netdev - Re: PROBLEM: smsc95xx loses config on link down/up

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: <CA+ZLECv7AcQSa1VZeeiOFJ43Vh=nfn_ptMB6XwXsfbRSz9VJ6A@mail.gmail.com>
Date:   Tue, 3 Dec 2019 17:19:25 +1100
From:   Sam Lewis <sam.vr.lewis@...il.com>
To:     Andrew Lunn <andrew@...n.ch>
Cc:     Steve Glendinning <steve.glendinning@...well.net>,
        UNGLinuxDriver@...rochip.com, netdev@...r.kernel.org
Subject: Re: PROBLEM: smsc95xx loses config on link down/up

On Tue, 3 Dec 2019 at 00:46, Andrew Lunn <andrew@...n.ch> wrote:
>
> On Thu, Nov 28, 2019 at 06:19:14PM +1100, Sam Lewis wrote:
> > I'm using a LAN9514 chip in my embedded Linux device and have noticed
> > that changing Ethernet configuration (with ethtool for example) does
> > not persist after putting the link up.
>
> Hi Sam
>
> Did you ever get a reply to this?
>

Hi Andrew - I hadn't, but I understand that it's a quiet time of year.
:) Thanks for getting back to me!

> > I've hacked through the driver code (without really knowing what I'm
> > doing, just adding various print statements) and I think this happens
> > because setting a link up causes the `smsc95xx_reset` function to be
> > called which seems to clear all configuration through:
> >
> > 1) Doing a PHY reset (with `smsc95xx_write_reg(dev, PM_CTRL, PM_CTL_PHY_RST_)`)
> > 2) Doing (another?) PHY reset (with `smsc95xx_mdio_write(dev->net,
> > dev->mii.phy_id, MII_BMCR, BMCR_RESET)`)
>
> In general, BMCR_RESET does not clear configuration registers such as
> auto-neg etc. It generally just gives the PHY a kick to restart itself
> using the configuration as set. So i would initially point a finger at
> PM_CTL_PHY_RST_.
>
> Is there a full datasheet somewhere?
>
> You might want to think about using PM_CTL_PHY_RST_ once during probe,
> and only BMCR_RESET in open.
>

Thanks, doing the PM_CTL_PHY_RST_ reset only during probing does sound
a good (& necessary) fix but it unfortunately doesn't look like it'll
get it all the way there.

I managed to track down the datasheet for the PHY, available here:
http://ww1.microchip.com/downloads/en//softwarelibrary/man-lan95xx-dat/lan9514_lan9514i%20databook%20rev.%201.2%20(03-01-12).pdf

Basically it looks as though doing a BMCR_RESET does, in fact, reset
every PHY R/W register bit except for those marked as "Not Affected by
Software Reset" (NASR). This means it will reset, to the default
value:

- Autonegotiation
- Speed
- Duplex
- Auto MDIX
- Energy Detect Power-Down
- Auto Negotiation Advertisement
- PHY Identification (although I don't know why you'd change this?)
- Power down
- Loopback

I tested this by checking the value of the BMCR register before and
after doing a BMCR_RESET and it did reset the BMCR register to its
default values.

The PHY does provide a NASR "Special Modes Register" which effectively
allows you to set 'default' values for the duplex, speed and autoneg
that are applied after a BMCR_RESET. See page 205 of the datasheet for
more details. Setting this to 0x0061 allowed me to set the PHY to be
full duplex, 100M and no autoneg on after a BMCR_RESET, for example.

However, given that the Special Modes register only allows saving a
subset of settings it perhaps isn't the best solution? (Auto MDIX is
something else I'd like to set differently to the default value in my
application and I imagine the other settings probably shouldn't be
lost either). The only other fix that I can think of is to save all
the PHY R/W registers before doing a BMCR_RESET and then restoring
them after the reset. Would this be an acceptable solution?

I do have to ask though - is it strictly necessary to do the
BMCR_RESET (& the PHY initialisation) every time on link up/open? What
is the reasoning behind doing it? Excuse my ignorance on this if it's
a dumb question! If it was only done on probing, it would make this
easier (well, easier for me at least :)).

Interested to know what you think - I'm still keen on seeing if I can
make a generic patch for this. I think I know enough now that I can
hack together a patch that'll work for my particular application, but
it would be cool to upstream something that will benefit everyone.

Sam