Netdev Archive on
help / color / mirror / Atom feed
From: Saravana Kannan <>
To: Marc Zyngier <>, Andrew Lunn <>
Cc: Kevin Hilman <>,
	Lee Jones <>,
	Neil Armstrong <>,
	Jerome Brunet <>,,
	linux-arm-kernel <>,
	open list <>,
	netdev <>,
	Android Kernel Team <>
Subject: Re: [PATCH 1/2] irqchip: irq-meson-gpio: make it possible to build as a module
Date: Wed, 4 Aug 2021 11:20:52 -0700	[thread overview]
Message-ID: <> (raw)
In-Reply-To: <>

On Wed, Aug 4, 2021 at 1:50 AM Marc Zyngier <> wrote:
> On Wed, 04 Aug 2021 02:36:45 +0100,
> Saravana Kannan <> wrote:
> Hi Saravana,
> Thanks for looking into this.

You are welcome. I just don't want people to think fw_devlink is broken :)

> [...]
> > > Saravana, could you please have a look from a fw_devlink perspective?
> >
> > Sigh... I spent several hours looking at this and wrote up an analysis
> > and then realized I might be looking at the wrong DT files.
> >
> > Marc, can you point me to the board file in upstream that corresponds
> > to the platform in which you see this issue? I'm not asking for [1],
> > but the actual final .dts (not .dtsi) file that corresponds to the
> > platform/board/system.
> The platform I can reproduce this on is described in
> arch/arm64/boot/dts/amlogic/meson-sm1-khadas-vim3l.dts. It is an
> intricate maze of inclusion, node merge and other DT subtleties. I
> suggest you look at the decompiled version to get a view of the
> result.

Thanks. After decompiling it, it looks something like (stripped a
bunch of reg and address properties and added the labels back):

eth_phy: mdio-multiplexer@4c000 {
        compatible = "amlogic,g12a-mdio-mux";
        clocks = <0x02 0x13 0x1e 0x02 0xb1>;
        clock-names = "pclk\0clkin0\0clkin1";
        mdio-parent-bus = <0x22>;

        ext_mdio: mdio@0 {
                reg = <0x00>;

                ethernet-phy@0 {
                        max-speed = <0x3e8>;
                        interrupt-parent = <0x23>;
                        interrupts = <0x1a 0x08>;
                        phandle = <0x16>;

        int_mdio: mdio@1 {

And phandle 0x23 refers to the gpio_intc interrupt controller with the
modular driver.

> > Based on your error messages, it's failing for mdio@0 which
> > corresponds to ext_mdio. But none of the board dts files in upstream
> > have a compatible property for "ext_mdio". Which means fw_devlink
> > _should_ propagate the gpio_intc IRQ dependency all the way up to
> > eth_phy.
> >
> > Also, in the failing case, can you run:
> > ls -ld supplier:*
> >
> > in the /sys/devices/....<something>/ folder that corresponds to the
> > "eth_phy: mdio-multiplexer@4c000" DT node and tell me what it shows?
> Here you go:
> root@tiger-roach:~# find /sys/devices/ -name 'supplier*'|grep -i mdio | xargs ls -ld
> lrwxrwxrwx 1 root root 0 Aug  4 09:47 /sys/devices/platform/soc/ff600000.bus/ff64c000.mdio-multiplexer/supplier:platform:ff63c000.system-controller:clock-controller -> ../../../../virtual/devlink/platform:ff63c000.system-controller:clock-controller--platform:ff64c000.mdio-multiplexer

As we discussed over chat, this was taken after the mdio-multiplexer
driver "successfully" probes this device. This will cause
SYNC_STATE_ONLY device links created by fw_devlink to be deleted
(because they are useless after a device probes). So, this doesn't
show the info I was hoping to demonstrate.

In any case, one can see that fw_devlink properly created the device
link for the clocks dependency. So fw_devlink is parsing this node
properly. But it doesn't create a similar probe order enforcing device
link between the mdio-multiplexer and the gpio_intc because the
dependency is only present in a grand child DT node (ethernet-phy@0
under ext_mdio). So fw_devlink is working as intended.

I spent several hours squinting at the code/DT yesterday. Here's what
is going on and causing the problem:

The failing driver in this case is
drivers/net/mdio/mdio-mux-meson-g12a.c. And the only DT node it's
handling is what I pasted above in this email. In the failure case,
the call flow is something like this:

-> mdio_mux_init()
-> of_mdiobus_register(ext_mdio DT node)
-> of_mdiobus_register_phy(ext_mdio DT node)
-> several calls deep fwnode_mdiobus_phy_device_register(ethernet_phy DT node)
-> Tried to get the IRQ listed in ethernet_phy and fails with
-EPROBE_DEFER because the IRQ driver isn't loaded yet.

The error is propagated correctly all the way up to of_mdiobus_register(), but
mdio_mux_init() ignores the -EPROBE_DEFER from of_mdiobus_register() and just
continues on with the rest of the stuff and returns success as long as
one of the child nodes (in this case int_mdio) succeeds.

Since the probe returns 0 without really succeeding, networking stuff
just fails badly after this. So, IMO, the real problem is with
mdio_mux_init() not propagating up the -EPROBE_DEFER. I gave Marc a
quick hack (pasted at the end of this email) to test my theory and he
confirmed that it fixes the issue (a few deferred probes later, things
work properly).

Andrew, I don't see any good reason for mdio_mux_init() not
propagating the errors up correctly (at least for EPROBE_DEFER). I'll
send a patch to fix this. Please let me know if there's a reason it
has to stay as-is.


index 110e4ee85785..d973a267151f 100644
--- a/drivers/net/mdio/mdio-mux.c
+++ b/drivers/net/mdio/mdio-mux.c
@@ -170,6 +170,9 @@ int mdio_mux_init(struct device *dev,
                        devm_kfree(dev, cb);
+                       /* Not a final fix. I think it can cause UAF issues. */
+                       mdio_mux_uninit(pb);
+                       return r;
                } else {
                        cb->next = pb->children;
                        pb->children = cb;

       reply	other threads:[~2021-08-04 18:21 UTC|newest]

Thread overview: 14+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <>
     [not found] ` <>
     [not found]   ` <>
     [not found]     ` <>
     [not found]       ` <>
     [not found]         ` <>
     [not found]           ` <>
     [not found]             ` <>
     [not found]               ` <>
2021-08-04 18:20                 ` Saravana Kannan [this message]
2021-08-04 21:47                   ` Saravana Kannan
2021-08-05  6:31                     ` Neil Armstrong
2021-08-06 23:55                       ` Saravana Kannan
2021-08-05  7:57                     ` Lee Jones
2021-08-16 12:47                       ` Lee Jones
2021-08-16 20:27                         ` Saravana Kannan
2021-08-16 20:46                           ` Andrew Lunn
2021-08-16 21:02                             ` Saravana Kannan
2021-08-16 21:18                               ` Andrew Lunn
2021-08-17  7:24                           ` Lee Jones
2021-08-17 18:12                             ` Saravana Kannan
2021-08-18 11:19                               ` Marc Zyngier
2021-09-02  9:28                                 ` Neil Armstrong

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \ \ \ \ \ \ \ \ \ \ \ \ \ \
    --subject='Re: [PATCH 1/2] irqchip: irq-meson-gpio: make it possible to build as a module' \

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).