netdev - Re: [PATCH net-next 1/3] net: phy: Support enabling clocks prior to bus probe

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <81972e9c-c664-08b4-a9f2-636df3cd801d@gmail.com>
Date:   Thu, 3 Sep 2020 14:43:54 -0700
From:   Florian Fainelli <f.fainelli@...il.com>
To:     Rob Herring <robh+dt@...nel.org>
Cc:     netdev <netdev@...r.kernel.org>, Andrew Lunn <andrew@...n.ch>,
        adam.rudzinski@....net.pl, Marco Felsch <m.felsch@...gutronix.de>,
        Heiner Kallweit <hkallweit1@...il.com>,
        Richard Leitner <richard.leitner@...data.com>,
        Dejin Zheng <zhengdejin5@...il.com>,
        devicetree@...r.kernel.org, Sascha Hauer <kernel@...gutronix.de>,
        Jakub Kicinski <kuba@...nel.org>
Subject: Re: [PATCH net-next 1/3] net: phy: Support enabling clocks prior to
 bus probe



On 9/3/2020 2:28 PM, Rob Herring wrote:
> On Wed, Sep 2, 2020 at 10:39 PM Florian Fainelli <f.fainelli@...il.com> wrote:
>>
>> Some Ethernet PHYs may require that their clock, which typically drives
>> their logic to respond to reads on the MDIO bus be enabled before
>> issusing a MDIO bus scan.
>>
>> We have a chicken and egg problem though which is that we cannot enable
>> a given Ethernet PHY's device clock until we have a phy_device instance
>> create and called the driver's probe function. This will not happen
>> unless we are successful in probing the PHY device, which requires its
>> clock(s) to be turned on.
>>
>> For DT based systems we can solve this by using of_clk_get() which
>> operates on a device_node reference, and make sure that all clocks
>> associaed with the node are enabled prior to doing any reads towards the
>> device. In order to avoid drivers having to know the a priori reference
>> count of the resources, we drop them back to 0 right before calling
>> ->probe() which is then supposed to manage the resources normally.
> 
> What if a device requires clocks enabled in a certain order or timing?
> It's not just clocks, you could have some GPIOs or a regulator that
> need enabling first. It's device specific, so really needs a per
> device solution. This is not just an issue with MDIO. I think we
> really need some sort of pre-probe hook in the driver model in order
> to do any non-discoverable init for discoverable buses. Or perhaps
> forcing probe when there are devices defined in DT if they're not
> discovered by normal means.

I like the pre-probe hook idea, and there are other devices that I have 
access to that would benefit from that, like PCIe end-points that 
require regulators to be turned on in order for them to be discoverable.

For MDIO we might actually be able to create the backing device 
reference without having read the device ID yet, provided that we know 
its address on the bus, which DT can tell us.

Bartosz attempted to do that not so long ago and we sort of stalled 
there, too:

https://lkml.org/lkml/2020/6/22/253

Let me see if I can just add a pre-probe hook, make use of it in the 
MDIO layer, and we see how we can apply it to other subsystems?
-- 
Florian