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]
Date:   Thu, 3 Feb 2022 09:52:03 -0800
From:   Tim Harvey <tharvey@...eworks.com>
To:     Andrew Lunn <andrew@...n.ch>
Cc:     Martin Schiller <ms@....tdt.de>, Hauke Mehrtens <hauke@...ke-m.de>,
        martin.blumenstingl@...glemail.com,
        Florian Fainelli <f.fainelli@...il.com>, hkallweit1@...il.com,
        Russell King - ARM Linux <linux@...linux.org.uk>,
        David Miller <davem@...emloft.net>, kuba@...nel.org,
        netdev <netdev@...r.kernel.org>,
        open list <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH net v3] net: phy: intel-xway: enable integrated led functions

On Thu, Feb 3, 2022 at 8:37 AM Andrew Lunn <andrew@...n.ch> wrote:
>
> > Andrew,
> >
> > I agree with the goal of having PHY drivers and dt-bindings in Linux
> > to configure everything but in the case I mention in the other thread
> > adding rgmii delay configuration which sets a default if a new dt
> > binding is missing is wrong in my opinion as it breaks backward
> > compatibility. If a new dt binding is missing then I feel that the
> > register fields those bindings act on should not be changed.
>
> I would like that understand this specific case in more detail.  We
> have seen a few cases were the DT is broken, yet works. This is often
> caused by having a wrong phy-mode, which historically the PHY driver
> was ignoring. Then support for honouring the phy-mode was added to the
> PHY driver, and all the boards with broken DT files actually break.
>
> So it could be that is what has happened here. Or it could be the
> driver is plan wrong. If i understand correctly, you say it is adding
> a default delay of 2ns. That would be correct for a phy-mode of
> rgmii-id, but wrong for a phy-mode of rgmii.
>
> > > LEDs are trickier. There is a slow on going effort to allow PHY LEDs
> > > to be configured as standard Linux LEDs. That should result in a DT
> > > binding which can be used to configure LEDs from DT.
> >
> > Can you point me to something I can look at? PHY LED bindings don't at
> > all behave like normal LED's as they are blinked internally depending
> > on a large set of rules that differ per PHY.
>
> Yes, this is what is slowing the work done, agreeing on details like
> this, and how the user space API would actually work. In the end, i
> suspect a subset of LED modes will be supported, covering the common
> blink patterns.
>
> > Completely off topic, but due to the chip shortage we have had to
> > redesign many of our boards with different PHY's that now have
> > different bindings for RGMII delays so I have to add multiple PHY
> > configurations to DT's if I am going to support the use of PHY
> > drivers. What is your suggestion there? Using DT overlays I suppose is
> > the right approach.
>
> I would try to only use phy-mode, and avoid all PHY specific tweaks.
> So long as the track lengths don't change too much on your redesign,
> and are kept about the same length, the standard 2ns delay should
> work.
>

Andrew,

The issue is that in an ideal world where all parts adhere to the
JEDEC standards 2ns would be correct but that is not reality. In my
experience with the small embedded boards I help design and support
not about trace lengths as it would take inches to skew 0.5ns but
instead differences in setup/hold values for various MAC/PHY parts
that are likely not adhering the standards.

Some examples from current boards I support:
- CN8030 MAC rgmii-rxid with intel-xway GPY111 PHY: we need to
configure this for rx_delay=1.5ns and tx_delay=0.5ns
- CN8030 MAC rgmii-rxid with dp83867 GPY111 PHY: we need to configure
this for rx_delay=2.0ns and tx_delay=2.0ns (as would be expected)
- IMX8MM FEC MAC rgmii-id with intel-xway PHY: we need to configure
this for rx_delay=2ns and tx_delay=2.5ns
- IMX8MM FEC MAC rgmii-id with dp83867 PHY: we need to configure this
for rx_delay=2.0ns and tx_delay=2.0ns (as would be expected)

The two boards above have identical well matched trace-lengths between
the two PHY variant designs so you can see here that the intel-xway
GPY111 PHY needs an extra 0.5ps delay to avoid RX errors on the
far-off board. I really hate this GPY111 PHY but it's the only PHY we
can source currently; it's full of errata and to make things worse the
only available pin strapping options are for 0ns or 1.5ns (how is
1.5ns useful?) so we must rely on software configuration. So having
the intel-xway PHY driver set the delays to 2.0ns delays in the
absence of dt props (vs leaving them untouched) is what bothered me
there. The LED blink configuration has much more flexible strapping
options which we are able to use until this driver goes and changes
them to something else.

The way I determine the correct delay values is by looking for MAC RX
errors on the RX side and by looking for RX errors on the off-board
receiving end (typically via a managed switch). I know of no better
way to do this because the timing happens inside the MAC and PHY thus
scope measurements don't help here.

Best regards,

Tim

Powered by blists - more mailing lists