfer <tsbogend@alpha.franken.de>,
	Huacai Chen <chenhc@lemote.com>,
	Lorenzo Pieralisi <lorenzo.pieralisi@arm.com>,
	Andrew Murray <amurray@thegoodpenguin.co.uk>,
	Paul Burton <paulburton@kernel.org>,
	devicetree@vger.kernel.org,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"open list:MIPS" <linux-mips@vger.kernel.org>
Subject: Re: [PATCH v5 1/6] PCI: OF: Don't remap iospace on unsupported platform
Date: Fri, 24 Apr 2020 14:05:23 -0500	[thread overview]
Message-ID: <CAL_JsqLEHZOZXkuW4_9KeUWjzahvP0OMtQUGVU_pASi9Ag=GAQ@mail.gmail.com> (raw)
In-Reply-To: <D5CC0902-F84F-43F1-88E5-C03018B25050@flygoat.com>

On Fri, Apr 24, 2020 at 1:03 PM Jiaxun Yang <jiaxun.yang@flygoat.com> wrote:
>
>
>
> 于 2020年4月25日 GMT+08:00 上午1:47:23, Rob Herring <robh+dt@kernel.org> 写到:
> >On Fri, Apr 24, 2020 at 8:09 AM Jiaxun Yang <jiaxun.yang@flygoat.com> wrote:
> >>
> >> There are some platforms that don't support I/O space remapping
> >> like MIPS. However, our PCI code will try to remap iospace
> >> unconditionally and reject io resources on these platforms.
> >>
> >> So we should remove I/O space remapping check and use a range
> >> check instead on these platforms.
> >>
> >> Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
> >> --
> >> v4: Fix a typo in commit message.
> >> v5: Commit message massage
> >> ---
> >>  drivers/pci/of.c | 9 +++++++++
> >>  1 file changed, 9 insertions(+)
> >>
> >> diff --git a/drivers/pci/of.c b/drivers/pci/of.c
> >> index 81ceeaa6f1d5..36e8761b66c6 100644
> >> --- a/drivers/pci/of.c
> >> +++ b/drivers/pci/of.c
> >> @@ -547,12 +547,21 @@ int pci_parse_request_of_pci_ranges(struct device *dev,
> >>
> >>                 switch (resource_type(res)) {
> >>                 case IORESOURCE_IO:
> >> +#if defined(PCI_IOBASE) && defined(CONFIG_MMU)
> >
> >We already have the same condition in pci_remap_iospace(). All this
> >does is suppress a WARN. If a WARN is not desired, then change it.
> >Perhaps just a single line rather than backtrace would be okay.
> >Whether to WARN or not shouldn't be a decision for firmware code.
> >
> >Though really, if I/O space can never be supported, then it's an error
> >in your DT to describe it.
>
> In MIPS world, we do have inb/oub family of I/O functions
> to support I/O space. But we're not using remap_iospace as it's breaking
> some of our assumptions.

I suspect that's largely because most MIPS drivers pre-date the common
iospace handling. Really MIPS should start using this.

> We have our own inb/outb implementation.

That's orthogonal to mapping the iospace.

> In that case, I think make a range check instead of remapping would
> be more practical.

Not the kernel's job to validate the DT especially if you aren't using
this data anywhere.

Just remove the WARN, make it a single line print, or add
!IS_ENABLED(CONFIG_MIPS) where the warning is.

Rob