Re: [PATCH v2 3/5] dt-bindings: net: add mac-address-increment option

From
Krzysztof Kozlowski <hidden>
Date
2023-05-17 19:27:44
Also in
linux-devicetree, lkml
Reply to
Ivan Mikhaylov, 2023-05-17
On 17/05/2023 23:38, Ivan Mikhaylov wrote:
On Wed, 2023-05-17 at 10:36 +0200, Krzysztof Kozlowski wrote:
quoted
On 16/05/2023 13:47, Ivan Mikhaylov wrote:
hy this is property of the hardware. I
quoted
quoted
quoted
quoted
understand
that this is something you want Linux to do, but DT is not
for
that
purpose. Do not encode system policies into DT and what above
commit
says is a policy.
Krzysztof, okay then to which DT subsystem it should belong? To
ftgmac100 after conversion?
To my understanding, decision to add some numbers to MAC address
does
not look like DT property at all. Otherwise please help me to
understand
- why different boards with same device should have different
offset/value?
I would like to remind this question.
"why different boards with same device should have different offset/value?"

It was literally ignored and you started explaining network cards and
BMC. I don't understand why, but it does not help your case.

Let me extend this question with one more:
"Why for all your boards of one type, so using the same DTS, would you
use one value of incrementing MAC address?"
quoted
quoted
quoted
Anyway, commit msg also lacks any justification for this.

Best regards,
Krzysztof
Krzysztof, essentially some PCIe network cards have like an
additional
*MII interface which connects directly to a BMC (separate SoC for
managing a motherboard) and by sending special ethernet type frames
over that connection (called NC-SI) the BMC can obtain MAC, get
link
parameters etc. So it's natural for a vendor to allocate two MACs
per
such a board with PCIe card intergrated, with one MAC "flashed
into"
the network card, under the assumption that the BMC should
Who makes the assumption that next MAC should differ by 1 or 2?
Krzysztof, in this above case BMC does, BMC should care about changing
it and doing it with current codebase without any options just by some
hardcoded numbers which is wrong.
But you hard-code the number, just in BMC DTS. How does it differ from
BMC hard-coding it differently?

You encode policy - or software decisions - into Devicetree.
quoted
quoted
automatically use the next MAC. So it's the property of the
hardware as
the vendor designs it, not a matter of usage policy.

Also at the nvmem binding tree is "nvmem-cell-cells" which is
literally
the same as what was proposed but on different level.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/Documentation/devicetree/bindings/nvmem?id=7e2805c203a6c8dc85c1cfda205161ed39ae82d5
How is this similar? This points the location of mac address on some
NV
storage. You add fixed value which should be added to the Ethernet.
It's not the points the location, this particular option provides this
increment for mac addresses to make use of them with multiple
interfaces. Just part of above commit:
"It's used as a base for calculating addresses for multiple interfaces.
It's done by adding proper values. Actual offsets are picked by
manufacturers and vary across devices."

It is same as we talked before about mac-address-increment in openwrt
project, if you want examples, you can look into their github. And same
as we trying to achieve here.

https://github.com/openwrt/openwrt/blob/master/target/linux/generic/pending-5.15/682-of_net-add-mac-address-increment-support.patch
Awesome... so if project added wrong property to bindings, e.g. SW
property, you find it as an argument for anyone else.

No, that's not how it works.
"Lots of embedded devices use the mac-address of other interface
extracted from nvmem cells and increments it by one or two. Add two
bindings to integrate this and directly use the right mac-address for
the interface. Some example are some routers that use the gmac
mac-address stored in the art partition and increments it by one for
the
wifi. mac-address-increment-byte bindings is used to tell what byte of
the mac-address has to be increased (if not defined the last byte is
increased) and mac-address-increment tells how much the byte decided
early has to be increased."

Don't you see similarity with nvmem commit?
Explanation is similar, but you are using wrong argument to justify the
property. The MAC address is stored in some NVMEM cell. There is such
NVMEM cell. That's the hardware property, thus it is justified in DT.

Now how MAC address will be modified - by 1, 2, 3, 252 - is not related
to that commit, because it is a software decision.

Again, we are back to the previous question to which you answered "BMC
will do it". I understand this is property for the BMC DTS, thus:
Why for all your boards of one type, so using one DTS, would you use one
value of incrementing MAC address?
Why devices with same board cannot use different values? One board "1"
and second "2" for MAC increments? I am sure that one customer could
have it different.

The choice how much you increment some MAC address is not a hardware
property. It does not even look like a firmware property. If playing
with this property was done by firmware, like we do for all MAC address
fields, then I would expect here some references to it. Which you did
not provide, I believe.


quoted
I might be missing the context but there is no DTS example nor user
of
this property, so how can I get such?
I don't see it either in linux kernel DTS tree but it in DTS doc.

Also, just a little bit history about older propositions
https://lore.kernel.org/all/?q=mac-address-increment
https://lore.kernel.org/all/20200919214941.8038-5-ansuelsmth@gmail.com/ (local)
I don't see any user there, except the same rejected proposal:

https://lore.kernel.org/all/CAL_JsqKhyeh2=pJcpBKkh+s3FM__DY+VoYSYJLRUErrujTLn9A@mail.gmail.com/ (local)

If you want to convince us, please illustrate it in a real world
upstreamed DTS (or explain why it cannot). Otherwise I don't see
justification as it is not a hardware property.

This is a NAK from me.

Feel free to ping Rob in some later time, as he might have different
opinion.

Best regards,
Krzysztof
Thread (22 messages) 22 messages, 4 authors, last reply 2023-06-04
Keyboard shortcuts
hback out one level
jnext message in thread / next series-diff position
kprevious message in thread / previous series-diff position
ldrill in
Escclose help / fold thread tree
?toggle this help