Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.debian.kernel > #93803

Opaque tables of numbers in r8169 driver

Show all headers | View raw

[Multipart message — attachments visible in raw view] - view raw

On Tue, 2026-08-18 at 18:28 +0000, John Scott wrote:
[...]

> On a different tangent, I looked into Realtek's driver which is
> currently in the non-free section and it is actually wholly licensed
> under the GNU GPL, but not included in main because it contains tables
> of random-looking bytes that are assumed to be firmware patches of
> some kind. However the mainline kernel driver is similar (even more so
> recently) which feels like a double standard.

I used to check each new kernel release for the introduction of large
tables of literal numbers that might be closed-source firmware.  I
challenged a few such additions, including in r8169, as a result of
which the PHY firmware used with r8169 is (at least mostly) separated
out into linux-firmware.git.

A smaller table that the driver copies to the device might be a
replacement for an incorrect lookup table, or for some register
initialisation values.  Even if it is a fragment of machine code, it may
be short enough to be "hand-assembled", with no separate source file
existing.  (I have had to do this once!  But I did put the assembly
source in a comment.)

It would certainly be better programming practice to properly comment
these magic numbers, and/or to use named constants.  But I don't think
we should jump to the conclusion that any mysterious table of numbers is
closed-source firmware.

As for what is a "large" or "small" tables, the script I wrote to look
for large tables of numbers had a threshold of 100 lines, but that was
not based on any scientific or legal principle.


> I was surprised to see that even Linux-libre doesn't remove that, so I
> sent the attached mail to them to learn more for the sake of
> curiosity. I don't know that we have a firm stance in Debian about
> "magic blobs" that are inline with the kernel source which aren't
> firmware (in the linux-firmware.git conventional sense).

If these tables do turn out to be closed-source firmware then we are
bound to remove them from the Debian linux package (and ideally we would
move them out to firmware-nonfree).  I am interested to see what
response you get.

But unless and until that happens, I'm going to assume that these tables
are only poorly documented, and are within the DFSG.

Ben.

-- 
Ben Hutchings
Tomorrow will be cancelled due to lack of interest.

Back to linux.debian.kernel | Previous | Next — Previous in thread | Find similar | Unroll thread

Thread

Bug#1143994: Ethernet driver r8169 is used for RTL8125 chipset by default and disconnects intermittently Uwe Kleine-König <ukleinek@debian.org> - 2026-08-18 14:00 +0200
  Opaque tables of numbers in r8169 driver Ben Hutchings <ben@decadent.org.uk> - 2026-08-19 20:40 +0200

csiph-web