Re: [PATCH v5 0/2] PCI: mediatek-gen3: Add 2-lanes mode support + clock

On Mon, Aug 10, 2026 at 6:55 PM Christian Marangi (Ansuel)
<[email protected]> wrote:
>
> Il giorno lun 10 ago 2026 alle ore 12:33 Chen-Yu Tsai
> <[email protected]> ha scritto:
> >
> > On Fri, Aug 7, 2026 at 11:21 PM Bjorn Helgaas <[email protected]> wrote:
> > >
> > > On Fri, Aug 07, 2026 at 11:24:09AM +0800, Chen-Yu Tsai wrote:
> > > > On Fri, Aug 7, 2026 at 12:53 AM Christian Marangi <[email protected]> wrote:
> > > > > This small series introduce support for 2-lanes mode for Airoha AN7581
> > > >
> > > > Just a nitpick, but I would probably name this something else, like
> > > > "cross-controller lane-bonding mode" or "lane stealing"? PCIe already
> > > > has 2x lanes as a standard feature, so this naming is a bit confusing.
> > > > It's not like dual-LVDS display in which LVDS is only a single lane.
> > >
> > > I suggested the "2-lane" and "x2" terminology because I assumed the
> > > result is what the PCIe spec would describe as a "x2 Link" consisting
> > > of two Lanes.
> > >
> > > If that's not the case, maybe "cross-controller lane-bonding mode" or
> > > "lane stealing" would be more accurate, but I don't know what those
> > > mean, so if we use them I would also like to know what the result
> > > looks like in standard PCIe terms.
> >
> > Benjamin pointed out later in the thread that it is just bifurcation.
> >
>
> Yes but as me, you, Benjamin and Bjorn people who weren't there with the HW
> team at times the SoC was created, and given how the internal documentation
> these are just assumption and observation... Everyone can be correct and wrong.
>
> > I would assume "PCIe bifurcation" is a standard term, so maybe we could
> > call this "Support un-bifurcated 2-lane operation"?
> >
>
> And how coming from PCIe bifurcation to un-bifurcated 2-lane operation keeps
> the standard meaning?
>
> This terms is equally custom as the previous "cross-controller
> lane-bonding mode"
> and "lane stealing"
>
> I want to be honest... I'm not following why we are trying in every
> way possible to find
> a standard meaning for an highly integrated PCIe controller for a vendor SoC.
> It's not like there are devices shipped with PCIe slot where you can
> insert whatever
> card you want. (there is only a single RFB board but that is really to
> debug stuff and
> develop)
>
> The SoC is designed to have WiFi cards attached and only a subset of them.
> And some of them require 2 lanes mode to work with this special configuration.
>
> Even the DT bindings are not that specific to the details... at the
> end of the day
> everything is modelled as simple as "num-lanes = <2>;"
>
> And that is effectively what the HW gets configured to. If then a
> robbery is done
> internally to the HW to get the second lane with probably some criminal act and
> some murder then those are just details. But unless we have a correct police
> report on that (Documentation) we can't assume someone got shot in the
> process.
> (sorry for the joke around the stealing thing)
>
> > On the implementation side, having to reach into the second PCIe controller
> > is a bit messy :(
> >
>
> I can assure you vendor hide all kind of bad stuff behind single register...
> It's just Airoha that loves to expose godzillion registers and configure
> everything clearly... Something that most of the time is better than hiding
> everything behind ATF with SCM commands (or even configure stuff
> in bootloader from HW strap)
>
> A bonding solution where 2 controllers are configured and then bonded might
> also be problematic as for 2 lanes mode the configuration needs to be done
> in the same time segment. (and that would introduce lots of complexity in
> both DT and driver for the sake of just having an additional reset and
> an additional reg)
>
> Aside from these 2 point... I'm open to any kind of naming and change
> but it seems sad to me to delay this feature for these details that would
> only affect the commit description and nothing else... This is currently needed
> for Eagle WiFi card to correctly works... With the current code it works but
> sometimes it doesn't as the reset and configuration phase is not well defined.
> This series fix exactly that to have stable functionality (tested on OpenWrt
> from at least half a year)

It was not my intention to block this. Having worked on other MediaTek
SoCs on which the PCIe gen4 controller, supported by the same driver,
can do up to 2 lanes natively, the subject of the cover letter was a bit
confusing for me.

Now that you've explained it, everything is clear.

> > > > > SoC. This is needed for correctly functionality of Eagle WiFi Card
> > > > > normally attached to this SoC that require a 2-line PCIe card to
> > > > > correctly work (and give the proper performance)
> > > > >
> > > > > The first 2 patch address a limitation of the PCIe implementation
> > > > > where the PERSTOUT reset were indirectly asserted and deasserted
> > > > > all at the same time (for all the 3 PCIe card) with PCIe
> > > > > enable and disable.
> > > > > The 2 patch address this and introduce correct reset to control
> > > > > reset line for the relevant PCIe line.
> > > > >
> > > > > The last 2 patch add additional logic and support to assert
> > > > > and deassert the PERSTOUT and also apply the required configuration
> > > > > for 2-lanes mode.
> > > > >
> > > > > 2-lanes mode is implemented in DT by adding the required property
> > > > > and by defining the "num-lanes" to 2.