Re: [PATCH] net: bcmgenet: Fix attaching to PYH failed on RPi 4B

[Jian-Hong Pan]

Heiner Kallweit <hkallweit1@xxxxxxxxx> 於 2021年6月22日 週二 下午2:50寫道：
>
> On 22.06.2021 08:29, Jian-Hong Pan wrote:
> > Florian Fainelli <f.fainelli@xxxxxxxxx> 於 2021年6月22日 週二 上午5:47寫道：
> >>
> >> On 6/21/21 1:15 PM, Stefan Wahren wrote:
> >>> Am 21.06.21 um 18:56 schrieb Peter Robinson:
> >>>> On Mon, Jun 21, 2021 at 5:39 PM Florian Fainelli <f.fainelli@xxxxxxxxx> wrote:
> >>>>> On 6/21/21 6:09 AM, Andrew Lunn wrote:
> >>>>>> On Mon, Jun 21, 2021 at 06:33:11PM +0800, Jian-Hong Pan wrote:
> >>>>>>> The Broadcom UniMAC MDIO bus comes too late. So, GENET cannot find the
> >>>>>>> ethernet PHY on UniMAC MDIO bus. This leads GENET fail to attach the
> >>>>>>> PHY.
> >>>>>>>
> >>>>>>> bcmgenet fd580000.ethernet: GENET 5.0 EPHY: 0x0000
> >>>>>>> ...
> >>>>>>> could not attach to PHY
> >>>>>>> bcmgenet fd580000.ethernet eth0: failed to connect to PHY
> >>>>>>> uart-pl011 fe201000.serial: no DMA platform data
> >>>>>>> libphy: bcmgenet MII bus: probed
> >>>>>>> ...
> >>>>>>> unimac-mdio unimac-mdio.-19: Broadcom UniMAC MDIO bus
> >>>>>>>
> >>>>>>> This patch makes GENET try to connect the PHY up to 3 times. Also, waits
> >>>>>>> a while between each time for mdio-bcm-unimac module's loading and
> >>>>>>> probing.
> >>>>>> Don't loop. Return -EPROBE_DEFER. The driver core will then probed the
> >>>>>> driver again later, by which time, the MDIO bus driver should of
> >>>>>> probed.
> >>>>> This is unlikely to work because GENET register the mdio-bcm-unimac
> >>>>> platform device so we will likely run into a chicken and egg problem,
> >>>>> though surprisingly I have not seen this on STB platforms where GENET is
> >>>>> used, I will try building everything as a module like you do. Can you
> >>>>> see if the following helps:
> >>>> For reference we have mdio_bcm_unimac/genet both built as modules in
> >>>> Fedora and I've not seen this issue reported using vanilla upstream
> >>>> kernels if that's a useful reference point.
> >>>
> >>> I was also unable to reproduce this issue, but it seems to be a known
> >>> issue [1], [2].
> >>>
> >>> Jian-Hong opened an issue in my Github repo [3], but before the issue
> >>> was narrowed down, he decided to send this workaround.
> >>
> >> The comment about changing the phy-mode property is not quite making
> >> sense to me, except if that means that in one case the Broadcom PHY
> >> driver is used and in the other case the Generic PHY driver is used.
> >>
> >> What is not clear to me from the debugging that has been done so far is
> >> whether the mdio-bcm-unimac MDIO controller was not loaded at the time
> >> of_phy_connect() was trying to identify the PHY device.
> >
> > MODULE_SOFTDEP("pre: mdio-bcm-unimac")  mentioned in the comment [1]
> > solves this issue.
> >
> > Tracing the code by following the debug message in comment #2 [2], I
> > learned the path bcmgenet_mii_probe()'s of_phy_connect() ->
> > of_phy_find_device() -> of_mdio_find_device() ->
> > bus_find_device_by_of_node().  And, bus_find_device_by_of_node()
> > cannot find the device on the mdio bus.
> >
> > So, I traced bcm2711-rpi-4-b's device tree to find out which one is
> > the mdio device and why it has not been prepared ready on the mdio bus
> > for genet.
> > Then, I found out it is mdio-bcm-unimac module as mentioned in comment
> > #4 [3].  Also, noticed "unimac-mdio unimac-mdio.-19: Broadcom UniMAC
> > MDIO bus" comes after "bcmgenet fd580000.ethernet eth0: failed to
> > connect to PHY" in the log.
> >
> > With these findings, I try to re-modprobe genet module again.  The
> > ethernet on RPi 4B works correctly!  Also, noticed mdio-bcm-unimac
> > module is loaded before I re-modprobe genet module.
> > Therefore, I try to make mdio-bcm-unimac built in kernel image,
> > instead of a module.  Then, genet always can find the mdio device on
> > the bus and the ethernet works as well.
> >
> > Consequently, the idea, loading mdio-bcm-unimac module earlier than
> > genet module comes in my head!  However, I don't know the key word
> > "MODULE_SOFTDEP" until Florian's guide.  That is why I have a loop to
> > connect the PHY in the original patch.  But, I understand
> > MODULE_SOFTDEP is a better solution now!

Forgot to place some reference as note:

* MODULE_SOFTDEP is defined in include/linux/module.h [1]

* modprobe.d has an example: [2]
  Assume "softdep c pre: a b post: d e" is provided in
  the configuration. Running "modprobe c" is now equivalent to
  "modprobe a b c d e" without the softdep.

[1] https://elixir.bootlin.com/linux/v5.13-rc7/source/include/linux/module.h#L170
[2] https://man7.org/linux/man-pages/man5/modprobe.d.5.html

> > I think this is like the module loading order situation mentioned in
> > commit 11287b693d03 ("r8169: load Realtek PHY driver module before
> > r8169") [4].
> >
> The reason in r8169 is different. When people add r8169 module to
> initramfs but not the Realtek PHY driver module then loading
> r8169 will fail. The MODULE_SOFTDEP is a hint to tools building
> initramfs.

Thanks for Heiner's quick clarification.
Maybe I missed some background of the commit ("r8169: load Realtek PHY
driver module before r8169").

Jian-Hong Pan

> > [1] https://bugzilla.kernel.org/show_bug.cgi?id=213485#c6
> > [2] https://bugzilla.kernel.org/show_bug.cgi?id=213485#c2
> > [3] https://bugzilla.kernel.org/show_bug.cgi?id=213485#c4
> > [4] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=11287b693d03830010356339e4ceddf47dee34fa
> >
> > Jian-Hong Pan
> >
>