Re: [net-next PATCH 1/5] net: dsa: tag_rtl4_a: Implement Realtek 4 byte A tag

From: Linus Walleij <linus.walleij@linaro.org>
To: Vladimir Oltean <olteanv@gmail.com>
Cc: Andrew Lunn <andrew@lunn.ch>,
	Vivien Didelot <vivien.didelot@gmail.com>,
	Florian Fainelli <f.fainelli@gmail.com>,
	netdev <netdev@vger.kernel.org>, DENG Qingfang <dqfext@gmail.com>,
	Mauri Sandberg <sandberg@mailfence.com>
Subject: Re: [net-next PATCH 1/5] net: dsa: tag_rtl4_a: Implement Realtek 4 byte A tag
Date: Wed, 17 Jun 2020 10:06:08 +0200	[thread overview]
Message-ID: <CACRpkda9kJUFwx-ASQfO-ThhgbV2fmT_tqT8zH0W_Jm23ZTVMg@mail.gmail.com> (raw)
In-Reply-To: <CA+h21hqNq6Xk5bMBsB884GZdH9h4pALr7nkn8yG+a16cXqfJsg@mail.gmail.com>

On Thu, Jun 4, 2020 at 1:23 PM Vladimir Oltean <olteanv@gmail.com> wrote:

> In the code you pointed to, there is a potentially relevant comment:
>
> 1532//CPU tag: Realtek Ethertype==0x8899(2 bytes)+protocol==0x9(4
> MSB)+priority(2 bits)+reserved(4 bits)+portmask(6 LSB)
>
> https://svn.dd-wrt.com/browser/src/linux/universal/linux-3.2/drivers/net/ethernet/raeth/rb/rtl_multicast_snooping.c#L1527
> https://svn.dd-wrt.com/browser/src/linux/universal/linux-3.2/drivers/net/ethernet/raeth/rb/rtl_multicast_snooping.c#L5224
>
> This strongly indicates to me that the insertion tag is the same as
> the extraction tag.

This code is a problem because it is Realtek-development style.
This style seems to be that the hardware people write the drivers
using copy/paste from the previous ASIC and ship is as soon as
possible. Keep this in mind.

The above tag is using protocol 9 and is actually even documented
in a PDF I have for RTL8306. The problem is that the RTL8366RB
(I suspect also RTL8366S) uses protocol "a" (as in hex 10).
Which is of course necessarily different.

I have *really* tried to figure out how the bits in protocol a works
when transmissing from the CPU port to any switch port.

When nothing else worked, I just tried all bit combinations with
0xannp where a is protocol and p is port. I looped through all
values several times trying to get a response from ping.

So this is really how far I can get right now, even with brute
force.

> It is completely opaque to me why in patch "[net-next PATCH 2/5] net:
> dsa: rtl8366rb: Support the CPU DSA tag" you are _disabling_ the
> injection of these tags via RTL8368RB_CPU_INSTAG. I think it's natural
> that the switch drops these packets when CPU tag insertion is
> disabled.

This is another Realtek-ism where they managed to invert the
meaning of a bit.

Bit 15 in register 0x0061 (RTL8368RB_CPU_CTRL_REG) can
be set to 1 and then the special (custom) CPU tag 0x8899
protocol a will be DISABLED. This value Realtek calls
"RTL8368RB_CPU_INSTAG" which makes you think that
the tag will be inserted, it is named "instag" right? But that
is not how it works.

That bit needs to be set to 0 to insert the tag and 1 to disable
insertion of the tag.

For this reason the patch also renames this bit to
RTL8368RB_CPU_NO_TAG which is more to the point.

Yours,
Linus Walleij